← 研踪 PaperTrailarXiv:2607.23379 · v2
When Activation Oracles Learn Not to Read: Concept-Specific Blind Spots in Fine-Tuned Oracles
首次提交 2026-07-26 05:58
版本更新 2026-11-05 09:47
雷达观测 2026-11-06 11:06
摘要
Activation Oracles (AOs) are language models trained to answer natural-language questions about another model's internal activations. They offer a flexible interface for reading hidden information from model states, especially when relevant information is internally represented but absent or incomplete in visible behavior. However, AOs are themselves learned systems: their answers are shaped by training data, objectives, and learned reporting behavior, rather than being neutral readouts of represented information. We study this in a controlled Taboo Word Guessing setting, where subject models are fine-tuned to internally use a hidden concept while avoiding direct disclosure. Contrary to the expectation that an AO trained on such a subject becomes a specialist reader, we find that fine-tuned AOs can become concept-specific anti-readers: they selectively fail to recover the concept persistently present during their own training. This failure is not simply explained by absence of the concept from the subject or oracle representations: the target remains decodable inside the oracle, while LogitLens and layer-ablation analyses indicate that the failure arises in the AO readout pathway. Our results show that behavioral leakage, representation-level decodability, and AO-verbalizability can come apart, raising a reliability concern for learned interpretability interfaces.
作者备注：A mistake was found in experimental results, so the results as presented in the current version of the paper are incorrect
观测到的版本
相邻记录可能跳过中间版本；首次发现 v2 及以上也会标为版本更新。
v2 · 2026-11-05 09:47 · 最新观测版本
When Activation Oracles Learn Not to Read: Concept-Specific Blind Spots in Fine-Tuned Oracles
Activation Oracles (AOs) are language models trained to answer natural-language questions about another model's internal activations. They offer a flexible interface for reading hidden information from model states, especially when relevant information is internally represented but absent or incomplete in visible behavior. However, AOs are themselves learned systems: their answers are shaped by training data, objectives, and learned reporting behavior, rather than being neutral readouts of represented information. We study this in a controlled Taboo Word Guessing setting, where subject models are fine-tuned to internally use a hidden concept while avoiding direct disclosure. Contrary to the expectation that an AO trained on such a subject becomes a specialist reader, we find that fine-tuned AOs can become concept-specific anti-readers: they selectively fail to recover the concept persistently present during their own training. This failure is not simply explained by absence of the concept from the subject or oracle representations: the target remains decodable inside the oracle, while LogitLens and layer-ablation analyses indicate that the failure arises in the AO readout pathway. Our results show that behavioral leakage, representation-level decodability, and AO-verbalizability can come apart, raising a reliability concern for learned interpretability interfaces.
查看此版本解读 →查看此版本原文 ↗