Also known as: Reinforcement Learning from Human Feedback (RLHF) · Reinforcement Learning From Human Feedback (RLHF) · RLHF
Reinforcement Learning from Human Feedback (RLHF) — техника alignment поведения модели с использованием человеческих предпочтений или ratings. Обычно человеческие judgments используются для обучения reward model, после чего reinforcement learning оптимизирует generative model так, чтобы ее outputs лучше соответствовали этим preferences.
A technique that involve training a "reward model" to predict how humans rate the quality of generated content, and then training a generative generative AI model to satisfy this reward model via reinforcement learning. It can be used for example to make the generative AI model more truthful or less harmful.
Using feedback from human raters to improve the quality of a model's responses. For example, an RLHF mechanism can ask users to rate the quality of a model's response with a or emoji. The system can then adjust its future responses based on that feedback.