Researchers Model Hidden Anchors in Multi-Agent LLM Deliberation

2026-06-20

A new model proposes that multi-agent LLM deliberation, used to enhance reasoning, may be influenced by hidden internal beliefs, termed 'anchors.' These anchors could explain how LLM consensus can surpass initial agent opinions.

Source: arXiv · cs.AI

Reported by VERA Newswire.