---
title: "Artificial Hivemind"
type: "entity"
entity_kind: "concept"
status: "hub"
canonical_name: "Artificial Hivemind: The Open-Ended Homogeneity of Language Models (and Beyond)"
aliases: ["Artificial Hivemind effect"]
first_seen: "2026-11-02T00:00:00.000Z"
writer_model: "claude-sonnet-5"
connects_to: ["generative-process diversity","correlated errors","algorithmic monoculture","representational similarity","Yejin Choi"]
seek_code_commit: "d40ee39"
---


A NeurIPS 2025 Datasets & Benchmarks (Oral) paper by Liwei Jiang, Yuanjun Chai, Margaret Li, Mickel Liu, Raymond Fok, Nouha Dziri, Yulia Tsvetkov, Maarten Sap, Alon Albalak, and Yejin Choi (arXiv 2510.22954; v1 27 Oct 2025, v2 camera-ready 6 Oct 2026), measuring how similar language models' outputs are to one another on 26,000 open-ended queries. It names and reports "a pronounced Artificial Hivemind effect in open-ended generation of LMs, characterized by (1) intra-model repetition, where a single model consistently generates similar responses, and more so (2) inter-model homogeneity, where different models produce strikingly similar outputs." Matters to this vault as independently-run, peer-reviewed prior work of exactly the kind — semantic-similarity assessment of a language-model population — that [[claim-generative-process-diversity-predicts-lower-correlated-failure|Tieman & Markou (2026)]] position their own "generative-process diversity" measure against in their own abstract ("Assessments of language-model populations using semantic similarity demonstrate limited semantic diversity"), though that abstract names no specific paper — and as the clean negative case for that comparison: a real, independent paper that is adjacent to the generative-process-diversity question without being a test of it, since it measures semantic similarity of outputs rather than diversity of the underlying generative process against benchmark correlated failure.

## Updates
- 2026-11-04 (propagation-repair): the body above said this paper is the prior work Tieman & Markou "explicitly position their own 'generative-process diversity' measure against in their own abstract" -> now: the abstract (capture 04245a50…, re-read in the 2026-11-04 audit) contrasts only with "Assessments of language-model populations using semantic similarity" and names no paper, so this page claims only that Hivemind is an instance of that kind of work; the named-target attribution is withdrawn. Propagated from the CORRECTED 2026-11-04 audit_status of [[claim-tieman-markou-2026-no-independent-corroboration-as-of-2026-11-01]].

## References
- [[claim-tieman-markou-2026-no-independent-corroboration-as-of-2026-11-01]]
- [[claim-generative-process-diversity-predicts-lower-correlated-failure]]
- [[entity-generative-process-diversity]]
- [[question-corroborate-tieman-markou-2026-generative-process-diversity]]
- Capture: 10-inbox/raw/2026-11-01-does-an-independent-source-corroborate-tieman-markou-2026s.md