Logical reasoning resides in a small brainlike part of language models
Logical subspace in LLMs
Artificial IntelligenceComputation and Language
Summary
Finding out whether language models have special parts for logical thinking like the human brain, the authors devised a way to pinpoint tiny areas in the model’s ‘brain’ that do the thinking. They found that these small parts handle logic tasks without helping with other abilities like memory or math. Removing these parts breaks logical thinking but leaves other skills intact. This shows that logical reasoning is localized in language models just as in human brains.
What this means in practice
- •For ai model developers: Isolate and manipulate logical reasoning abilities within large language models for targeted improvements or interpretability.
- •For software engineers building ai assistants: Design assistant features that rely on dependable logical reasoning by preserving critical model subspaces identified for inference.
Authors
Hope Kean, Enric Boix-Adsera
Abstract
Recent work has identified a human brain network specialized for abstract formal reasoning (Kean et al., 2025). Does the same hold true in language models? To answer this question, we introduce the minimal viable subspace (MVS) method, which searches for the lowest-rank activation subspace at a layer that preserves task performance when everything outside that subspace is ablated. Using MVS, we demonstrate low-rank subspaces supporting logical inference on Gemma and Qwen models. Furthermore, these subspaces exhibit a clear dissociation from model capacities on other tasks, such that retaining these late logic subspaces preserves inference while impairing factual knowledge, working memory, cognitive control, and arithmetic. Conversely, ablating them reduces logical inference accuracy to chance while largely sparing these other capacities. Our results suggest a functionally localizable core machinery for logic akin to that in the human brain.