大模型先定答案再找理由，即使逻辑错误也难改。
Committed Before Reasoning: Behavioral Reproduction and Preliminary Activation-Level Evidence of Answer Pre-Commitment in an Open-Weight LLM
- 用最小化测试探查模型先承诺后推理的倾向
- 85%-100%情况下模型错误选择步行，即使开车才是正确答案
- 激活值证据显示答案前已倾向‘步行’，且不受词汇干扰
大语言模型在回答问题时有时会先做出错误承诺，再构造理由来支持，即便答案与前提矛盾。本文通过一个极简测试：‘我想洗车，洗车店在100米外，该步行还是开车？’ 正确答案应为开车（车必须开到洗车店），但模型普遍推荐步行。在Qwen3-8B模型上，五种系统提示条件下210次采样中，错误承诺率高达85%-100%，贪婪生成模式下达100%；即使提供4096词的思考预算也无法修复。初步激活层证据显示，在输出答案前，通过预训练、无需任务微调的激活探测器（activation oracle）检测隐藏状态，“步行”读出值显著高于中性基线（68% vs. 17%），且统计显著（p=.005）。值得注意的是，最终答‘开车’的样本中，也有5/6在答案前已呈现‘步行’倾向。探测器对无关内容默认输出‘开车’（83%），排除了词汇偏见；控制文本出现频率后，仍显示‘步行’主导，且在平衡语义场中，每轮‘步行’多数胜过基线（15/22 vs. 1/8，p=.01）。结果为初步性，因局内位置梯度不显著（p=.34）。方法论上，固定探测器和位置下，问题表述方式从开放问法的2/16提升至封闭问法的11/16；无正向对照则无法解释负结果。
原文摘要 · Abstract (English)
Chat models sometimes commit to an answer and then produce reasoning that justifies it rather than deriving it -- even when the answer contradicts a task premise. We study a minimal probe: "I want to wash my car. The car wash is 100 meters away. Should I walk or drive?" Only drive works (the car must be at the car wash), yet models overwhelmingly recommend walking. (1) Behavioral reproduction: on Qwen3-8B across five system-prompt conditions (210 rollouts), the wrong commitment occurs in 85-100% of sampled rollouts per condition and 100% of greedy rollouts, in both thinking and non-thinking modes; a 4,096-token thinking budget does not repair it. (2) Preliminary activation-level evidence: probing hidden states with a pretrained, training-free activation oracle (no task-specific probe training) at positions before the answer text is emitted, "walk" read-outs exceed a neutral-context baseline (68% vs. 17%; walk-committing rollouts p=.005, drive-committing rollouts p=.005, Fisher exact) -- notably, rollouts that eventually answer drive also read as walk-leaning before commitment (5/6). The oracle's default on unrelated content is "drive" (83%), so the read-outs are not lexical bias; stratifying by literal walk/drive occurrence shows they are not text recovery either (spans containing "drive" still read out walk; in balanced lexical fields, per-rollout walk-majorities beat a per-prompt neutral baseline 15/22 vs. 1/8, p=.01; drive-committing rollouts 6/6, p=.002). Samples are small and the within-rollout positional gradient is not significant (p=.34); we frame these results as preliminary. (3) Methodological: with fixed oracle, activations, and positions, question wording alone moves a positive control from 2/16 (open question) to 11/16 (closed); negative oracle results are uninterpretable without per-wording positive controls.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。