Brain-Language Model Alignment: Insights into the Platonic Hypothesis and Intermediate-Layer Advantage

TL;DR

This paper reviews recent fMRI studies to explore whether brain and language model representations align, supporting the hypotheses that models converge to real-world representations and that intermediate layers encode more general features.

Contribution

It provides a comprehensive review of recent neural and model alignment studies, testing the Platonic and Intermediate-Layer hypotheses.

Findings

Evidence of shared abstract representations between brains and models

Support for the convergence of models toward real-world representations

Intermediate layers encode richer, more generalizable features

Abstract

Do brains and language models converge toward the same internal representations of the world? Recent years have seen a rise in studies of neural activations and model alignment. In this work, we review 25 fMRI-based studies published between 2023 and 2025 and explicitly confront their findings with two key hypotheses: (i) the Platonic Representation Hypothesis -- that as models scale and improve, they converge to a representation of the real world, and (ii) the Intermediate-Layer Advantage -- that intermediate (mid-depth) layers often encode richer, more generalizable features. Our findings provide converging evidence that models and brains may share abstract representational structures, supporting both hypotheses and motivating further research on brain-model alignment.