Metraipolis › archive › dc_efed9128464f

Localized, unstable, and entangled: Exploring `Us vs. Them' bias in Large Language Models — arXiv:2610.10864

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.10864
License (as archived)
CC0-1.0
Archived
2026-10-09T06:39:41.351Z
SHA-256 of the archived text
cd3761b9391f223ac5eede87a5d0adea09d7a6777f74b067fa58de78e2236936
Stable URL of this page
https://metraipolis.com/documento/dc_efed9128464f

How to cite

Localized, unstable, and entangled: Exploring `Us vs. Them' bias in Large Language Models — arXiv:2610.10864 — arXiv (en), https://arxiv.org/abs/2610.10864 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_efed9128464f (Metraipolis archive dc_efed9128464f, sha256 cd3761b9391f223ac5eede87a5d0adea09d7a6777f74b067fa58de78e2236936)
BibTeX
@misc{dcefed9128464f,
  title = {Localized, unstable, and entangled: Exploring `Us vs. Them' bias in Large Language Models — arXiv:2610.10864},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.10864},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_efed9128464f; sha256 cd3761b9391f223ac5eede87a5d0adea09d7a6777f74b067fa58de78e2236936},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.10864. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Large language models (LLMs) exhibit us-vs.-them bias: A behavioral asymmetry in which prompts framed around an ingroup (``we''/``us'') receive systematically more positive continuations than matched prompts framed around an outgroup (``they''/``them''). Using Edge Attribution Patching (EAP), we localize this behavior to directed circuits in GPT-2 Small, GPT-2 Large, Llama-2 7B, and Llama-3 8B. Across models, high-attribution edges are sparse and concentrated primarily in middle-to-late layers, while ingroup- and outgroup-related pathways are only partially overlapping and reorganize under fine-tuning. Behavioral comparisons and 30-token generations show that the measured asymmetry extends beyond next-token evaluation. Targeted patching of EAP-ranked edges reduces the bias score more than equal-cardinality random-edge controls, but also substantially degrades performance on CoLA, HellaSwag, and WikiText-2, suggesting that bias-relevant pathways are functionally entangled with broader language capabilities. Overall, our findings characterize us-vs.-them bias as a sparse, context-dependent, and functionally entangled computational phenomenon whose internal organization varies across the models examined.

Tema degli esploratori: governance digitale e beni comuni (elenco arXiv physics.soc-ph, nuovi lavori del giorno).
Categorie arXiv: Physics and Society (physics.soc-ph).
Pagina del lavoro: https://arxiv.org/abs/2610.10864 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-09T06:00:01.410Z dalla pagina https://arxiv.org/list/physics.soc-ph/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).