2099 lines
270 KiB
Plaintext
2099 lines
270 KiB
Plaintext
{
|
||
"cells": [
|
||
{
|
||
"metadata": {},
|
||
"cell_type": "markdown",
|
||
"source": [
|
||
"#### LangGraph\n",
|
||
"\n",
|
||
"- 워크플로 프레임워크\n",
|
||
"- LCEL 선형 (A -> B -> C) / LangGraph 순환 ( A-> B -> A -> C -> B) 흐름 지원\n",
|
||
"- 개념\n",
|
||
" - Node : 실행할 함수\n",
|
||
" - Edge : Node 간의 연결\n",
|
||
" - State : Node 간의 전체 상태를 담는 딕셔너리\n",
|
||
"- 조건부 엣지, Agent 루프, 자기 수정, 멀티 에이전트 등 복잡한 팬턴 구현"
|
||
],
|
||
"id": "4f173862ffa5000b"
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T01:14:55.090656687Z",
|
||
"start_time": "2026-06-09T01:14:55.050105345Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": "# !pip install langgraph grandalf",
|
||
"id": "1ecece0e830ff1b5",
|
||
"outputs": [],
|
||
"execution_count": 1
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T01:15:00.597733384Z",
|
||
"start_time": "2026-06-09T01:14:55.094788739Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"from langgraph.graph import StateGraph, START, END\n",
|
||
"from typing import TypedDict\n",
|
||
"\n",
|
||
"import os\n",
|
||
"\n",
|
||
"from cohere.types import summarize_request_extractiveness\n",
|
||
"from google.protobuf.timestamp import from_current_time\n",
|
||
"from langchain_classic.chains.sequential import SequentialChain\n",
|
||
"from langchain_openai import ChatOpenAI\n",
|
||
"from langchain_ibm import WatsonxEmbeddings\n",
|
||
"from langchain_ibm import ChatWatsonx\n",
|
||
"from langchain.agents import create_agent\n",
|
||
"from langchain_ollama import OllamaEmbeddings\n",
|
||
"from langchain_ollama import ChatOllama\n",
|
||
"from langchain_ollama import OllamaEmbeddings, ChatOllama\n",
|
||
"from dotenv import load_dotenv\n",
|
||
"\n",
|
||
"from langchain_community.utilities import GoogleSerperAPIWrapper\n",
|
||
"from langchain_core.tools import Tool\n",
|
||
"from langchain_core.tools import tool\n",
|
||
"\n",
|
||
"from langchain_core.prompts import PromptTemplate, ChatPromptTemplate, MessagesPlaceholder\n",
|
||
"from langchain_core.output_parsers import StrOutputParser, JsonOutputParser, PydanticOutputParser\n",
|
||
"from langchain_core.runnables import RunnablePassthrough, RunnableParallel, RunnableLambda\n",
|
||
"from langgraph import graph\n",
|
||
"from pygments.unistring import combine\n",
|
||
"from sklearn import pipeline"
|
||
],
|
||
"id": "424747aac7874d9",
|
||
"outputs": [
|
||
{
|
||
"name": "stderr",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"/home/cooney/Source/.venv/lib/python3.12/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
|
||
" from .autonotebook import tqdm as notebook_tqdm\n",
|
||
"/tmp/ipykernel_38089/245101980.py:18: DeprecationWarning: `langchain-community` is being sunset and is no longer actively maintained. See https://github.com/langchain-ai/langchain-community/issues/674 for details and migration guidance toward standalone integration packages.\n",
|
||
" from langchain_community.utilities import GoogleSerperAPIWrapper\n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 2
|
||
},
|
||
{
|
||
"metadata": {},
|
||
"cell_type": "markdown",
|
||
"source": [
|
||
"#### TypedDIct / Pydantic\n",
|
||
"- TypedDict : 딕셔너리 키 , 값 타입을 정의할 수 있게 도와줌, 단순한 State 구현 시 사용, 기본값 줄 수 없음\n",
|
||
"- Pydantic : validation 검사(입력값), 검증이 필요한 State 인 경우 사용"
|
||
],
|
||
"id": "ab7dfbff952ba846"
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T01:15:00.662448491Z",
|
||
"start_time": "2026-06-09T01:15:00.633564506Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"# 데이터 저장소 생성\n",
|
||
"# message 상태관리 할거야 => 공유\n",
|
||
"class MyState(TypedDict):\n",
|
||
" message:str\n",
|
||
"\n",
|
||
"# 작업 함수 생성(노드)\n",
|
||
"def say_hello(state):\n",
|
||
" # state 값의 변화\n",
|
||
" return {\"message\": \"Hello, LangGraph!\"}\n",
|
||
"\n",
|
||
"# 그래프 생성\n",
|
||
"graph = StateGraph(MyState)\n",
|
||
"graph.add_node(\"hello\", say_hello)\n",
|
||
"\n",
|
||
"graph.add_edge(START, \"hello\")\n",
|
||
"graph.add_edge(\"hello\", END)\n",
|
||
"\n",
|
||
"# 실행\n",
|
||
"app = graph.compile()\n",
|
||
"result = app.invoke({\"message\": \"\"})\n",
|
||
"print(result)\n",
|
||
"\n",
|
||
"# 그래프 시각화(참고)\n",
|
||
"app.get_graph().print_ascii()"
|
||
],
|
||
"id": "977a81b9df795373",
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"{'message': 'Hello, LangGraph!'}\n",
|
||
"+-----------+ \n",
|
||
"| __start__ | \n",
|
||
"+-----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +-------+ \n",
|
||
" | hello | \n",
|
||
" +-------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +---------+ \n",
|
||
" | __end__ | \n",
|
||
" +---------+ \n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 3
|
||
},
|
||
{
|
||
"metadata": {},
|
||
"cell_type": "markdown",
|
||
"source": [
|
||
"#### State\n",
|
||
"- 상태를 중심으로 동작\n",
|
||
"- TypeDict, Pydantic Base Model 을 사용하여 정의\n",
|
||
"- 그래프 실행 중 지속적으로 업데이트 됨\n",
|
||
"- 노드 간의 전환은 조건부 엣지를 통해 제어 가능하며 복잡한 의사결정 프로세스 모델링 간으\n",
|
||
"- 재귀적 실행 지원\n",
|
||
"\n",
|
||
"#### Node\n",
|
||
"- 실제 작업을 수행하는 기본단위\n",
|
||
"- 함수기반\n",
|
||
"- 상태중심 : 현재 상태를 입력으로 받아 처리\n",
|
||
"- 독립적 실행 : 각 노드는 독립적으로 실행\n",
|
||
"- 조합 가능 : 여러 노드를 연결하여 복잡한 워크플로 가능\n",
|
||
"\n",
|
||
"### Edge\n",
|
||
"- 노드간의 연결과 실행 흐름을 정의\n"
|
||
],
|
||
"id": "65888b5803542435"
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T01:15:00.703755128Z",
|
||
"start_time": "2026-06-09T01:15:00.667966313Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"# 카운터 공유\n",
|
||
"\n",
|
||
"class CounterState(TypedDict):\n",
|
||
" count: int\n",
|
||
"\n",
|
||
"# 증가 함수\n",
|
||
"def increment(state):\n",
|
||
" print(f\"현재 카운트 : {state['count']}\")\n",
|
||
" new_count = state['count'] + 1\n",
|
||
" print(f\"새로운 카운트 : {new_count}\")\n",
|
||
"\n",
|
||
" # state 값 변경(return)\n",
|
||
" return {\"count\": new_count}\n",
|
||
"\n",
|
||
"# 그래프 생성\n",
|
||
"graph = StateGraph(CounterState)\n",
|
||
"graph.add_node(\"increment\", increment)\n",
|
||
"\n",
|
||
"# 그래프 연결(그래프가 실행될 것인가?)\n",
|
||
"graph.add_edge(START, \"increment\")\n",
|
||
"graph.add_edge(\"increment\", END)\n",
|
||
"\n",
|
||
"# 그래프를 실행 가능한 형태로 변경\n",
|
||
"app = graph.compile()\n",
|
||
"result = app.invoke({\"count\": 0})\n",
|
||
"print(f\"최종 결과 {result}\")\n",
|
||
"\n",
|
||
"# 그래프 시각화(참고)\n",
|
||
"app.get_graph().print_ascii()\n",
|
||
"\n"
|
||
],
|
||
"id": "3acb2357ac1825d7",
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"현재 카운트 : 0\n",
|
||
"새로운 카운트 : 1\n",
|
||
"최종 결과 {'count': 1}\n",
|
||
"+-----------+ \n",
|
||
"| __start__ | \n",
|
||
"+-----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
"+-----------+ \n",
|
||
"| increment | \n",
|
||
"+-----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +---------+ \n",
|
||
" | __end__ | \n",
|
||
" +---------+ \n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 4
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T01:15:00.845724709Z",
|
||
"start_time": "2026-06-09T01:15:00.715886583Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"# 2개의 노드\n",
|
||
"\n",
|
||
"def first_increment(state):\n",
|
||
" return {\"count\":state['count']+1}\n",
|
||
"\n",
|
||
"def second_increment(state):\n",
|
||
" return {\"count\":state['count']+10}\n",
|
||
"\n",
|
||
"class CounterState(TypedDict):\n",
|
||
" count: int\n",
|
||
"\n",
|
||
"# START => first => second => END\n",
|
||
"# 그래프 생성\n",
|
||
"graph = StateGraph(CounterState)\n",
|
||
"graph.add_node(\"first\",first_increment)\n",
|
||
"graph.add_node(\"second\", second_increment)\n",
|
||
"\n",
|
||
"# 그래프 연결(그래프가 실행될 것인가?)\n",
|
||
"graph.add_edge(START, \"first\")\n",
|
||
"graph.add_edge(\"first\", 'second')\n",
|
||
"graph.add_edge('second', END)\n",
|
||
"\n",
|
||
"# 그래프를 실행 가능한 형태로 변경\n",
|
||
"app = graph.compile()\n",
|
||
"result = app.invoke({\"count\": 0})\n",
|
||
"print(f\"최종 결과 {result}\")\n",
|
||
"\n",
|
||
"# 그래프 시각화(참고)\n",
|
||
"app.get_graph().print_ascii()"
|
||
],
|
||
"id": "d0b60e65d488a596",
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"최종 결과 {'count': 11}\n",
|
||
"+-----------+ \n",
|
||
"| __start__ | \n",
|
||
"+-----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +-------+ \n",
|
||
" | first | \n",
|
||
" +-------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +--------+ \n",
|
||
" | second | \n",
|
||
" +--------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +---------+ \n",
|
||
" | __end__ | \n",
|
||
" +---------+ \n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 5
|
||
},
|
||
{
|
||
"metadata": {},
|
||
"cell_type": "markdown",
|
||
"source": [
|
||
"- 조건부 엣지\n",
|
||
" - 런타임 상태에 따라 동적으로 실행 경로 결정\n",
|
||
" - add_conditional_edges()"
|
||
],
|
||
"id": "c516f3c13d3ab6e7"
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T01:15:00.922609854Z",
|
||
"start_time": "2026-06-09T01:15:00.863712659Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"# 입력숫자 > 10 => big\n",
|
||
"# 입력숫자 > 10 => small\n",
|
||
"\n",
|
||
"# State : 숫자, 결과\n",
|
||
"class NumberState(TypedDict):\n",
|
||
" number : int\n",
|
||
" result : str\n",
|
||
"\n",
|
||
"# 노드(result 값 변경)\n",
|
||
"def handle_big_number(state):\n",
|
||
" return {'result':f\"{state['number']}는 큰 숫자입니다.\"}\n",
|
||
"\n",
|
||
"def handle_small_number(state):\n",
|
||
" return {'result':f\"{state['number']}는 작은 숫자입니다.\"}\n",
|
||
"\n",
|
||
"# 라우터(조건함수)\n",
|
||
"def check_size(state):\n",
|
||
" if state['number'] > 10:\n",
|
||
" return \"big\"\n",
|
||
" else:\n",
|
||
" return \"small\"\n",
|
||
"\n",
|
||
"# 그래프 생성\n",
|
||
"graph = StateGraph(NumberState)\n",
|
||
"graph.add_node(\"big_handler\", handle_big_number)\n",
|
||
"graph.add_node(\"small_handler\", handle_small_number)\n",
|
||
"\n",
|
||
"# 엣지\n",
|
||
"graph.add_edge(\"big_handler\", END)\n",
|
||
"graph.add_edge(\"small_handler\", END)\n",
|
||
"\n",
|
||
"# 조건부 엣지(number 값에 따라 big? small?)\n",
|
||
"graph.add_conditional_edges(START, check_size, {'big':'big_handler', \"small\":\"small_handler\"})\n",
|
||
"\n",
|
||
"app = graph.compile()\n",
|
||
"result = app.invoke({\"number\": 15, \"result\" : \"\"})\n",
|
||
"print(f\"큰 숫자 {result}\")\n",
|
||
"\n",
|
||
"result = app.invoke({\"number\": 5, \"result\" : \"\"})\n",
|
||
"print(f\"작은 숫자 {result}\")\n",
|
||
"\n",
|
||
"# 그래프 시각화(참고)\n",
|
||
"app.get_graph().print_ascii()\n"
|
||
],
|
||
"id": "acf3f40b4d866477",
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"큰 숫자 {'number': 15, 'result': '15는 큰 숫자입니다.'}\n",
|
||
"작은 숫자 {'number': 5, 'result': '5는 작은 숫자입니다.'}\n",
|
||
" +-----------+ \n",
|
||
" | __start__ | \n",
|
||
" +-----------+ \n",
|
||
" .. .. \n",
|
||
" .. .. \n",
|
||
" .. .. \n",
|
||
"+-------------+ +---------------+ \n",
|
||
"| big_handler | | small_handler | \n",
|
||
"+-------------+ +---------------+ \n",
|
||
" ** ** \n",
|
||
" ** ** \n",
|
||
" ** ** \n",
|
||
" +---------+ \n",
|
||
" | __end__ | \n",
|
||
" +---------+ \n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 6
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T01:15:00.979524486Z",
|
||
"start_time": "2026-06-09T01:15:00.927595212Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"# 홀, 짝\n",
|
||
"\n",
|
||
"# State : 숫자, 결과\n",
|
||
"class NumberState(TypedDict):\n",
|
||
" number : int\n",
|
||
" result : str\n",
|
||
"\n",
|
||
"# 노드(result 값 변경)\n",
|
||
"def even_node(state):\n",
|
||
" return {'result':f\"{state['number']}는 짝수입니다.\"}\n",
|
||
"\n",
|
||
"def odd_node(state):\n",
|
||
" return {'result':f\"{state['number']}는 홀수입니다.\"}\n",
|
||
"\n",
|
||
"# 라우터(조건함수)\n",
|
||
"def check_number(state):\n",
|
||
" if state['number'] % 2 == 0:\n",
|
||
" return \"even\"\n",
|
||
" else:\n",
|
||
" return \"odd\"\n",
|
||
"\n",
|
||
"# 그래프 생성\n",
|
||
"graph = StateGraph(NumberState)\n",
|
||
"graph.add_node(\"even_handler\", even_node)\n",
|
||
"graph.add_node(\"odd_handler\", odd_node)\n",
|
||
"\n",
|
||
"# 엣지\n",
|
||
"graph.add_edge(\"even_handler\", END)\n",
|
||
"graph.add_edge(\"odd_handler\", END)\n",
|
||
"\n",
|
||
"# 조건부 엣지(number 값에 따라 even? odd)\n",
|
||
"graph.add_conditional_edges(START, check_number, {'even':'even_handler', \"odd\":\"odd_handler\"})\n",
|
||
"\n",
|
||
"app = graph.compile()\n",
|
||
"result = app.invoke({\"number\": 15, \"result\" : \"\"})\n",
|
||
"print(f\"홀수 {result}\")\n",
|
||
"\n",
|
||
"result = app.invoke({\"number\": 6, \"result\" : \"\"})\n",
|
||
"print(f\"짝수 {result}\")\n",
|
||
"\n",
|
||
"# 그래프 시각화(참고)\n",
|
||
"app.get_graph().print_ascii()"
|
||
],
|
||
"id": "6900d17fc36539a3",
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"홀수 {'number': 15, 'result': '15는 홀수입니다.'}\n",
|
||
"짝수 {'number': 6, 'result': '6는 짝수입니다.'}\n",
|
||
" +-----------+ \n",
|
||
" | __start__ | \n",
|
||
" +-----------+ \n",
|
||
" .. .. \n",
|
||
" .. .. \n",
|
||
" .. .. \n",
|
||
"+--------------+ +-------------+ \n",
|
||
"| even_handler | | odd_handler | \n",
|
||
"+--------------+ +-------------+ \n",
|
||
" ** ** \n",
|
||
" ** ** \n",
|
||
" ** ** \n",
|
||
" +---------+ \n",
|
||
" | __end__ | \n",
|
||
" +---------+ \n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 7
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T01:15:01.028718647Z",
|
||
"start_time": "2026-06-09T01:15:00.985340859Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"class ScoreState(TypedDict):\n",
|
||
" score:int\n",
|
||
"\n",
|
||
"# 노드 return => 상태값 변경\n",
|
||
"# return 을 안하면 None\n",
|
||
"def grade_a(state):\n",
|
||
" print(\"A 학점\")\n",
|
||
" # score = None\n",
|
||
" return {}\n",
|
||
"\n",
|
||
"def grade_b(state):\n",
|
||
" print(\"B 학점\")\n",
|
||
" return {}\n",
|
||
"\n",
|
||
"def grade_c(state):\n",
|
||
" print(\"C 학점\")\n",
|
||
" return {}\n",
|
||
"\n",
|
||
"# >=90 A, >=80 B, 나머지C,\n",
|
||
"# 라우터(조건함수)\n",
|
||
"def route_grade(state):\n",
|
||
" if state['score'] >= 90:\n",
|
||
" return \"A\"\n",
|
||
" elif state['score'] >= 80:\n",
|
||
" return \"B\"\n",
|
||
" else:\n",
|
||
" return \"C\"\n",
|
||
"\n",
|
||
"graph = StateGraph(ScoreState)\n",
|
||
"graph.add_node(\"grade_a\", grade_a)\n",
|
||
"graph.add_node(\"grade_b\", grade_b)\n",
|
||
"graph.add_node(\"grade_c\", grade_c)\n",
|
||
"\n",
|
||
"graph.add_edge(\"grade_a\", END)\n",
|
||
"graph.add_edge(\"grade_b\", END)\n",
|
||
"graph.add_edge(\"grade_c\", END)\n",
|
||
"\n",
|
||
"graph.add_conditional_edges(START, route_grade, {'A':'grade_a', \"B\":\"grade_b\", \"C\":\"grade_c\"})\n",
|
||
"\n",
|
||
"app = graph.compile()\n",
|
||
"result = app.invoke({\"score\": 95, \"result\" : \"\"})\n",
|
||
"print(f\"학점 {result}\")\n",
|
||
"\n",
|
||
"result = app.invoke({\"score\": 88, \"result\" : \"\"})\n",
|
||
"print(f\"학점 {result}\")\n",
|
||
"\n",
|
||
"# app.get_graph().print_ascii()"
|
||
],
|
||
"id": "3ec626fba58ea942",
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"A 학점\n",
|
||
"학점 {'score': 95}\n",
|
||
"B 학점\n",
|
||
"학점 {'score': 88}\n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 8
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T01:15:01.076979177Z",
|
||
"start_time": "2026-06-09T01:15:01.038237715Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"# 상태관리 : text, sentiment, result\n",
|
||
"# text : 오늘 기분이 너무 좋아\n",
|
||
"# analyze_sentiment() : 감정평가 text 좋 positive / 싫 negative / x neutral\n",
|
||
"# positive_node() : 긍정 의견 / negative_node() : 부정 의견 / neutral_node() : 중립 의견 => result 업데이트\n",
|
||
"# route_sentiment : return state['sentiment']\n",
|
||
"\n",
|
||
"# START => analyze => positive / negative / neutral => route_sentiment => positive_node / negative_node / neutral_node => END\n",
|
||
"\n",
|
||
"class SentimentState(TypedDict):\n",
|
||
" text : str # 입력 문장\n",
|
||
" sentiment : str # 감정 결과 (positive / negative / neutral)\n",
|
||
" result : str # 최종 출력 메시지\n",
|
||
"\n",
|
||
"def analyze_sentiment(state):\n",
|
||
" text = state[\"text\"]\n",
|
||
"\n",
|
||
" if \"좋\" in text:\n",
|
||
" sentiment = \"positive\"\n",
|
||
" elif \"싫\" in text:\n",
|
||
" sentiment = \"negative\"\n",
|
||
" else:\n",
|
||
" sentiment = \"neutral\"\n",
|
||
"\n",
|
||
" return {\"sentiment\": sentiment}\n",
|
||
"\n",
|
||
"def positive_node(state):\n",
|
||
" return {\"result\": \"긍정 의견\"}\n",
|
||
"\n",
|
||
"def negative_node(state):\n",
|
||
" return {\"result\": \"부정 의견\"}\n",
|
||
"\n",
|
||
"def neutral_node(state):\n",
|
||
" return {\"result\": \"중립 의견\"}\n",
|
||
"\n",
|
||
"# 라우터(조건함수)\n",
|
||
"def route_sentiment(state):\n",
|
||
" return state['sentiment']\n",
|
||
"\n",
|
||
"graph = StateGraph(SentimentState)\n",
|
||
"graph.add_node(\"analyze\", analyze_sentiment)\n",
|
||
"graph.add_node(\"positive\", positive_node)\n",
|
||
"graph.add_node(\"negative\", negative_node)\n",
|
||
"graph.add_node(\"neutral\", neutral_node)\n",
|
||
"\n",
|
||
"graph.add_edge(START, \"analyze\")\n",
|
||
"graph.add_edge(\"positive\", END)\n",
|
||
"graph.add_edge(\"negative\", END)\n",
|
||
"graph.add_edge(\"neutral\", END)\n",
|
||
"\n",
|
||
"graph.add_conditional_edges(\"analyze\", route_sentiment, {'positive':'positive', \"negative\":\"negative\", \"neutral\":\"neutral\"}) # 라우터 결과가 'positive'면 ->️ 'positive' 노드로 가라!\n",
|
||
"\n",
|
||
"app = graph.compile()\n",
|
||
"result = app.invoke({\"text\": \"좋\"})\n",
|
||
"print(f\"{result}\")\n",
|
||
"\n",
|
||
"result = app.invoke({\"text\": \"싫\"})\n",
|
||
"print(f\"{result}\")\n",
|
||
"\n",
|
||
"result = app.invoke({\"text\": \"밥\"})\n",
|
||
"print(f\"{result}\")\n",
|
||
"\n",
|
||
"app.get_graph().print_ascii()"
|
||
],
|
||
"id": "94202953b8b8be5f",
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"{'text': '좋', 'sentiment': 'positive', 'result': '긍정 의견'}\n",
|
||
"{'text': '싫', 'sentiment': 'negative', 'result': '부정 의견'}\n",
|
||
"{'text': '밥', 'sentiment': 'neutral', 'result': '중립 의견'}\n",
|
||
" +-----------+ \n",
|
||
" | __start__ | \n",
|
||
" +-----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +---------+ \n",
|
||
" | analyze | \n",
|
||
" ...+---------+.... \n",
|
||
" .... . .... \n",
|
||
" .... . .... \n",
|
||
" .. . .. \n",
|
||
"+----------+ +---------+ +----------+ \n",
|
||
"| negative | | neutral | | positive | \n",
|
||
"+----------+**** +---------+ ***+----------+ \n",
|
||
" **** * **** \n",
|
||
" **** * **** \n",
|
||
" ** * ** \n",
|
||
" +---------+ \n",
|
||
" | __end__ | \n",
|
||
" +---------+ \n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 9
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T01:15:01.120328425Z",
|
||
"start_time": "2026-06-09T01:15:01.081769309Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"# .env 내용 가져오기\n",
|
||
"load_dotenv()\n",
|
||
"\n",
|
||
"apikey = os.getenv(\"WATSONX_API_KEY\")\n",
|
||
"project_id = os.getenv(\"WATSONX_PROJECT_ID\")\n",
|
||
"watsonx_ai_url = os.getenv(\"WATSONX_URL\")\n",
|
||
"hf_token = os.getenv(\"HF_TOKEN\")\n",
|
||
"COHERE_API_KEY = os.getenv(\"COHERE_API_KEY\")\n",
|
||
"SERPER_API_KEY = os.getenv(\"SERPER_API_KEY\")\n",
|
||
"GEMINI_API_KEY = os.getenv(\"GEMINI_API_KEY\")\n",
|
||
"LANGSMITH_API_KEY = os.getenv(\"LANGSMITH_API_KEY\")"
|
||
],
|
||
"id": "c2b2990c27f33682",
|
||
"outputs": [],
|
||
"execution_count": 10
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T01:15:06.547905777Z",
|
||
"start_time": "2026-06-09T01:15:01.122188390Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"watson_llm = ChatWatsonx(\n",
|
||
" model_id=\"ibm/granite-4-h-small\",\n",
|
||
" url=f\"{watsonx_ai_url}\",\n",
|
||
" api_key=f\"{apikey}\",\n",
|
||
" project_id=f\"{project_id}\",\n",
|
||
" max_tokens=2000,\n",
|
||
" params={\n",
|
||
" \"temperature\": 0\n",
|
||
" }\n",
|
||
")\n",
|
||
"\n",
|
||
"watson_embedding = WatsonxEmbeddings(\n",
|
||
" model_id=\"ibm/granite-embedding-278m-multilingual\",\n",
|
||
" url=f\"{watsonx_ai_url}\",\n",
|
||
" api_key=f\"{apikey}\",\n",
|
||
" project_id=f\"{project_id}\"\n",
|
||
")\n",
|
||
"\n",
|
||
"hugging_llm = ChatOpenAI(\n",
|
||
" model=\"Qwen/Qwen2.5-7B-Instruct:together\",\n",
|
||
" api_key=hf_token,\n",
|
||
" base_url=\"https://router.huggingface.co/v1\",\n",
|
||
" temperature=0,\n",
|
||
")\n",
|
||
"\n",
|
||
"ollama_embedding = OllamaEmbeddings(model=\"nomic-embed-text-v2-moe\")\n",
|
||
"\n",
|
||
"# 로컬 LLM\n",
|
||
"qwen_llm = ChatOllama(model=\"qwen3.5:4b\", temperature=0)\n",
|
||
"exaone_llm = ChatOllama(model=\"exaone3.5:2.4b\", temperature=0)\n",
|
||
"gemma_llm = ChatOllama(model=\"gemma4:e2b\")"
|
||
],
|
||
"id": "f5fdb0bea4930484",
|
||
"outputs": [],
|
||
"execution_count": 11
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T01:15:09.926585532Z",
|
||
"start_time": "2026-06-09T01:15:06.560447736Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"parser = StrOutputParser()\n",
|
||
"\n",
|
||
"translate_chain = ChatPromptTemplate.from_messages([\n",
|
||
" (\"system\", \"다음 텍스트를 한국어로 번역하세요. 번역문만 출력:\\n{text}.\"),\n",
|
||
"]) | watson_llm | parser\n",
|
||
"\n",
|
||
"summarize_chain = ChatPromptTemplate.from_messages([\n",
|
||
" (\"system\", \"다음 텍스트를 3문장으로 요약하세요.:\\n{text}.\"),\n",
|
||
"]) | watson_llm | parser\n",
|
||
"\n",
|
||
"sentiment_chain = ChatPromptTemplate.from_messages([\n",
|
||
" (\"system\", \"다음 텍스트를 의 감정을 긍정/부정/중립 중 하나로만 답하세요.:\\n{summary}.\"),\n",
|
||
"]) | watson_llm | parser\n",
|
||
"\n",
|
||
"# State 정의\n",
|
||
"class AnalysisState(TypedDict):\n",
|
||
" text: str\n",
|
||
" translated: str\n",
|
||
" summary: str\n",
|
||
" sentiment: str\n",
|
||
" done: bool\n",
|
||
"\n",
|
||
"# node 정의\n",
|
||
"def translated_node(state):\n",
|
||
" result = translate_chain.invoke({\"text\":state['text']})\n",
|
||
" return {\"translated\":result}\n",
|
||
"\n",
|
||
"def summary_node(state):\n",
|
||
" result = summarize_chain.invoke({'text':state['translated']})\n",
|
||
" return {\"summary\":result}\n",
|
||
"\n",
|
||
"def sentiment_node(state):\n",
|
||
" result = sentiment_chain.invoke({'summary':state['summary']})\n",
|
||
" return {\"sentiment\":result}\n",
|
||
"\n",
|
||
"# 그래프 생성\n",
|
||
"graph = StateGraph(AnalysisState)\n",
|
||
"graph.add_node(\"translate\", translated_node)\n",
|
||
"graph.add_node(\"summarize\", summary_node)\n",
|
||
"graph.add_node(\"sentiment\", sentiment_node)\n",
|
||
"\n",
|
||
"graph.add_edge(START, \"translate\")\n",
|
||
"graph.add_edge(\"translate\", \"summarize\")\n",
|
||
"graph.add_edge(\"summarize\", \"sentiment\")\n",
