Rule-based reward model
A language model used as a judge that checks an answer against a written rubric, such as "a refusal in the desired style", and gives a reward during reinforcement learning.
Introduced by
GPT-4, 2023
A language model used as a judge that checks an answer against a written rubric, such as "a refusal in the desired style", and gives a reward during reinforcement learning.
GPT-4, 2023