Constitutional AI: RLHF On Steroids
astralcodexten.substack.com · 1,689 words · saved by 2 readers
...
AIs like GPT-4 go through several different1 types of training. First, they train on giant text corpuses in order to work at all. Later, they go through a process called “reinforcement learning through human feedback” (RLHF) which trains them to be “nice”. RLHF is why they (usually) won’t make up fake answers to your questions, tell you how to make a bomb, or rank all human races from best to worst. RLHF is hard. The usual method is to make human crowdworkers rate thousands of AI responses as good or bad, then train the AI towards the good answers and away from the bad answers. But having…
saved by
related reading
- Constitutional AI: Harmlessness from AI Feedbackarxiv.org
- Constitutional AI: Harmlessness from AI feedbackanthropic.com
- [2212.08073] Constitutional AI: Harmlessness from AI Feedbackar5iv.labs.arxiv.org
- Constitutional AI vs. RLHF vs. Deliberative Alignment — LessWronglesswrong.com
- [2310.13798] Specific versus General Principles for Constitutional AIarxiv.org
- RLHF: Reinforcement Learning from Human Feedbackhuyenchip.com
- Claude's Constitutional Structure - by Zvi Mowshowitzthezvi.substack.com
- Constitutional AI | Tracking Anthropic's AI Revolutionconstitutional.ai
- Claude’s Constitution \ Anthropicanthropic.com
- Video and transcript of talk on writing AI constitutionsjoecarlsmith.com
- Inverse Constitutional AI: Compressing Preferences into Principlesarxiv.org
- Teaching Claude Whyalignment.anthropic.com