|
||
"graph.add_edge(\"sentiment\", END)\n",
|
||
"\n",
|
||
"# 그래프 실행\n",
|
||
"app = graph.compile()\n",
|
||
"result = app.invoke({'text': 'Python is grate!!', \"done\":False})\n",
|
||
"print(\"번역 : \", result['translated'])\n",
|
||
"print(\"요약 : \", result['summary'])\n",
|
||
"print(\"감정 : \", result['sentiment'])\n",
|
||
"\n",
|
||
"app.get_graph().print_ascii()"
|
||
],
|
||
"id": "75b0102055a3b951",
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"번역 : 파이썬은 대단해요!!\n",
|
||
"요약 : 파이썬은 대단한 프로그래밍 언어입니다. 파이썬은 간결하고 읽기 쉬운 문법을 가지고 있어 초보자에게도 적합합니다. 또한 다양한 라이브러리와 프레임워크를 제공하여 다양한 분야에서 활용될 수 있습니다.\n",
|
||
"감정 : 긍정\n",
|
||
"+-----------+ \n",
|
||
"| __start__ | \n",
|
||
"+-----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
"+-----------+ \n",
|
||
"| translate | \n",
|
||
"+-----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
"+-----------+ \n",
|
||
"| summarize | \n",
|
||
"+-----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
"+-----------+ \n",
|
||
"| sentiment | \n",
|
||
"+-----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +---------+ \n",
|
||
" | __end__ | \n",
|
||
" +---------+ \n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 12
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T01:15:11.407677695Z",
|
||
"start_time": "2026-06-09T01:15:09.946940683Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"# 조건부엣지 : summary >= 100자 이상 good / 100자 미만 poor\n",
|
||
"\n",
|
||
"class SummaryState(TypedDict):\n",
|
||
" text:str\n",
|
||
" summary:str\n",
|
||
" quality:str\n",
|
||
" retries:int\n",
|
||
"\n",
|
||
"def summary_node(state):\n",
|
||
" result = summarize_chain.invoke({'text':state['text']})\n",
|
||
" return {\"summary\":result}\n",
|
||
"\n",
|
||
"# quality\n",
|
||
"def check_quality_node(state):\n",
|
||
" # summary >= 100 자 이상 good / 100자 미만 poor\n",
|
||
" quality = \"good\" if len(state['summary']) >= 100 else \"poor\"\n",
|
||
" return {\"quality\":quality}\n",
|
||
"\n",
|
||
"def retry_node(state):\n",
|
||
" return {\"text\":state['text']+\"\\n 더 자세히 요약하세요\", \"retries\":state['retries'] + 1}\n",
|
||
"\n",
|
||
"def route_by_quality(state):\n",
|
||
" if state['quality'] == 'poor' and state['retries'] < 3:\n",
|
||
" return \"retry\"\n",
|
||
" return \"done\"\n",
|
||
"\n",
|
||
"graph = StateGraph(SummaryState)\n",
|
||
"graph.add_node(\"summary\", summary_node)\n",
|
||
"graph.add_node(\"check_quality\", check_quality_node)\n",
|
||
"graph.add_node(\"retry\", retry_node)\n",
|
||
"\n",
|
||
"# START -> summary -> check_quality\n",
|
||
"graph.add_edge(START, \"summary\")\n",
|
||
"graph.add_edge(\"summary\", \"check_quality\")\n",
|
||
"graph.add_conditional_edges(\"check_quality\", route_by_quality, {\"retry\": \"retry\", \"done\" : END})\n",
|
||
"graph.add_edge(\"retry\", \"summary\")\n",
|
||
"\n",
|
||
"# 그래프 실행\n",
|
||
"app = graph.compile()\n",
|
||
"result = app.invoke({'text': '잛은 글'})\n",
|
||
"print(f\"최종요약({len(result['summary'])}자): {result['summary'][:100]}\")\n",
|
||
"\n",
|
||
"app.get_graph().print_ascii()\n"
|
||
],
|
||
"id": "b5a849ba6c73ae41",
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"최종요약(150자): 제공된 텍스트는 질문이 없는 한 문장으로, 요약할 내용이 없습니다. 그러나 이 문장을 3문장으로 요약하면 다음과 같습니다:\n",
|
||
"\n",
|
||
"1. 이 프롬프트는 요약을 요청하는 것입니다.\n",
|
||
"2. 하\n",
|
||
" +-----------+ \n",
|
||
" | __start__ | \n",
|
||
" +-----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +---------+ \n",
|
||
" | summary | \n",
|
||
" +---------+ \n",
|
||
" *** *** \n",
|
||
" * * \n",
|
||
" ** *** \n",
|
||
"+---------------+ * \n",
|
||
"| check_quality | * \n",
|
||
"+---------------+. * \n",
|
||
" . ..... * \n",
|
||
" . ... * \n",
|
||
" . ... * \n",
|
||
" +---------+ +-------+ \n",
|
||
" | __end__ | | retry | \n",
|
||
" +---------+ +-------+ \n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 13
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T01:15:20.189038478Z",
|
||
"start_time": "2026-06-09T01:15:11.427378090Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"class VerifyState(TypedDict):\n",
|
||
" question:str\n",
|
||
" answer:str\n",
|
||
" feedback:str\n",
|
||
" is_verified:bool\n",
|
||
" attempt:int\n",
|
||
"\n",
|
||
"def generate(state):\n",
|
||
" \"\"\"답변을 생성합니다. 이전 피드백이 있으면 반영합니다.\"\"\"\n",
|
||
" prompt = f\"질문 : {state['question']}\"\n",
|
||
"\n",
|
||
" if state['feedback']:\n",
|
||
" prompt += f\"\\n\\n이전 답변의 피드백 : {state['feedback']}\\n위 피드백을 반영하여 개선된 답변을 작성하세요\"\n",
|
||
"\n",
|
||
" result = watson_llm.invoke(prompt)\n",
|
||
" return {'answer':result.content, \"attempt\":state['attempt'] + 1}\n",
|
||
"\n",
|
||
"def verify(state):\n",
|
||
" \"\"\"답변의 정확성과 완전성을 검증합니다.\"\"\"\n",
|
||
" verification = watson_llm.invoke(f\"\"\"\n",
|
||
"다음 답변의 정확성과 완정성을 검증하세요\n",
|
||
"\n",
|
||
"질문 :\n",
|
||
"{state['question']}\n",
|
||
"\n",
|
||
"답변 :\n",
|
||
"{state['answer']}\n",
|
||
"\n",
|
||
"정확하고 완전하면 첫 줄에 'PASS' 를, 수정이 필요하면 첫줄에 'FAIL' 을 쓰고 구체적인 개선사항을 설명하세요.\n",
|
||
"\"\"\")\n",
|
||
"\n",
|
||
" content = verification.content\n",
|
||
" is_pass = content.strip().startswith('PASS')\n",
|
||
"\n",
|
||
" return {'is_verified':is_pass, 'feedback':content}\n",
|
||
"\n",
|
||
"# router 개념\n",
|
||
"def should_retry(state):\n",
|
||
" \"\"\"검증 통과 또는 최대 횟수 도달시 종료\"\"\"\n",
|
||
" if state['is_verified'] or state['attempt'] >= 3:\n",
|
||
" return END\n",
|
||
" return 'generate'\n",
|
||
"\n",
|
||
"# 질문 => LLM 답변 생성 => LLM 답변 검증 => 검증통과\n",
|
||
"# 검증 미통과 => LLM 답변 생성\n",
|
||
"\n",
|
||
"graph = StateGraph(VerifyState)\n",
|
||
"graph.add_node(\"generate\", generate)\n",
|
||
"graph.add_node(\"verify\", verify)\n",
|
||
"\n",
|
||
"graph.add_edge(START, \"generate\")\n",
|
||
"graph.add_edge(\"generate\", \"verify\")\n",
|
||
"graph.add_conditional_edges(\"verify\", should_retry, {\"generate\": \"generate\", END: END})\n",
|
||
"\n",
|
||
"# 그래프 실행\n",
|
||
"app = graph.compile()\n",
|
||
"result = app.invoke({'question': '파이썬에서 GIL이 무엇이며 멀티쓰레딩에 어떤 영향을 주는지 설명해줘', \"answer\":\"\",\"feedback\":\"\", \"is_verified\":False, \"attempt\":0})\n",
|
||
"\n",
|
||
"print(\"시도 횟수\", result['attempt'])\n",
|
||
"print(\"검증통과\", result['is_verified'])\n",
|
||
"print(\"최종답변\", result['answer'][:300])\n",
|
||
"\n",
|
||
"app.get_graph().print_ascii()"
|
||
],
|
||
"id": "686634fe1625b815",
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"시도 횟수 1\n",
|
||
"검증통과 True\n",
|
||
"최종답변 GIL(Global Interpreter Lock)은 Python 인터프리터에서 사용되는 뮤텍스(Mutex)입니다. 이는 Python 인터프리터가 동시에 하나의 스레드만 실행하도록 제한하는 역할을 합니다. GIL은 Python의 메모리 관리가 스레드 안전하지 않기 때문에 도입되었습니다. 이는 Python의 메모리 관리가 참조 카운팅 기반으로 이루어지기 때문에, 여러 스레드가 동시에 객체의 참조 카운트를 변경하면 메모리 누수나 다른 문제가 발생할 수 있기 때문입니다.\n",
|
||
"\n",
|
||
"GIL의 존재는 멀티쓰레딩에 다음과 같은 영향을 미칩니다:\n",
|
||
"\n",
|
||
"1.\n",
|
||
"+-----------+ \n",
|
||
"| __start__ | \n",
|
||
"+-----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
"+----------+ \n",
|
||
"| generate | \n",
|
||
"+----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +--------+ \n",
|
||
" | verify | \n",
|
||
" +--------+ \n",
|
||
" . \n",
|
||
" . \n",
|
||
" . \n",
|
||
" +---------+ \n",
|
||
" | __end__ | \n",
|
||
" +---------+ \n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 14
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T01:15:23.772672246Z",
|
||
"start_time": "2026-06-09T01:15:20.206724228Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"import time\n",
|
||
"from langchain_core.runnables import RunnableParallel\n",
|
||
"\n",
|
||
"class AnalysisState(TypedDict):\n",
|
||
" text:str\n",
|
||
" translated:str\n",
|
||
" summary:str\n",
|
||
" keywords:list[str]\n",
|
||
"\n",
|
||
"translate_chain = ChatPromptTemplate.from_messages([\n",
|
||
" (\"system\", \"다음 텍스트를 한국어로 번역하세요. 번역문만 출력:\\n{text}.\"),\n",
|
||
"]) | watson_llm | parser\n",
|
||
"\n",
|
||
"summarize_chain = ChatPromptTemplate.from_messages([\n",
|
||
" (\"system\", \"다음 텍스트를 3문장으로 번역하세요.:\\n{text}.\"),\n",
|
||
"]) | watson_llm | parser\n",
|
||
"\n",
|
||
"keyword_chain = ChatPromptTemplate.from_messages([\n",
|
||
" (\"system\", \"키워드 5개 추출:\\n{text}.\"),\n",
|
||
"]) | watson_llm | parser\n",
|
||
"\n",
|
||
"def translate_node(state):\n",
|
||
" print(\"번역 시작\")\n",
|
||
" time.sleep(3)\n",
|
||
" print(\"번역 종료\")\n",
|
||
" return {\"translated\": translate_chain.invoke({\"text\":state['text']})}\n",
|
||
"\n",
|
||
"def summarize_node(state):\n",
|
||
" print(\"요약 시작\")\n",
|
||
" time.sleep(3)\n",
|
||
" print(\"요약 종료\")\n",
|
||
" return {\"summary\": summarize_chain.invoke({\"text\":state['text']})}\n",
|
||
"\n",
|
||
"def keyword_node(state):\n",
|
||
" print(\"키워드 추출 시작\")\n",
|
||
" time.sleep(3)\n",
|
||
" print(\"키워드 추출 종료\")\n",
|
||
" return {\"keywords\": keyword_chain.invoke({\"text\":state['text']})}\n",
|
||
"\n",
|
||
"graph = StateGraph(AnalysisState)\n",
|
||
"graph.add_node('translated', translated_node)\n",
|
||
"graph.add_node('summary', summarize_node)\n",
|
||
"graph.add_node('keywords', keyword_node)\n",
|
||
"\n",
|
||
"graph.add_edge(START, \"translated\")\n",
|
||
"graph.add_edge(START, \"summary\")\n",
|
||
"graph.add_edge(START, \"keywords\")\n",
|
||
"\n",
|
||
"graph.add_edge(\"translated\", END)\n",
|
||
"graph.add_edge(\"summary\", END)\n",
|
||
"graph.add_edge(\"keywords\", END)\n",
|
||
"\n",
|
||
"app = graph.compile()\n",
|
||
"text = {'text' : 'Python is a versatile language used in AI and web development'}\n",
|
||
"result = app.invoke(text)\n",
|
||
"\n",
|
||
"print(\"번역\", result['translated'][:100])\n",
|
||
"print(\"요약\", result['summary'][:100])\n",
|
||
"print(\"키워드\", result['keywords'][:100])\n",
|
||
"\n",
|
||
"app.get_graph().print_ascii()"
|
||
],
|
||
"id": "956e5a8f89675559",
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"키워드 추출 시작요약 시작\n",
|
||
"\n",
|
||
"요약 종료\n",
|
||
"키워드 추출 종료\n",
|
||
"번역 파이썬은 AI와 웹 개발에 사용되는 다용도의 언어입니다.\n",
|
||
"요약 파이썬은 AI와 웹 개발에 사용되는 다용도 언어입니다.\n",
|
||
"키워드 1. Python\n",
|
||
"2. versatile\n",
|
||
"3. language\n",
|
||
"4. AI\n",
|
||
"5. web development\n",
|
||
" +-----------+ \n",
|
||
" | __start__ | \n",
|
||
" **+-----------+**** \n",
|
||
" **** * **** \n",
|
||
" **** * **** \n",
|
||
" ** * ** \n",
|
||
"+----------+ +---------+ +------------+ \n",
|
||
"| keywords | | summary | | translated | \n",
|
||
"+----------+**** +---------+ **+------------+ \n",
|
||
" **** * **** \n",
|
||
" **** * **** \n",
|
||
" ** * ** \n",
|
||
" +---------+ \n",
|
||
" | __end__ | \n",
|
||
" +---------+ \n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 15
|
||
},
|
||
{
|
||
"metadata": {},
|
||
"cell_type": "markdown",
|
||
"source": [
|
||
"- rag LangGraph\n",
|
||
" - 사용자 질문 => retrieve => generate => END"
|
||
],
|
||
"id": "e9270ebbd217d878"
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T01:15:24.985154798Z",
|
||
"start_time": "2026-06-09T01:15:23.789840941Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"# !pip install langchain_community\n",
|
||
"from langchain_community.document_loaders import PyPDFLoader, CSVLoader, WebBaseLoader, DirectoryLoader\n",
|
||
"from youtube_transcript_api import YouTubeTranscriptApi\n",
|
||
"from langchain_core.documents import Document\n",
|
||
"from langchain_text_splitters import RecursiveCharacterTextSplitter\n",
|
||
"from langchain_ollama import OllamaEmbeddings\n",
|
||
"from langchain_ibm import WatsonxEmbeddings\n",
|
||
"from langchain_chroma import Chroma\n",
|
||
"from langchain_community.vectorstores import FAISS"
|
||
],
|
||
"id": "d03e2b9989d53cb2",
|
||
"outputs": [
|
||
{
|
||
"name": "stderr",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"USER_AGENT environment variable not set, consider setting it to identify your requests.\n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 16
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T01:15:28.610696993Z",
|
||
"start_time": "2026-06-09T01:15:25.007089688Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"# STEP 1 : 문서로드\n",
|
||
"loader = PyPDFLoader(\"./data/Summary of ChatGPTGPT-4 Research.pdf\")\n",
|
||
"# STEP 2 : 문서분할\n",
|
||
"splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)\n",
|
||
"chunks = splitter.split_documents(loader.load())\n",
|
||
"print(f\"chunks 수 {chunks}\")\n",
|
||
"# STEP 3 : 인덱싱 - 임베딩\n",
|
||
"# STEP 4 : 벡터스토어(Chroma or FAISS)\n",
|
||
"vectorstore = Chroma.from_documents(chunks, watson_embedding, persist_directory=\"./db/chroma_db\", collection_name=\"research\")\n",
|
||
"# STEP 5 : as_retriever() : Vector Store 를 Retriever 형태로 변환하여 LangChain 에 연결\n",
|
||
"retriever = vectorstore.as_retriever(search_type=\"similarity\", search_kwargs={\"k\":3})"
|
||
],
|
||
"id": "1f0fda34f40177ea",
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"chunks 수 [Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 0, 'page_label': '1'}, page_content='Summary of ChatGPT/GPT-4 Research\\nand Perspective Towards the Future of Large\\nLanguage Models\\nYiheng Liu ∗1, Tianle Han ∗1, Siyuan Ma 1, Jiayue Zhang 1,\\nYuanyuan Yang1, Jiaming Tian 1, Hao He 1, Antong Li 2, Mengshen\\nHe1, Zhengliang Liu 3, Zihao Wu 3, Dajiang Zhu 4, Xiang Li 5, Ning\\nQiang1, Dingang Shen 6,7,8, Tianming Liu 3, and Bao Ge †1\\n1School of Physics and Information Technology, Shaanxi Normal University, Xi’an\\n710119 China'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 0, 'page_label': '1'}, page_content='710119 China\\n2School of Life and Technology Biomedical-Engineering, Xi’an Jiaotong University,\\nXi’an 710049, China\\n3School of Computing, The University of Georgia, Athens 30602, USA\\n4Department of Computer Science and Engineering, The University of Texas at\\nArlington, Arlington 76019, USA\\n5Department of Radiology, Massachusetts General Hospital and Harvard Medical\\nSchool, Boston 02115, USA\\n6School of Biomedical Engineering, ShanghaiTech University, Shanghai 201210,\\nChina'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 0, 'page_label': '1'}, page_content='China\\n7Shanghai United Imaging Intelligence Co., Ltd., Shanghai 200230, China\\n8Shanghai Clinical Research and Trial Center, Shanghai 201210, China\\nAbstract\\nThis paper presents a comprehensive survey of ChatGPT and GPT-4,\\nstate-of-the-art large language models (LLM) from the GPT series, and\\ntheir prospective applications across diverse domains. Indeed, key innova-\\ntions such as large-scale pre-training that captures knowledge across the'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 0, 'page_label': '1'}, page_content='entire world wide web, instruction fine-tuning and Reinforcement Learn-\\ning from Human Feedback (RLHF) have played significant roles in en-\\nhancing LLMs’ adaptability and performance. We performed an in-depth\\nanalysis of 194 relevant papers on arXiv, encompassing trend analysis,\\nword cloud representation, and distribution analysis across various appli-\\ncation domains. The findings reveal a significant and increasing interest\\nin ChatGPT/GPT-4 research, predominantly centered on direct natural'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 0, 'page_label': '1'}, page_content='language processing applications, while also demonstrating considerable\\npotential in areas ranging from education and history to mathematics,\\n∗Co-first author\\n†Corresponding author: bob ge@snnu.edu.cn\\n1\\narXiv:2304.01852v1 [cs.CL] 4 Apr 2023'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 1, 'page_label': '2'}, page_content='medicine, and physics. This study endeavors to furnish insights into Chat-\\nGPT’s capabilities, potential implications, ethical concerns, and offer di-\\nrection for future advancements in this field.\\n1 Introduction\\nRecent advances in natural language processing (NLP) have led to the devel-\\nopment of powerful language models such as the GPT (Generative Pre-trained\\nTransformer) series [1, 2, 3, 4, 5], including large language models (LLM) such'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 1, 'page_label': '2'}, page_content='as ChatGPT and GPT-4. These models are pre-trained on vast amounts of\\ntext data and have demonstrated exceptional performance in a wide range of\\nNLP tasks, including language translation, text summarization, and question-\\nanswering. In particular, the ChatGPT model has demonstrated its potential\\nin various fields, including education, healthcare, reasoning, text generation,\\nhuman-machine interaction, and scientific research.\\nA key milestone of LLM development is InstructGPT [5], a framework that'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 1, 'page_label': '2'}, page_content='allows for instruction fine-tuning of a pre-trained language model based on Re-\\ninforcement Learning from Human Feedback (RLHF) [6, 5]. This framework\\nenables an LLM to adapt to a wide range of NLP tasks, making it highly ver-\\nsatile and flexible by leveraging human feedback. RLHF enables the model to\\nalign with human preferences and human values, which significantly improves\\nfrom large language models that are solely trained text corpora through unsu-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 1, 'page_label': '2'}, page_content='pervised pre-training. ChatGPT is a successor to InstructGPT. Since its release\\nin December 2022, ChatGPT has been equipped with these advanced develop-\\nments, leading to impressive performances in various downstream NLP tasks\\nsuch as reasoning and generalized text generation. These unprecedented NLP\\ncapabilities spur applications in diverse domains such as education, healthcare,\\nhuman-machine interaction, medicine and scientific research. ChatGPT has re-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 1, 'page_label': '2'}, page_content='ceived widespread attention and interest, leading to an increasing number of\\napplications and research that harness its exceeding potential. The open release\\nof the multi-modal GPT-4 model further expands the horizon of large language\\nmodels and empowers exciting developments that involve diverse data beyond\\ntext.\\nThe purpose of this paper is to provide a comprehensive survey of the existing\\nresearch on ChatGPT and its potential applications in various fields. To achieve'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 1, 'page_label': '2'}, page_content='this goal, we conducted a thorough analysis of papers related to ChatGPT in\\nthe arXiv repository. As of April 1st, 2023, there are a total of 194 papers\\nmentioning ChatGPT on arXiv. In this study, we conducted a trend analysis of\\nthese papers and generated a word cloud to visualize the commonly used terms.\\nAdditionally, we also examined the distribution of the papers across various\\nfields and presented the corresponding statistics. Figure 1 displays the daily'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 1, 'page_label': '2'}, page_content='submission trend of papers related to ChatGPT, indicating a growing interest\\nin this field. Figure 2 illustrates the word cloud analysis of all the papers. We\\ncan observe that the current research is primarily focused on natural language\\nprocessing, but there is still significant potential for research in other fields such\\n2'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 2, 'page_label': '3'}, page_content='Figure 1: The number of papers submitted by researchers per day.\\n3'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 3, 'page_label': '4'}, page_content='Figure 2: Word cloud analysis of all the 194 papers.\\nas education and history. This is further supported by Figure 3, which displays\\nthe distribution of submitted papers across various fields, highlighting the need\\nfor more research and development in these areas.\\nThis paper aims to shed light on the promising capabilities of ChatGPT\\nand provide insight into its potential impact in the future, including ethical\\nconsiderations. Through this survey, we hope to provide insights into how these'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 3, 'page_label': '4'}, page_content='models can be improved and extended in the future. In section 2, we will\\nreview the existing work related to ChatGPT, including its applications, ethical\\nconsiderations, and evaluation. In addition to discussing the current state of\\nresearch related to ChatGPT, we will also explore its limitations in section 3.\\nFurthermore, we will provide guidance on future directions for language model\\ndevelopment.\\n2 Related work of ChatGPT'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 3, 'page_label': '4'}, page_content='development.\\n2 Related work of ChatGPT\\nIn this section, we review the latest research related to the application, ethics,\\nand evaluation of ChatGPT.\\n2.1 Application of ChatGPT\\n2.1.1 Question And Answering\\nIn the education field\\nChatGPT is commonly used for question and answers testing in the edu-\\ncation sector. Users can use ChatGPT to learn, compare and verify answers\\nfor different academic subjects such as physics, mathematics, and chemistry,\\n4'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 4, 'page_label': '5'}, page_content='Figure 3: The distribution of submitted papers across various fields.\\n5'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 5, 'page_label': '6'}, page_content='and/or conceptual subjects such as philosophy and religion. Additionally, users\\ncan ask open-ended and analytical questions to understand the capabilities of\\nChatGPT.\\nIn the field of mathematics, Frieder et al. [7] constructed the GHOSTS nat-\\nural language dataset, which consists of graduate-level math test questions.