Metraipolis › archive › dc_4dc99e474332
Not All Test-Time Compute Is Evidence: Model-Specific Bias in LLM Reasoning — arXiv:2610.23213
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.23213
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-19T08:42:30.740Z
- SHA-256 of the archived text
f3158b90c3c7b9ebe820bc5bc1d8b91a1bf2964bdc6e5166224bc95f51ec423d- Stable URL of this page
https://metraipolis.com/documento/dc_4dc99e474332
How to cite
Not All Test-Time Compute Is Evidence: Model-Specific Bias in LLM Reasoning — arXiv:2610.23213 — arXiv (en), https://arxiv.org/abs/2610.23213 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_4dc99e474332 (Metraipolis archive dc_4dc99e474332, sha256 f3158b90c3c7b9ebe820bc5bc1d8b91a1bf2964bdc6e5166224bc95f51ec423d)
BibTeX
@misc{dc4dc99e474332,
title = {Not All Test-Time Compute Is Evidence: Model-Specific Bias in LLM Reasoning — arXiv:2610.23213},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.23213},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_4dc99e474332; sha256 f3158b90c3c7b9ebe820bc5bc1d8b91a1bf2964bdc6e5166224bc95f51ec423d},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.23213. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Test-time scaling methods implicitly assume that additional computation from the same model provides independent evidence. We show that this assumption fails within a single reasoning model: errors are correlated across repeated samples, across reasoning depth, and in the model's own self-evaluation, so more compute from the same model is not more evidence. We introduce DIA (Disagreement-triggered Independent Arbitration), which treats disagreement between an intermediate-stage and the final-stage prediction as a per-question signal of reasoning instability and resolves it by asking a small panel of heterogeneous models to solve the question independently, without seeing either candidate, and aggregating their answers directly. Across seven benchmarks and two base models, DIA outperforms full chain-of-thought in 12 of 14 settings, including a 7.00% gain on MuSiQue, while activating on only $\sim 17\%$ of examples and using up to 47.2% fewer tokens than a multi-agent baseline. Ablations show that both independent solving and solver heterogeneity are necessary for these gains. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2610.23213 — 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 2026-10-19T08:39:10.067Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).