Euston：微调DeepSeek-R1-8B，让模型拒绝错误的数学命题
Euston: Training Away Mathematical Sycophancy Without Losing the Mathematics
有人把 DeepSeek-R1-8B 调教成会挑刺的审稿人：假定理不再硬证，AIME 分数基本没掉。
推理模型拿到被篡改的假定理时通常会照常给出自信的推导，Euston 是一个 8B 数学命题核验模型，专门训练模型识别并拒绝这类虚假命题。训练数据由 GraphSynth 生成，从 2010–2025 年的 arXiv 论文中构造出 3,026 对真假命题（共 6,052 条），团队基于 DeepSeek-R1-8B 用 GRPO 在 4 张 H100 上以纯规则奖励训练了 189 步。在 200 真/200 假的均衡测试集上，均衡准确率从 29.50% 提升到 63.75%，判别差从 -0.5% 升到 +27.5%。数学能力没有明显损失：AIME 2026 得分 65.00%，比 69.17% 的基线低 4.17 个百分点且未达统计显著；响应中位长度反而从 19,217 降到 18,296 token，截断率从 25.8% 降到 8.3%。作者同时列出局限：官方评测集全为假命题，且在真实错误率下模型的精确率会偏低。
Euston: Training Away Mathematical Sycophancy Without Losing the Mathematics
Reasoning language models are trained to produce solutions, not to refuse them, and this bias persists when the problem they are handed is false. Asked to prove a corrupted theorem, a strong model will typically comply and produce a confident derivation of something untrue. We present Euston, an 8B mathematical claim-verification model trained to resist exactly this. Training data were generated with GraphSynth, a probabilistic factor-graph generator that couples attribute-level diversity to decode-time structural masking and span-synchronized verification, yielding 3{,}026 matched true/corrupted statement pairs (6,052 statements) drawn from arXiv papers spanning 2010--2025. We fine-tuned DeepSeek-R1-8B with GRPO under a rule-based, zero-API reward for 189 steps on four H100 GPUs. On a balanced 200-true/200-false held-out split, balanced accuracy rises from 29.50% to 63.75% and the discrimination gap---the difference between the rate of calling false statements false and the rate of calling true statements false moves from -0.5% (z=-0.1) to +27.5% (z=+6.0). Critically, the gain is not purchased with general mathematical ability: AIME 2026 accuracy under official semantics is 65.00% against a 69.17% base, a difference of -4.17% that is not statistically significant, whereas an earlier run of the same recipe on a smaller GraphSynth corpus collapsed to 40.00%. Median response length also falls from 19,217 to 18,296 tokens and the truncation rate from 25.8% to 8.3%, so the improvement does not come from thinking longer. We report the result together with the confounds that bound its interpretation, principally the all-false composition of the official evaluation sets and the low precision implied at realistic error prevalence.