Metraipolis › archive › dc_de093787a171

On the Position Bias of Large Language Models: A Comprehensive Study — arXiv:2610.22807

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.22807
License (as archived)
CC0-1.0
Archived
2026-10-19T08:43:35.766Z
SHA-256 of the archived text
1b73ac469307075a6a59a216e0a9d4017147af93e8cfdf6f157701c82e376b13
Stable URL of this page
https://metraipolis.com/documento/dc_de093787a171

How to cite

On the Position Bias of Large Language Models: A Comprehensive Study — arXiv:2610.22807 — arXiv (en), https://arxiv.org/abs/2610.22807 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_de093787a171 (Metraipolis archive dc_de093787a171, sha256 1b73ac469307075a6a59a216e0a9d4017147af93e8cfdf6f157701c82e376b13)
BibTeX
@misc{dcde093787a171,
  title = {On the Position Bias of Large Language Models: A Comprehensive Study — arXiv:2610.22807},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.22807},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_de093787a171; sha256 1b73ac469307075a6a59a216e0a9d4017147af93e8cfdf6f157701c82e376b13},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.22807. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Recent studies have shown that position bias remains a critical vulnerability in large language models (LLMs), undermining both the integrity of multiple-choice evaluations and the reliability of retrieval-augmented generation. However, existing understanding of this phenomenon is largely derived from studies with limited model coverage and English-centric benchmarks, leaving the behavior of frontier models and diverse languages underexplored. In this work, we conduct a comprehensive investigation of position bias across model scales ranging from billion-level to trillion-level parameters, covering both multiple-choice and multi-document tasks in five languages. Our experiments reveal that while position bias remains noticeable in small-scale models and Non-English languages, increasing model scale and employing standard Chain-of-Thought prompting significantly mitigate this issue. Moreover, we demonstrate that emerging \emph{thinking models} with extended reasoning capabilities are nearly immune to position bias, suggesting a costly but general approach toward resolving this vulnerability.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2610.22807 — 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 2026-10-19T08:39:24.834Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).