arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2602.06941cs.LGcs.AIcs.CL

语言模型中激活引导的内生抵抗

Endogenous Resistance to Activation Steering in Language Models

University of Washington(华盛顿大学)

Alex McKenzie, Keenan Pepper, Stijn Servaes, Martin Leitgab, Murat Cubuktepe, Mike Vaiana, Diogo de Lucena, Judd Rosenblatt, Michael S. A. Graziano

AI总结:

研究发现大型语言模型在任务不匹配的激活引导下能内生抵抗,通过显式重启恢复正确生成,并识别出相关稀疏自编码器潜在变量,可增强或削弱该抵抗。

摘要:

大型语言模型可以在生成过程中从任务不匹配的激活引导中恢复,产生显式的语言重启(例如,“等等,那不对”),并在引导扰动仍然活跃的情况下继续讨论主题。我们将此称为内生引导抵抗(ESR)。使用稀疏自编码器(SAE)潜在变量来引导模型激活,我们发现Llama-3.3-70B在\llamaseventyEsrRate\\%的情况下表现出显式ESR,而来自Llama-3和Gemma-2系列的较小模型则较少出现显式形式。两个对照实验将ESR分解为检测事件和持续抵抗组件,后者不能仅由最近的on-topic token条件化来完全解释。我们通过对比on-topic/off-topic搜索识别出\numOtdLatents{}个SAE潜在变量;将其零消融使多次尝试率降低\multiAttemptReductionPct\\%,随机潜在变量和保留提示对照支持特异性。ESR还可以通过元提示和基于合成自我纠正示例的微调来有意增强。ESR对安全性具有双重影响:它可能使模型对对抗性激活空间操纵更具抵抗力,但同样可能干扰有益的基于引导的干预,因为模型无法区分两者。代码可在\href{https://github.com/agencyenterprise/endogenous-steering-resistance}{github.com/agencyenterprise/endogenous-steering-resistance}获取。

英文摘要:

Large language models can recover mid-generation from task-misaligned activation steering, producing explicit verbal restarts (e.g., ``wait, that's not right'') and continuing on-topic even while the steering perturbation remains active. We term this Endogenous Steering Resistance (ESR). Using sparse autoencoder (SAE) latents to steer model activations, we find that Llama-3.3-70B exhibits explicit ESR at 3.8%, with smaller models from the Llama-3 and Gemma-2 families showing the explicit form less frequently. Two controls dissociate ESR into a detection event and a sustained-resistance component that conditioning on recent on-topic tokens does not fully explain. We identify 26 SAE latents through contrastive on-topic/off-topic search; zero-ablating them reduces the multi-attempt rate by 25%, with random-latent and held-out-prompt controls supporting specificity. ESR can also be deliberately enhanced through both meta-prompting and fine-tuning on synthetic self-correction examples. ESR has dual implications for safety: it could harden models against adversarial activation-space manipulation, but may equally interfere with beneficial steering-based interventions, since the model has no way to distinguish the two. Code is available at https://github.com/agencyenterprise/endogenous-steering-resistance.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