---
title: "Qwen3 中的跨期偏好引导：通过对比激活添加实现"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.03892v1"
permalink: "https://feed.astralor.com/items/itm_70fe4c6acd87ebc68fe445b2"
published: "2026-08-04T16:25:38.000Z"
score: 77
channel: "ai"
tags: ["AI 安全","Qwen3","对比激活添加","线性探针","跨期选择"]
---

# Qwen3 中的跨期偏好引导：通过对比激活添加实现

## AI 点评

> 这篇论文有意思的地方在于，它把“时间偏好”这种心理概念变成了模型内部可操控的向量。对于做对齐的人来说，提供了一个比 RLHF 更轻量的偏好调节手段；对于做产品的人，也许未来能根据用户性格动态调整 AI 的建议视野，比如让理财助手更耐心或更果断。我尤其欣赏那个意外发现：适度引导时间偏好居然能提升规划能力，这说明某些偏好方向其实和能力是耦合的，值得进一步深挖。

## 速览

该研究探索大型语言模型 Qwen3-32B 中时间视野的线性表示，并利用对比线性探针在模型残差流中识别短期与长期偏好方向。通过对比激活添加（CAA）进行引导，作者在保留二元时间选择任务、分布外货币跨期选择任务及 TravelPlanner 基准上验证了效果。结果表明，该方法能双向显著改变模型对“更小更早”与“更大更晚”奖励的偏好阈值，并在适度引导下提升规划相关能力指标。这些发现意味着模型的跨期偏好可测量且可引导，对涉及延迟成本收益的 AI 建议系统和长程规划安全具有重要参考价值。

## 关键点

- 使用对比线性探针在 Qwen3-32B 残差流中找到时间视野方向，实现双向偏好操纵。
- 在分布外货币跨期选择任务上，引导显著移动模型的漠视阈值，改变 smaller-sooner 与 larger-later 的选择偏好。
- 适度的时间引导能够提升 TravelPlanner 基准上的规划相关能力指标。
- 研究表明 LLM 的跨期偏好具有线性可表示性，为对齐和安全研究提供了新工具。

## 全文

We study linear representations of temporal horizon in the large language model Qwen3-32B and use them to change the model's time-related preferences, recommendations, and capabilities. We train contrastive linear probes on teacher-forced temporal-choice answers to find a short-term versus long-term direction in the model's residual stream, and evaluate contrastive activation-addition steering on a held-out binary temporal-choice task, an out-of-distribution monetary intertemporal-choice task, and a TravelPlanner capability benchmark. The central result is that temporal-horizon directions can be identified with simple contrastive linear probes and then used for steering to induce large, bidirectional preference changes. On an out-of-distribution monetary choice task that varies reward size and delay, steering strongly shifts the model's indifference threshold between smaller-sooner and larger-later rewards in both directions. We further show improvements on a planning-related capability metric under moderate temporal steering. These results suggest that model intertemporal preferences are measurable and steerable, which is relevant for AI systems that give advice involving delayed costs and benefits, and for safety questions about long-horizon planning.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_70fe4c6acd87ebc68fe445b2