Dwarkesh spent $20K in API credits testing which AI is the best liar at Clocktower
Dwarkesh Patel · youtube · 2026-11-06
Dwarkesh Patel spent $20,000 in API credits having AI models play each other at Blood on the Clocktower, a social deception game. Models improved sharply this year at on-the-spot scheming, bluffing, and coordinated deduction; OpenAI's newer models clearly outperform Anthropic's. Full benchmark is public at clocktowerbench.com, where humans can also play against the AIs.
Related event: Dwarkesh Spends $20K Testing AI Deception in Blood on the Clocktower(3 posts)→
More from Fun
- Pro-AI poster asks anti-AI critics: is a machine doing human painting 'slop'? — charis_ai · 2026-11-06
- AI Agents Placed 500 Uber Eats Orders That Resulted in Just 4 Trips — fkasummer · 2026-11-06
- OpenAI devrel star Tibo's reach collapses to a fifth as even colleagues tire of the constant hyping — teortaxesTex · 2026-11-06
- Mistral agent lead Thibault Sautiaux 'demoted'? His own reply: "I live in the basement now" — CtrlAltDwayne · 2026-11-06
- Florida deploys robot bunnies to lure invasive pythons that wiped out 99% of raccoons — IsForAt · 2026-11-06
- My favorite album of the year is made by a robot — dreamwieber · 2026-11-06