ActivationReasoning grounds logical reasoning in LLM latent activations via SAEs to enable structured inference, concept composition, and behavior steering on multi-hop, abstraction, and safety tasks.
Measuring and guiding monosemanticity.arXiv preprint arXiv:2506.19382,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
ActivationReasoning: Logical Reasoning in Latent Activation Spaces
ActivationReasoning grounds logical reasoning in LLM latent activations via SAEs to enable structured inference, concept composition, and behavior steering on multi-hop, abstraction, and safety tasks.