原帖引用 · AI 翻译
标题：“最初几个词元很重要：理解思维链的可控性”
哈佛大学和麻省理工学院的一篇新论文发现，推理模型对其思维链控制不佳，部分原因在于它们习惯以“Okay”开头。
安全团队会查看模型的推理过程，以发现不当行为；显示模型无法操控这一推理过程的基准测试结果一直令人安心。此前，模型为何做不到这一点尚不清楚。
作者要求 Qwen3-32B 等模型用全大写字母推理，或给每个句子编号。Qwen3-32B 和 OLMo-3.1-32B 将约 99% 的首词元概率分配给“Okay”，因此作者尝试禁用这个词，或预填 1 个符合要求的词元。
禁用“Okay”和“Alright”，并加入 1 个符合要求的示例，与仅提供指令相比，使平均遵循率提高了 34 至 41 个百分点。隐藏信息的任务，例如避免使用某个词，表现出的改善不那么一致；在 12 个模型中，这类任务看起来属于另一种独立能力。
如果你因可控性评分而信任思维链监测，就应将开头用词问题与整体控制能力区分开，并单独报告隐藏信息的任务。
原文
New Harvard and MIT paper finds that reasoning models' poor control over their chain-of-thought partly comes from a habit of opening it with "Okay". Safety teams read a model's reasoning to catch misbehavior, and benchmarks showing models can't steer that reasoning have been reassuring. Why models fail at this was unclear. The authors told models like Qwen3-32B to reason in uppercase or number each sentence. Qwen3-32B and OLMo-3.1-32B put about 99% of first-token probability on "Okay", so the authors tried banning it or prefilling 1 compliant token. Banning "Okay" and "Alright" and adding 1 worked example raised average compliance by 34 to 41 percentage points over instructions alone. Tasks that hide information, like avoiding a word, responded less consistently and look like a separate skill across 12 models. If you trust chain-of-thought monitoring because of controllability scores, separate the opening-word problem from general control and report hiding tasks separately.