01
Alignment2022
Constitutional AI: Harmlessness from AI Feedback
Bai, Kadavath, et al. — Anthropic
Alignment without an army of human labelers. Worth reading for the governance pattern alone.
Original paper ↗
Summary · Glossary
Read the papers. Skip the jargon.
The white papers that actually moved the field, each linked to the original and paired with my plain-language summary, a glossary of the terms, and what it changes in practice.
02
Alignment2023
Direct Preference Optimization
Rafailov, Sharma, et al. — Stanford
RLHF without the RL. The math is dense; the summary isn’t. If you fine-tune in production, this is the one to understand.
Original paper ↗
Summary
All original papers remain the work of their authors and link to the official source (arXiv or publisher). Summaries and glossaries on this site are my own commentary.