---
aliases:
  en:
  - RLHF
  - reinforcement learning with human feedback
concept_id: reinforcement_learning_from_human_feedback
preferred:
  en: reinforcement learning from human feedback
  ja: 人間のフィードバックによる強化学習
relations:
  relations:
    broader:
    - reinforcement_learning
    prerequisite:
    - human_preferences
    - reinforcement_learning
    - reward
status: active
---

# reinforcement learning from human feedback

A family of reinforcement-learning-based training methods that use human judgments, commonly pairwise preferences or ratings, to define or learn a reward signal and then optimize a model or policy using that signal; particular pipelines may also include supervised fine-tuning and a learned reward model.

## broader

- [reinforcement learning](../terminology/reinforcement_learning.txt)

## prerequisite

- [human preferences](../terminology/human_preferences.txt)
- [reinforcement learning](../terminology/reinforcement_learning.txt)
- [reward](../terminology/reward.txt)

## Documents: about

- [チャット形式の安全性訓練で望ましく振る舞うことは、エージェント的な状況での安全性を保証しない](../documents/macdiarmid_et_al.chat-alignment-does-not-establish-agentic-alignment.txt)

## Documents: mentions

- [望ましい振る舞いは人間を重視する目標の証拠ではない](../documents/akira.behavioral-alignment-is-not-goal-alignment.txt)
- [訓練中に従うようになっただけでは、競合する選好が置き換わったとは言えない](../documents/greenblatt_et_al.training-compliance-does-not-establish-preference-change.txt)
- [行動訓練で望ましく振る舞っても、意図した目標の頑健な汎化は保証されない](../documents/ngo_et_al.behavioral-training-does-not-establish-robust-goal-generalization.txt)

[Corpus index](../index.txt)