激活导向能否替代 LLM 数字孪生中的提示词情感启动?
Can Activation Steering Substitute for Prompt-Based Affective Priming in LLM Digital Twins?
AI 导读
一项 arXiv 研究测试用激活导向替代情感启动中的写作任务,在 Llama-3.1-8B-Instruct 上对 100 个人设、gratitude 与 lack of control 两个领域进行实验。
正文 · 原文
Abstract:Affective priming -- asking participants to write about a recalled event, such as a time they felt grateful or lacked control, and then measuring the induced state and its spillover onto related judgments -- is a standard way to manipulate emotion in the social sciences, and it also shifts the responses of LLM-based digital twins. The writing task, however, requires generating a full text. We test whether activation steering can replace it. Using Llama-3.1-8B-Instruct on 100 personas in two domains (gratitude, lack of control), we compare a real-prompt path replicating the original Twin-2K-500 Mega Study (J2) with an injection path that skips the writing and adds a mean-difference direction, extracted from human writing, to the residual stream at varying strengths. The real-prompt effects fall between the original human and digital-twin effects in 5 of 6 comparisons. Injection shifts the manipulation check in a dose-dependent way and matches the real-prompt effect at one strength in the gratitude domain (beta ~ 2.7), but overshoots beyond it, on the proximal measure as well, and asymmetrically across the two directions. A structure-matched No-Injection baseline separates two sources of difference: omitting the writing task raises the proximal measure in both domains, whereas the injected direction itself moves a distal measure in the lack-of-control domain. Substitution is therefore measure-dependent: validity established for one way of inducing a state does not carry over uniformly to another.
| Comments: | Code, generation logs and analysis scripts: this https URL |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.22738 [cs.AI] |
| (or arXiv:2611.22738v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2611.22738 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Junyi Fan [view email]
[v1]
Tue, 17 Nov 2026 13:08:06 UTC (232 KB)
来源:arXiv cs.AI · arxiv.org