analiza aiqbench
This paper studies whether LLMs can track implicit emotional state across multi-turn conversations, and proposes SIAB, a training-free method that uses causal activation patching to find emotion-relevant attention heads and boosts the tokens they already attend to. Reported gains of up to +10.3 macro-F1 on ECC and +5.5 on GoEmotions are measured against prompt-matched baselines on three LLMs, so they are useful evidence but not yet a general claim. For readers, the main takeaway is that mechanistic interpretability is starting to yield inference-time control knobs, though transfer beyond these benchmarks remains untested.
Emotion Is Not External Knowledge: Internal Attention Signals for Affective State Reasoning
arXiv:2610.21347v1 Announce Type: new Abstract: Large Language Models (LLMs) increasingly operate in multi-turn interactions that require maintaining latent conversational states, not only explicit facts. We study affective state as an implicit state inferred from sparse contextual cues that remain relevant after topic shifts, and ask whether LLMs expose internal attention signals for routing such evidence at inference time. Prior attention interventions rely on labeled in-domain head profiles and externally specified target spans, limiting transfer across affective-evidence distributions. We propose Self-targeted Internal Attention Boosting (SIAB), which localizes sparse affective heads via causal localization (in detail activation patching) on synthetic minimal pairs and boosts, at the answer position, the content tokens these heads already attend to; both edited heads and target tokens are derived from the model's own behavior, enabling the same head set to be reused across tasks.We evaluate on ECC and GoEmotions across three LLMs and multiple prompting settings. Across settings, SIAB improves prompt-matched baselines by up to +10.3 macro-F1 on ECC and +5.5 on GoEmotions. Combined with PASTA-style external supervision, it achieves up to +13.84 accuracy points and +12.51 macro-F1 points, while mechanistic analyses trace the gains to sparse, emotion-enriched heads and show that layer proximity affects additivity.
Abstract:Large Language Models (LLMs) increasingly operate in multi-turn interactions that require maintaining latent conversational states, not only explicit facts. We study affective state as an implicit state inferred from sparse contextual cues that remain relevant after topic shifts, and ask whether LLMs expose internal attention signals for routing such evidence at inference time. Prior attention interventions rely on labeled in-domain head profiles and externally specified target spans, limiting transfer across affective-evidence distributions. We propose Self-targeted Internal Attention Boosting (SIAB), which localizes sparse affective heads via causal localization (in detail activation patching) on synthetic minimal pairs and boosts, at the answer position, the content tokens these heads already attend to; both edited heads and target tokens are derived from the model's own behavior, enabling the same head set to be reused across this http URL evaluate on ECC and GoEmotions across three LLMs and multiple prompting settings. Across settings, SIAB improves prompt-matched baselines by up to +10.3 macro-F1 on ECC and +5.5 on GoEmotions. Combined with PASTA-style external supervision, it achieves up to +13.84 accuracy points and +12.51 macro-F1 points, while mechanistic analyses trace the gains to sparse, emotion-enriched heads and show that layer proximity affects additivity.
Computation and Language (cs.CL)
arXiv:2610.21347 [cs.CL]
(or arXiv:2610.21347v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2610.21347 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 · 291 cuvinte
Subiecte conexe
[AINews] not much happened today
Latent Space · 10 sept. 2026
Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face
AI Alignment Forum · 3 aug. 2026
State of the Field: AI for Epistemics and Coordination
LessWrong · 2 oct. 2026
Some Intuitions on Steering Vectors
LessWrong · 29 sept. 2026