通过对齐行为表征与神经表征来引导大语言模型中的风险偏好
Steering Risk Preferences in Large Language Models by Aligning Behavioral and Neural Representations
AI总结：
本文提出将行为方法引出的潜在表征与 LLM 内部神经表征对齐以系统发现引导向量，并验证其可可靠调控模型的风险偏好输出。
摘要：
改变大语言模型（LLM）的行为，可以像使用适当构造的“引导向量”编辑 Transformer 的残差流一样直接。这些对内部神经激活的修改是表征工程的一种形式，提供了一种有效且有针对性的方式来影响模型行为，而无需重新训练或微调模型。但是，如何才能系统地识别此类引导向量呢？我们提出了一种有原则的方法，通过将行为方法（具体而言是结合 LLM 的 Markov chain Monte Carlo）引出的潜在表征与其对应的神经表征对齐，来揭示引导向量。为评估该方法，我们专注于从 LLM 中提取潜在风险偏好，并使用对齐后的表征作为引导向量来引导其与风险相关的输出。我们表明，所得引导向量能够成功且可靠地调节 LLM 输出，使其符合目标行为。
英文摘要：
Changing the behavior of large language models (LLMs) can be as straightforward as editing the Transformer's residual streams using appropriately constructed "steering vectors." These modifications to internal neural activations, a form of representation engineering, offer an effective and targeted means of influencing model behavior without retraining or fine-tuning the model. But how can such steering vectors be systematically identified? We propose a principled approach for uncovering steering vectors by aligning latent representations elicited through behavioral methods (specifically, Markov chain Monte Carlo with LLMs) with their neural counterparts. To evaluate this approach, we focus on extracting latent risk preferences from LLMs and steering their risk-related outputs using the aligned representations as steering vectors. We show that the resulting steering vectors successfully and reliably modulate LLM outputs in line with the targeted 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。
