hackertube.io

Training AI Without Writing A Reward Function, with Reward Modelling

Robert Miles AI Safety · explainer · Watch on YouTube

AI safety, AGI & impactMachine learning & deep learning

What it is

Robert Miles explains reward modelling, in which an AI learns what humans want from their preference comparisons instead of a hand-written reward function. He walks through the 2017 paper on deep reinforcement learning from human preferences.

Why Hacker News recommends it

Seen as the clearest introduction to the idea behind RLHF, and also borrowed by commenters as a metaphor for human motivation.

  • 3 people recommended it
  • 3 different months
  • 2 strong recommendations
  • first 2019, last 2023

What HN commenters said

  1. “I would highly recommend as a way of framing the problem.”

    Frames ML problem-solving options like human feedback.

    sillysaurusx, 2019-12 · Strong recommendation · in “Autonomous DeLorean drives sideways to move forward”

  2. “I find this video by Rob Miles still the best presentation of the ingenious original 2017(!) paper”

    Best presentation of the original 2017 RLHF paper.

    cubefox, 2023-07 · Strong recommendation · in “What We Know About LLMs (A Primer)”

  3. “To really understand what I'm getting at here, watch this video”

    Reward model concept applied to understanding personal motivation.

    afarrell, 2020-07 · Recommendation · in “Ask HN: How do I get out of the procrastination abyss?”

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

Topics: rlhf, reinforcement learning, ai safety, reward function, human preferences

Also recommended by the same people