analiza aiqbench
This research introduces EmoVec, a lightweight framework that enables precise control over emotional intensity in Large Language Models without retraining weights. By extracting emotion-specific directions from latent vectors and injecting them during inference, the method effectively counters the emotionally flattened responses often seen in aligned models while preserving semantic coherence. This approach offers a practical solution for deploying affect-sensitive AI applications that require nuanced emotional expression.
Controllable Affective Generation via Latent Vector Steering
arXiv:2608.25569v1 Announce Type: cross Abstract: Large Language Models (LLMs) often produce emotionally flattened responses after alignment, limiting their effectiveness in affect-sensitive applications. In this paper, we propose EmoVec, a lightweight framework for controllable affective generation via latent vector steering. EmoVec extracts emotion-specific directions from paired neutral and emotion-conditioned responses using contrastive activation addition, and further refines them through task-specific debiasing and principal subspace removal. During inference, these vectors are injected into the final residual stream with static or scenario-adaptive scaling, enabling continuous control over emotional intensity without updating model weights. Experiments across three LLMs and eight emotions show that EmoVec consistently improves emotional salience while largely preserving semantic content, fluency, and coherence. Ablation studies and human evaluation further confirm the effectiveness of vector purification and adaptive scaling, establishing EmoVec as a practical inference-time method for affective control in deployed LLMs.
Abstract:Large Language Models (LLMs) often produce emotionally flattened responses after alignment, limiting their effectiveness in affect-sensitive applications. In this paper, we propose EmoVec, a lightweight framework for controllable affective generation via latent vector steering. EmoVec extracts emotion-specific directions from paired neutral and emotion-conditioned responses using contrastive activation addition, and further refines them through task-specific debiasing and principal subspace removal. During inference, these vectors are injected into the final residual stream with static or scenario-adaptive scaling, enabling continuous control over emotional intensity without updating model weights. Experiments across three LLMs and eight emotions show that EmoVec consistently improves emotional salience while largely preserving semantic content, fluency, and coherence. Ablation studies and human evaluation further confirm the effectiveness of vector purification and adaptive scaling, establishing EmoVec as a practical inference-time method for affective control in deployed LLMs.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2608.25569 [cs.CL]
(or arXiv:2608.25569v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2608.25569 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 233 cuvinte
Subiecte conexe
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
LessWrong · 8 oct. 2026
Evaluating task vectors, unlearning and inoculation
LessWrong · 20 sept. 2026
CoCoEmo: Composable and Controllable Human-Like Emotional TTS via Activation Steering
arXiv · 17 iun. 2026
PIRAMID: Progress and Plans
LessWrong · 27 iul. 2026