Post

This is also how I tend to think about it! As Olivia says though ANNs can simulate pretty much anything. So if an ANN succeeds through training, that doesn't say much about whether the same happens in humans. It's more of a theoretical limit as you say

But ANNs give us a chance to think in terms of architecture vs training which is much more concrete than the hand wavy nature vs nurture dichotomy. We can think of what architectures and curricula are more plausible (though at a very abstract level) and in that direction things get more interesting

I fully agree with that! Now I'm curious about what you think about this www.nature.com/articles/s41... Is it full M? Kinda M? (sorry if you already commented on it before and I missed it...)

even just from name you can probably guess what I think ☺️ — but what you do think if you have read both works and my previous work?

I think it's full M but the only thing that throws me off is that they still argue for treating architecture, training data, objective and learning rule as manipulable hypotheses. But I'm probably misunderstanding something.. Or is it that they say it but in practice they don't do it enough?

Can I take a minute to register my distaste for capital mathfrak S, I didn't have the faintest idea what that fucking letter was supposed to be until grad school

🫠 I'm so sorry it's a C

This font will never leave me alone