Source record

Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2610.22882.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

Researchers tested five production LLM models in a six-player collective-risk dilemma across ten rounds. Four contributed roughly equal shares at every positive risk level, and four also contributed at zero risk. Cooperative self-play was not evidence of strategic adaptability, with selection outcomes depending on population composition.

Shows that LLM cooperation in self-play can mask fixed, non-adaptive strategies, which is a concrete caution for anyone designing or evaluating multi-agent LLM workflows.

Cited by

No citations on record.