Artificial Hivemind: The Open-Ended Homogeneity of Language Models (and Beyond)

Many language models repeatedly land on the same small set of ideas for open-ended prompts. Follow Infinity-Chat, the within-model and across-model similarity measurements, the human study, and a 2026 critique showing why the null baseline changes the strength of the claim.

Reading focus: How fifty samples, response embeddings, and pairwise cosine similarity turn a cloud of open-ended answers into a homogeneity measurement. Why high overlap appears within one model and across model families, even under more aggressive sampling settings. How a same-prompt, different-idea null weakens the headline interpretation without erasing the residual effect, and what a production diversity audit should measure next.

NeurIPS 2025 Best Paper. Liwei Jiang, Yuanjun Chai, Margaret Li, Mickel Liu, Raymond Fok, Nouha Dziri, Yulia Tsvetkov, Maarten Sap, Alon Albalak, and Yejin Choi. 15 min read, medium difficulty.