build1 publisher
The best model on CommentBench rediscovered 8.3% of the points human reviewers raised
CommentBench splits human comments on AI-safety posts and drafts into target points, filters out the ones a model could not reach without extra context, and has Opus 5 judge which of the rest a model hit.
Publishers:lesswrong.com
Reality
- Evidence45
- Adoption
- Insufficient
- Hype gap+15
- Incentives55
- Confidence50