Sign inSign in
ConceptAI & ML

Reward model

A separate neural network trained to imitate human preferences and scores for model rollouts.
as Andrej Karpathy defines itThe wording closest to all 6 definitions
Watch it explained

Know first

Named as a prerequisite by the lectures that define this. The number is how many say so. A grey one has no page yet.

Opens up

Concepts that name this one as a prerequisite.

Lecturers

5

Each in their own terms, from the easiest level.

  • Matt Jones

    Gresham College · Introductory

    A model built up from human rankings of possible outputs to nudge post-training mathematics toward more agreeable responses.
    • 21:44AI as Therapist? Intimacy Without Understanding
  • Andrej Karpathy

    Undergraduate

    • 1:53:32Let's Build GPT: From Scratch, in Code, Spelled Out
  • Emma Brunskill

    Stanford University · Undergraduate

    A function that defines the scalar feedback signal received by an agent for taking a specific action in a specific state.
    • 42:00Introduction to Reinforcement Learning (CS234, 2024, Lecture 1)
  • Jacob Andreas

    MIT · Undergraduate

    A model trained to reproduce human annotators' judgments about which outputs are good and which are bad.
    • 1:13:03Language Models: Prompting, Chain-of-Thought, and Instruction-Tuning
  • Sergey Levine

    UC Berkeley · Graduate

    A neural network function used to score any possible completion or answer for a given prompt in an open-vocabulary setting.
    • 12:23RL with Sequence Models & Language Models, Part 2

Lectures that assume it

1

Connections

17

The lectures that define it or assume it, and the concepts next to it.

5 lecturers. Definitions are written by AI from each lecture’s transcript. A time with a play mark starts the lecture at the moment the lecturer says it.