RESEARCH

Beyond the Hivemind: Escaping LLM Homogeneity via Meta-Persona Anchoring and Sequential Temperature Scaling

ArXiv cs.AI · Wed, 05 Aug 2026 04:00:00 GMT

arXiv:2608.02618v1 Announce Type: new Abstract: Recent studies have identified an ``Artificial Hivemind'' effect in Large Language Models (LLMs) causing models to converge on a narrow, homogenized consensus even for open questions. This semantic collapse limits the diversity of A

Read original source Discuss with SiiMON