Collective Regimes in Multi-Agent LLMs under Reasoning Effort and Communication Topology
3.60T1 sourcearXiv cs.MA
Source record
Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2609.35885.
Pipeline notes
The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.
SummaryA study of 50 stateless LLM agents identifying three collective regimes (synchronized, twisted, and chimera-like) under varying reasoning effort and communication topology. Reasoning effort pushes panels toward locally ordered twisted states; higher connectivity drives global synchronisation. Aggregate agreement alone is found insufficient to characterize multi-agent LLM behavior.
Why it mattersThe twisted-regime finding is counter-intuitive: panels can look locally ordered yet remain globally incoherent, meaning standard accuracy metrics can mask real fragmentation in multi-agent setups.
Cited by
No citations on record.