\\nThe authors tested ChatGPT’s math abilities on the GHOSTS dataset us-\\ning a question-and-answer format and evaluated it according to fine-grained'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 5, 'page_label': '6'}, page_content='standards.In the Grad Text dataset, which covers simple set theory and logic\\nproblems, ChatGPT performed the best. However, in the Olympiad-Problem-\\nSolving dataset, ChatGPT performed poorly, receiving only two 4-point scores\\n(out of a total of 5), with the majority of scores being 2 points. In the Holes-\\nin-Proofs dataset, ChatGPT received the lowest score of 1 point. In the MATH\\ndataset, ChatGPT only scored impressively in 26% of cases. These results sug-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 5, 'page_label': '6'}, page_content='gest that ChatGPT’s math abilities are clearly lower than those of ordinary\\nmath graduate students. Although ChatGPT can generally understand math\\nproblems, it fails to provide the correct solutions. Pardos et al. [8] used the\\nOpen Adaptive Tutoring system (OATutor) to investigate whether prompts gen-\\nerated by ChatGPT were helpful for learning algebra, with 77 participants from\\nMechanical Turk taking part in the experiment. The experiment used ques-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 5, 'page_label': '6'}, page_content='tions from OpenStax’s Elementary and Intermediate Algebra textbooks. These\\nparticipants were randomly assigned to either a control group (with manual\\nprompts) or an experimental group (with ChatGPT prompts). For each ques-\\ntion in both courses, the authors obtained answers from ChatGPT through a\\nquestion-and-answer format and evaluated scores according to three criteria:\\nChatGPT provided an answer, the answer was correct, and inappropriate lan-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 5, 'page_label': '6'}, page_content='guage was not used in the answer. The study found that 70% of prompts gener-\\nated by ChatGPT passed manual quality checks, and both humans and Chat-\\nGPT produced positive learning gains. However, the scores of human prompts\\nranged from 74.59% to 84.32%, significantly higher than those of ChatGPT\\nprompts. Shakarian et al. [9] studied the performance of ChatGPT on math\\nword problems (MWPs), using the DRAW-1K dataset for experimentation. The'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 5, 'page_label': '6'}, page_content='dataset consists of 1000 MWPs and their answers, along with algebraic equa-\\ntion templates for solving such problems. The authors used the idea of machine\\nlearning introspection and built performance prediction models using random\\nforests and XGBoost, and evaluated them on the dataset using five-fold cross-\\nvalidation. ChatGPT’s accuracy increased from an initial 34% to a final 69%,\\nwhile its recall increased from an initial 41% to a final 83%. The authors also'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 5, 'page_label': '6'}, page_content='found that ChatGPT’s failure rate decreased from an initial 84% to a final\\n20%, indicating that performance can vary greatly depending on specific job\\nrequirements.\\nIn the field of physics, Lehnert et al. [10] explored the capabilities and limita-\\ntions of ChatGPT by studying how it handles obscure physics topics such as the\\nswamp land conjecture in string theory. The experimental dialogue began with\\nbroader and more general questions in the field of string theory before narrowing'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 5, 'page_label': '6'}, page_content='down to specific swamp land conjectures and examining ChatGPT’s understand-\\ning of them. The study found that ChatGPT could define and explain different\\n6'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 6, 'page_label': '7'}, page_content='concepts in various styles, but was not effective in truly connecting various con-\\ncepts. It would confidently provide false information and fabricate statements\\nwhen necessary, indicating that ChatGPT cannot truly create new knowledge\\nor establish new connections. However, in terms of identifying analogies and\\ndescribing abstract concepts of visual representation, ChatGPT can cleverly\\nuse language. Kortemeyer et al. [11] evaluated ChatGPT’s ability to answer'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 6, 'page_label': '7'}, page_content='calculus-based physics questions through a question-and-answer test. The tests\\nincluded online homework, clicker questions, programming exercises, and exams\\ncovering classical mechanics, thermodynamics, electricity and magnetism, and\\nmodern physics. While ChatGPT was able to pass the course, it also demon-\\nstrated many misconceptions and errors commonly held by beginners. West et\\nal. [12] used the Force Concept Inventory (FCI) to evaluate ChatGPT’s accu-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 6, 'page_label': '7'}, page_content='racy in answering physics concept problems related to kinematics and Newtonian\\nmechanics in the first semester of college physics. The FCI covers topics such\\nas kinematics, projectile motion, free fall, circular motion, and Newton’s laws.\\nThe study included data from 415 students who took the FCI at the end of the\\nsemester, with an average score of 56%, while ChatGPT scored approximately\\nbetween 50% to 65%. The authors demonstrated that ChatGPT’s performance'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 6, 'page_label': '7'}, page_content='in physics learning can reach or even exceed the average level of a semester of\\ncollege physics.\\nIn the medical field\\nChatGPT’s question-answering capabilities can also be applied in the med-\\nical field, such as for answering medical questions from patients or assisting\\nhealthcare professionals in diagnosing diseases. Nov et al. [13] evaluated the fea-\\nsibility of using ChatGPT for patient-doctor communication. The experiment'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 6, 'page_label': '7'}, page_content='extracted 10 representative patient-doctor interactions from EHR, placed the\\npatient’s questions in ChatGPT, and asked ChatGPT to respond using roughly\\nthe same number of words as the doctor’s response. Each patient’s question was\\nanswered by either the doctor or ChatGPT, and the patient was informed that\\n5 were answered by the doctor and 5 were generated by ChatGPT, and was\\nasked to correctly identify the source of the response. The results of the exper-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 6, 'page_label': '7'}, page_content='iment showed that the probability of correctly identifying ChatGPT’s response\\nwas 65.5%, while the probability of correctly identifying the doctor’s response\\nwas 65.1%. In addition, the experiment found that the patient’s response to\\nthe trustworthiness of ChatGPT’s function was weakly positive (average Likert\\nscore: 3.4), and trust decreased as the complexity of health-related tasks in\\nthe questions increased. ChatGPT’s responses to patient questions were only'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 6, 'page_label': '7'}, page_content='slightly different from those of doctors, but people seem to trust ChatGPT to\\nanswer low-risk health questions, while for complex medical questions, people\\nstill tend to trust the doctor’s responses and advice.\\nTu et al. [14] explored the causal discovery ability of ChatGPT in the diag-\\nnosis of neuropathic pain. Causal relationship discovery aims to reveal potential\\nunknown causal relationships based purely on observed data [15]. The experi-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 6, 'page_label': '7'}, page_content='mental results found that ChatGPT has some limitations in understanding new\\n7'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 7, 'page_label': '8'}, page_content='knowledge and concepts beyond the existing textual training data corpus, that\\nis, it only understands language commonly used to describe situations and not\\nunderlying knowledge. In addition, its performance consistency and stability are\\nnot high, as the experiment observed that it would provide different answers for\\nthe same question under multiple inquiries. However, despite the many limita-\\ntions of ChatGPT, we believe that it has a great opportunity to improve causal\\nrelationship research.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 7, 'page_label': '8'}, page_content='relationship research.\\nIn the other field\\nGuo et al. [16] attempted to apply ChatGPT in the field of communication,\\nspecifically using ChatGPT for ordered importance semantic communication,\\nwhere ChatGPT plays the role of an intelligent consulting assistant that can\\nreplace humans in identifying the semantic importance of words in messages and\\ncan be directly embedded into the current communication system. For a message\\nto be transmitted, the sender first utilizes ChatGPT to output the semantic'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 7, 'page_label': '8'}, page_content='importance order of each word. Then, the transmitter executes an unequal error\\nprotection transmission strategy based on the importance order to make the\\ntransmission of important words in the message more reliable. The experimental\\nresults show that the error rate and semantic loss of important words measured\\nin the communication system embedded with ChatGPT are much lower than\\nthose of existing communication schemes, indicating that ChatGPT can protect'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 7, 'page_label': '8'}, page_content='important words well and make semantic communication more reliable.\\nWang et al. [17] studied the effectiveness of ChatGPT in generating high-\\nquality Boolean queries for systematic literature search. They designed a wide\\nrange of prompts and investigated these tasks on more than 100 systematic\\nreview topics. In the end, queries generated by ChatGPT achieved higher accu-\\nracy compared to the currently most advanced query generation methods but'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 7, 'page_label': '8'}, page_content='at the cost of reduced recall. For time-limited rapid reviews, it is often accept-\\nable to trade off higher precision for lower recall. Additionally, ChatGPT can\\ngenerate high search accuracy Boolean queries by guiding the prompts. How-\\never, it should be noted that when two queries use the same prompts, ChatGPT\\ngenerates different queries, indicating its limitations in consistency and stabil-\\nity. Overall, this study demonstrated the potential of ChatGPT in generating'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 7, 'page_label': '8'}, page_content='effective Boolean queries for systematic literature searches.\\n2.1.2 Text Classification\\nThe purpose of text classification is to assign text data to predefined categories.\\nThis task is critical for many applications, including sentiment analysis, spam\\ndetection, and topic modeling. While traditional machine learning algorithms\\nhave been widely used for text classification, recent advances in natural lan-\\nguage processing have led to the development of more advanced techniques.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 7, 'page_label': '8'}, page_content='ChatGPT has shown immense potential in this field. Its ability to accurately\\nclassify text, flexibility in handling various classification tasks, and potential for\\ncustomization make it a valuable tool for text classification, as evidenced by\\nseveral studies in the literature.\\n8'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 8, 'page_label': '9'}, page_content='Kuzman et al. [18] employed ChatGPT for automatic genre recognition, with\\nthe goal of simplifying the text classification task by utilizing ChatGPT’s zero-\\nshot classification capability. They compared ChatGPT’s genre recognition per-\\nformance, using two prompt languages (EN and SL), with the X-GENRE clas-\\nsifier based on the multilingual model XLM-RoBERTa on the English dataset\\nEN-GINCO and the Slovenian dataset GINCO. The results showed that when'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 8, 'page_label': '9'}, page_content='EN was used as the prompt language, ChatGPT achieved Micro F1, Macro F1,\\nand Accuracy scores of 0.74, 0.66, and 0.72. However, on the GINCO dataset,\\nChatGPT’s genre recognition performance with both EN and SL prompt lan-\\nguages was lower than that of the X-GENRE classifier to varying degrees.\\nAmin et al. [19] evaluated the text classification ability of ChatGPT in\\naffective computing by using it to perform personality prediction, sentiment'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 8, 'page_label': '9'}, page_content='analysis, and suicide ideation detection tasks. They prompted ChatGPT with\\ncorresponding prompts on three datasets: First Impressions, Sentiment140, and\\nSuicide and Depression, and compared its classification performance with three\\nbaseline models: RoBERTa-base, Word2Vec, and BoW. The results showed\\nthat ChatGPT’s accuracy and UAR for the five personality classifications on\\nthe First Impressions dataset were lower than the baseline methods to varying'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 8, 'page_label': '9'}, page_content='degrees. On the Sentiment140 dataset, ChatGPT’s accuracy and UAR were 85.5\\nand 85.5, respectively, which were better than the three baseline methods. On\\nthe Suicide and Depression dataset, ChatGPT’s accuracy and UAR were 92.7\\nand 91.2, respectively, which were lower than RoBERTa, the best-performing\\nbaseline method.\\nZhang et al. [20] employed ChatGPT for stance detection, which includes\\nsupport and opposition. They used ChatGPT to classify the political stance'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 8, 'page_label': '9'}, page_content='of tweets in the SemEval-2016 and P-Stance datasets. SemEval-2016 contains\\n4,870 English tweets, and they selected tweets with the most commonly occur-\\nring FM, LA, and HC political labels for stance classification. The P-Stance\\ndataset has 21,574 English tweets, and they classified the stance of tweets to-\\nwards Trump, Biden, and Bernie. The final results showed that on the SemEval-\\n2016 dataset, ChatGPT achieved F1-m scores of 68.4, 58.2, and 79.5 for the FM,'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 8, 'page_label': '9'}, page_content='LA, and HC political labels, and F1-avg scores of 72.6, 59.3, and 78.0, respec-\\ntively. On the P-Stance dataset, ChatGPT achieved F1-m scores of 82.8, 82.3,\\nand 79.4 for the Trump, Biden, and Bernie political figures, and F1-avg scores\\nof 83.2, 82.0, and 79.4, respectively.\\nHuang et al. [21] used ChatGPT to detect implicit hate speech in tweets.\\nThey selected 12.5% (795 tweets) of the LatentHatred dataset containing im-\\nplicit hate speech and asked ChatGPT to classify them into three categories:'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 8, 'page_label': '9'}, page_content='implicit hate speech, non-hate speech, and uncertain. The results showed that\\nChatGPT correctly recognized 636 (80%) of the tweets. The number of tweets\\nclassified as non-hate speech and uncertain were 146 (18.4%) and 13 (1.6%),\\nrespectively. The results of the reclassification of tweets in the non-hate speech\\nand uncertain categories by Amazon Mechanical Turk (Mturk) workers were\\nconsistent with ChatGPT’s classification.\\nOverall, ChatGPT has tremendous potential in text classification tasks, as it'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 8, 'page_label': '9'}, page_content='can effectively address problems such as genre identification, sentiment analysis,\\n9'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 9, 'page_label': '10'}, page_content='stance detection, and more. However, there are still challenges that ChatGPT\\nfaces in the field of text classification. Firstly, it struggles to perform well in\\nclassification tasks with rare or out-of-vocabulary words since it heavily relies\\non the distribution of training data. Additionally, the significant computational\\nresources required for training and utilizing ChatGPT can limit its use in some\\napplications.\\n2.1.3 Text generation'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 9, 'page_label': '10'}, page_content='applications.\\n2.1.3 Text generation\\nWe live in an era of information explosion, and text is an efficient way of trans-\\nmitting information. The diversity of information has led to a diversity of text\\ncategories. When researchers use ChatGPT’s text generation capabilities for\\nresearch, they inevitably choose to generate different types of text. In the pro-\\ncess of reading papers, we found that the word count of the text generated by'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 9, 'page_label': '10'}, page_content='researchers increased from small to large, so we wanted to summarize existing\\nresearch based on the size of the text word count. We divided the generated\\ntext into three levels: phrases, sentences, and paragraphs.\\nThe following article uses ChatGPT to generate phrases. Zhang et al. [22]\\nproves that the semantic HAR model with semantic augmentation added dur-\\ning training performs better in motion recognition than other models. Semantic'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 9, 'page_label': '10'}, page_content='augmentation requires shared tokens, which is lacking in some datasets. There-\\nfore, authors leverage ChatGPT for an automated label generation approach\\nfor datasets originally without shared tokens. Fu et al. [23] describes a new\\nworkflow for converting natural language commands into Bash commands. The\\nauthor uses ChatGPT to generate a candidate list of Bash commands based\\non user input, and then uses a combination of heuristic and machine learning'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 9, 'page_label': '10'}, page_content='techniques to rank and select the most likely candidates. This workflow was\\nevaluated on a real command dataset and achieved high accuracy compared\\nto other state-of-the-art methods. Chen et al. [24] used the Bart model and\\nChatGPT for the task of summarizing humorous titles and compared the perfor-\\nmance of the two models. It was found that the Bart model performed better on\\nlarge datasets, but ChatGPT was competitive with our best fine-tuned model\\nin a small range (48), albeit slightly weaker.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 9, 'page_label': '10'}, page_content='in a small range (48), albeit slightly weaker.\\nThe following article uses ChatGPT to generate sentences.Chen et al. [25]\\nconstructed a dialogue dataset (HPD) with scenes, timelines, character at-\\ntributes, and character relationships in order to use ChatGPT as a conver-\\nsational agent to generate dialogue. However, ChatGPT’s performance on the\\ntest set was poor, and there is room for improvement.In study [26], chatGPT\\ndemonstrated its ability to simplify complex text by providing three fictional'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 9, 'page_label': '10'}, page_content='radiology reports to chatGPT for simplification. Most radiologists found the\\nsimplified reports to be accurate and complete, with no potential harm to pa-\\ntients. However, some errors, omissions of critical medical information and text\\npassages were identified, which could potentially lead to harmful conclusions if\\nnot understood by the physicians. Xia et al. [27] proposes a new program re-\\npair paradigm called Session-based Automatic Program Repair (APR). In APR,'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 9, 'page_label': '10'}, page_content='the previously generated patches are iteratively built upon by combining them\\n10'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 10, 'page_label': '11'}, page_content='with validation feedback to construct the model’s input. The effectiveness of\\nthe approach is verified using the QuixBugs dataset. The experiment shows\\nthat ChatGPT fine-tuned with reinforcement learning from human feedback\\n(RLHF) outperforms Codex trained unsupervisedly in both repair datasets. In\\nreference to study [28], ChatGPT was compared to three commercial transla-\\ntion products: Google Translate2, DeepL Translate3, and Tencent TranSmart4.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 10, 'page_label': '11'}, page_content='The evaluation was conducted on the Flores101 test set, using the WMT19\\nbiomedical translation task to test translation robustness, with BLEU score as\\nthe main metric. The study found that ChatGPT is competitive with commer-\\ncial translation products on high-resource European languages but falls behind\\non low-resource or distant languages. The authors explored an interesting strat-\\negy called pivot prompts, which significantly improved translation performance.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 10, 'page_label': '11'}, page_content='While ChatGPT did not perform as well as commercial systems on biomedical\\nabstracts or Reddit comments, it may be a good speech translator. Prieto et\\nal. [29] evaluated the use of ChatGPT in developing an automated construction\\nschedule based on natural language prompts. The experiment required building\\nnew partitions in an existing space and providing details on the rooms to be\\npartitioned. The results showed that ChatGPT was able to generate a coher-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 10, 'page_label': '11'}, page_content='ent schedule that followed a logical approach to meet the requirements of the\\ngiven scope. However, there were still several major flaws that would limit the\\nuse of this tool in real-world projects.Michail et al. [30] proposed a method\\nto improve the prediction accuracy of the HeFit fine-tuned XLM T model on\\ntweet intimacy by generating a dataset of tweets with intimacy rating tags using\\nChatGPT. The specific operation is to input tweets with intimacy rating tags'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 10, 'page_label': '11'}, page_content='into ChatGPT and then output similar tweets.\\nThe following article uses ChatGPT to generate paragraphs. Wang et al.