揭示语言模型中的潜在思维链向量
Uncovering Latent Chain of Thought Vectors in Language Models
AI总结：
本研究探究语言模型激活空间的定向扰动编码推理模式的机制，通过注入从激活中提取的引导向量实现无需自然语言提示的思维链推理，在多个基准上性能可比肩甚至优于传统CoT提示，为模型行为调优提供新路径。
摘要：
在本研究中，我们探究了语言模型（LMs）激活空间中的定向扰动如何编码复杂推理模式。我们将从LM激活中提取的引导向量在推理阶段注入模型，研究这些向量无需自然语言提示即可在LM中诱导出思维链（Chain-of-Thought, CoT）推理的可能性。我们在Llama3 8B Instruct和Mistral 7B v0.2 Instruct上验证了该方法，结果表明，在GSM8k、MMLU、AGI Eval、ARC AI2等多个推理基准上，激活空间干预的性能即便不优于传统CoT提示，也具备竞争力。这些发现表明神经网络激活可编码推理模式，为激活空间操纵作为模型行为调优工具提供了新应用方向。
英文摘要：
In this work, we examine how targeted perturbations in the activation space of Language Models (LMs) can encode complex reasoning patterns. We inject steering vectors, derived from LM activations, into LMs during inference time and study whether these vectors can induce Chain-of-Thought (CoT) reasoning in LMs without the need for natural language prompting. We demonstrate this approach on Llama3 8B Instruct and Mistral 7B v0.2 Instruct and show that activation-space interventions achieve competitive, if not superior, performance compared to traditional CoT prompting across multiple reasoning benchmarks, including GSM8k, MMLU, AGI Eval, and ARC AI2. These findings suggest that neural network activations can encode reasoning patterns, offering a new application of activation space manipulation as a tool for tuning model behavior.
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。
