通过对比对偶搜索进行定向神经元调节
Targeted Neuron Modulation via Contrastive Pair Search
AI总结:
研究通过对比神经元属性识别有害与良性提示的区分神经元,实现无梯度的神经元干预,有效降低拒绝率并保持输出流畅性,展示了神经层面干预在行为引导中的优势。
摘要:
语言模型经过指令微调以拒绝有害请求,但其机制尚不明确。流行的方向方法在残差流上操作,高干预强度下导致输出不连贯,限制了实际应用。我们引入对比神经元属性(CNA),识别出0.1%的MLP神经元,其激活最能区分有害与良性提示,仅需前向传递无梯度或辅助训练。在指令模型中,消去发现的回路使拒绝率降低超过50%,在标准 jailbreak 测试中保持流畅性和非退化性。将CNA应用于Llama和Qwen架构(从1B到72B参数)的基模型和指令模型,发现基模型包含相似的后期层区分结构,但干预这些神经元仅导致内容变化,而非行为改变。这些结果表明,神经层面干预能可靠地引导行为,而无需残差流方法的质量折中。更广泛地说,我们的发现表明,对齐微调将预存的区分结构转化为稀疏、可靶向的拒绝门。
英文摘要:
Language models are instruction-tuned to refuse harmful requests, but the mechanisms underlying this behavior remain poorly understood. Popular steering methods operate on the residual stream and degrade output coherence at high intervention strengths, limiting their practical use. We introduce contrastive neuron attribution (CNA), which identifies the 0.1% of MLP neurons whose activations most distinguish harmful from benign prompts, requiring only forward passes with no gradients or auxiliary training. In instruct models, ablating the discovered circuit reduces refusal rates by over 50% on a standard jailbreak benchmark while preserving fluency and non-degeneracy across all steering strengths. Applying CNA to matched base and instruct models across Llama and Qwen architectures (from 1B to 72B parameters), we find that base models contain similar late-layer discrimination structures but steering these neurons produces only content shifts, not behavioral change. These results demonstrate that neuron-level intervention enables reliable behavioral steering without the quality tradeoffs of residual-stream methods. More broadly, our findings suggest that alignment fine-tuning transforms pre-existing discrimination structure into a sparse, targetable refusal gate.
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。