Volver a Artículos
Académico Ciencia y Salud inglés

Reading Without a Reader: Large Language Models Collapse Reading and Writing into a Single Entangled Code

Autor: Diego Saldana Ulloa

Fecha de publicación: 02 de julio de 2026

Visualizaciones: 14


Resumen

In the literate human brain, reading and writing doubly dissociate: a ventral decoding route (pure alexia) and a fronto-parietal encoding route (pure agraphia), sharing a partial orthographic core. A decoder-only large language model (LLM) drives both from one autoregressive path optimized on text (a \emph{cultural} invention, not an evolved instinct). We ask how entangled it is, comparing an input-side ``reading code'' WE with an output-side ``writing code'' WU via an index E∈[0,1] (CKA, Procrustes residual, mutual k-NN) calibrated against an independent-init floor and tied ceiling. On GPT-2, OPT and Pythia (14M--1.4B), untied models hold one \emph{coupled but sub-ceiling} code (E=0.23--0.35, far above floor) on a non-monotonic couple-then-differentiate trajectory, WU drifting ∼3.2× farther than WE in every decile. Equally informative is a negative: the matching behavioural test, that comprehension and production fail together rather than dissociate, cannot be run. For minimal pairs the alexia analogue is em

Palabras clave
large language models mechanistic interpretability cognitive neuroscience of reading embeddings
Información del Artículo

Tipo:
Académico

Categoría:
Ciencia y Salud

Idioma:
inglés

Visualizaciones:
14

Sobre el Autor
Diego Saldana Ulloa

Director Data Science

Moneypool

Ver Perfil Completo