模型写的推理过程真的被后续计算使用了吗？实验发现能用的才有效。
Do Models Read What They Write? Causal Registers in Scratchpad Reasoning
- 通过编辑中间状态测试模型是否真正依赖写出的内容
- 写中间状态的模型对修改后结果预测准确率达80%~91%
- 适合关注推理可解释性与对齐训练的研究者
过程监督的核心假设是模型能暴露对其后续行为有影响的中间变量。为验证这一假设，我们设计了一个具有已知更新规则的状态追踪任务，对比仅输出最终结果的模型与提前写出中间状态的模型。评估时，我们修改某个已写入状态的内部表示，但保持可见的草稿文本不变。由于转换规则已知，修改应导致唯一正确后果。在Qwen2.5-Coder-7B中，状态写作模型在两种任务变体上分别对下一阶段比特的预测准确率达到80%和91%，而预训练模型及仅输出最终答案的对照组仍接近基线水平。额外对照实验排除了通用下一个词引导或复制延续的可能性：预测同时依赖于编辑后的状态和当前动作。该因果使用模式在多个模型族中复现。结果表明，草稿监督的目标应从单纯可读性，转向训练模型真正使用所写状态进行计算。
原文摘要 · Abstract (English)
A central hope behind process supervision is that models can expose intermediate variables that matter for their later behavior. For this to help with alignment, a scratchpad must be tied to the computation: when the model writes a state, later steps should compute from that state. To test this requirement, we use a controlled state-tracking task with a known update rule, comparing models trained to report only the final state with models trained to write intermediate states before giving the final answer. At evaluation, we edit the internal representation of one written state while leaving the visible scratchpad text fixed. Because the transition rule is known, the edit has a single correct downstream consequence. In Qwen2.5-Coder-7B, the state-writing model predicts the next phase bit implied by the edited state on 80% and 91% of held-out examples across the two task variants, while pretrained and final-answer-only controls remain near baseline. Additional controls rule out generic next-token steering and copying another continuation: the prediction depends on both the edited state and the current move. The same causal-use pattern replicates across model families. Together, these results suggest a sharper goal for scratchpad oversight: not just to make intermediate reasoning legible, but to train written states that the model uses as part of its computation.
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。