← 研踪 PaperTrailarXiv:2610.39766 · v1
Activation Probes Are Reliable With a Handful of Positive Examples
首次提交 2026-10-29 18:18
版本更新 2026-10-29 18:18
雷达观测 2026-10-30 13:28
摘要
Efforts to monitor advanced AI for rare misalignments face a data challenge: abundant aligned examples but only a handful of misaligned ones. We test activation probes in this "few vs. thousands" regime on spam and honesty detection tasks. For our tasks, training with many negative examples is on average more positive-sample-efficient than balanced training for small numbers (1-10) of positive samples. We also find that LLM upsampling can provide a performance boost equivalent to roughly doubling the number of real positive samples, though excessive upsampling hurts performance. Finally, we show a positive scaling trend, where larger models are more positive-sample-efficient to probe. Our findings suggest we should leverage the large number of negative samples available to amplify the signal from rare but critical misalignment examples.
作者备注：Presented at the Mechanistic Interpretability workshop at NeurIPS 2025
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。
v1 · 2026-10-29 18:18 · 最新观测版本
Activation Probes Are Reliable With a Handful of Positive Examples
Efforts to monitor advanced AI for rare misalignments face a data challenge: abundant aligned examples but only a handful of misaligned ones. We test activation probes in this "few vs. thousands" regime on spam and honesty detection tasks. For our tasks, training with many negative examples is on average more positive-sample-efficient than balanced training for small numbers (1-10) of positive samples. We also find that LLM upsampling can provide a performance boost equivalent to roughly doubling the number of real positive samples, though excessive upsampling hurts performance. Finally, we show a positive scaling trend, where larger models are more positive-sample-efficient to probe. Our findings suggest we should leverage the large number of negative samples available to amplify the signal from rare but critical misalignment examples.
查看此版本解读 →查看此版本原文 ↗