https://pangyd.com/tags/cs.cy/Recent content in Cs.CY on 庞玉栋个人博客Hugozh-cnThu, 17 Sep 2026 23:11:08 +0800https://pangyd.com/post/radar-27/Thu, 17 Sep 2026 23:11:08 +0800https://pangyd.com/post/radar-27/<blockquote> <p><strong>原文</strong>：<a href="https://arxiv.org/abs/2609.20543v1">Language-model groups overstate consensus when replaying human deliberation on a reasoning task</a></p> <p><strong>作者</strong>：Tengfei Shao</p> <p><strong>来源</strong>：arXiv cs.AI（人工智能）</p> </blockquote> <h2 id="正文">正文</h2> <p>Computer Science &gt; Artificial Intelligence</p> <p>arXiv:2609.20543v1 (cs)</p> <p>[Submitted on 17 Sep 2026]</p> <p>Title:Language-model groups overstate consensus when replaying human deliberation on a reasoning task</p> <p>Authors:Tengfei Shao</p> <p>View a PDF of the paper titled Language-model groups overstate consensus when replaying human deliberation on a reasoning task, by Tengfei Shao</p> <p>View PDF</p> <p>Abstract:Full-consensus rates are often treated as indicators of collective cognition, yet depend on how participation and final states are operationalized. We replayed 100 held-out human Wason groups with matched large language model (LLM) agent groups, seeding one belief-anchored agent per participant&rsquo;s pre-discussion answer and scoring agents and people with the same code. Across human scoring definitions, estimates ranged from 24.0% to 57.0%; about one fifth of participants never posted, whereas agents almost always did. Agent groups remained more consensual in two post-unblinding sensitivity analyses: the submit-based comparison (n = 98) yielded gaps of 34.0 and 43.9 percentage points for chat and reasoning modes, and the participation-matched comparison (n = 45) yielded gaps of 34.1 and 44.4 points. These complementary routes reduced different measurement asymmetries yet converged within 0.5 percentage points. The gap persisted without early stopping and under a reparameterization removing the memorizable answer; reasoning-mode groups then agreed nearly unanimously, mostly on incorrect answers. Simulated consensus did not track collective accuracy, and belief-anchored agent groups were biased estimators of the human group-outcome distribution in this setting. These analyses provide a scoring-explicit basis for assessing simulated-group estimates of human deliberative outcomes.</p>