Reinforcement Learning from Human Feedback Basics

Home
GitHub Repository
PDF
Order a copy (Soon)

Introductions

Introduction
What are preferences?
Optimization and RL
Seminal (Recent) Works

Problem Setup

Definitions
Preference Data
Reward Modeling
Regularization

Optimization

Instruction Tuning
Rejection Sampling
Policy Gradients
Direct Alignment Algorithms

Advanced (TBD)

Constitutional AI
Synthetic Data
Evaluation

Open Questions (TBD)

Over-optimization
Style

Chapter Contents

Direct Alignment Algorithms

Direct Alignment Algorithms

Citation:
@book{rlhf2024,
  author = {Nathan Lambert},
  title = {Reinforcement Learning from Human Feedback},
  year = {2024},
  publisher = {Online},
  url = {https://rlhfbook.com},
}