Skip to content

Latest commit

 

History

History
191 lines (115 loc) · 9.21 KB

README.md

File metadata and controls

191 lines (115 loc) · 9.21 KB
Raw

discuss-chatgpt

Hits

Rationale

GPT3 consists of 175 billion learnable parameters ChatGPT is of comparable size to GPT3 (OpenAI, 2022)
image image

ChatGPT is of comparable size to GPT3. Thus, in the regime of the scaling laws for neural language models (Kaplan et al., 2020), it is reasonable to assume that their overall performance should be more or less the same.

GPT3 (text-danvinci-003) and ChatGPT return more or less the same results
Prompt: Write a 300-word value-packed book summary of the book “The 4-Hour Workweek”
image
Prompt: If I could read only 10 self-help books in 2023, which ones would you suggest are the best?
image
Prompt: Write an 800-word Statement Of Purpose to apply to the University Of Texas for a Master’s in I/O Psychology
image image
ChatGPT Sprints to One Million Users (Statista, 2023)
image

And yet, ChatGPT's userbase grew in an unprecedented rate. It attracted 1 million users in merely 5 days. GPT3's attraction was nowhere near this when it was released to the public 3 years ago.

Question

This begs the question:

❓사람들은 왜 ChatGPT에 열광하는가?

GPT3나 ChatGPT나 성능은 비슷하다. ChatGPT는 무엇이 다르길래 사람들이 열광하는가?

Task

팀 별로 토의 후 위 질문에 답해주세요. 팀에서 도출한 결론을 발표해주시면 됩니다. 대화형 검색의 관점도 좋고, 심리학의 관점도 좋고, 경제학의 관점도 좋습니다. 다양한 관점에서 자유롭게 사고하여 팀별로 설득력있는 주장을 펼쳐주세요.

⚠️ 뇌피셜을 지양하기 위해 반드시 직/간접적 데이터를 (논문, 통계, 인터뷰, 이론 등) 인용해 주장을 뒷받침해주세요.

Discussions

Course leaders - HCI / ethics / productivity

13:51 ~ 26:48 🗣️ Eu-Bin KIM
image

HCI / ethics / productivity

  • Why is ChatGPT so successful?
  • Shneiderman’s eight “golden rules” of interface design — a human-computer interaction perspective
  • But why did Tay / 이루다(2021) fail? — ethics and productivity

contributors:

Ha Hyeon Choi Eu-Bin KIM
image image

Team 1️⃣ - Diffusion of innovations

1:16:21 ~ 1:24:26 🗣️ Marc Zen
image

Why the heck did ChatGPT succeed?

  • Diffusion of Innovations
    • Perceived Advantage
    • Compatibility with existing values and practices
    • We are comfortable with text
    • Simple to use
    • Trialability
    • Observability
  • Then, why not GPT3?
    • UX!
    • Affordance
  • GenAI에 대한 Hype이 옮겨붙었다?

contributors:

강병준 (Marc Zen) 오효민 정순호 조주환
image image image image

Team 2️⃣ - marketing and the product itself

27:32 ~ 40:28 🗣️ Minyoung Kang
image

Slides for <DiscussGPT> Why ChatGPT?

<왜 사람들은 ChatGPT에 열광하는가>

  1. GPT3 vs ChatGPT
  2. 성공요인 1: 마케팅적 우위
  3. 성공요인 2: 제품적 우위
  • 데이터 타입: 이미지 vs 텍스트
  • GPT3, GPT3.5, ChatGPT 기술적 성능 차이 - PPO
  • “Chat”이라는 핵심에 집중한 제품적 우위(코딩 및 교육)
  1. 부록: <제 2의 기계시대>

contributors:

Minyoung Kang Sungkyung Kim Yuri Kim Hyunjoon-Kim JunYoung Park
image image image image image

Team 3️⃣ - interface adoption

43:40 ~ 51:54 🗣️ Yujong Lee
image

The success of ChatGPT: the era of interface adoption

  1. Ready from day1
  2. Expectation gap
  3. Intellectual, creative task
  4. Message interface

contributors:

이유종 박윤아 박준우 오주상 오채영
image image image image image

Team 4️⃣ - analysis / performance / future

54:40 ~ 1:13:05 🗣️ Ki Sang Lee
image

"Why ChatGPT?"

  1. Why ChatGPT?
    1. Analysis
    2. Performance
  2. Future LLMs (Gen.AI)
    1. Products
    2. Impacts
    3. Key Factors

contributors:

이아담 이기상 김용환 김태형
image image image image

Bringing it all together

Everything goes 에 그치지 말고 좀 더 고차원적인 사고를 해보자. 그래서 결론은? 모든 관점과 데이터를 관통하는 하나의 결론이 있다면 무엇인가?

(🚧 공사중 ... )