aiqbench analysis
This study empirically validates that foundation model representations converge toward shared physical realities when scaled sufficiently, even without domain-specific fine-tuning. By using astronomy as a rigorous external reference, the research demonstrates that local embedding alignment correlates with astrophysical performance across diverse architectures and modalities. These findings provide critical insights into the limits of generalization for large models applied to scientific data analysis.
The Platonic Universe: Do Foundation Models See the Same Sky?
arXiv:2509.19453v2 Announce Type: replace-cross Abstract: We investigate when foundation models converge towards shared representations, and how this convergence depends on model capacity, training regime, and model architecture. We take a `science-for-AI' approach, using astronomy as an experimental instrument to test the Platonic Representation Hypothesis and its Aristotelian refinement against an external physical reference. The historical success of astrophysics is evidence that a compact, modality-invariant description of galaxy observables exists, and so representation convergence toward reality should be measurable against the physical parameters astronomers already use. Given this framework, we evaluate eleven foundation model families (spanning classification, self-distillation, joint-embedding prediction, autoencoding, vision-language pre-training, and astro-specific architectures from $\mathcal{O}$(10M)${\to}\mathcal{O}$(10B) parameters) on crossmatched JWST, HSC, and Legacy imagery, and DESI spectroscopy. All models are evaluated frozen, with no astronomy-specific fine-tuning. We probe redshift, stellar mass, and sSFR via linear probes, and local (MKNN) and global (CKA) embedding geometry within families, between modalities, and across architectures. We find that physics performance scales predictably with capacity; probe directions align consistently with expected astrophysical correlations and selection effects; and local (not global) embedding alignment tracks physics performance, including between DESI spectra and HSC imagery---modalities that share essentially no low-level statistics. Our results support the ARH over the strict PRH, demonstrate astronomy's value as an experimental framework for neural representation learning, and suggest that astro-foundation models can build on general-purpose pre-trained architectures, capitalizing on the broader open machine learning community's already-spent computational investment.
Authors:UniverseTBD: Trinidad Borrell, Steven Dillmann, Kshitij Duraphe, Furkan Eris, Kartheik Iyer, Ashod Khederlarian, Aman Kumar, Giovanni Marraffini, Michael J. Smith, Shashwat Sourav, Rocco Di Tella, John F. Wu
Abstract:We investigate when foundation models converge towards shared representations, and how this convergence depends on model capacity, training regime, and model architecture. We take a `science-for-AI' approach, using astronomy as an experimental instrument to test the Platonic Representation Hypothesis and its Aristotelian refinement against an external physical reference. The historical success of astrophysics is evidence that a compact, modality-invariant description of galaxy observables exists, and so representation convergence toward reality should be measurable against the physical parameters astronomers already use. Given this framework, we evaluate eleven foundation model families (spanning classification, self-distillation, joint-embedding prediction, autoencoding, vision-language pre-training, and astro-specific architectures from $\mathcal{O}$(10M)${\to}\mathcal{O}$(10B) parameters) on crossmatched JWST, HSC, and Legacy imagery, and DESI spectroscopy. All models are evaluated frozen, with no astronomy-specific fine-tuning. We probe redshift, stellar mass, and sSFR via linear probes, and local (MKNN) and global (CKA) embedding geometry within families, between modalities, and across architectures. We find that physics performance scales predictably with capacity; probe directions align consistently with expected astrophysical correlations and selection effects; and local (not global) embedding alignment tracks physics performance, including between DESI spectra and HSC imagery---modalities that share essentially no low-level statistics. Our results support the ARH over the strict PRH, demonstrate astronomy's value as an experimental framework for neural representation learning, and suggest that astro-foundation models can build on general-purpose pre-trained architectures, capitalizing on the broader open machine learning community's already-spent computational investment.
Instrumentation and Methods for Astrophysics (astro-ph.IM); Machine Learning (cs.LG)
arXiv:2509.19453 [astro-ph.IM]
(or arXiv:2509.19453v2 [astro-ph.IM] for this version)
Full text under the publisher's open licence · 276 words
Related coverage
Toward a Kantian refutation of Agent Foundations
LessWrong · Jun 17, 2026
Import AI 474: Platonic mindspace; TPUs in space; Zhipu starts an outer RSI loop
Import AI · Sep 28, 2026
3D Ising criticality with Platonic lattice superconducting qubits
arXiv · Jun 16, 2026
News
Quantum Computing Report · Jul 25, 2026