Friends and Grandmothers in Silico: Localizing Entity Cells in Language Models

TL;DR

This paper identifies and manipulates sparse, entity-specific neurons in language models, demonstrating their causal role in factual recall and revealing insights into how models encode entity knowledge.

Contribution

It introduces a method to localize entity cells in language models and provides causal evidence of their role in factual recall, highlighting their stability and surface form invariance.

Findings

Localized neurons cluster in early layers across models.

Suppressing a neuron erases recall for its entity; activating it recovers knowledge.

Entity cells encode canonical identity, not surface tokens, and are stable across tuning.

Abstract

How do language models retrieve entity-specific facts from their parameters? We investigate this question by searching for sparse, entity-selective MLP neurons - which we call entity cells, by analogy to the "grandmother cell" hypothesis in neuroscience - and testing whether they play a causal role in factual recall. We localize candidate entity cells by ranking MLP neurons for activation consistency across varied prompts about the same entity, applying this procedure across seven models on a curated subset of PopQA. In all models, localized neurons cluster predominantly in early layers, an empirical pattern not imposed by the architecture. Using Qwen2.5-7B base as a model organism, we find the clearest causal evidence: suppressing a localized cell selectively erases recall for its matched entity while leaving others intact, and activating a single cell is sufficient to recover correct knowledge for most entities - even when the entity is absent from the context. The same cells are recovered under aliases, acronyms, misspellings, and multilingual surface forms, and remain stable through instruction tuning, suggesting they encode canonical entity identity rather than surface token patterns. Causal signals vary across model families, pointing to architectural differences in how entity knowledge is organized. These findings offer concrete, interpretable access points for understanding, controlling, and correcting factual knowledge in language models, and draw a surprising empirical parallel to longstanding questions in neuroscience about sparse coding of concepts.