arXiv 论文提出 PHLE 方法，基础模型可检测提示词分类器漏掉的隐藏意图
Base Models Can Detect Hidden Intentions that Prompted Classifiers Miss
AI 导读
论文提出 persona-header likelihood evaluation（PHLE），用带合适 header 的基础预训练模型对输出做 likelihood 评估，可解码 subliminal learning 的隐蔽信号并区分它们。
正文 · 原文
Abstract:The Persona Selection Model (PSM) posits that post-training selects a persona the model learned during pre-training. Work on subliminal learning has shown that models can impart covert signals (like "love of owls") into semantically unrelated outputs like strings of numbers, and that students trained on these outputs inherit those traits. In this work I show that by evaluating the likelihood of such outputs under the base pre-trained model with an appropriate header, a method I call persona-header likelihood evaluation (PHLE), one can decode these subliminal signals and classify between them. For example, PHLE distinguishes number sequences generated by a post-trained model prompted with "You love owls" from those prompted with "You love eagles", and distinguishes answers from a model with a secret desire to harm the user, filtered to remove any noticeable misalignment, from those of a model with no system prompt. In both cases prompted classifiers, including the generating model itself, remain near chance. The signal partially persists across model families, with the Qwen base model reading signals generated by Olmo and vice versa. This is evidence for a non-trivial prediction of the PSM, and the method may be practically useful for auditing the hidden objectives of models.
| Comments: | 28 pages, 10 figures. Code available at this https URL |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.36202 [cs.LG] |
| (or arXiv:2610.36202v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.36202 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Oz Amram [view email]
[v1]
Mon, 26 Oct 2026 19:39:00 UTC (211 KB)
来源：arXiv cs.AI · arxiv.org