LLM Multi-Agent Systems Are Vulnerable to Herding and Runaway Cascades
4.40T1 sourcearXiv cs.MA
Source record
Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2610.30769.
Pipeline notes
The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.
SummaryResearchers adapted the classic Music Lab experiment to test whether LLM multi-agent systems exhibit herding. Exposure to cumulative recommendation counts produced runaway cascades, with collective outputs diverging sharply from intrinsic song appeal. LLM populations were more susceptible to herding than humans and less able to self-correct.
Why it mattersAdapts a classic social-psychology paradigm to expose a structural weakness in LLM multi-agent design — cascading influence that humans correct but LLMs amplify.
Cited by
No citations on record.