Metraipolis › archive › dc_ae81dc23ec10
One Axis, No Brake: Self-Knowledge Limits the Filtering of Harmful Peer Conformity in LLMs — arXiv:2609.18998
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.18998
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-15T13:50:02.875Z
- SHA-256 of the archived text
7707121c027c26be30d162a88276904201168db218b8372ad0118244b32272bc- Stable URL of this page
https://metraipolis.com/documento/dc_ae81dc23ec10
How to cite
One Axis, No Brake: Self-Knowledge Limits the Filtering of Harmful Peer Conformity in LLMs — arXiv:2609.18998 — arXiv (en), https://arxiv.org/abs/2609.18998 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ae81dc23ec10 (Metraipolis archive dc_ae81dc23ec10, sha256 7707121c027c26be30d162a88276904201168db218b8372ad0118244b32272bc)
BibTeX
@misc{dcae81dc23ec10,
title = {One Axis, No Brake: Self-Knowledge Limits the Filtering of Harmful Peer Conformity in LLMs — arXiv:2609.18998},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.18998},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ae81dc23ec10; sha256 7707121c027c26be30d162a88276904201168db218b8372ad0118244b32272bc},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.18998. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Multi-agent LLM systems are expected to be more reliable because agents can catch each other's mistakes. But peer pressure cuts both ways: the same correction that fixes a wrong answer can overturn a right one. The tempting safeguard is a brake that keeps the beneficial revisions and blocks the harmful ones. We show this brake is hard to build, for a simple reason: a revision is harmful exactly when the original answer was right, so deciding whether to block it is the same as knowing whether the model was already correct. This turns the open-ended hunt for a brake into one measurable quantity, the model's self-knowledge: any brake built from a deploy-time signal is a correctness probe in disguise, and self-knowledge is far from perfect (AUROC $\approx 0.64$--$0.89$ across six model families). We call this ceiling the wall. Even white-box steering of the model's own correctness direction does not breach it: it changes how often the model revises, but harmful and beneficial revisions move together. At population scale the wall becomes the cliff: when most agents start wrong, debate amplifies the shared mistake into a confident, wrong consensus. In our multiple-choice societies, more agents, more model diversity, and a stronger member do not fix it. What helps is adding information before the revision, not filtering after it. Local agreement is not global correctness. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-17 in cs.CL). Categorie arXiv: Machine Learning (cs.LG) ; Computation and Language (cs.CL); Multiagent Systems (cs.MA). Pagina del lavoro: https://arxiv.org/abs/2609.18998 — 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 2026-10-15T13:48:24.443Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-17?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).