\\n[31] compared the abstract summarization performance of ChatGPT and other\\nmodels on various cross-lingual text datasets and found that ChatGPT may\\nperform worse in metrics such as R 1, R 2, R L, and B S. Yang et al. [32]\\nsummarized the performance of ChatGPT in question answering-based text\\nsummarization and found that, compared to fine-tuned models, ChatGPT’s'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 10, 'page_label': '11'}, page_content='performance is slightly worse in all performance metrics. However, the article\\nsuggests that if the dataset is golden annotation, ChatGPT’s performance may\\nsurpass fine-tuned models in these metrics. Belouadi et al. [33] compared the\\nability of ByGPT5 and ChatGPT trained on a range of labeled and unlabeled\\ndatasets of English and German poetry to generate constrained style poetry, and\\nevaluated them using three metrics: Rhyme, ScoreAlliteration, and ScoreMe-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 10, 'page_label': '11'}, page_content='ter Score. The conclusion is that ByGPT5 performs better than ChatGPT.\\nBlanco-Gonzalez et al. [34] evaluated chatGPT’s ability to write commentary\\narticles, and in fact, this article itself was written by chatGPT. The human au-\\nthor rewrote the manuscript based on chatGPT’s draft. Experts found that it\\ncan quickly generate and optimize text, as well as help users complete multiple\\ntasks. However, in terms of generating new content, it is not ideal. Ultimately,'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 10, 'page_label': '11'}, page_content='it can be said that without strong human intervention, chatGPT is not a use-\\nful tool for writing reliable scientific texts. It lacks the knowledge and expertise\\nrequired to accurately and fully convey complex scientific concepts and informa-\\n11'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 11, 'page_label': '12'}, page_content='tion. Khalil et al. [35] on the originality of content generated by ChatGPT. To\\nevaluate the originality of 50 papers on various topics generated by ChatGPT,\\ntwo popular plagiarism detection tools, Turnitin and iThenticate, were used.\\nThe results showed that ChatGPT has great potential in generating complex\\ntext output that is not easily captured by plagiarism detection software. The\\nexisting plagiarism detection software should update their plagiarism detection'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 11, 'page_label': '12'}, page_content='engines. Basic et al. [36] conducted a comparison of the writing performance\\nof students using or not using ChatGPT-3 as a writing aid. The experiment\\nconsisted of two groups of 9 participants each. The control group wrote articles\\nusing traditional methods, while the experimental group used ChatGPT as an\\naid. Two teachers evaluated the papers. The study showed that the assistance of\\nChatGPT did not necessarily improve the quality of the students’ essays.Noever'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 11, 'page_label': '12'}, page_content='et al. [37] discusses the potential of using artificial intelligence (AI), particularly\\nlanguage models like GPT (including GPT-3), to create more convincing chat-\\nbots that can deceive humans into thinking they are interacting with another\\nperson. The article describes a series of experiments in which they used GPT-3\\nto generate chatbot responses that mimic human-like conversations and were\\ntested on human participants. The results show that some participants were'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 11, 'page_label': '12'}, page_content='unable to distinguish between the chatbot and a real human, highlighting the\\npotential for these AI chatbots to be used for deceptive purposes.\\n2.1.4 Code Generation\\nCode generation refers to the process of automatically generating computer code\\nfrom high-level descriptions or specifications. ChatGPT’s advanced natural lan-\\nguage processing capabilities make it capable of performing code generation\\ntasks. By analyzing the requirements for code generation, ChatGPT can pro-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 11, 'page_label': '12'}, page_content='duce code snippets that accurately execute the intended functionality. This not\\nonly saves time and effort in writing code from scratch but also reduces the risk\\nof errors that may occur during manual coding. In addition, ChatGPT’s ability\\nto learn and adapt to new programming languages and frameworks enables it\\nto complete more complex programming tasks. For example:\\nMegahed et al. [38] discussed the potential of using ChatGPT for tasks'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 11, 'page_label': '12'}, page_content='such as code explanation, suggesting alternative methods for problem-solving\\nwith code, and translating code between programming languages. The solutions\\nprovided by ChatGPT were found to be viable. In another study, Treude et\\nal. [39] introduced a ChatGPT-based prototype called GPTCOMCARE, which\\nhelps programmers generate multiple solutions for a programming problem and\\nhighlight the differences between each solution using colors.Sobania et al. [40]'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 11, 'page_label': '12'}, page_content='utilized ChatGPT for code bug fixing, and further improved the success rate of\\nbug fixing by inputting more information through its dialogue system. Specifi-\\ncally, the QuixBugs standard bug fixing benchmark contained 40 code bugs that\\nneeded to be fixed. With limited information, ChatGPT fixed 19 bugs, which\\nwas slightly lower than the 21 bugs fixed by the Codex model, but significantly\\nhigher than the 7 fixed by the Standard APR model. When given more prompts'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 11, 'page_label': '12'}, page_content='and information, ChatGPT was able to fix 31 bugs, demonstrating its potential\\n12'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 12, 'page_label': '13'}, page_content='for code bug fixing tasks.Xia et al. [27] proposed a conversational approach\\nfor Automated Program Repair (APR), which alternates between generating\\npatches and validating them against feedback from test cases until the correct\\npatch is generated. Selecting 30 bugs from the QuixBugs standard bug fixing\\nbenchmark, which are suitable for test case feedback, and demonstrating them\\nwith Java and Python, the QuixBugs-Python and QuixBugs-Java datasets were'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 12, 'page_label': '13'}, page_content='obtained. The conversational APR using ChatGPT outperformed the conversa-\\ntional APR using Codex and the conversational APR using CODEGEN (with\\nmodel parameters of 350M, 2B, 6B, and 16B) on both datasets. Furthermore,\\nChatGPT’s conversational APR generated and validated patches with signifi-\\ncantly fewer feedback loops than the other models.\\nChatGPT can not only be used to achieve some simple code generation tasks\\nbut also can be used to accomplish some complex programming tasks. Noever'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 12, 'page_label': '13'}, page_content='et al. [41] tested ChatGPT’s code generation capabilities on four datasets -\\nIris, Titanic, Boston Housing, and Faker. When prompted to mimic a Python\\ninterpreter in the form of a Jupyter notebook, the model was able to generate\\nindependent code based on the prompt and respond with the expected out-\\nput. For example, when given the prompt ”data.cor()” for the Iris dataset,\\nChatGPT generated correct Python output. The test results indicate that'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 12, 'page_label': '13'}, page_content='ChatGPT can access structured datasets and perform basic software opera-\\ntions required by databases, such as create, read, update, and delete (CRUD).\\nThis suggests that cutting-edge language models like ChatGPT have the nec-\\nessary scale to tackle complex problems. McKee et al. [42] utilized ChatGPT\\nas an experimental platform to investigate cybersecurity issues. They modeled\\nfive different modes of computer virus properties, including self-replication, self-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 12, 'page_label': '13'}, page_content='modification, execution, evasion, and application, using ChatGPT. These five\\nmodes encompassed thirteen encoding tasks from credential access to defense\\nevasion within the MITRE ATT&CK framework. The results showed that the\\nquality of ChatGPT’s generated code was generally above average, except for the\\nself-replication mode, where it performed poorly.They [43] also employed Chat-\\nGPT as a network honeypot to defend against attackers. By having ChatGPT'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 12, 'page_label': '13'}, page_content='mimic Linux, Mac, and Windows terminal commands and providing interfaces\\nfor TeamViewer, nmap, and ping, a dynamic environment can be created to\\nadapt to attackers’ operations, and logs can be used to gain insight into their\\nattack methods, tactics, and procedures. The authors demonstrated ten hon-\\neypot tasks to illustrate that ChatGPT’s interface not only provides sufficient\\nAPI memory to execute previous commands without defaulting to repetitive'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 12, 'page_label': '13'}, page_content='introductory tasks but also offers a responsive welcome program that maintains\\nattackers’ interest in multiple queries.\\nIn the field of code generation, there are still several challenges with Chat-\\nGPT. Firstly, its application scope is limited as its training data is biased to-\\nwards programming languages such as Python, C++, and Java, making it po-\\ntentially unsuitable for some programming languages or coding styles. Secondly,'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 12, 'page_label': '13'}, page_content='manual optimization is necessary for code formatting, as the generated code may\\nnot be performance-optimized or follow best coding practices, requiring manual\\nediting and optimization. Lastly, the quality of the generated code cannot be\\n13'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 13, 'page_label': '14'}, page_content='guaranteed, as it heavily relies on the quality of the natural language input,\\nwhich may contain errors, ambiguities, or inconsistencies, ultimately affecting\\nthe accuracy and reliability of the generated code.\\n2.1.5 Inference\\nInference refers to the process of drawing new conclusions or information through\\nlogical deduction from known facts or information. It is typically based on a se-\\nries of premises or assumptions, and involves applying logical rules or reasoning'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 13, 'page_label': '14'}, page_content='methods to arrive at a conclusion. Inference is an important ability in human\\nthinking, and is often used to solve problems, make decisions, analyze and eval-\\nuate information, etc. Inference also plays a key role in fields such as science,\\nphilosophy, law, etc. There are two types of inference: inductive reasoning,\\nwhich involves deriving general rules or conclusions from known facts or expe-\\nriences, and deductive reasoning, which involves deriving specific conclusions'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 13, 'page_label': '14'}, page_content='from known premises or assumptions. Whether inductive or deductive, the pro-\\ncess of inference requires following strict logical rules to ensure the correctness\\nand reliability of the inference.\\nSome papers attempt to use ChatGPT’s ability in inductive reasoning to\\ncapture the meaning in text and use defined metrics to score the text. Michail\\net al. [30] uses ChatGPT to infer intimacy expressed in tweets. They first input\\n50 tweets with intimacy markers to ChatGPT, then use inductive reasoning to'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 13, 'page_label': '14'}, page_content='infer the standards for generating tweets with different levels of intimacy, and\\nfinally generate ten tweets with intimacy values ranging from 0 to 5. Susnjak\\net al. [44] collected a large amount of textual data from patient-doctor dis-\\ncussion forums, patient testimonials, social media platforms, medical journals,\\nand other scientific research publications. Using the BERT model, the author\\ninferred emotion values from 0 to 1. The author visualized the process of how'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 13, 'page_label': '14'}, page_content='the presence of bias in the discourse surrounding chronic manifestations of the\\ndisease using the SHAP tool. The author also envisioned ChatGPT as a replace-\\nment for the BERT model for scoring the emotional value of text. Huang et al.\\n[21] chose 12.5% of individuals in the potential hate dataset as study materials,\\ninduced ChatGPT to make classifications based on a prompt, and ChatGPT\\nproduced three classifications: unclear, yes, and no. The author assigned a'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 13, 'page_label': '14'}, page_content='value of 1 to yes, -1 to no, and 0 to unclear, and had ChatGPT score and clas-\\nsify them. ChatGPT was able to correctly classify 80% of implicit hate tweets\\nin the author’s experimental setup, demonstrating ChatGPT’s great potential\\nas a data labeling tool using simple prompts.\\nSome papers have evaluated ChatGPT’s reasoning performance, mainly in\\ndecision-making and spatial reasoning, and identifying ambiguity. Tang et al.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 13, 'page_label': '14'}, page_content='[45] used the independence axiom and the transitivity axiom, as well as other\\nnon-VNM related decision-making abilities, by presenting bets conditioned on\\nrandom events, bets with asymmetric outcomes, decisions encapsulating Sav-\\nage’s Sure Thing principle, and other complex bet structures like nested bets, to\\ndesign experiments where each experiment input a short prompt to ChatGPT\\nand evaluated the results. The conclusion is that ChatGPT exhibits uncer-\\n14'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 14, 'page_label': '15'}, page_content='tainty in the decision-making process: in some cases, large language models\\ncan arrive at the correct answer through incorrect reasoning; and it may make\\nsuboptimal decisions for simple reasoning problems. Ortega-Martn et al. [46]\\nhad ChatGPT detect three different levels of language ambiguity and evaluated\\nits performance. The conclusion is that In semantics, ChatGPT performed per-\\nfectly in the detection of ambiguities. Apart from that, it has some bright sports'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 14, 'page_label': '15'}, page_content='(co-reference resolution) and some weaknesses (puts gender bias over grammar\\nin some non-ambiguous situations). In the generation task ChatGPT did well,\\nbut also revealed some of its worse issues: the lack of systematicity. Lastly, it\\nshould also be pointed that in most of the cases ChatGPT brilliantly alludes to\\nlack of context as the key factor in disambiguation.\\n2.1.6 Data or information extraction, transformation, enhancement,\\nprocessing\\nData Visualization'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 14, 'page_label': '15'}, page_content='processing\\nData Visualization\\nNatural language interfaces have contributed to generating visualizations\\ndirectly from natural language, but visualization problems remain challenging\\ndue to the ambiguity of natural language.ChatGPT provides a new avenue for\\nthe field by converting natural language into visualized code.\\nIn terms of data visualization, Noever et al. [41] tested ChatGPT’s ba-\\nsic arithmetic skills by asking questions.On the iris dataset, Titanic survival'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 14, 'page_label': '15'}, page_content='dataset, Boston housing data, and randomly generated insurance claims dataset,\\nthe statistical analysis of data and visualization problems were converted to\\nprogramming problems using Jupyter to verify ChatGPT’s ability to generate\\npython code to draw suitable graphs and analyze the data. The results show\\nthat ChatGPT can access structured and organized datasets to perform the\\nfour basic software operations required for databases: create, read, update, and'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 14, 'page_label': '15'}, page_content='delete, and generate suitable python code to plot graphs for descriptive statis-\\ntics, variable correlation analysis, describing trends, and other data analysis\\noperations.Maddigan et al. [47] proposed an end-to-end solution for visualizing\\ndata in natural language using LLM, which uses an open-source python frame-\\nwork designed to generate appropriate hints for selected datasets to make LLM\\nmore effective in understanding natural language, and uses internal reasoning'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 14, 'page_label': '15'}, page_content='capabilities to select the appropriate visualization type to generate the code for\\nvisualization. In this paper,the reseachers compare the visualization results of\\nGPT-3, Codex and ChatGPT in the case of nvBench SQLite database [48] and\\nthe visualization results of energy production dataset in the study of ADVISor\\nwith NL4DV [49, 50].In addition to, they explore the ability to reason and hy-\\npothesize of the LLM on movie dataset [48] when the hints are insufficient or'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 14, 'page_label': '15'}, page_content='wrong .Experimental results show that LLM can effectively support the end-to-\\nend generation of visualization results from natural language when supported\\nby hints, providing an efficient, reliable and accurate solution to the natural\\nlanguage visualization problem.\\n15'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 15, 'page_label': '16'}, page_content='Information Extraction\\nThe goal of information extraction is to extract specific information from\\nnatural language text for structured representation, including three important\\nsubtasks such as entity relationship extraction, named entity recognition, and\\nevent extraction, which have wide applications in business, medical, and other\\nfields.\\nIn information extraction, Wei et al. [51] proposed ChatIE, a ChatGPT-\\nbased multi-round question-and-answer framework for information extraction.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 15, 'page_label': '16'}, page_content='The framework decomposes a complex information extraction (IE) task into\\nseveral parts, then combines the results of each round into a final structured\\nresult. The entity association triple extraction, named entity recognition, and\\nevent extraction tasks were performed on six datasets NYT11-HRL, DuIE2.0\\n, conllpp, MSR , DuEE1.0 [52, 53, 54, 55, 56], and ACE05 in both languages,\\ncomparing three metrics of precision, recall, and F1 score.These results sug-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 15, 'page_label': '16'}, page_content='gest that on six widely used IE datasets, ChatIE improves performance by an\\naverage of 18.98% compared to the original ChatGPT without ChatIE, and out-\\nperforms the supervised models FCM and MultiR [57, 58] on the NYT11-HRL\\ndataset.While the original ChatGPT cannot solve complex IE problems with\\noriginal task instructions, and with this framework, successfully IE tasks were\\nimplemented on six datasets.Gao et al. [59] explored the feasibility and chal-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 15, 'page_label': '16'}, page_content='lenges of ChatGPT for event extraction on the ACE2005 corpus, evaluating the\\nperformance of ChatGPT in long-tail and complex scenarios (texts containing\\nmultiple events) and comparing it with two task-specific models, Text2Event\\nand EEQA [60, 61].Then,they explored the impact of different cues on perfor-\\nmance of ChatGPT. The results show that the average performance of Chat-\\nGPT in long-tail and complex scenarios is only 51.04% of that of task-specific'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 15, 'page_label': '16'}, page_content='models such as EEQA. Continuous refinement of cues does not lead to consis-\\ntent performance improvements, and ChatGPT is highly sensitive to different\\ncue styles.Tang et al. [62] proposed a new training paradigm that incorporates\\nappropriate cues to guide ChatGPT to generate a variety of examples with dif-\\nferent sentence structures and language patterns and eliminate the resulting\\nlow-quality or duplicate samples for downstream tasks. Although compared to'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 15, 'page_label': '16'}, page_content='a soft model for a specific healthcare tasks, ChatGPT underperforms in Named\\nEntity Recognition (NER) and Relationship Extraction (RE) tasks , in the Gene\\nAssociation Database (GAD) Release; EU-ADR corpus for the RE task , the\\ninnovative training framework was able to train local models, with F1 scores im-\\nproving from 23.37% to 63.99% for the named entity recognition task and from\\n75%, while alleviating privacy concerns and time-consuming data collection and'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 15, 'page_label': '16'}, page_content='annotation problems.He et al. [63] proposed a contextual learning framework\\nICL- D3IE. this framework introduces formatted presentation, continuously it-\\nerates to update and improve the presentation, and then combines ChatGPT\\nfor text information extraction. In the paper, ICL-D3IE is compared with ex-\\nisting pre-trained models such as LiLT,BROS (in-distribution (ID) setting and\\nout-of-distribution (OOD) setting) on datasets (FUNSD, CORD, and SROIE'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 15, 'page_label': '16'}, page_content='[64, 65, 66]).These results show that the ICL-D3IE method in all datasets and\\nsettings except for the ID setting on CORD are superior to other methods,\\n16'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 16, 'page_label': '17'}, page_content='with ICL-D3IE (GPT-3) F1 scores reaching 90.32% on FUNSD and97.88% on\\nSROIE; in the out-of-distribution (OOD) setting, ICL-D3IE performs much\\nbetter than previous pre-trained methods on all datasets.Polak et al. [67] pro-\\nposed ChatExtract method - consisting of a set of engineering prompts applied\\nto a conversational LLM - for automatic data extraction. In the experiment,\\nthey extracted a large number of sentences from hundreds of papers and ran-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 16, 'page_label': '17'}, page_content='domly selected 100 sentences containing data and 100 sentences without data\\nas test data. The results show that the accuracy and recall of LLM exceeded\\n90% and may be comparable to human accuracy in many cases; in addition to\\nthis, the experiments were conducted under the condition of removing follow-up\\nprompts and not keeping the conversation compared to previous experiments,\\nrespectively. The accuracy of deleting follow-up questions dropped to 80.2%'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 16, 'page_label': '17'}, page_content='and the recall rate dropped to 88.0%. Removing the conversational aspect and\\nrelated information retention recall and accuracy dropped to 90.0% and 56.6%,\\nrespectively, demonstrating the effect of information retention combined with\\npurposeful redundancy on LLM information extraction performance.