Stanford's multi-agent LLM information-design paper now on arXiv
rohanpaul_ai · x · 2026-11-08
rohanpaulai follows up with the arXiv link for the Stanford paper: arXiv:2611.07667, by Mete Erdogan, Mirac Suzgun, James Zou, et al., from Stanford, University of Maryland, and All Purpose AI.
The paper introduces "empirical information design," systematically comparing disclosure policies while holding interaction rules and models fixed, using ground-truth utilities to separate identified opportunities from realized outcomes. In double auctions, full disclosure triggered market breakdown: Gemini 3.1 Pro fell from 97% to 78%, Claude Opus 5.5 from 94% to 64%, and Grok 4.7 from 97% to 62%.
Related event: Stanford Paper: More Information Can Hurt LLM Trading Agents(2 posts)→
More from Research
- MuJoCo 4.0 ships with speculative contacts, permanent soft-body deformation and MuJoCo Studio — ZoubinGhahrama1 · 2026-11-08
- mitsuhiko: djs' JS "heap dumps" may beat real dumps for durable coding agents — mitsuhiko · 2026-11-08
- Salesforce paper: RL-trained experts merged via OPD save 42% FLOPs vs external-teacher distillation — rohanpaul_ai · 2026-11-08
- Albert Gu: Bitter Lesson isn't 'just scale Transformers' — algorithmic innovation matters — sewon__min · 2026-11-08
- Autoware open-sources an off-road driving simulator built on NVIDIA Isaac Sim — rsasaki0109 · 2026-11-08
- The meta-question as agent swarms scale: can models beat gradient descent at its own game? — JensHonack · 2026-11-08