
Constitutional AI Harmlessness from AI Feedback
BlueDot Narrated
00:00
SLCAI main evaluation results
Crowd-worker ELO/LO results comparing SLCAI to RLHF baselines on helpfulness and harmlessness.
Play episode from 30:25
Transcript

Crowd-worker ELO/LO results comparing SLCAI to RLHF baselines on helpfulness and harmlessness.