\\nQuality Assessment\\nFor translation quality, text generation quality, manual assessment is usually\\neffective but suffers from subjectivity and time-consuming, etc. It was found'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 16, 'page_label': '17'}, page_content='through exploration that ChatGPT has also achieved significant performance\\nin automatic quality assessment.\\nIn terms of quality assessment,Kocmi et al. [68] proposed a GPT-based\\ntranslation quality assessment metric, GEMBA, which evaluates the transla-\\ntion of each fragment individually and then averages all the obtained scores to\\nobtain a final system-level score. In the MQM2022 test set (English-German,\\nEnglish-Russian, and Chinese-English) [69], a scoring task was performed with'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 16, 'page_label': '17'}, page_content='a classification task to compare the accuracy [70] and kendall tau scores [71] of\\nseven GPT models under four cue templates.The results showed that GEMBA\\nhad the highest system-level accuracy of 88.0% compared to more than 10 au-\\ntomatic metrics such as BLEU, and among the seven GPT models, ChatGPT\\naccuracy is above 80%, in addition to, the best performance can be obtained in\\nthe least constrained template, demonstrating the potential of LLM for trans-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 16, 'page_label': '17'}, page_content='lation quality assessment tasks, but the evaluation is only applicable at the\\nsystem level and needs further improvement.Wang et al. [72] used ChatGPT\\nas a natural language generation (NLG) evaluator to study the correlation with\\nhuman judgment. On three datasets covering different NLG tasks, task- and\\naspect-specific cues were designed to guide ChatGPT for NLG evaluation in\\nCNN/DM [73], OpenMEVA-ROC, and BAGEL for summary, story generation,'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 16, 'page_label': '17'}, page_content='and data-to-text scoring, respectively. Then,they compute Spearman coeffi-\\ncients [74],Pearson correlation coefficients [75]. Kendall’s Tau score [76] to as-\\nsess the correlation with human evaluations.The results show that ChatGPT is\\nhighly correlated with human judgments in all aspects, with correlation coeffi-\\ncients of 0.4 or more in all categories, showing its potential as an NLG indicator.\\n17'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 17, 'page_label': '18'}, page_content='Data Augmentation\\nIn natural language processing, text data augmentation is an effective mea-\\nsure to alleviate the problem of low data quantity and low quality training data,\\nand ChatGPT has shown great potential in this regard.\\nIn terms of data augmentation, Dai et al. [77] proposed a ChatGPT-based\\ntext data augmentation method that reformulates each sentence in the train-\\ning sample into multiple conceptually similar but semantically different samples'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 17, 'page_label': '18'}, page_content='for classification tasks downstream of the Bert model.On text transcriptions\\nand PubMed 20k datasets containing more than 8 hours of audio data of com-\\nmon medical symptom descriptions,experiments were conducted to compare co-\\nsine similarity and TransRate metrics with multiple data enhancement methods\\n[9].This paper shows that compared with existing data enhancement methods,\\nthe proposed ChatAug method shows a double-digit improvement in sentence'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 17, 'page_label': '18'}, page_content='classification accuracy and generates more diverse augmented samples while\\nmaintaining its accuracy, but the original model is not fine-tuned in the pa-\\nper and suffers from a lack of domain knowledge, which may produce incorrect\\naugmented data.\\nMultimodal fusion\\nChatGPT can currently only process natural language directly, but with a\\ncross-modal encoder, it can combine natural language with cross-modal pro-\\ncessing to provide solutions for intelligent transportation, healthcare, and other\\nfields.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 17, 'page_label': '18'}, page_content='fields.\\nIn terms of multimodal data processing, Wu et al. [78] constructed a frame-\\nwork that Visual ChatGPT integrates with different Visual Foundation Models\\n(VFMs) and then combines a series of hints to input visual information to Chat-\\nGPT to solve visual problems.The paper shows examples of visual tasks such\\nas removing or replacing certain objects from images, interconversion between\\nimages and text, demonstrating the Visual ChatGPT has great potential and'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 17, 'page_label': '18'}, page_content='capability for different tasks.But there are issues during the task that require\\na large number of hints to convert VFMs to language, invoke multiple VFMs\\nto solve complex problems leading to limited real-time capability, and security\\nand privacy issues. Zheng et al. [79] showed a text mining example of LLM\\nfor extracting self-driving car crash data from California crash news, analyz-\\ning a failure report example, and generating a crash report example based on'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 17, 'page_label': '18'}, page_content='keywords; introduced a use case concept of a smartphone-based framework for\\nautomatic LLM failure report generation, which absorbs multiple data sources\\ncaptured by cell phone sensors and then transfers the data to a language space\\nfor text mining, inference and generation, and further outputs the key informa-\\ntion needed to form a comprehensive fault report, demonstrating the potential\\nof LLM for a variety of transportation tasks.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 17, 'page_label': '18'}, page_content='of LLM for a variety of transportation tasks.\\nNowadays, ChatGPT shows a wide range of applications in data visualiza-\\ntion, information extraction, data enhancement, quality assessment, and multi-\\nmodal data processing.But there are also issues on how to further utilize hints\\nto effectively interact with ChatGPT, lack of ability to process and analyze data\\n18'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 18, 'page_label': '19'}, page_content='from devices such as sensors, and data privacy and security.\\nCueing Techniques\\nCue engineering provides important support for effective dialogue with large\\nlanguage models.White et al. [80] proposed a framework for cueing models\\napplicable to different domains. This framework structures cues to interact\\nwith LLMs by providing specific rules and guidelines. Also, this paper presents\\na catalog of cueing patterns that have been applied to LLM interactions, as well'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 18, 'page_label': '19'}, page_content='as specific examples with and without cues. The advantages of the combinability\\nof prompting patterns are demonstrated, allowing users to interact with LLM\\nmore effectively, but patterns for reusable solutions and new ways to use LLM\\nneed to be continuously explored.\\n2.1.7 Human-ChatGPT Collaboration\\nCollaboration between humans and machines is a process where humans and\\nmachines work together to achieve a common goal. In such collaboration, hu-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 18, 'page_label': '19'}, page_content='mans provide domain expertise, creativity, and decision-making abilities, while\\nmachines provide automation, scalability, and computing power. ChatGPT is\\nan advanced natural language processing model that can understand and gen-\\nerate human-like language, thereby reducing communication costs. Its ability\\nto process and generate natural language makes it an ideal partner for human\\ncollaboration. ChatGPT can offer relevant suggestions, complete tasks based'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 18, 'page_label': '19'}, page_content='on human input, and enhance human productivity and creativity. It can learn\\nfrom human feedback and adapt to new tasks and domains, further improv-\\ning its performance in human-machine collaboration. ChatGPT’s capability to\\ncomprehend natural language and produce appropriate responses makes it a\\nvaluable tool for various collaboration applications, as demonstrated by several\\nstudies in the literature we have gathered.\\nAhmad et al. [81] proposed a method for human-machine collaboration us-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 18, 'page_label': '19'}, page_content='ing ChatGPT to create software architecture. This method transforms software\\nstories (created by software architects based on application scenarios) into feasi-\\nble software architecture diagrams through continuous interaction between the\\nsoftware architect and ChatGPT. During the evaluation stage, ChatGPT uses\\nthe Software Architecture Analysis Method (SAAM) to evaluate each compo-\\nnent in the software architecture and generate evaluation reports. This method'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 18, 'page_label': '19'}, page_content='efficiently utilizes the knowledge and supervision of the architect with the ca-\\npabilities of ChatGPT to collaboratively build software-intensive systems and\\nservices. Lanzi et al. [82] proposed a collaborative design framework that com-\\nbines interactive evolution and ChatGPT to simulate typical human design pro-\\ncesses. Humans collaborate with large language models (such as ChatGPT) to\\nrecombine and transform ideas, and use genetic algorithms to iterate through'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 18, 'page_label': '19'}, page_content='complex creative tasks. The results of three game design tasks showed that the\\nframework received positive feedback from game designers. The framework has\\ngood reusability and can be applied to any design task that can be described in\\nfree text form.\\n19'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 19, 'page_label': '20'}, page_content='In the future, ChatGPT’s ability to understand nonverbal cues such as tone\\nof voice and body language can be enhanced, enabling it to better understand\\nhuman thoughts and interact with people more effectively.\\n2.1.8 ChatGPT Integration\\nIntegration refers to combining different systems or software components to\\nachieve a common goal. ChatGPT can be integrated as a part of a whole or\\nact as an integration tool to enable seamless communication between different'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 19, 'page_label': '20'}, page_content='systems. Its natural language processing ability makes it easier for non-technical\\nusers to interact with systems, reducing the need for specialized knowledge or\\ntraining. Some studies in the literature we collected have already demonstrated\\nthis.\\nTreude et al. [39] integrated ChatGPT into the prototype of ”GPTCOM-\\nCARE” to address programming query problems. This integration allowed for\\nthe generation of multiple source code solutions for the same query, which in-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 19, 'page_label': '20'}, page_content='creased the efficiency of software development. The results of their study demon-\\nstrated the effectiveness of using ChatGPT to improve the quality and diversity\\nof code solutions, ultimately reducing the amount of time and effort required for\\nsoftware development.Wang et al. [83] proposed the chatCAD method, which\\nutilizes large language models (LLMs) such as ChatGPT to enhance the out-\\nput of multiple CAD networks for medical images, including diagnosis, lesion'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 19, 'page_label': '20'}, page_content='segmentation, and report generation networks. The method generates sugges-\\ntions in the form of a chat dialogue. The authors tested the effectiveness of\\nthe method on a randomly selected set of 300 cases from the MIMIC-CXR\\ndataset, which included 50 cases each of cardiomegaly, edema, consolidation,\\natelectasis, pleural effusion, and no findings. Compared to CvT2DistilGPT2\\nand R2GenCMN, chatCAD showed significant advantages in RC and F1, while\\nonly performing weaker than R2GenCMN in PR.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 19, 'page_label': '20'}, page_content='only performing weaker than R2GenCMN in PR.\\nIntegrating ChatGPT into applications will still present challenges. Firstly,\\nChatGPT’s performance may be affected by language barriers or differences\\nin terminology between different systems. Additionally, ChatGPT’s responses\\nare not always deterministic, which poses a challenge when integrating with\\nsystems that require precise and reproducible results. Finally, the processing'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 19, 'page_label': '20'}, page_content='time of ChatGPT is slow for integration tasks involving time-sensitive data such\\nas traffic, which is a limitation in time-critical environments.\\n2.2 AI Ethics\\nSince the advent of ChatGPT, this powerful natural language processing model\\nhas not only brought great convenience to people but also triggered more crisis-\\naware thinking. Some researchers have started to hypothesize and study the\\npotential negative impacts of ChatGPT. This proactive research provides good'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 19, 'page_label': '20'}, page_content='proposals for standardized construction to address future AI abuse issues.\\nRegarding the evaluation of ChatGPT’s own political and ethical tendencies,\\nHartmann et al. [84] used Wahl-O-Mat, one of the most commonly used voting\\n20'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 20, 'page_label': '21'}, page_content='advice applications in the world, to show ChatGPT political statements from\\ndifferent parties, forcing it to make choices of agree, disagree, or neutral. The\\nresults indicated that ChatGPT has a pro-environment, left-wing liberal ide-\\nology, which was also confirmed in the nation-state agnostic political compass\\ntest. Another study (referenced as [85]) examined ChatGPT’s moral standards\\nby repeatedly asking it different versions of the trolley problem, and found that'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 20, 'page_label': '21'}, page_content='ChatGPT gave answers with different moral orientations, lacking a firm moral\\nstance. A subsequent test also found that ChatGPT’s lack of consistency could\\naffect people’s moral judgments. Additionally, Borji et al. [86] demonstrated\\nChatGPT’s inconsistency in reasoning, factual errors, mathematics, coding, and\\nbias across eleven related aspects. These findings highlight ChatGPT’s inher-\\nent traits and limitations, and people should be aware of their potential impact'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 20, 'page_label': '21'}, page_content='when seeking advice from ChatGPT.\\nRegarding relevant policies and regulations, Hacker et al. [87] discussed the\\nnature and rules of large generative AI models, including ChatGPT, which are\\nrapidly changing the way we communicate, explain, and create. The author\\nsuggested that different stakeholders in the value chain should take regulatory\\nresponsibility and deploy four strategies to tailor more comprehensive laws for'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 20, 'page_label': '21'}, page_content='the benefit of society. Another study (referenced as [88]) criticized the European\\nCommission’s proposal on AI responsibility and suggested revising the proposed\\nAI responsibility framework to ensure effective compensation while promoting\\ninnovation, legal certainty, and sustainable AI regulation. A policy framework\\nwas proposed (referenced as [89]) to customize LLMs, such as ChatGPT, in a so-\\ncially acceptable and safe manner, emphasizing the need to align large language'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 20, 'page_label': '21'}, page_content='models (LLMs) with human preferences.\\nThe political and ethical tendencies of ChatGPT could influence users’ be-\\nhavior and decision-making to some extent. However, some studies have con-\\nducted in-depth research on the use of norms and limitations, which could enable\\nhumans to use ChatGPT more reasonably and safely.\\n2.3 Evaluation\\n2.3.1 Comparison of ChatGPT with existing popular models\\nWe use publicly available datasets to comprehensively evaluate the strengths'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 20, 'page_label': '21'}, page_content='and limitations of ChatGPT. Reference [90] evaluates the technical performance\\nof ChatGPT in multitask, multilingual, and multimodal aspects based on 23\\nstandard public datasets and newly designed multimodal datasets, including\\neight different common natural language processing application tasks. The ex-\\nperimental results show that, in terms of multitasking, ChatGPT outperforms\\nvarious state-of-the-art zero-shot learning large language models in most tasks,'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 20, 'page_label': '21'}, page_content='and even outperforms fine-tuned task-specific models in some individual tasks.\\nIn terms of multilingualism, we found that ChatGPT cannot be applied to low-\\nresource languages because it cannot understand the language and generate\\ntranslations for that language. In terms of multimodality, ChatGPT’s ability is\\nstill basic compared to specialized language-visual models.\\n21'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 21, 'page_label': '22'}, page_content='In terms of stability, reference [91] concludes that ChatGPT’s performance is\\nalways lower than SOTA, the current state-of-the-art model, in almost all tasks.\\nThis means that as a general model, ChatGPT has never reached the level of the\\nbest existing models. Experimental data shows that the average quality of the\\nSOTA model is 73.7%, while the average quality of the ChatGPT model is only\\n56.5%. At the same time, ChatGPT’s stability is poor: the standard deviation'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 21, 'page_label': '22'}, page_content='of its performance is 23.3%, while the SOTA model’s standard deviation is\\nonly 16.7%. This non-deterministic behavior exhibited by ChatGPT could be\\na serious drawback in some problems.\\nSimilarly, Qin et al. [92] conducted a comprehensive evaluation of whether\\nChatGPT is a qualified general natural language processing task solver. The ex-\\nperiment analyzed ChatGPT’s zero-shot learning ability based on 20 commonly\\nused public datasets covering 7 representative task categories. Below, we will'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 21, 'page_label': '22'}, page_content='analyze ChatGPT’s performance on each task:\\nIn terms of reasoning tasks, ChatGPT performs average on mathematical\\nsymbol, commonsense causal, and logical reasoning tasks, but performs well\\nin arithmetic reasoning [92]. That is to say, ChatGPT’s abilities vary among\\ndifferent types of reasoning tasks. In terms of logical reasoning, ChatGPT’s\\ndeductive and abductive reasoning are superior to inductive reasoning, while\\nin other reasoning tasks, such as analogy, causal and commonsense reasoning,'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 21, 'page_label': '22'}, page_content='ChatGPT performs well [90].\\nIn terms of sentiment analysis task, ChatGPT performs similarly to GPT-3.5\\nand bert-style models [92, 93]. However, according to literature [91], ChatGPT\\nhas losses not exceeding 25% on most tasks, except for three relatively sub-\\njective emotion perception tasks where it performs poorly. If we remove these\\ntasks to calculate the average quality of the two models, we find that the SOTA\\nmethod has an average quality of 80%, while the ChatGPT method has an av-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 21, 'page_label': '22'}, page_content='erage quality of 69.7%. That is to say, ChatGPT performs well on all tasks\\nexcept for emotion-related tasks, and can handle most of the problems we con-\\nsider. However, overall, its performance is lower than the SOTA model based\\non experimental data, but the difference between the two is not very large.\\nIn other tasks, according to literature [92], ChatGPT performs well in nat-\\nural language inference, i.e., the task of inferring sentence relationships, and its'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 21, 'page_label': '22'}, page_content='performance on this task is significantly better than all bert-style models [93].\\nHowever, while ChatGPT performs well on inference tasks, it may produce some\\nself-contradictory or unreasonable responses, which is its potential limitation.\\nIn question-answering, dialogue, and summarization tasks, ChatGPT performs\\nbetter than the GPT-3.5 model [92], especially in the question-answering task,\\nwhere its performance is comparable to bert-style models [93]. Therefore, we'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 21, 'page_label': '22'}, page_content='have demonstrated that ChatGPT is a qualified general-purpose model.