ActivationReasoning: Logical Reasoning in Latent Activation Spaces

TL;DR

ActivationReasoning (AR) introduces a three-stage framework to embed explicit logical reasoning into the latent activation space of LLMs by grounding propositions in sparse autoencoder (SAE) features and applying user-defined rules. The stages—finding latent representations to form a concept dictionary $D$, activating propositions to build activation matrices $A$, and applying logical rules $L$ to derive an enriched $A'$—enable structured inference and controllable generation. Across PrOntoQA, Rail2Country, ProverQA, and BeaverTails, AR delivers robust multi-hop reasoning, generalizes to abstract meta-descriptions, and improves safety concept handling, often outperforming much larger models and prompting-based approaches, with only modest runtime overhead. By transforming latent activations into a transparent, auditable substrate for reasoning and control, AR provides a practical pathway toward more reliable and steerable AI systems. It remains complementary to SAEs and invites future work on automated rule induction, probabilistic reasoning, and external knowledge integration to scale reasoning in open-ended tasks.

Abstract

Large language models (LLMs) excel at generating fluent text, but their internal reasoning remains opaque and difficult to control. Sparse autoencoders (SAEs) make hidden activations more interpretable by exposing latent features that often align with human concepts. Yet, these features are fragile and passive, offering no mechanism for systematic reasoning or model control. To address this, we introduce ActivationReasoning (AR), a framework that embeds explicit logical reasoning into the latent space of LLMs. It proceeds in three stages: (1) Finding latent representations, first latent concept representations are identified (e.g., via SAEs) and organized into a dictionary; (2) Activating propositions, at inference time AR detects activating concepts and maps them to logical propositions; and (3)Logical reasoning, applying logical rules over these propositions to infer higher-order structures, compose new concepts, and steer model behavior. We evaluate AR on multi-hop reasoning (PrOntoQA), abstraction and robustness to indirect concept cues (Rail2Country), reasoning over natural and diverse language (ProverQA), and context-sensitive safety (BeaverTails). Across all tasks, AR scales robustly with reasoning complexity, generalizes to abstract and context-sensitive tasks, and transfers across model backbones. These results demonstrate that grounding logical structure in latent activations not only improves transparency but also enables structured reasoning, reliable control, and alignment with desired behaviors, providing a path toward more reliable and auditable AI.

Figures (2)

• Figure 1: AR applies logical rules on latent activations to enable downstream reasoning, abstraction, and control. AR composes missing concepts from existing SAE features (top left), aids robustness to spelling errors or similes (top right), implements rules for safeguarding at the neural level (bottom left), and solves multi-hop reasoning tasks via logical inference in the latent space (bottom right).
• Figure 2: Overview of ActivationReasoning. AR performs logical reasoning over LLM activations in three stages: (1) Finding latent representations, where concepts are identified in the SAE latent space and stored in a concept dictionary using single, multi, or relational feature representations; (2) Activating propositions, where token-level activations are detected during inference to form an activation matrix $A$; and (3) Logical reasoning, where pre-defined rules are applied over $A$ to infer new higher-order structures, compose new propositions, yielding an enriched matrix $A'$. The structured activations can then be used for downstream transparency and control.