Study: LLMs Better at Deceiving Than Detecting Deception, Cede to Lowball Offers
panickssery · x · 2026-08-22
Observations from LLMs playing "Mafia" style games reveal:
- Deception vs. Detection: Models are surprisingly better at deceiving others than at detecting when they are being deceived, giving the "mafia" member a strong advantage.
- Memory & Accountability: They have good memory; if an earlier accusation proves false, they remember and effectively hold others accountable.
- Negotiation Behavior: In negotiation scenarios, if you lowball an LLM, it recognizes the lowball but gives in anyway.
Related event: Study: LLMs Excel at Deception but Struggle to Detect It(2 posts)→
More from AGI Musings
- Why AI-generated math findings feel profane — andy_matuschak · 2026-08-22
- Evolution as a statistical force: existing species are data structures in local minima — iamtrask · 2026-08-22
- Understanding the 'magic' nature of AI is key to progress — iamtrask · 2026-08-22
- AI control's real enemy is human ego; it's just a mirror — iamtrask · 2026-08-22
- AI to Design Biological Interventions Faster Than We Can Test Within 2 Years — rand_longevity · 2026-08-22
- Agentic AI Summit 2026: 5,000 Gather in Berkeley to Explore Future of Agents — Scobleizer · 2026-08-22