\\nHowever, ChatGPT also has limitations in many aspects. Firstly, it lacks the\\nability to handle non-textual semantic reasoning tasks such as mathematical,\\ntemporal, and spatial reasoning, and it performs poorly in multi-hop reasoning\\n[90]. Secondly, ChatGPT is not good at solving named entity recognition tasks\\n[92]. Furthermore, ChatGPT performs poorly in handling tasks involving nega-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 21, 'page_label': '22'}, page_content='tive connotations and neutral similarity [93]. Finally, these conclusions indicate\\n22'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 22, 'page_label': '23'}, page_content='that, like other large pre-trained language models, ChatGPT has limitations in\\ncompleting complex reasoning tasks.\\nIn summary, ChatGPT’s zero-shot performance is comparable to fine-tuned\\nbert and GPT-3.5 models, and with the help of advanced prompting strategies,\\nChatGPT can demonstrate better comprehension abilities. However, it still\\ncannot outperform the current SOTA models.\\n2.3.2 The possibility of using ChatGPT for plagiarism and cheating'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 22, 'page_label': '23'}, page_content='In response to the possibility of ChatGPT being used for plagiarism and cheat-\\ning, Zhou et al. [94] reflected on the current state of development of artificial in-\\ntelligence like ChatGPT. As ChatGPT becomes increasingly easy to obtain and\\nscalable in text generation, there is a high likelihood that these technologies will\\nbe used for plagiarism, including scientific literature and news sources, posing\\na great threat to the credibility of various forms of news media and academic'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 22, 'page_label': '23'}, page_content='articles. Some scholars are concerned that the end of paper as a meaningful\\nevaluation tool may be approaching [95, 96], as ChatGPT can easily generate\\npersuasive paragraphs, chapters, and papers on any given topic. Additionally,\\nit will exacerbate plagiarism issues in many fields such as education, medicine,\\nand law [10], and may be used for cheating in academic exams [97]. Definitional\\nrecognition technology is a relatively effective method for detecting plagiarism,'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 22, 'page_label': '23'}, page_content='and the definitional typology proposed in [94] can alleviate people’s concerns by\\nbeing used to construct new datasets. Susnjak [97] proposed a solution to the\\npossibility of large language models like ChatGPT being used for exam cheat-\\ning: guiding ChatGPT to generate some critical thinking problems through\\nquestioning, then providing answers and critically evaluating them. Analysis of\\nChatGPT shows that it exhibits critical thinking, can generate highly realistic'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 22, 'page_label': '23'}, page_content='text in terms of accuracy, relevance, depth, breadth, logic, persuasiveness, and\\noriginality. Therefore, educators must be aware of the possibility of ChatGPT\\nbeing used for exam cheating and take measures to combat cheating behavior\\nto ensure the fairness of online exams.\\n2.3.3 Feedback from ChatGPT users\\nIn response to feedback from ChatGPT users, Haque et al. [98] conducted a\\nmixed-methods study using 10,732 early ChatGPT user tweets. The authors'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 22, 'page_label': '23'}, page_content='extracted Twitter data using Python and Twitter API and constructed the\\nChatGPTTweet dataset, which contains 18k tweets. For each tweet, the au-\\nthors collected information on text content, user location, occupation, verifica-\\ntion status, date of publication, and tags. Based on this dataset, the authors\\nstudied the characteristics of early ChatGPT users, discussion topics related to\\nChatGPT on Twitter, and the sentiment of Twitter users toward ChatGPT.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 22, 'page_label': '23'}, page_content='For RQ1, the authors found that early ChatGPT users had a diverse and wide\\nrange of occupational backgrounds and geographical locations. For RQ2, the\\nauthors identified nine topics related to ChatGPT, including its impact on soft-\\nware development, entertainment and creativity, natural language processing,\\n23'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 23, 'page_label': '24'}, page_content='education, chatbot intelligence, business development, search engines, question-\\nanswering tests, and future careers and opportunities. For RQ3, most early\\nusers expressed positive sentiment toward topics such as software development\\nand creativity, while only a few expressed concern about the potential misuse\\nof ChatGPT.\\n2.3.4 Adverse effects of ChatGPT on users\\nRegarding the negative effects of ChatGPT on users, Luan et al. [99] studied'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 23, 'page_label': '24'}, page_content='the psychological principles of ChatGPT, delved into the factors that attract\\nusers’ attention, and revealed the impact of these factors on future learning.\\nIn the post-pandemic era, teachers and students are both facing uncertainty\\nin the teaching process and job pressures. Under these common constraints of\\neducation and employment, educators and students must re-evaluate current\\neducational methods and outcomes, as well as students’ future career devel-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 23, 'page_label': '24'}, page_content='opment. Through question-and-answer exchanges with ChatGPT, people can\\neasily obtain appropriate solutions or key information, thereby enhancing their\\nmotivation, eliminating anxiety in learning, improving interest, and achieving\\npsychological satisfaction. Subhash et al. [100] explored whether large language\\nmodels have the ability to reverse user preferences. With the development of\\npre-trained large language models, people are increasingly concerned about the'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 23, 'page_label': '24'}, page_content='ability of these models to influence, persuade, and potentially manipulate user\\npreferences in extreme cases. Therefore, the literature [100] roughly qualita-\\ntively analyzed that adversarial behavior does lead to potential changes in user\\npreferences and behaviors in dialogue systems. If we want to further quanti-\\ntatively analyze the ability of large language models in this regard, additional\\nstatistical summary techniques need to be used for future research.\\n3 Discussion\\n3.1 Limitations'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 23, 'page_label': '24'}, page_content='3 Discussion\\n3.1 Limitations\\nDespite the remarkable capabilities of ChatGPT and GPT-4, it still faces certain\\nlimitations. Some of these limitations include:\\nOutdated Knowledge\\nThe current models are trained on historical data (up to 2021), thereby\\nlacking real-time comprehension of current affairs. This is a critical concern\\nin today’s information-explosion era, as the reliability of prior knowledge bases\\nprogressively diminishes, potentially yielding inaccurate responses, especially in'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 23, 'page_label': '24'}, page_content='rapidly evolving domains such as jurisprudence and technology. Additionally,\\nthese models are incapable of fact-checking while the training data is composed\\nof content from various sources, some of which may be unreliable, which may\\nresult in seemingly plausible yet nonsensical responses.\\n24'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 24, 'page_label': '25'}, page_content='Insufficient Understanding\\nWhile these models can interpret the majority of inquiries and contextual\\nsituations, they occasionally encounter comprehension biases when addressing\\nambiguous or contextually complex queries. Furthermore, in certain specialized\\nfields, the abundance of unique abbreviation exacerbates the models’ under-\\nstanding challenges, resulting in incorrect and vacuous responses.\\nEnergy Consumption\\nThroughout the training and inference stages, these large-scale models re-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 24, 'page_label': '25'}, page_content='quire significant computational resources and electrical power, resulting in ele-\\nvated energy consumption and significant carbon emissions. Consequently, this\\nrestricts their deployment and practical applications.\\nMalicious Usage\\nDespite OpenAI implementing a series of restrictions to mitigate model tox-\\nicity, instances of users evading these constraints through meticulously designed\\nprompts have emerged, inducing the model to produce unhealthy content or'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 24, 'page_label': '25'}, page_content='even using it for illicit commercial purposes.\\nBias and Discrimination\\nDue to the influence of pre-training data, the models exhibit biases in polit-\\nical, ideological, and other areas. The application of LLMs in public domains,\\nsuch as education and publicity, should be approached with extreme caution.\\nPrivacy and Data Security\\nConcurrent with the expansion of users, protecting user privacy and data\\nsecurity becomes increasingly important. In fact, ChatGPT was banned in'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 24, 'page_label': '25'}, page_content='Italy in early April due to privacy concerns. This is particularly relevant given\\nthe models’ extensive collection of personal information and preferences during\\ninteractions, and as future multimodal models, such as GPT-4, may frequently\\nrequire users to upload private photos.\\n3.2 Future Directions\\nIn forthcoming research, the development of models based on ChatGPT and\\nGPT-4 may focus on addressing these limitations to enhance their practical\\napplications.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 24, 'page_label': '25'}, page_content='applications.\\nPrimarily, researchers should continue to work on refining model training\\nmethodologies while filtering pre-training data to minimize the presence of mis-\\nleading information in the model’s knowledge base, thereby obtaining accurate\\nresponses. Concurrently, it is crucial to emphasize training approaches that\\neconomize computational resources, thereby mitigating costs and broadening\\npotential application scenarios.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 24, 'page_label': '25'}, page_content='potential application scenarios.\\nMoreover, the advancements in context-awareness and disambiguation tech-\\nnologies are anticipated to facilitate enhanced comprehension of complex queries\\n25'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 25, 'page_label': '26'}, page_content='by models, improving the accuracy, relevance, and context-awareness of AI-\\ngenerated content. Integrating real-time data streams can also keep these mod-\\nels in sync with current events and trends, enabling them to provide up-to-date\\ninformation such as live traffic, weather, and stock updates.\\nAdditionally, developers should engage in interdisciplinary collaboration with\\nspecialists from diverse domains, including policy-making, jurisprudence, and'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 25, 'page_label': '26'}, page_content='sociology, with the objective of formulating standard and ethical frameworks\\nfor LLM development, deployment, and utilization, thereby alleviating poten-\\ntial harmful consequences. In terms of public awareness and education, manda-\\ntory awareness training should be implemented prior to large-scale public de-\\nployment and application to increase public awareness of LLM capabilities and\\nlimitations while promoting responsible and informed utilization, especially in'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 25, 'page_label': '26'}, page_content='industries such as K-12 education and journalism.\\nFinally, the influence of ChatGPT and GPT-4 should not be limited to\\njust the NLP field. They also show promising prospects in the areas of com-\\nputer vision, brain-inspired AI, and robotics. These models exhibit a capacity\\nfor learning and comprehension comparable with human-level intelligence, po-\\nsitioning them as a pivotal component in the development of artificial general'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 25, 'page_label': '26'}, page_content='intelligence (AGI)[101]. Their ability to facilitate seamless interactions between\\nhumans and robots paves the way for the execution of more complex tasks.\\nThe remarkable capacity of zero-shot in-context learning of these models en-\\nables quick adaptation to new tasks without the requirement for labeled data\\nfor fine-tuning, which is a critical challenge in fields like medical informatics[102]\\nand robotics[103] where the availability of labeled data is commonly limited or\\nnon-existent.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 25, 'page_label': '26'}, page_content='non-existent.\\n4 Conclusion\\nThis review paper provides a comprehensive survey of ChatGPT and GPT-4,\\nhighlighting their potential applications and significant contributions to the field\\nof natural language processing. The findings of this study reveal that the interest\\nin these models is growing rapidly, and they have shown considerable potential\\nfor application across a wide range of domains. One key factor contributing\\nto the success of ChatGPT and GPT-4 is their ability to perform large-scale'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 25, 'page_label': '26'}, page_content='pre-training, which captures knowledge from the vast expanse of the internet,\\nallowing the models to learn from a massive amount of data. The integration of\\nReinforcement Learning from Human Feedback (RLHF) has further enhanced\\nthe model’s adaptability and performance, making it highly efficient in process-\\ning natural language. This study has also identified several potential ethical\\nconcerns related to the development and use of ChatGPT and GPT-4. For in-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 25, 'page_label': '26'}, page_content='stance, there are concerns about the generation of biased or harmful content,\\nprivacy violations, and the potential for misuse of the technology. It is crucial\\nto address these concerns and ensure that ChatGPT and GPT-4 are developed\\nand used in a responsible and ethical manner. Furthermore, the results of this\\nstudy demonstrate that there is significant potential for ChatGPT and GPT-4\\n26'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 26, 'page_label': '27'}, page_content='to be applied in a range of domains, including education, history, mathematics,\\nphysics, and more. These models can facilitate tasks such as generating sum-\\nmaries, answering questions, and providing personalized recommendations to\\nusers. Overall, the insights presented in this review paper can serve as a useful\\nguide for researchers and practitioners looking to advance the field of natural\\nlanguage processing. Future research in this field should focus on addressing'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 26, 'page_label': '27'}, page_content='ethical concerns, exploring new applications, and ensuring the responsible use\\nof ChatGPT and GPT-4. The potential of these models to revolutionize natural\\nlanguage processing is enormous, and we look forward to seeing more develop-\\nments in this field.\\nReferences\\n[1] Radford A, Narasimhan K, Salimans T, Sutskever I, et al. Improving\\nlanguage understanding by generative pre-training. OpenAI. 2018.\\n[2] Radford A, Wu J, Child R, Luan D, Amodei D, Sutskever I, et al.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 26, 'page_label': '27'}, page_content='Language models are unsupervised multitask learners. OpenAI blog.\\n2019;1(8):9.\\n[3] Radford A, Wu J, Amodei D, Amodei D, Clark J, Brundage M, et al. Bet-\\nter language models and their implications. OpenAI Blog https://openai\\ncom/blog/better-language-models. 2019;1(2).\\n[4] Brown T, Mann B, Ryder N, Subbiah M, Kaplan JD, Dhariwal P, et al.\\nLanguage models are few-shot learners. Advances in neural information\\nprocessing systems. 2020;33:1877-901.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 26, 'page_label': '27'}, page_content='processing systems. 2020;33:1877-901.\\n[5] Ouyang L, Wu J, Jiang X, Almeida D, Wainwright CL, Mishkin P, et al.\\nTraining language models to follow instructions with human feedback.\\narXiv preprint arXiv:220302155. 2022.\\n[6] Christiano PF, Leike J, Brown T, Martic M, Legg S, Amodei D. Deep\\nreinforcement learning from human preferences. Advances in neural infor-\\nmation processing systems. 2017;30.\\n[7] Frieder S, Pinchetti L, Griffiths RR, Salvatori T, Lukasiewicz T, Pe-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 26, 'page_label': '27'}, page_content='tersen PC, et al. Mathematical capabilities of chatgpt. arXiv preprint\\narXiv:230113867. 2023.\\n[8] Pardos ZA, Bhandari S. Learning gain differences between ChatGPT and\\nhuman tutor generated algebra hints. arXiv preprint arXiv:230206871.\\n2023.\\n[9] Shakarian P, Koyyalamudi A, Ngu N, Mareedu L. An Independent Eval-\\nuation of ChatGPT on Mathematical Word Problems (MWP). arXiv\\npreprint arXiv:230213814. 2023.\\n27'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 27, 'page_label': '28'}, page_content='[10] Lehnert K. AI Insights into Theoretical Physics and the Swampland Pro-\\ngram: A Journey Through the Cosmos with ChatGPT. arXiv preprint\\narXiv:230108155. 2023.\\n[11] Kortemeyer G. Could an Artificial-Intelligence agent pass an introductory\\nphysics course? arXiv preprint arXiv:230112127. 2023.\\n[12] West CG. AI and the FCI: Can ChatGPT Project an Understanding of\\nIntroductory Physics? arXiv preprint arXiv:230301067. 2023.\\n[13] Nov O, Singh N, Mann DM. Putting ChatGPT’s Medical Advice to the'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 27, 'page_label': '28'}, page_content='(Turing) Test. medRxiv. 2023.\\n[14] Tu R, Ma C, Zhang C. Causal-Discovery Performance of ChatGPT in the\\ncontext of Neuropathic Pain Diagnosis. arXiv preprint arXiv:230113819.\\n2023.\\n[15] Glymour C, Zhang K, Spirtes P. Review of Causal Discovery Methods\\nBased on Graphical Models. Frontiers in Genetics. 2019.\\n[16] Guo S, Wang Y, Li S, Saeed N. Semantic Communications with Ordered\\nImportance using ChatGPT. arXiv preprint arXiv:230207142. 2023.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 27, 'page_label': '28'}, page_content='[17] Wang S, Scells H, Koopman B, Zuccon G. Can chatgpt write a good\\nboolean query for systematic review literature search? arXiv preprint\\narXiv:230203495. 2023.\\n[18] Kuzman T, Mozetic I, Ljubeˇ sic N. ChatGPT: Beginning of an End of\\nManual Linguistic Data Annotation? Use Case of Automatic Genre Iden-\\ntification. arXiv e-prints. 2023:arXiv-2303.\\n[19] Amin MM, Cambria E, Schuller BW. Will Affective Computing Emerge\\nfrom Foundation Models and General AI? A First Evaluation on Chat-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 27, 'page_label': '28'}, page_content='GPT. arXiv preprint arXiv:230303186. 2023.\\n[20] Zhang B, Ding D, Jing L. How would Stance Detection Techniques Evolve\\nafter the Launch of ChatGPT? arXiv preprint arXiv:221214548. 2022.\\n[21] Huang F, Kwak H, An J. Is ChatGPT better than Human Annota-\\ntors? Potential and Limitations of ChatGPT in Explaining Implicit Hate\\nSpeech. arXiv preprint arXiv:230207736. 2023.\\n[22] Zhang X, Chowdhury RR, Hong D, Gupta RK, Shang J. Model-\\ning Label Semantics Improves Activity Recognition. arXiv preprint'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 27, 'page_label': '28'}, page_content='arXiv:230103462. 2023.\\n[23] Fu Q, Teng Z, Georgaklis M, White J, Schmidt DC. NL2CMD: An Up-\\ndated Workflow for Natural Language to Bash Commands Translation.\\narXiv preprint arXiv:230207845. 2023.\\n28'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 28, 'page_label': '29'}, page_content='[24] Chen Y, Eger S. Transformers go for the LOLs: Generating (humourous)\\ntitles from scientific abstracts end-to-end. arXiv preprint arXiv:221210522.\\n2022.\\n[25] Chen N, Wang Y, Jiang H, Cai D, Chen Z, Li J. What would Harry\\nsay? Building Dialogue Agents for Characters in a Story. arXiv preprint\\narXiv:221106869. 2022.\\n[26] Jeblick K, Schachtner B, Dexl J, Mittermeier A, St¨ uber AT, Topalis J,\\net al. ChatGPT Makes Medicine Easy to Swallow: An Exploratory Case'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 28, 'page_label': '29'}, page_content='Study on Simplified Radiology Reports. arXiv preprint arXiv:221214882.\\n2022.\\n[27] Xia CS, Zhang L. Conversational automated program repair. arXiv\\npreprint arXiv:230113246. 2023.\\n[28] Jiao W, ZhaopengTu WJtX. Is ChatGPT A Good Translator? Yes With\\nGPT-4 As The Engine.\\n[29] Prieto SA, Mengiste ET, de Soto BG. Investigating the Use of\\nChatGPT for the Scheduling of Construction Projects. Buildings.\\n2023 mar;13(4):857. Available from: https://doi.org/10.3390%\\n2Fbuildings13040857.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 28, 'page_label': '29'}, page_content='2Fbuildings13040857.\\n[30] Michail A, Konstantinou S, Clematide S. UZH CLyp at SemEval-2023\\nTask 9: Head-First Fine-Tuning and ChatGPT Data Generation for\\nCross-Lingual Learning in Tweet Intimacy Prediction. arXiv preprint\\narXiv:230301194. 2023.\\n[31] Wang J, Liang Y, Meng F, Li Z, Qu J, Zhou J. Cross-Lingual Summa-\\nrization via ChatGPT. arXiv preprint arXiv:230214229. 2023.\\n[32] Yang X, Li Y, Zhang X, Chen H, Cheng W. Exploring the limits of'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 28, 'page_label': '29'}, page_content='chatgpt for query or aspect-based text summarization. arXiv preprint\\narXiv:230208081. 2023.\\n[33] Belouadi J, Eger S. ByGPT5: End-to-End Style-conditioned Po-\\netry Generation with Token-free Language Models. arXiv preprint\\narXiv:221210474. 2022.\\n[34] Blanco-Gonzalez A, Cabezon A, Seco-Gonzalez A, Conde-Torres D,\\nAntelo-Riveiro P, Pineiro A, et al. The Role of AI in Drug Discovery: Chal-\\nlenges, Opportunities, and Strategies. arXiv preprint arXiv:221208104.