Activity (Process)

Training the Value Function with a Reward Model

In Reinforcement Learning from Human Feedback (RLHF) and similar actor-critic alignment frameworks, the training of the value function depends on a separate, pre-trained reward model. Because the environment does not provide explicit intrinsic rewards, this reward model provides the essential reward signal, rtr_t, which serves as the basis for computing the value function's learning target, such as within the Temporal Difference (TD) error calculation.

0

1

Updated 2026-06-27

Contributors are:

Who are from:

Tags

Ch.4 Alignment - Foundations of Large Language Models

Foundations of Large Language Models

Foundations of Large Language Models Course

Computing Sciences