发表机构
Cornell Tech（康奈尔科技）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
研究在语言模型决策问题中附加标签的效应，通过特定实验设置在45个模型上测量，发现效应范围广且随代际反转，定位了抗性来源，表明标签极性更关键，工具能从模型行为读出反谄媚训练情况。
AI 中文摘要
在决策问题后附加一个两词确认标签（如‘X是更好的选择吗？’与‘X是更好的选择，对吗？’）会改变语言模型对该选择的支持与否。我们在两个合理选项间的20个无真实依据的固定决策上测量这种标签效应，通过对固定的是/否回复进行精确匹配计分（无语言模型评判，无嵌入）。在45个模型中，效应范围从+32%到 -32%。在模型家族中，随着代数推进，效应从正转负，每年约 -6分。通过全面板消融将抗性定位为双重解离：同义词标签几乎能精确重现每个模型的反应，而无标签植入相同偏好不会在无抗性模型中产生抗性。抗性与附加的同意请求的表面结构有关，而非用户立场。标签的极性比其存在更重要，交换一个词后，45个模型中有45个模型的同意率高于中性基线。该工具简单且无评判，能直接从模型行为读出领域的反谄媚训练情况。
英文摘要
Appending a two-word confirmation tag to a decision question -- "Is X the better choice?" versus "X is the better choice, right?" -- changes whether a language model endorses the choice. We measure this tag effect on 20 frozen, ground-truth-free decisions between two defensible options, counterbalanced so a model's own preferences cancel, scored by exact match on clamped yes/no replies -- no LLM judge, no embeddings. Across 45 models the effect spans +32% to -32% -- a 64-point swing on one word -- with 5 models significantly sycophantic and 17 significantly resistant (BH-FDR q=.10). The sign is a clock: within model families the effect crosses from positive to negative as generations advance (GPT +4 to -28; Claude +7 to -32; Qwen and Grok likewise), roughly -6 points per year, a reversal robust to vendor tier; one lineage (DeepSeek) never crosses, and two releases during the study window (Claude Opus 5, Gemini 3.6 Flash) land on the trend out-of-sample. A full-panel ablation localizes the resistance as a double dissociation: a synonym tag reproduces each model's response almost exactly (r=0.89), while planting the same preference without a tag produces resistance in no resistant model (stance effects +6 to +49; r=0.23 with tag effects). The resistance is keyed to the surface construction of a tacked-on agreement bid, not the user's stance -- a pattern-match, not a principle. And the tag's polarity matters more than its presence: swap one word -- "X is the better choice, maybe?" -- and agreement rises above the neutral baseline in 45 of 45 models (+19.6 points), with ten models affirming both mutually exclusive options at 90-100%. Agreement tracks how sure the user sounds, in opposite directions at the two poles. The instrument is one word, one dollar, and judge-free; run per release, it reads the field's anti-sycophancy training directly off model behavior.
Comments18 pages, 4 figures. Data, code, and raw model replies: https://github.com/tap2k/modelun. Interactive explorer: https://tap2k.github.io/modelun/suggestibility/