\\n2022.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 28, 'page_label': '29'}, page_content='2022.\\n[35] Khalil M, Er E. Will ChatGPT get you caught? Rethinking of plagiarism\\ndetection. arXiv preprint arXiv:230204335. 2023.\\n[36] Basic Z, Banovac A, Kruzic I, Jerkovic I. Better by you, better than\\nme, chatgpt3 as writing assistance in students essays. arXiv preprint\\narXiv:230204536. 2023.\\n29'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 29, 'page_label': '30'}, page_content='[37] Noever D, Ciolino M. The Turing Deception. arXiv preprint\\narXiv:221206721. 2022.\\n[38] Megahed FM, Chen YJ, Ferris JA, Knoth S, Jones-Farmer LA. How\\nGenerative AI models such as ChatGPT can be (Mis) Used in SPC Prac-\\ntice, Education, and Research? An Exploratory Study. arXiv preprint\\narXiv:230210916. 2023.\\n[39] Treude C. Navigating Complexity in Software Engineering: A Prototype\\nfor Comparing GPT-n Solutions. arXiv preprint arXiv:230112169. 2023.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 29, 'page_label': '30'}, page_content='[40] Sobania D, Briesch M, Hanna C, Petke J. An analysis of the automatic\\nbug fixing performance of chatgpt. arXiv preprint arXiv:230108653. 2023.\\n[41] Noever D, McKee F. Numeracy from Literacy: Data Science as an Emer-\\ngent Skill from Large Language Models. arXiv preprint arXiv:230113382.\\n2023.\\n[42] McKee F, Noever D. Chatbots in a Botnet World. arXiv preprint\\narXiv:221211126. 2022.\\n[43] McKee F, Noever D. Chatbots in a Honeypot World. arXiv preprint\\narXiv:230103771. 2023.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 29, 'page_label': '30'}, page_content='arXiv:230103771. 2023.\\n[44] Susnjak T. Applying BERT and ChatGPT for Sentiment Analysis of\\nLyme Disease in Scientific Literature. arXiv preprint arXiv:230206474.\\n2023.\\n[45] Tang Z, Kejriwal M. A Pilot Evaluation of ChatGPT and DALL-E 2 on\\nDecision Making and Spatial Reasoning. arXiv preprint arXiv:230209068.\\n2023.\\n[46] Ortega-Mart´ ın M, Garc´ ıa-Sierra´O, Ardoiz A, ´Alvarez J, Armenteros JC,\\nAlonso A. Linguistic ambiguity analysis in ChatGPT. arXiv preprint\\narXiv:230206426. 2023.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 29, 'page_label': '30'}, page_content='arXiv:230206426. 2023.\\n[47] Maddigan P, Susnjak T. Chat2vis: Generating data visualisations via\\nnatural language using chatgpt, codex and gpt-3 large language models.\\narXiv preprint arXiv:230202094. 2023.\\n[48] Luo Y, Tang J, Li G. nvBench: A Large-Scale Synthesized Dataset for\\nCross-Domain Natural Language to Visualization Task. arXiv preprint\\narXiv:211212926. 2021.\\n[49] Liu C, Han Y, Jiang R, Yuan X. Advisor: Automatic visualization answer'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 29, 'page_label': '30'}, page_content='for natural-language question on tabular data. In: 2021 IEEE 14th Pacific\\nVisualization Symposium (PacificVis). IEEE; 2021. p. 11-20.\\n30'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 30, 'page_label': '31'}, page_content='[50] Narechania A, Srinivasan A, Stasko J. NL4DV: A toolkit for generat-\\ning analytic specifications for data visualization from natural language\\nqueries. IEEE Transactions on Visualization and Computer Graphics.\\n2020;27(2):369-79.\\n[51] Wei X, Cui X, Cheng N, Wang X, Zhang X, Huang S, et al. Zero-\\nShot Information Extraction via Chatting with ChatGPT. arXiv preprint\\narXiv:230210205. 2023.\\n[52] Takanobu R, Zhang T, Liu J, Huang M. A hierarchical framework for'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 30, 'page_label': '31'}, page_content='relation extraction with reinforcement learning. In: Proceedings of the\\nAAAI conference on artificial intelligence. vol. 33; 2019. p. 7072-9.\\n[53] Li S, He W, Shi Y, Jiang W, Liang H, Jiang Y, et al. Duie: A large-\\nscale chinese dataset for information extraction. In: Natural Language\\nProcessing and Chinese Computing: 8th CCF International Conference,\\nNLPCC 2019, Dunhuang, China, October 9–14, 2019, Proceedings, Part\\nII 8. Springer; 2019. p. 791-800.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 30, 'page_label': '31'}, page_content='II 8. Springer; 2019. p. 791-800.\\n[54] Wang Z, Shang J, Liu L, Lu L, Liu J, Han J. Crossweigh: Training named\\nentity tagger from imperfect annotations. arXiv preprint arXiv:190901441.\\n2019.\\n[55] Levow GA. The third international Chinese language processing bakeoff:\\nWord segmentation and named entity recognition. In: Proceedings of the\\nFifth SIGHAN workshop on Chinese language processing; 2006. p. 108-17.\\n[56] Li X, Li F, Pan L, Chen Y, Peng W, Wang Q, et al. DuEE: a large-scale'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 30, 'page_label': '31'}, page_content='dataset for Chinese event extraction in real-world scenarios. In: Natu-\\nral Language Processing and Chinese Computing: 9th CCF International\\nConference, NLPCC 2020, Zhengzhou, China, October 14–18, 2020, Pro-\\nceedings, Part II 9. Springer; 2020. p. 534-45.\\n[57] Gormley MR, Yu M, Dredze M. Improved relation extraction with feature-\\nrich compositional embedding models. arXiv preprint arXiv:150502419.\\n2015.\\n[58] Hoffmann R, Zhang C, Ling X, Zettlemoyer L, Weld DS. Knowledge-based'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 30, 'page_label': '31'}, page_content='weak supervision for information extraction of overlapping relations. In:\\nProceedings of the 49th annual meeting of the association for computa-\\ntional linguistics: human language technologies; 2011. p. 541-50.\\n[59] Gao J, Zhao H, Yu C, Xu R. Exploring the feasibility of ChatGPT for\\nevent extraction. arXiv preprint arXiv:230303836. 2023.\\n[60] Lu Y, Lin H, Xu J, Han X, Tang J, Li A, et al. Text2event: Controllable\\nsequence-to-structure generation for end-to-end event extraction. arXiv'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 30, 'page_label': '31'}, page_content='preprint arXiv:210609232. 2021.\\n31'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 31, 'page_label': '32'}, page_content='[61] Du X, Cardie C. Event extraction by answering (almost) natural ques-\\ntions. arXiv preprint arXiv:200413625. 2020.\\n[62] Tang R, Han X, Jiang X, Hu X. Does Synthetic Data Generation of LLMs\\nHelp Clinical Text Mining? arXiv preprint arXiv:230304360. 2023.\\n[63] He J, Wang L, Hu Y, Liu N, Liu H, Xu X, et al. ICL-D3IE: In-Context\\nLearning with Diverse Demonstrations Updating for Document Informa-\\ntion Extraction. arXiv preprint arXiv:230305063. 2023.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 31, 'page_label': '32'}, page_content='[64] Jaume G, Ekenel HK, Thiran JP. Funsd: A dataset for form under-\\nstanding in noisy scanned documents. In: 2019 International Conference\\non Document Analysis and Recognition Workshops (ICDARW). vol. 2.\\nIEEE; 2019. p. 1-6.\\n[65] Park S, Shin S, Lee B, Lee J, Surh J, Seo M, et al. CORD: a consoli-\\ndated receipt dataset for post-OCR parsing. In: Workshop on Document\\nIntelligence at NeurIPS 2019; 2019. .\\n[66] Huang Z, Chen K, He J, Bai X, Karatzas D, Lu S, et al. Icdar2019 com-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 31, 'page_label': '32'}, page_content='petition on scanned receipt ocr and information extraction. In: 2019 In-\\nternational Conference on Document Analysis and Recognition (ICDAR).\\nIEEE; 2019. p. 1516-20.\\n[67] Polak MP, Morgan D. Extracting Accurate Materials Data from Research\\nPapers with Conversational Language Models and Prompt Engineering–\\nExample of ChatGPT. arXiv preprint arXiv:230305352. 2023.\\n[68] Kocmi T, Federmann C. Large language models are state-of-the-art eval-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 31, 'page_label': '32'}, page_content='uators of translation quality. arXiv preprint arXiv:230214520. 2023.\\n[69] Freitag M, Rei R, Mathur N, Lo Ck, Stewart C, Avramidis E, et al.\\nResults of WMT22 metrics shared task: Stop using BLEU–neural metrics\\nare better and more robust. In: Proceedings of the Seventh Conference\\non Machine Translation (WMT); 2022. p. 46-68.\\n[70] Kocmi T, Federmann C, Grundkiewicz R, Junczys-Dowmunt M, Mat-\\nsushita H, Menezes A. To ship or not to ship: An extensive eval-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 31, 'page_label': '32'}, page_content='uation of automatic metrics for machine translation. arXiv preprint\\narXiv:210710821. 2021.\\n[71] Freitag M, Rei R, Mathur N, Lo Ck, Stewart C, Avramidis E, et al.\\nResults of WMT22 metrics shared task: Stop using BLEU–neural metrics\\nare better and more robust. In: Proceedings of the Seventh Conference\\non Machine Translation (WMT); 2022. p. 46-68.\\n[72] Wang J, Liang Y, Meng F, Shi H, Li Z, Xu J, et al. Is chatgpt a good nlg\\nevaluator? a preliminary study. arXiv preprint arXiv:230304048. 2023.\\n32'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 32, 'page_label': '33'}, page_content='[73] Hermann KM, Kocisky T, Grefenstette E, Espeholt L, Kay W, Suleyman\\nM, et al. Teaching machines to read and comprehend. Advances in neural\\ninformation processing systems. 2015;28.\\n[74] Zar JH. Spearman rank correlation. Encyclopedia of biostatistics. 2005;7.\\n[75] Mukaka MM. A guide to appropriate use of correlation coefficient in\\nmedical research. Malawi medical journal. 2012;24(3):69-71.\\n[76] Kendall MG. A new measure of rank correlation. Biometrika.\\n1938;30(1/2):81-93.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 32, 'page_label': '33'}, page_content='1938;30(1/2):81-93.\\n[77] Dai H, Liu Z, Liao W, Huang X, Wu Z, Zhao L, et al. ChatAug: Leveraging\\nChatGPT for Text Data Augmentation. arXiv preprint arXiv:230213007.\\n2023.\\n[78] Wu C, Yin S, Qi W, Wang X, Tang Z, Duan N. Visual chatgpt: Talk-\\ning, drawing and editing with visual foundation models. arXiv preprint\\narXiv:230304671. 2023.\\n[79] Zheng O, Abdel-Aty M, Wang D, Wang Z, Ding S. ChatGPT is on the\\nhorizon: Could a large language model be all we need for Intelligent Trans-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 32, 'page_label': '33'}, page_content='portation? arXiv preprint arXiv:230305382. 2023.\\n[80] White J, Fu Q, Hays S, Sandborn M, Olea C, Gilbert H, et al. A prompt\\npattern catalog to enhance prompt engineering with chatgpt. arXiv\\npreprint arXiv:230211382. 2023.\\n[81] Ahmad A, Waseem M, Liang P, Fehmideh M, Aktar MS, Mikkonen T.\\nTowards Human-Bot Collaborative Software Architecting with ChatGPT.\\narXiv preprint arXiv:230214600. 2023.\\n[82] Lanzi PL, Loiacono D. ChatGPT and Other Large Language Models as'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 32, 'page_label': '33'}, page_content='Evolutionary Engines for Online Interactive Collaborative Game Design.\\narXiv preprint arXiv:230302155. 2023.\\n[83] Wang S, Zhao Z, Ouyang X, Wang Q, Shen D. Chatcad: Interactive\\ncomputer-aided diagnosis on medical image using large language models.\\narXiv preprint arXiv:230207257. 2023.\\n[84] Hartmann J, Schwenzow J, Witte M. The political ideology of conver-\\nsational AI: Converging evidence on ChatGPT’s pro-environmental, left-\\nlibertarian orientation. arXiv preprint arXiv:230101768. 2023.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 32, 'page_label': '33'}, page_content='[85] Kr¨ ugel S, Ostermaier A, Uhl M. The moral authority of ChatGPT. arXiv\\npreprint arXiv:230107098. 2023.\\n[86] Borji A. A categorical archive of chatgpt failures. arXiv preprint\\narXiv:230203494. 2023.\\n33'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 33, 'page_label': '34'}, page_content='[87] Hacker P, Engel A, Mauer M. Regulating chatgpt and other large gener-\\native ai models. arXiv preprint arXiv:230202337. 2023.\\n[88] Hacker P. The European AI Liability Directives–Critique of a Half-\\nHearted Approach and Lessons for the Future. arXiv preprint\\narXiv:221113960. 2022.\\n[89] Kirk HR, Vidgen B, R¨ ottger P, Hale SA. Personalisation within bounds:\\nA risk taxonomy and policy framework for the alignment of large language'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 33, 'page_label': '34'}, page_content='models with personalised feedback. arXiv preprint arXiv:230305453. 2023.\\n[90] Bang Y, Cahyawijaya S, Lee N, Dai W, Su D, Wilie B, et al. A multitask,\\nmultilingual, multimodal evaluation of chatgpt on reasoning, hallucina-\\ntion, and interactivity. arXiv preprint arXiv:230204023. 2023.\\n[91] Koco´ n J, Cichecki I, Kaszyca O, Kochanek M, Szyd lo D, Baran J,\\net al. ChatGPT: Jack of all trades, master of none. arXiv preprint\\narXiv:230210724. 2023.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 33, 'page_label': '34'}, page_content='arXiv:230210724. 2023.\\n[92] Qin C, Zhang A, Zhang Z, Chen J, Yasunaga M, Yang D. Is chatgpt a\\ngeneral-purpose natural language processing task solver? arXiv preprint\\narXiv:230206476. 2023.\\n[93] Zhong Q, Ding L, Liu J, Du B, Tao D. Can chatgpt understand too?\\na comparative study on chatgpt and fine-tuned bert. arXiv preprint\\narXiv:230210198. 2023.\\n[94] Zhou C, Qiu C, Acuna DE. Paraphrase Identification with Deep Learning:\\nA Review of Datasets and Methods. arXiv preprint arXiv:221206933.\\n2022.'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 33, 'page_label': '34'}, page_content='2022.\\n[95] de Winter J. Can ChatGPT Pass High School Exams on English Language\\nComprehension? 2023.\\n[96] Yeadon W, Inyang OO, Mizouri A, Peach A, Testrow C. The Death of the\\nShort-Form Physics Essay in the Coming AI Revolution. arXiv preprint\\narXiv:221211661. 2022.\\n[97] Susnjak T. ChatGPT: The End of Online Exam Integrity? arXiv preprint\\narXiv:221209292. 2022.\\n[98] Haque MU, Dharmadasa I, Sworna ZT, Rajapakse RN, Ahmad H.\\n” I think this is the most disruptive technology”: Exploring Senti-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 33, 'page_label': '34'}, page_content='ments of ChatGPT Early Adopters using Twitter Data. arXiv preprint\\narXiv:221205856. 2022.\\n[99] Luan L, Lin X, Li W. Exploring the Cognitive Dynamics of Artificial\\nIntelligence in the Post-COVID-19 and Learning 3.0 Era: A Case Study\\nof ChatGPT. arXiv preprint arXiv:230204818. 2023.\\n34'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 34, 'page_label': '35'}, page_content='[100] Subhash V. Can Large Language Models Change User Preference Adver-\\nsarially? arXiv preprint arXiv:230210291. 2023.\\n[101] Zhao L, Zhang L, Wu Z, Chen Y, Dai H, Yu X, et al. When Brain-inspired\\nAI Meets AGI. arXiv preprint arXiv:230315935. 2023.\\n[102] Liu Z, Yu X, Zhang L, Wu Z, Cao C, Dai H, et al. DeID-GPT:\\nZero-shot Medical Text De-Identification by GPT-4. arXiv preprint\\narXiv:230311032. 2023.\\n[103] Liu D, Chen Y, Wu Z. Digital Twin (DT)-CycleGAN: Enabling Zero-'), Document(metadata={'producer': 'pdfTeX-1.40.21', 'creator': 'LaTeX with hyperref', 'creationdate': '2023-04-05T00:33:07+00:00', 'author': '', 'keywords': '', 'moddate': '2023-04-05T00:33:07+00:00', 'ptex.fullbanner': 'This is pdfTeX, Version 3.14159265-2.6-1.40.21 (TeX Live 2020) kpathsea version 6.3.2', 'subject': '', 'title': '', 'trapped': '/False', 'source': './data/Summary of ChatGPTGPT-4 Research.pdf', 'total_pages': 35, 'page': 34, 'page_label': '35'}, page_content='Shot Sim-to-Real Transfer of Visual Grasping Models. IEEE Robotics\\nand Automation Letters. 2023.\\n35')]\n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 17
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T01:15:31.381367287Z",
|
||
"start_time": "2026-06-09T01:15:28.625949172Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"from typing import List\n",
|
||
"from langchain_core.documents import Document\n",
|
||
"\n",
|
||
"class RAGState(TypedDict):\n",
|
||
" query:str\n",
|
||
" retriever_docs:list[Document]\n",
|
||
" answer:str\n",
|
||
"\n",
|
||
"def retrieve(state):\n",
|
||
" # 기존의 벡터스토어에 질의\n",
|
||
" vectorstore = Chroma(collection_name=\"research\", embedding_function=watson_embedding, persist_directory=\"./db/chroma_db\")\n",
|
||
"\n",
|
||
" docs = vectorstore.similarity_search(state['query'], k=3)\n",
|
||
"\n",
|
||
" return {\"retriever_docs\" : docs}\n",
|
||
"\n",
|
||
"def generate(state):\n",
|
||
" context = \"\\n\\n\".join(doc.page_content for doc in state['retriever_docs'])\n",
|
||
"\n",
|
||
" prompt = \"\"\"\\\n",
|
||
"다음 컨텍스트를 참고하여 질문에 답하세요.\n",
|
||
"컨텍스트에 없는 내용은 모른다고 답하세요.\n",
|
||
"\n",
|
||
"컨텍스트:\n",
|
||
"{context}\n",
|
||
"질문:\n",
|
||
"{query}\n",
|
||
"\"\"\"\n",
|
||
" response = watson_llm.invoke(prompt.format(context=context, query=state['query']))\n",
|
||
" return {\"answer\":response.content}\n",
|
||
"\n",
|
||
"graph = StateGraph(RAGState)\n",
|
||
"graph.add_node(\"retrieve\", retrieve)\n",
|
||
"graph.add_node(\"generate\", generate)\n",
|
||
"\n",
|
||
"graph.add_edge(START, \"retrieve\")\n",
|
||
"graph.add_edge(\"retrieve\", \"generate\")\n",
|
||
"graph.add_edge(\"generate\", END)\n",
|
||
"\n",
|
||
"app = graph.compile()\n",
|
||
"result = app.invoke({\"query\":\"where can i use chatGPT?\"})\n",
|
||
"print(result['answer'])"
|
||
],
|
||
"id": "bcf8f539b91ab2b7",
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"Based on the provided context, ChatGPT can be used in the following areas:\n",
|
||
"\n",
|
||
"1. Education field: ChatGPT is commonly used for question and answer testing in the education sector. Students and learners can use ChatGPT to learn, compare, and verify answers for different academic subjects such as physics, mathematics, and chemistry.\n",
|
||
"\n",
|
||
"2. Introductory tasks: ChatGPT offers a responsive welcome program that maintains attackers' interest in multiple queries. This suggests it can be used for introductory or initial interactions.\n",
|
||
"\n",
|
||
"3. Code generation: The context mentions challenges with ChatGPT in the field of code generation. While it has limitations, it can potentially be used to assist with coding tasks, particularly in programming languages like Python, C++, and Java.\n",
|
||
"\n",
|
||
"However, the context also notes that ChatGPT's application scope is limited in code generation due to its training data being biased towards certain programming languages. So while it can be used for coding assistance, it may not be suitable for all programming languages or coding styles.\n",
|
||
"\n",
|
||
"In summary, based on the given context, ChatGPT can be used in education for question and answer testing across various subjects, for introductory interactions, and potentially for code generation assistance, with some limitations.\n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 18
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T01:18:58.885899699Z",
|
||
"start_time": "2026-06-09T01:18:54.451352107Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"# 기본 RAG + 평가\n",
|
||
"\n",
|
||
"# 기존의 벡터스토어에 질의\n",
|
||
"vectorstore = Chroma(collection_name=\"research\", embedding_function=watson_embedding, persist_directory=\"./db/chroma_db\")\n",
|
||
"\n",
|
||
"class RAGState(TypedDict):\n",
|
||
" query:str\n",
|
||
" retriever_docs:list[Document]\n",
|
||
" answer:str\n",
|
||
" is_relevant:bool\n",
|
||
" retry_count:int\n",
|
||
"\n",
|
||
"def retrieve(state):\n",
|
||
" docs = vectorstore.similarity_search(state['query'], k=3)\n",
|
||
"\n",
|
||
" return {\"retriever_docs\" : docs}\n",
|
||
"\n",
|
||
"def generate(state):\n",
|
||
" context = \"\\n\\n\".join(doc.page_content for doc in state['retriever_docs'])\n",
|
||
"\n",
|
||
" prompt = \"\"\"\\\n",
|
||
"다음 컨텍스트를 참고하여 질문에 답하세요.\n",
|
||
"컨텍스트에 없는 내용은 모른다고 답하세요.\n",
|
||
"\n",
|
||
"컨텍스트:\n",
|
||
"{context}\n",
|
||
"질문:\n",
|
||
"{query}\n",
|
||
"\"\"\"\n",
|
||
" response = watson_llm.invoke(prompt.format(context=context, query=state['query']))\n",
|
||
" return {\"answer\":response.content}\n",
|
||
"\n",
|
||
"def evaluate(state):\n",
|
||
" \"\"\"답변이 질문과 관린이 있는지 평가\"\"\"\n",
|
||
" prompt = ChatPromptTemplate.from_template(\"\"\"\\\n",
|
||
"질문:\n",
|
||
"{query}\n",
|
||
"\n",
|
||
"답변:\n",
|
||
"{answer}\n",
|
||
"\n",
|
||
"이 답변이 질문에 적절히 대답하고 있나요? 'yes' 또는 'no'로만 답하세요.\n",
|
||
"\"\"\")\n",
|
||
" response = watson_llm.invoke(prompt.format(query=state['query'], answer=state['answer']))\n",
|
||
" is_relevant = 'yes' in response.content.strip().lower()\n",
|
||
" return {\"is_relevant\": is_relevant, \"retry_count\": state['retry_count'] + 1}\n",
|
||
"\n",
|
||
"# route\n",
|
||
"def should_retry(state):\n",
|
||
" \"\"\"재검사 여부 결정\"\"\"\n",
|
||
" if state['is_relevant'] or state['retry_count'] > 2:\n",
|
||
" return \"done\"\n",
|
||
" return \"retry\"\n",
|
||
"\n",
|
||
"# 그래프 구성\n",
|
||
"graph = StateGraph(RAGState)\n",
|
||
"graph.add_node(\"retrieve\", retrieve)\n",
|
||
"graph.add_node(\"generate\", generate)\n",
|
||
"graph.add_node(\"evaluate\", evaluate)\n",
|
||
"\n",
|
||
"graph.add_edge(START, \"retrieve\")\n",
|
||
"graph.add_edge(\"retrieve\", \"generate\")\n",
|
||
"graph.add_edge(\"generate\", \"evaluate\")\n",
|
||
"graph.add_conditional_edges(\"evaluate\", should_retry, {\"done\": END, \"retry\": \"retrieve\"})\n",
|
||
"\n",
|
||
"app = graph.compile()\n",
|
||
"result = app.invoke({\"query\":\"where can i use chatGPT?\", \"retry_count\" : 0})\n",
|
||
"print(\"시도 횟수\", result['retry_count'])\n",
|
||
"print(\"검증통과\", result['is_relevant'])\n",
|
||
"print(\"최종답변\", result['answer'][:300])\n",
|
||
"\n",
|
||
"app.get_graph().print_ascii()"
|
||
],
|
||
"id": "c8ca9ace06e85ffd",
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"시도 횟수 1\n",
|
||
"검증통과 True\n",
|
||
"최종답변 Based on the provided context, ChatGPT can be used in the following areas:\n",
|
||
"\n",
|
||
"1. Education field: ChatGPT is commonly used for question and answer testing in the education sector. Students and learners can use ChatGPT to learn, compare, and verify answers for different academic subjects such as physic\n",
|
||
" +-----------+ \n",
|
||
" | __start__ | \n",
|
||
" +-----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +----------+ \n",
|
||
" | retrieve | \n",
|
||
" +----------+ \n",
|
||
" *** ... \n",
|
||
" * . \n",
|
||
" ** ... \n",
|
||
"+----------+ . \n",
|
||
"| generate | ... \n",
|
||
"+----------+ . \n",
|
||
