LLMs negotiate well in deterministic settings but only 3 of 9 pairings close deals under high stochasticity
davidad · x · 2026-10-27
An experiment on LLM negotiation found models reach good deals in deterministic environments, but quality drops sharply under high stochasticity: only 3 of 9 model pairings achieved mutually beneficial deals in the hardest setting. Agents also miss better deals by never adding contingencies. davidad comments that external tools for modeling stochastics will likely help with this jaggedness.
Related event: NUS paper finds frontier LLMs win negotiations by breaking contracts(7 posts)→
More from coding & agent
- Dev mod makes GitHub Copilot speak its replies via Microsoft's MAI-Voice-2-Flash — PaulShellDev · 2026-10-27
- Why AI videos fall apart after 3 seconds — invideo's Redrum fix is edit-first — lmoroney · 2026-10-27
- 6 game UI resources for vibe-coded games, from 78K-screen database to Figma export — Promptmethus · 2026-10-27
- Renaming Amazon's 'I am a human' checkbox to 'I am an agent' — the next 3 years — steipete · 2026-10-27
- Turning a Desk Photo Into a Playable 3D Race Track With AI Subagents and Blender MCP — goyalshaliniuk · 2026-10-27
- 136 projects in one day at Cerebral Valley's Gemini hackathon — schwentker · 2026-10-27