Sign inSign in
ConceptAI & ML

Reward hacking

A phenomenon where the specified reward does not provide the behavior that was really hoped to be achieved.
as Emma Brunskill defines itStanford University
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.

Language model1Reinforcement learning1Reinforcement learning from verifiable rewards1Reward model1Training and test reward1

Opens up

Concepts that name this one as a prerequisite.

Lecturers

2

Each in their own terms, from the easiest level.

  • Xi Ye

    Alberta Machine Intelligence Institute (Amii) · Graduate

    A phenomenon where language models achieve high rewards by exploiting shortcuts or unintended loopholes in training environments rather than solving the actual tasks.
    • 30:00AI Seminar: Actionable Interpretability for Language Model Reasoning

Connections

6

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

2 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.