VERS：通过编辑激活解释来引导推理模型
VERS: Steering Reasoning Models by Editing Their Activation Explanations
一篇有意思的论文：不改提示词，直接在模型推理链的怀疑点上编辑激活解释，就能让 GSM8K 少生成近一半 token，准确率几乎不掉。
论文提出 VERS 方法，利用自然语言自编码器（NLA）在推理链的怀疑点编辑解释并转化为固定方向，在段落分隔处注入来引导推理模型。在 GSM8K 全量测试上，VERS 减少 45.8% 生成 token，准确率仅下降 0.82 个百分点；在 MATH-500 上节省 31.3% token，准确率下降 3.05 个百分点。该方法可迁移到 Qwen3 235B-A22B 等更大模型，节省 30-35% token，token 与准确率的权衡介于已发表的 SEAL 和 DEER 之间。但在竞赛级难题上，各模型尺寸均损失 3-7 个百分点，单一固定方向难以区分必要检查与多余检查。
VERS: Steering Reasoning Models by Editing Their Activation Explanations
Reasoning language models write a chain of thought before they answer, and many checks in it do not change the final answer. Natural language autoencoders (NLAs) explain a model's activations in text and rebuild activations from text, and editing an explanation can steer a model at a single token. We study whether such edits can also steer how a model reasons. We build VERS (Verbalize, Edit, Reconstruct, Steer): it turns explanation edits at the doubt points of one reasoning trace into one fixed direction, and adds this direction at paragraph breaks. On the full GSM8K test, VERS cuts generated tokens by 45.8% at an accuracy change of -0.82 percentage points (pp); on MATH-500 it saves 31.3% at -3.05 pp. At half strength, its trade-off between tokens and accuracy lies between those of the published methods SEAL and DEER. With fitted affine maps and a strength tuned for each model, the direction transfers to larger Qwen3 models, up to 235B-A22B, and saves 30-35% of tokens. On competition problems, however, VERS loses 3-7 pp at every model size, and a fixed direction did not separate necessary checks from unneeded ones.
- IT之家02:00原文
