LeCun et al. prove Platonic Representation Theorem for multi-modal AI models
Researchers turn a hypothesis about AI convergence into a mathematical theorem, showing Joint-Embedding Predictive Architectures learn the same world representation across modalities.
What to know
- Researchers proved that Joint-Embedding Predictive Architectures (JEPAs) trained on different modalities mathematically converge to shared representations, converting a longstanding hypothesis into a formal theorem.
- The work provides theoretical grounding for multi-modal AI training and has implications for how smaller models can be trained more efficiently.
- Community discussion centers on whether the theorem is philosophically profound or merely formalizes existing intuitions about optimal dimensionality reduction and task-specific representations.
The dispute Whether the theorem expresses a genuine insight about AI representation or merely formalizes straightforward principles about optimal dimensionality reduction. · positions read across 20 posts and comments
The theorem merely formalizes intuitive ideas about optimal dimensionality reduction without adding genuine insight.
-
“part of me thinks it might just be saying 'There is an optimal way to perform dimensionality reduction and as things get bigger it is easier to appoximate'”
Lerc · Hacker News ↗
The theorem is practically valuable for training smaller, more efficient multi-modal models.
-
“Im bullish on jepa models. Success of autoregressive next token prediction feels like a stroke of luck.”
kooi · Hacker News ↗
The proof may rest on assumptions too strong to capture the original hypothesis's ambition across all model types.
-
“I thought the platonic representation hypothesis was much more ambitious: wasn't it supposed to apply for any model across modalities (e.g. including human minds), rather than just the same model but with different modalities?”
suddenlybananas · Hacker News ↗
“Whether models trained on different modalities converge to the same representation of the world is central to AI.”
Yann LeCun and Randall Balestriero · arXiv abstract ↗
Yann LeCun Research authorRandall Balestriero Research author
How it unfolded 2 developments, newest first · click a bar or a number to jump articlespostscomments
-
2
Paper surfaces on Hacker News frontpage
The arXiv preprint reached Hacker News, generating discussion among researchers and practitioners about the implications of the theorem and its practical relevance.
“I'm not sure how I feel about the Platonic Representation Hypothesis. In some ways it seems like it should be profound, but part of me thinks it might just be saying 'There is an optimal way to perform dimensionality reduction…'”
— Lerc, Hacker News commenter · source -
All strands of scientific studies eventually are bound by "physical" reality. This assumption means all these research would gradually become more and more relevant, like math problems are highly relevant to physics, computer science, psychology is similar to experimental physics (observation and conjecture theorizing eventually crystalize into…
2 more of the top 3 · 20 posts in this stretch
-
L
ive been wondering why the frontier open video/image models can be run on consumer hardware (minimax h3 runs on a 4090) while frontier open llms are wayy larger and cant run on consumer hardware think this paper by lecun explains why that is?
-
Haven’t thought too deeply about this, but as a physicist we’d do things like switch from Euclidean to Spherical coordinates or a completely different dynamical representation like the Langrangian, and this makes certain problems dissolve (or at least become much simpler to solve analytically).Does it even make sense to claim there is a “platonic”…
-
-
1
LeCun and Balestriero prove Platonic Representation Theorem
Researchers published a proof converting the Platonic Representation Hypothesis into a formal theorem. The paper shows that Joint-Embedding Predictive Architectures trained on multiple modalities provably converge to the same representation of the world, moving from conjecture to mathematical guarantee.
-
first by HN Frontpage, 10d ago · also arXiv cs.AI
-
What people are saying 17 voices from 1 site · best of 20 · verbatim
- Can JEPAs formulate a scale-equivalent training approach to the success of autoregressive next-token prediction on unstructured text?
- Does the proof require assumptions that limit its applicability beyond JEPAs to broader classes of multi-modal models?
- Nov 4
-
I think I figured something interesting out regarding the NTP baseline in this paper. Changing how you read out the hidden states seems to matter a lot for how well the representations align and imo changes the papers outlook a bit on the sample efficiency of NTP vs JEPA. See my blog for experiment results
- Nov 2
-
For those reading the paper: the positives of a good-view JEPA are described as “two observations of one scene,” and yet the paper also states that “A good-view JEPA's positives are (g_m(z, ν), g_m(z′, ν′)) with (z, z′) ∼ π and independent nuisances.” My question is: why doesn't “two observations of one scene” mean that both observations share the…
-
Yeah, it's neither spooky not really that profound. Perhaps a more layman-friendly explanation:A good encoder has to capture real relationships between things in the world (what’s similar to what, what goes with what). Since there’s only one world, different good encoders converge to the same relational structure, which shows up as geometric…
- Nov 1
-
Different mathematical framework, but I wonder if this could be better understood using sheaf theory: each modality captures part of the same underlying reality, and those different views can be glued into a consistent bigger picture(Very basic intro:
-
I don't get this intuitively. So in the end, does this mean words are a kind of conceptualization of images? (Since images would come before language.) More than that, I'm curious whether this can improve the performance of vision-language models.
-
I haven't read this paper in details, but I have read papers about the platonic hypothesis in the past. Usually one of their weakness is about the design an experiment proving that "neural networks converges towards the same latent space when scaling". Usually authors are effectively proving a subset of the hypothesis, that can be explained by the…
-
I am sick of theory papers that adds human interpretation to connect a mathematical result to some fancy terms that everyone seeks after in real-world ML applications. Ironically, chatGPT is the most productive way to strip off those advertisement so the essence is better exposed.
-
how come all the examples in demo have wrong answershttps://huggingface.co/spaces/galilai-group/prh_demo> A man in a black shirt and hat with a yellow neckerchief has his hand raised.no hand raised> A white sheep sitting on top of a cake.no its just cake sheep . sheep is not sitting> A man is standing on the beach while holding his surfboard.Its…
-
I guess the fundamental question would be, for what domains can we pretrain with “good views”?Part of the magic of next token prediction is it works to extract intelligence from the unstructured web corpus.Can JEPA formulate a similar scale training set of “good views”?
-
> 33M-parameter text JEPA reaches .125 on the PRH paper’s benchmark, against .034 on Flickr captions, level with Qwen2.5-3B and -7B. They need fewer samples.Im bullish on jepa models. Success of autoregressive next token prediction feels like a stroke of luck.
-
The Platonic Representation Hypothesis initially feels profound but actually seems inevitable after some thought because all neural networks are trying to model the same reality. It would actually be weirder if it wasn't true and neural networks could form completely incompatible representations of the same reality. In some sense it seems similar…
-
What if we ultimately find there are two different, internally consistent but mutually incompatible, representations?What if one of them is a Republican and one is a Democrat? We already see this in some systems today.More worrying, what if one is (for lack of a better word, what we might describe as) God and one is the Devil?
-
Do memes have a concept of proto memes? Like primes are the proto numbers that all other numbers are made from? Could there be similar ontological primes for all concepts / semantic space? Base reality concepts that are irreducible so to speak.
-
I thought the platonic representation hypothesis was much more ambitious: wasn't it supposed to apply for any model across modalities (e.g. including human minds), rather than just the same model but with different modalities? I'll have to actually read the paper but I suspect there to be very strong assumptions which defeat the interesting…
-
It’s less profound sounding if instead of talking about “representations of the world” you talk about compressed representations of data or decision boundaries in a classifier over data.
-
Seems to me that it matters because it has implications for how we can train smaller models?
-
I'm not sure how I feel about the Platonic Representation Hypothesis.In some ways it seems like it should be profound, but part of me thinks it might just be saying 'There is an optimal way to perform dimensionality reduction and as things get bigger it is easier to appoximate'