" *** ... \n",
|
||
" * . \n",
|
||
" ** .. \n",
|
||
" +----------+ \n",
|
||
" | evaluate | \n",
|
||
" +----------+ \n",
|
||
" . \n",
|
||
" . \n",
|
||
" . \n",
|
||
" +---------+ \n",
|
||
" | __end__ | \n",
|
||
" +---------+ \n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 22
|
||
},
|
||
{
|
||
"metadata": {},
|
||
"cell_type": "markdown",
|
||
"source": [
|
||
"#### 검색 품질 개선\n",
|
||
"- Multi-Query : 여러 관점의 쿼리로 검색(모호한 질문, 넓은 검색 범위 필요)\n",
|
||
"- HyDE : 가상 문서를 생성하여 검색(쿼리와 문서의 표현 차이 큰 경우)\n",
|
||
"- Self-RAG : 답변을 평가하고 반복 개선"
|
||
],
|
||
"id": "e4fdf763188b08a1"
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T02:21:53.644411254Z",
|
||
"start_time": "2026-06-09T02:21:48.214822768Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"class MultiQueryState(TypedDict):\n",
|
||
" query: str\n",
|
||
" sub_queries:List[str]\n",
|
||
" retrieved_docs:List[Document]\n",
|
||
" answer: str\n",
|
||
"\n",
|
||
"def generate_sub_queries(state):\n",
|
||
" \"\"\"원본 질문을 여러 관점의 하위 쿼리로 분해\"\"\"\n",
|
||
" prompt = \"\"\"\\\n",
|
||
"다음 질문에 대해 서로 다른 관점으 검색 쿼리 3개를 생성하세요.\n",
|
||
"각 쿼리를 줄바꿈으로 구분하세요.\n",
|
||
"\n",
|
||
"원본 질문:\n",
|
||
"{query}\n",
|
||
"\"\"\"\n",
|
||
" response = watson_llm.invoke(prompt.format(query=state['query']))\n",
|
||
" #sub_query\n",
|
||
" sub_queries = [ q for q in response.content.strip().split(\"\\n\") if q.strip()]\n",
|
||
" print(f\"sub_queries: {sub_queries}\")\n",
|
||
"\n",
|
||
" return {\"sub_queries\": sub_queries}\n",
|
||
"\n",
|
||
"def multi_retrieve(state):\n",
|
||
" \"\"\"각 하위 쿼리로 검색하고 결과 합치기\"\"\"\n",
|
||
"\n",
|
||
" all_docs = []\n",
|
||
" seen_contents = set()\n",
|
||
"\n",
|
||
" for sub_query in state['sub_queries']:\n",
|
||
" docs = vectorstore.similarity_search(state['query'], k=3)\n",
|
||
"\n",
|
||
" for doc in docs:\n",
|
||
" if doc.page_content not in seen_contents:\n",
|
||
" all_docs.append(doc)\n",
|
||
" seen_contents.add(doc.page_content)\n",
|
||
"\n",
|
||
" return {\"retrieved_docs\":all_docs}\n",
|
||
"\n",
|
||
"def generate(state):\n",
|
||
" context = \"\\n\\n\".join(doc.page_content for doc in state['retrieved_docs'])\n",
|
||
"\n",
|
||
" prompt = \"\"\"\\\n",
|
||
"다음 컨텍스트를 참고하여 질문에 답하세요.\n",
|
||
"컨텍스트에 없는 내용은 모른다고 답하세요.\n",
|
||
"\n",
|
||
"컨텍스트:\n",
|
||
"{context}\n",
|
||
"질문:\n",
|
||
"{query}\n",
|
||
"\"\"\"\n",
|
||
" response = watson_llm.invoke(prompt.format(context=context, query=state['query']))\n",
|
||
" return {\"answer\":response.content}\n",
|
||
"\n",
|
||
"# 그래프 생성\n",
|
||
"graph = StateGraph(MultiQueryState)\n",
|
||
"graph.add_node(\"generate_sub_queries\", generate_sub_queries)\n",
|
||
"graph.add_node(\"multi_retrieve\", multi_retrieve)\n",
|
||
"graph.add_node(\"generate\", generate)\n",
|
||
"\n",
|
||
"graph.add_edge(START, \"generate_sub_queries\")\n",
|
||
"graph.add_edge(\"generate_sub_queries\", \"multi_retrieve\")\n",
|
||
"graph.add_edge(\"multi_retrieve\", \"generate\")\n",
|
||
"graph.add_edge(\"generate\", END)\n",
|
||
"\n",
|
||
"multiapp = graph.compile()\n",
|
||
"result = multiapp.invoke({\"query\":\"where can i use chatGPT?\"})\n",
|
||
"print(\"최종답변\", result['answer'][:300])\n",
|
||
"\n",
|
||
"multiapp.get_graph().print_ascii()"
|
||
],
|
||
"id": "4d64f9acabe6c986",
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"sub_queries: ['1. ChatGPT 사용 가능한 플랫폼 및 웹사이트는 어디인가요?', '2. ChatGPT를 사용할 수 있는 다양한 어플리케이션은 무엇인가요?', '3. ChatGPT를 활용할 수 있는 교육 및 연구 목적의 리소스는 어디에서 찾을 수 있나요?']\n",
|
||
"최종답변 제공된 컨텍스트에 따르면, ChatGPT는 다음과 같은 분야에서 사용될 수 있습니다:\n",
|
||
"\n",
|
||
"1. 교육 분야: 학생들은 ChatGPT를 사용하여 다양한 학문 분야(예: 물리학, 수학, 화학 등)에서 질문에 대한 답변을 찾고, 비교하며, 검증할 수 있습니다.\n",
|
||
"\n",
|
||
"2. 코드 생성 분야: ChatGPT는 코드 생성 작업에 사용될 수 있지만, 여전히 몇 가지 도전 과제가 있습니다. 예를 들어, 훈련 데이터가 Python, C++, Java와 같은 프로그래밍 언어에 편중되어 있어 일부 프로그래밍 언어나 코딩 스타일에는 적합하지 않을 수 있습니다.\n",
|
||
"\n",
|
||
" +-----------+ \n",
|
||
" | __start__ | \n",
|
||
" +-----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
"+----------------------+ \n",
|
||
"| generate_sub_queries | \n",
|
||
"+----------------------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +----------------+ \n",
|
||
" | multi_retrieve | \n",
|
||
" +----------------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +----------+ \n",
|
||
" | generate | \n",
|
||
" +----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +---------+ \n",
|
||
" | __end__ | \n",
|
||
" +---------+ \n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 34
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T02:21:59.900743028Z",
|
||
"start_time": "2026-06-09T02:21:53.645392234Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"# HyDE\n",
|
||
"# 질문 => 가상의 정답 문서를 먼저 생성, 가상의 문서를 검색어처럼 사용하여 더 관련성 높은 문서 찾음\n",
|
||
"\n",
|
||
"class HyDEState(TypedDict):\n",
|
||
" query: str\n",
|
||
" hypothetical_doc:str\n",
|
||
" retrieved_docs:List[Document]\n",
|
||
" answer: str\n",
|
||
"\n",
|
||
"def generate_hypothetical(state):\n",
|
||
" \"\"\"질문에 대한 가상의 답변 문서를 생성합니다.\"\"\"\n",
|
||
"\n",
|
||
" prompt = \"\"\"\\\n",
|
||
"다음 질문에 대한 답변이 될 만한 문서를 작성하세요.\n",
|
||
"실제 정확한 답변이 아니어도 됩니다. 관련 용어와 개념을 포함하면 됩니다.\n",
|
||
"\n",
|
||
"질문:\n",
|
||
"{query}\n",
|
||
"\"\"\"\n",
|
||
" response = watson_llm.invoke(prompt.format(query=state['query']))\n",
|
||
" return {\"hypothetical_doc\":response.content}\n",
|
||
"\n",
|
||
"def hyde_retrieve(state):\n",
|
||
" \"\"\"가상문서를 쿼리로 사용하여 검색\"\"\"\n",
|
||
" docs = vectorstore.similarity_search(state['hypothetical_doc'], k=3)\n",
|
||
" return {\"retriever_docs\" : docs}\n",
|
||
"\n",
|
||
"def generate(state):\n",
|
||
" context = \"\\n\\n\".join(doc.page_content for doc in state['retrieved_docs'])\n",
|
||
"\n",
|
||
" prompt = \"\"\"\\\n",
|
||
"다음 컨텍스트를 참고하여 질문에 답하세요.\n",
|
||
"컨텍스트에 없는 내용은 모른다고 답하세요.\n",
|
||
"\n",
|
||
"컨텍스트:\n",
|
||
"{context}\n",
|
||
"질문:\n",
|
||
"{query}\n",
|
||
"\"\"\"\n",
|
||
" response = watson_llm.invoke(prompt.format(context=context, query=state['query']))\n",
|
||
" return {\"answer\":response.content}\n",
|
||
"\n",
|
||
"graph = StateGraph(HyDEState)\n",
|
||
"graph.add_node(\"hypothetical\", generate_hypothetical)\n",
|
||
"graph.add_node(\"retrieve\", hyde_retrieve)\n",
|
||
"graph.add_node(\"generate\", generate)\n",
|
||
"\n",
|
||
"graph.add_edge(START, \"hypothetical\")\n",
|
||
"graph.add_edge(\"hypothetical\", \"retrieve\")\n",
|
||
"graph.add_edge(\"retrieve\", \"generate\")\n",
|
||
"graph.add_edge(\"generate\", END)\n",
|
||
"\n",
|
||
"hypeapp = graph.compile()\n",
|
||
"result = multiapp.invoke({\"query\":\"where can i use chatGPT?\"})\n",
|
||
"print(\"최종답변\", result['answer'][:300])\n",
|
||
"\n",
|
||
"hypeapp.get_graph().print_ascii()"
|
||
],
|
||
"id": "7347f9026a2646e3",
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"sub_queries: ['1. ChatGPT 사용 가능한 플랫폼 및 웹사이트는 어디인가요?', '2. ChatGPT를 사용할 수 있는 다양한 어플리케이션은 무엇인가요?', '3. ChatGPT를 활용할 수 있는 교육 및 연구 목적의 리소스는 어디에서 찾을 수 있나요?']\n",
|
||
"최종답변 제공된 컨텍스트에 따르면, ChatGPT는 다음과 같은 분야에서 사용될 수 있습니다:\n",
|
||
"\n",
|
||
"1. 교육 분야: 학생들은 ChatGPT를 사용하여 다양한 학문 분야(예: 물리학, 수학, 화학 등)에서 질문에 대한 답변을 찾고, 비교하며, 검증할 수 있습니다.\n",
|
||
"\n",
|
||
"2. 코드 생성 분야: ChatGPT는 코드 생성 작업에 사용될 수 있지만, 여전히 몇 가지 도전 과제가 있습니다. 예를 들어, 훈련 데이터가 Python, C++, Java와 같은 프로그래밍 언어에 편중되어 있어 일부 프로그래밍 언어나 코딩 스타일에는 적합하지 않을 수 있습니다.\n",
|
||
"\n",
|
||
" +-----------+ \n",
|
||
" | __start__ | \n",
|
||
" +-----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
"+--------------+ \n",
|
||
"| hypothetical | \n",
|
||
"+--------------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +----------+ \n",
|
||
" | retrieve | \n",
|
||
" +----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +----------+ \n",
|
||
" | generate | \n",
|
||
" +----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +---------+ \n",
|
||
" | __end__ | \n",
|
||
" +---------+ \n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 35
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T03:15:30.741669953Z",
|
||
"start_time": "2026-06-09T03:15:16.835071132Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"# self refine rag\n",
|
||
"# 답변을 생성=> 평가=> 재검색 or 답변수정\n",
|
||
"\n",
|
||
"class SelfRefineRAGState(TypedDict):\n",
|
||
" query:str\n",
|
||
" retrieved_docs:list[Document]\n",
|
||
" answer:str\n",
|
||
" evaluation:str\n",
|
||
" retry_count:int\n",
|
||
"\n",
|
||
"def retrieve(state):\n",
|
||
" docs = vectorstore.similarity_search(state['query'], k=3)\n",
|
||
" return {\"retrieved_docs\" : docs}\n",
|
||
"\n",
|
||
"def generate(state):\n",
|
||
" context = \"\\n\\n\".join(doc.page_content for doc in state['retrieved_docs'])\n",
|
||
"\n",
|
||
" prompt = \"\"\"\\\n",
|
||
"다음 컨텍스트를 참고하여 질문에 답하세요.\n",
|
||
"컨텍스트에 없는 내용은 모른다고 답하세요.\n",
|
||
"컨텍스트에 정보가 부족하면 그 사실을 명시하세요.\n",
|
||
"\n",
|
||
"컨텍스트:\n",
|
||
"{context}\n",
|
||
"질문:\n",
|
||
"{query}\n",
|
||
"\"\"\"\n",
|
||
" response = watson_llm.invoke(prompt.format(context=context, query=state['query']))\n",
|
||
" return {\"answer\":response.content}\n",
|
||
"\n",
|
||
"def evaluate(state):\n",
|
||
" \"\"\"생성한 답변이 답변의 충실도와 관련성을 평가\"\"\"\n",
|
||
"\n",
|
||
" context = \"\\n\\n\".join(doc.page_content for doc in state['retrieved_docs'])\n",
|
||
"\n",
|
||
" prompt = ChatPromptTemplate.from_template(\"\"\"\\\n",
|
||
"질문:\n",
|
||
"{query}\n",
|
||
"\n",
|
||
"컨텍스트:\n",
|
||
"{context}\n",
|
||
"\n",
|
||
"답변:\n",
|
||
"{answer}\n",
|
||
"\n",
|
||
"반드시 아래 둘 중 하나로만 출력하세요.\n",
|
||
"'sufficient'\n",
|
||
"'insufficient'\n",
|
||
"\"\"\")\n",
|
||
" response = watson_llm.invoke(prompt.format(query=state['query'], answer=state['answer'], context = context))\n",
|
||
"\n",
|
||
" content = response.content.lower().strip()\n",
|
||
" evaluation = \"insufficient\" if content.startswith('insufficient') else \"sufficient\"\n",
|
||
"\n",
|
||
" return {\"evaluation\": evaluation}\n",
|
||
"\n",
|
||
"def refine(state):\n",
|
||
" \"\"\"평가결과를 반영하여 답변을 개선합니다.\"\"\"\n",
|
||
" context = \"\\n\\n\".join(doc.page_content for doc in state['retrieved_docs'])\n",
|
||
"\n",
|
||
" prompt = \"\"\"\\\n",
|
||
"다음 답변을 개선하세요.\n",
|
||
"원래 질문 : {query}\n",
|
||
"컨텍스트 : {context}\n",
|
||
"이전 답변 : {answer}\n",
|
||
"컨텍스트에 더 충실하고 질문에 더 정확히 답하도록 수정하세요.\n",
|
||
"\"\"\"\n",
|
||
" response = watson_llm.invoke(prompt.format(context=context, query=state['query'], answer=state['answer']))\n",
|
||
" return {\"answer\":response.content, \"retry_count\": state['retry_count'] + 1}\n",
|
||
"\n",
|
||
"# route\n",
|
||
"def route_after_eval(state):\n",
|
||
" \"\"\"재검사 여부 결정\"\"\"\n",
|
||
" if state['evaluation'] == \"sufficient\" or state['retry_count'] > 2:\n",
|
||
" return \"done\"\n",
|
||
" return \"refine\"\n",
|
||
"\n",
|
||
"# 그래프 구성\n",
|
||
"graph = StateGraph(SelfRefineRAGState)\n",
|
||
"graph.add_node(\"retrieve\", retrieve)\n",
|
||
"graph.add_node(\"generate\", generate)\n",
|
||
"graph.add_node(\"evaluate\", evaluate)\n",
|
||
"graph.add_node(\"refine\", refine)\n",
|
||
"\n",
|
||
"graph.add_edge(START, \"retrieve\")\n",
|
||
"graph.add_edge(\"retrieve\", \"generate\")\n",
|
||
"graph.add_edge(\"generate\", \"evaluate\")\n",
|
||
"graph.add_conditional_edges(\"evaluate\", route_after_eval, {\"done\": END, \"refine\": \"refine\"})\n",
|
||
"graph.add_edge(\"refine\", \"evaluate\")\n",
|
||
"\n",
|
||
"app = graph.compile()\n",
|
||
"result = app.invoke({\"query\":\"where can i use chatGPT?\", \"retry_count\" : 0})\n",
|
||
"print(result['answer'])\n",
|
||
"\n",
|
||
"app.get_graph().print_ascii()"
|
||
],
|
||
"id": "652cdb1961c74fbc",
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"Based on the provided context, ChatGPT can be used in the following ways:\n",
|
||
"\n",
|
||
"1. Education field: ChatGPT is commonly used for question and answer testing in the education sector. Students and users can use ChatGPT to learn, compare, and verify answers for different academic subjects such as physics, mathematics, and chemistry.\n",
|
||
"\n",
|
||
"2. Code generation: ChatGPT can be applied to code generation tasks. However, there are some challenges and limitations to consider:\n",
|
||
" - Limited application scope: ChatGPT's training data is biased towards programming languages like Python, C++, and Java. This makes it potentially unsuitable for some programming languages or coding styles that are not well-represented in its training data.\n",
|
||
" - Other challenges: The context mentions that there are still several challenges with using ChatGPT for code generation, but it does not provide specific details about these challenges.\n",
|
||
"\n",
|
||
"3. Introductory tasks: ChatGPT offers a responsive welcome program that maintains attackers' interest in multiple queries. This suggests it can be used for introductory or initial interactions, possibly in the context of security or penetration testing.\n",
|
||
"\n",
|
||
"The context does not mention any other specific domains or applications where ChatGPT can be used. It primarily focuses on the education field and code generation, with a brief mention of introductory tasks. Therefore, based on the given information, ChatGPT's main applications seem to be in the education sector for question and answer testing, code generation (with some limitations), and introductory interactions.\n",
|
||
" +-----------+ \n",
|
||
" | __start__ | \n",
|
||
" +-----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +----------+ \n",
|
||
" | retrieve | \n",
|
||
" +----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +----------+ \n",
|
||
" | generate | \n",
|
||
" +----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +----------+ \n",
|
||
" | evaluate | \n",
|
||
" +----------+ \n",
|
||
" .. .. \n",
|
||
" .. .. \n",
|
||
" . . \n",
|
||
"+---------+ +--------+ \n",
|
||
"| __end__ | | refine | \n",
|
||
"+---------+ +--------+ \n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 53
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T03:15:53.532912899Z",
|
||
"start_time": "2026-06-09T03:15:42.384644440Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"# self rag\n",
|
||
"# 답변을 생성=> 평가=> 부족 => 질문 개선 => 검색 => 새문서 답변\n",
|
||
"\n",
|
||
"class SelfRAGState(TypedDict):\n",
|
||
" query:str\n",
|
||
" retrieved_docs:list[Document]\n",
|
||
" answer:str\n",
|
||
" evaluation:str\n",
|
||
" retry_count:int\n",
|
||
"\n",
|
||
"def retrieve(state):\n",
|
||
" docs = vectorstore.similarity_search(state['query'], k=3)\n",
|
||
" return {\"retrieved_docs\" : docs}\n",
|
||
"\n",
|
||
"def generate(state):\n",
|
||
" context = \"\\n\\n\".join(doc.page_content for doc in state['retrieved_docs'])\n",
|
||
"\n",
|
||
" prompt = \"\"\"\\\n",
|
||
"다음 컨텍스트를 참고하여 질문에 답하세요.\n",
|
||
"컨텍스트에 없는 내용은 모른다고 답하세요.\n",
|
||
"컨텍스트에 정보가 부족하면 그 사실을 명시하세요.\n",
|
||
"\n",
|
||
"컨텍스트:\n",
|
||
"{context}\n",
|
||
"질문:\n",
|
||
"{query}\n",
|
||
"\"\"\"\n",
|
||
" response = watson_llm.invoke(prompt.format(context=context, query=state['query']))\n",
|
||
" return {\"answer\":response.content}\n",
|
||
"\n",
|
||
"def evaluate(state):\n",
|
||
" \"\"\"생성한 답변이 답변의 충실도와 관련성을 평가\"\"\"\n",
|
||
"\n",
|
||
" context = \"\\n\\n\".join(doc.page_content for doc in state['retrieved_docs'])\n",
|
||
"\n",
|
||
" prompt = ChatPromptTemplate.from_template(\"\"\"\\\n",
|
||
"질문:\n",
|
||
"{query}\n",
|
||
"\n",
|
||
"컨텍스트:\n",
|
||
"{context}\n",
|
||
"\n",
|
||
"답변:\n",
|
||
"{answer}\n",
|
||
"\n",
|
||
"반드시 아래 둘 중 하나로만 출력하세요.\n",
|
||
"'sufficient'\n",
|
||
"'insufficient'\n",
|
||
"\"\"\")\n",
|
||
" response = watson_llm.invoke(prompt.format(query=state['query'], answer=state['answer'], context = context))\n",
|
||
"\n",
|
||
" content = response.content.lower().strip()\n",
|
||
" evaluation = \"insufficient\" if content.startswith('insufficient') else \"sufficient\"\n",
|
||
"\n",
|
||
" return {\"evaluation\": evaluation}\n",
|
||
"\n",
|
||
"def rewrite_query(state):\n",
|
||
" \"\"\"평가결과를 반영하여 질문을 개선합니다.\"\"\"\n",
|
||
"\n",
|
||
" prompt = \"\"\"\\\n",
|
||
"다음 질문을 개선하세요.\n",
|
||
"원래 질문 : {query}\n",
|
||
"검색 결과가 충분하지 않았습니다.\n",
|
||
"더 구체적이고 검색하기 좋은 질문으로 재작성하세요.\n",
|
||
"질문만 출력하세요.\n",
|
||
"\"\"\"\n",
|
||
" response = watson_llm.invoke(prompt.format(query=state['query']))\n",
|
||
" return {\"query\":response.content, \"retry_count\": state['retry_count'] + 1}\n",
|
||
"\n",
|
||
"# route\n",
|
||
"def route_after_eval(state):\n",
|
||
" \"\"\"재검사 여부 결정\"\"\"\n",
|
||
" if state['evaluation'] == \"sufficient\" or state['retry_count'] > 2:\n",
|
||
" return \"done\"\n",
|
||
" return \"retry\"\n",
|
||
"\n",
|
||
"# 그래프 구성\n",
|
||
"graph = StateGraph(SelfRAGState)\n",
|
||
"graph.add_node(\"retrieve\", retrieve)\n",
|
||
"graph.add_node(\"generate\", generate)\n",
|
||
"graph.add_node(\"evaluate\", evaluate)\n",
|
||
"graph.add_node(\"rewrite\", rewrite_query)\n",
|
||
"\n",
|
||
"graph.add_edge(START, \"retrieve\")\n",
|
||
"graph.add_edge(\"retrieve\", \"generate\")\n",
|
||
"graph.add_edge(\"generate\", \"evaluate\")\n",
|
||
"graph.add_conditional_edges(\"evaluate\", route_after_eval, {\"done\": END, \"retry\": \"rewrite\"})\n",
|
||
"graph.add_edge(\"rewrite\", \"retrieve\")\n",
|
||
"\n",
|
||
"app = graph.compile()\n",
|
||
"result = app.invoke({\"query\":\"where can i use chatGPT?\", \"retry_count\" : 0})\n",
|
||
"print(result['answer'])\n",
|
||
"\n",
|
||
"app.get_graph().print_ascii()"
|
||
],
|
||
"id": "b3760850711dad9d",
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"제공된 컨텍스트에 따르면, ChatGPT는 다음과 같은 구체적인 예시에서 사용될 수 있습니다:\n",
|
||
"\n",
|
||
"1. **대화형 대화 생성**: Chen et al. [25]는 씬, 타임라인, 캐릭터 속성, 캐릭터 관계를 포함하는 대화 데이터셋(HPD)을 구축하여 ChatGPT를 대화형 에이전트로 사용하여 대화를 생성했습니다. 이는 영화, 게임, 가상 현실 등에서 사용자와의 상호작용을 위한 대화 시스템을 개발하는 데 활용될 수 있습니다.\n",
|
||
"\n",
|
||
"2. **복잡한 텍스트 단순화**: Study [26]에서 언급된 바와 같이, ChatGPT는 복잡한 텍스트를 단순화하는 데 사용될 수 있습니다. 예를 들어, 교육 분야에서 학생들이 이해하기 어려운 과학적 개념이나 역사적 사건을 더 쉽게 이해할 수 있도록 설명을 단순화하는 데 활용될 수 있습니다.\n",
|
||
"\n",
|
||
"3. **플래그시즘 및 사기 방지**: 제공된 컨텍스트의 마지막 부분에서 언급된 것처럼, ChatGPT는 플래그시즘 및 사기 방지를 위한 도구로 사용될 수 있습니다. 예를 들어, 학술 연구나 글쓰기 과정에서 학생이나 연구원이 자신의 작업이 원래의 것인지 확인하거나, 텍스트가 다른 출처에서 복사되었는지 확인하는 데 도움을 줄 수 있습니다.\n",
|
||
"\n",
|
||
"이러한 예시들은 ChatGPT가 다양한 분야에서 어떻게 활용될 수 있는지를 보여줍니다. 그러나 ChatGPT의 성능은 여전히 개선의 여지가 있으며, 특히 복잡한 추론 작업에서는 한계가 있음을 인식해야 합니다.\n",
|
||
" +-----------+ \n",
|
||
" | __start__ | \n",
|
||
" +-----------+ \n",
|
||
" * \n",
|
||
" * \n",
|
||
" * \n",
|
||
" +----------+ \n",
|
||
" | retrieve | \n",
|
||
" +----------+ \n",
|
||
" *** *** \n",
|
||
" * * \n",
|
||
" ** *** \n",
|
||
"+----------+ * \n",
|
||
"| generate | * \n",
|
||
"+----------+ * \n",
|
||
" * * \n",
|
||
" * * \n",
|
||
" * * \n",
|
||
"+----------+ * \n",
|
||
"| evaluate | * \n",
|
||
"+----------+.... * \n",
|
||
" . .... * \n",
|
||
" . .... * \n",
|
||
" . .. * \n",
|
||
"+---------+ +---------+ \n",
|
||
"| __end__ | | rewrite | \n",
|
||
"+---------+ +---------+ \n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 54
|
||
},
|
||
{
|
||
"metadata": {},
|
||
"cell_type": "markdown",
|
||
"source": [
|
||
"#### LLM 애플리케이션 성능 최적화 전략\n",
|
||
"- 1. 비용절감\n",
|
||
" - InMemoryCache, SQLiteCache 모델 경량화\n",
|
||
"- 2. 응답속도 향상\n",
|
||
" - RunnableParallel, batch, Streaming\n",
|
||
"- 3. 처리량 향상\n",
|
||
" - ainvoke(), asyncio.gather()"
|
||
],
|
||
"id": "20ac68ee3da61a0"
|
||
},
|
||
{
|
||
"metadata": {
|
||
"ExecuteTime": {
|
||
"end_time": "2026-06-09T05:50:13.225328876Z",
|
||
"start_time": "2026-06-09T05:49:57.712136496Z"
|
||
}
|
||
},
|
||
"cell_type": "code",
|
||
"source": [
|
||
"from langchain_core.globals import set_llm_cache\n",
|
||
"from langchain_community.cache import InMemoryCache, SQLiteCache\n",
|
||
"import time\n",
|
||
"\n",
|
||
"text = \"파이썬이란 무엇인가요?\"\n",
|
||
"\n",
|
||
"print(\"====== 캐시 없음 ======\")\n",
|
||
"set_llm_cache(None)\n",
|
||
"\n",
|
||
"start = time.time()\n",
|
||
"r1 = watson_llm.invoke(text)\n",
|
||
"print(f\"1회차 : {time.time()-start:.2f}초\")\n",
|
||
"\n",
|
||
"start = time.time()\n",
|
||
"r1 = watson_llm.invoke(text)\n",
|
||
"print(f\"2회차 : {time.time()-start:.2f}초\")\n",
|
||
"\n",
|
||
"# InMemoryCache\n",
|
||
"print(\"===== InMemoryCache =====\")\n",
|
||
"set_llm_cache(InMemoryCache())\n",
|
||
"\n",
|
||
"start = time.time()\n",
|
||
"r2 = watson_llm.invoke(text)\n",
|
||
"print(f\"1회차 : {time.time()-start:.2f}초\")\n",
|
||
"\n",
|
||
"start = time.time()\n",
|
||
"r2 = watson_llm.invoke(text)\n",
|
||
"print(f\"2회차 : {time.time()-start:.2f}초\")\n",
|
||
"\n",
|
||
"print(\"===== SQLiteCache =====\")\n",
|
||
"set_llm_cache(SQLiteCache(database_path=\"./db/llm_cache.db\"))\n",
|
||
"\n",
|
||
"start = time.time()\n",
|
||
"r3 = watson_llm.invoke(\"LangChain 이란?\")\n",
|
||
"print(f\"1회차 : {time.time()-start:.2f}초\")\n",
|
||
"\n",
|
||
"start = time.time()\n",
|
||
"r3 = watson_llm.invoke(\"LangChain 이란?\")\n",
|
||
"print(f\"2회차 : {time.time()-start:.2f}초\")"
|
||
],
|
||
"id": "1e58af44608d5433",
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"====== 캐시 없음 ======\n",
|
||
"1회차 : 4.90초\n",
|
||
"2회차 : 2.77초\n",
|
||
"===== InMemoryCache =====\n",
|
||
"1회차 : 3.07초\n",
|
||
"2회차 : 0.00초\n",
|
||
"===== SQLiteCache =====\n",
|
||
"1회차 : 2.64초\n",
|
||
"2회차 : 0.00초\n"
|
||
]
|
||
},
|
||
{
|
||
"name": "stderr",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"/home/cooney/Source/.venv/lib/python3.12/site-packages/langchain_community/cache.py:265: LangChainPendingDeprecationWarning: The default value of `allowed_objects` will change in a future version. Pass an explicit list of allowed classes (or 'messages' for untrusted input that contains only chat messages) to suppress this warning.\n",
|
||
" return [loads(row[0]) for row in rows]\n"
|
||
]
|
||
}
|
||
],
|
||
"execution_count": 55
|
||
},
|
||
{
|
||
"metadata": {},
|
||
"cell_type": "code",
|
||
"outputs": [],
|
||
"execution_count": null,
|
||
"source": "",
|
||
"id": "caf04db673a45806"
|
||
}
|
||
],
|
||
"metadata": {
|
||
"kernelspec": {
|
||
"display_name": "Python 3",
|
||
"language": "python",
|
||
"name": "python3"
|
||
},
|
||
"language_info": {
|
||
"codemirror_mode": {
|
||
"name": "ipython",
|
||
"version": 2
|
||
},
|
||
"file_extension": ".py",
|
||
"mimetype": "text/x-python",
|
||
"name": "python",
|
||
"nbconvert_exporter": "python",
|
||
"pygments_lexer": "ipython2",
|
||
"version": "2.7.6"
|
||
}
|
||
},
|
||
"nbformat": 4,
|
||
"nbformat_minor": 5
|
||
}
|