Noam Shazeer's models now beat the friend he taught BOTV to, at Blood on the Clocktower

_sholtodouglas · x · 2026-11-06

sholtodouglas jokes that Noam taught him Blood on the Clocktower, and now Noam's models are schooling theirs on a game-playing eval benchmark—Noam praises the methodology and half-jokes about recruiting Dwarkesh to design evals at OpenAI.

Related event: Dwarkesh Spent $20K on API Credits to Test LLMs at Blood on the Clocktower; OpenAI Overtakes Anthropic(7 posts)→

Original post →

More from Fun

Fun channel →