What Converges in the Platonic Representation Hypothesis? Structure over Geometry
Abstract
The Platonic Representation Hypothesis suggests that increasingly capable models converge toward shared representations. Recent work narrows this claim to shared local neighborhood relationships, finding that capacity-dependent trends in several global similarity measures largely disappear after calibration. We challenge this interpretation by showing that prior local-global comparisons confound structural scale (local versus global) with what is compared: relational structure, defined by which samples are related, versus metric geometry, characterized by quantitative relations such as distances, similarities, or correlations. To disentangle these factors, we construct a controlled 2×2 framework that evaluates both relational structure and metric geometry at local and global scales. We introduce H0 skeleton overlap as a global counterpart to mutual k-nearest neighbors, together with matched distance-aware variants. Across vision-language models, relational structure exhibits robust representational convergence at both scales after calibration, whereas increasingly stringent distance agreement substantially weakens alignment and progressively flattens the capacity-dependent trend. We further extend the analysis beyond ambient Euclidean geometry by evaluating distance agreement under a Riemannian metric approximation and recover the same structure-geometry pattern. The pattern is also reproduced in video-text representations. Together, these results show that relational convergence extends beyond local neighborhoods to global spanning structure, whereas metric geometry exhibits substantially weaker convergence.
Related
- Proving and Improving the Platonic Representation Hypothesis with JEPAs
- The Operator–Spectral Framework of ZEBTS: A Unified Structure for Geometry, Arithmetic, and the Riemann Hypothesis
- Rethinking Representation Geometry in Supervised Contrastive Learning under Long-Tailed Data
- Elimination Geometry II: Integrability, Representation, and Structural Limits
- Geometry-Aware Latent Representation Learning for Drag Coefficient Prediction of Irregular Particles
BibTeX
@misc{you2026what,
title = {{What Converges in the Platonic Representation Hypothesis? Structure over Geometry}},
author = {Junwon You and Mihyun Jang and Sangwoo Mo and Jae-Hun Jung},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.27252},
eprint = {2609.27252},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.27252},
}
RIS
TY - UNPB
TI - What Converges in the Platonic Representation Hypothesis? Structure over Geometry
AU - Junwon You
AU - Mihyun Jang
AU - Sangwoo Mo
AU - Jae-Hun Jung
PY - 2026
DA - 2026/09/23
PB - arXiv
DO - 10.48550/arxiv.2609.27252
UR - https://arxiv.org/abs/2609.27252
AB - The Platonic Representation Hypothesis suggests that increasingly capable models converge toward shared representations. Recent work narrows this claim to shared local neighborhood relationships, finding that capacity-dependent trends in several global similarity measures largely disappear after calibration. We challenge this interpretation by showing that prior local-global comparisons confound structural scale (local versus global) with what is compared: relational structure, defined by which samples are related, versus metric geometry, characterized by quantitative relations such as distances, similarities, or correlations. To disentangle these factors, we construct a controlled 2×2 framework that evaluates both relational structure and metric geometry at local and global scales. We introduce H_0 skeleton overlap as a global counterpart to mutual k-nearest neighbors, together with matched distance-aware variants. Across vision-language models, relational structure exhibits robust representational convergence at both scales after calibration, whereas increasingly stringent distance agreement substantially weakens alignment and progressively flattens the capacity-dependent trend. We further extend the analysis beyond ambient Euclidean geometry by evaluating distance agreement under a Riemannian metric approximation and recover the same structure-geometry pattern. The pattern is also reproduced in video-text representations. Together, these results show that relational convergence extends beyond local neighborhoods to global spanning structure, whereas metric geometry exhibits substantially weaker convergence.
ER -