Critique claims the Platonic Representation Hypothesis is a statistical artifact, but co-author Phillip Isola clarifies both papers agree

Wait for the next week's episode of the Information Bottleneck podcast to see what @phillip_isola thinks about this work...

@giffmana Perhaps I'm just not looking in the right places, but where are AI researchers taking classical philosophy most seriously? I'm thinking specifically of the distinction between the rational nature of the soul vs. the organ of the brain and how most AI research skips past all that.

@giffmana Platonic representational hypothesis may not be entirely wrong, if we focus on the “convergence” idea. But the paper itself is a different story yeah.

@giffmana I’m currently trying to explore these things https://www.alpernebikanli.com/blog/posts/representational-convergence-1

@abursuc @giffmana Yeah it's a nice paper! I think people are assuming it disagrees with PRH, but that's not quite right.

Both papers find that it's local neighbors, not global structure, that is converging.

New paper nicely elaborates on this point, but it's also in original PRH:

@trmtoo @giffmana nowhere! most are functionalists -- if philosophy doesnt guide concrete directions nor leads to production phenomenas they dont care.

@alpernebikanli @giffmana Exactly, we did not try to claim there is no global convergence. But using the current experimental setting, we should not come to that conclusion.

I explore convergence of several AI models via emotion scoring to write papers and articles and complaints.

It’s wild but works really well.

While there are similarities overall the models have enough differences that Gradient descent on their convergence brings meaningful results.

@t_d_tr @giffmana My extreme lack of surprise equals my disappointment!

@sanmking Ah that would explain it :)

@trmtoo @giffmana do you have any concrete concerns between the distinction?

@giffmana In fairness, @phillip_isola did share it on his timeline at that time

@sanmking @giffmana That is also one reason we chose the name :)

@t_d_tr @giffmana Yes. For everyone before Materialism happened, rationality was a power of the soul, which perhaps *used* the organ of the brain, but was not produced solely by the brain's structure. It seems to me that most AI people think if you build a brain you achieve rationality.

@giffmana Good, but isn’t the platonic representation hypothesis obviously false? I thought it was something like a joke paper, or overreach by some philosophers new to the field

@giffmana This why I follow the GOAT

@giffmana It is a massive burn. They proved the entire Platonic hypothesis was just a math error from using uncalibrated similarity metrics.

@giffmana Happens all the time — impact often inversely correlated with how politely worded the critique is.

@giffmana wait, which paper is it? i have to see this.

@giffmana To be fair, Aristotle only achieved fame far later than Plato.