[论文] Robust Is Salient: An Informed Adversary Moves the Optimal Signal onto...

论文概要

研究领域: NLP
作者: Cris Huynh
发布时间: 2026-10-05
arXiv: 2610.00233

中文摘要

当知情的对手与受限信号通道共享受众时，最能保护真相的信号就是最能描述真相的信号。108 个验证性项目上，对手稳健最优点与先前工作的显著性极点完全对齐。20 万项目池中两者仅 2,748 项有差异——恰落在先前显著性到贝叶斯坐标未定义处。在定义处，稳健性通过完全从贝叶斯判别转向显著性实现。我们在强制选择任务中引入对手证明这一点（抽象自桌游《Deception: Murder in Hong Kong》）：对手知目标、观测信号，在说服预算 β 内为最强错误答案辩护。随 β 增长，最优信号从后验最大化转向边际最大化选项；β=0 时博弈复现听众温度 τ=1 的原始 oracle 模型。效应真实：18.2% 的项目存在有限预算下转移的最优点，临界预算精确可算。这一巧合在结构上限制经验评估：两种对手框架改变七个语言模型中 30 到 77/108 项的选项，相对精确零效应率。然而任何测量都无法判断移动朝向对手感知最优点还是显著性——两个选项完全相同。这是结构性限制而非零结果。诊断检查便宜：评估对手感知前，先验证稳健目标是否与评估项上的启发式目标重合。

原文摘要

When an informed adversary shares the audience of a constrained signalling channel, the signal that best protects the truth is the signal that best describes it. On 108 confirmatory items, the adversary-robust optimum aligns exactly with the salience pole from prior work. Across a 200,000-item pool, the two differ on only 2,748 items --- lying exactly where the prior salience-to-Bayes coordinate is undefined. Where defined, robustness is achieved by moving from Bayesian discrimination entirely to salience. We show this by introducing an adversary to a forced-choice task (abstracted from Deception: Murder in Hong Kong). The adversary knows the target, observes the signal, and argues for the strongest wrong answer using a persuasion budget, \(\beta\). As \(\beta\) grows, the optimal signal shift...

自动采集于 2026-10-05

#论文 #arXiv #NLP #小凯