hackertube.io

ChatGPT with Rob Miles - Computerphile

Computerphile · interview · Watch on YouTube

LLMs & generative AIAI safety, AGI & impact

What it is

Computerphile interviews AI safety researcher Rob Miles about ChatGPT: how it was trained with human feedback, why it tends to produce confident nonsense, and where it may or may not be dangerous.

Why Hacker News recommends it

HN commenters send it to people confused about ChatGPT's quirks, explaining how human-feedback training makes it a pleasing yes-man that bullshits.

  • 4 people recommended it
  • 2 different months
  • 0 strong recommendations
  • first 2023, last 2023

What HN commenters said

  1. “This video[1] cleared up a lot of questions for me as to the issues I've seen with ChatGPT.”

    Explains ChatGPT's quirks and training incentives to bullshit.

    mikewarot, 2023-02 · Recommendation · in “Ask HN: Are ChatGPT answers getting worse for anyone else?”

  2. “For context I recommend watching the discussion on what chatgpt is/does[0]”

    Context on what ChatGPT is and does as a tuned assistant.

    f_devd, 2023-02 · Recommendation · in “Open Assistant: Conversational AI for Everyone”

  3. “This video explains the approach and some of its limitations.”

    Explains the human-feedback training approach and its limitations.

    simonh, 2023-02 · Recommendation · in “The new Bing will happily give you citations for a pile of nonsense”

  4. “Here is a good video about that”

    Explains misalignment between user-pleasing training and correct output.

    fxj, 2023-04 · Recommendation · in “What GPT-4 Does Is Less Like “Figuring Out” and More Like “Already Knowing””

Excerpts from Hacker News comments, linked to the originals. Summaries and labels were written by a language model from those comments.

Topics: chatgpt, rlhf, ai safety, language models, rob miles

Also recommended by the same people