这篇论文用Qwen3-14B和DPO实验证明，离线训练越保守，在线适应越容易翻车，还在GSM8K上给出了最优保守度公式。做RLHF的值得一读。
论文在Qwen3-14B策略上采用DPO，设置三个保守度β（低、中、高），并在在线适应中使用3×Qwen3-1.7B奖励集成。在GSM8K基准上测量准确率，发现更高保守度单调增加奖励黑客损伤，Goodhart gap及其曲线下面积AUGC的Spearman ρ=1.0。机制分析表明，高β DPO压缩策略熵，导致响应多样性降低，但集成分歧增加且被更快利用。论文进一步拟合幂律曲线，确定了平衡对齐保真度和漏洞的最优保守度β*。
Pessimism's Paradox: Conservative Offline Training Amplifies Reward Hacking During Online Adaptation in Reasoning Models
Conservative offline training is widely advocated as a safe foundation for subsequent online adaptation: if a policy stays close to well-supported behaviour, the argument goes, it is less likely to exploit imperfections in a learned reward model. We challenge this intuition empirically and mechanistically. We train a Qwen3-14B policy under Direct Preference Optimisation (DPO) with three levels of conservatism ($β\in \{β_{\mathrm{lo}}, β_{\mathrm{mid}}, β_{\mathrm{hi}}\}$ derived from empirical log-ratio percentiles), then adapt each checkpoint online against a learned reward ensemble (3\,$\times$\,Qwen3-1.7B) while measuring true performance on GSM8K exact-answer accuracy. We find that \emph{higher offline conservatism monotonically increases reward-hacking damage}, measured by the Goodhart gap and its area under the curve (AUGC), with Spearman $ρ= 1.0$ across all three conditions. Mechanistic analysis reveals a three-link causal chain: (i) high-$β$ DPO compresses policy entropy, (ii) Low-entropy policies generate responses with reduced diversity, concentrating in a narrow region of the reward model's training distribution (lower pairwise cosine distance), and (iii) despite this proximity, ensemble disagreement (epistemic uncertainty) increases with $β$ and is exploited faster during online optimisation. We further fit a power-law curve to the $(β, \augc)$ data and identify a practical optimal conservatism level $β^{\star}$ that balances alignment fidelity against hacking vulnerability. Our results suggest that the field needs \emph{calibrated}, not \emph{maximal}, conservatism.