The Platonic Representation Hypothesis

TL;DR

This paper explores the convergence of representations in AI models across domains and modalities, proposing a shared 'platonic' ideal model of reality that emerges as models grow larger and more aligned.

Contribution

It introduces the concept of the platonic representation hypothesis, highlighting the convergence of neural network representations across data types and discussing its implications.

Findings

Neural network representations are becoming increasingly aligned across domains.

Larger models measure data distances in more similar ways.

Convergence suggests a shared statistical model of reality.

Abstract

We argue that representations in AI models, particularly deep networks, are converging. First, we survey many examples of convergence in the literature: over time and across multiple domains, the ways by which different neural networks represent data are becoming more aligned. Next, we demonstrate convergence across data modalities: as vision models and language models get larger, they measure distance between datapoints in a more and more alike way. We hypothesize that this convergence is driving toward a shared statistical model of reality, akin to Plato's concept of an ideal reality. We term such a representation the platonic representation and discuss several possible selective pressures toward it. Finally, we discuss the implications of these trends, their limitations, and counterexamples to our analysis.