Comment by gruez
1 day ago
>Human learning is materially different from LLM training [...] but the similarities end there.
Specifically what "material differences" are there? The only arguments I heard are are around human exceptionalism (eg. "brains are different, because... they just are ok?"), or giving humans a pass because they're not evil corporations.
We don't understand human brains well enough to answer this question specifically: if we understood the mechanisms of the human brain, we could replicate them in software, and AI would be that much more advanced. We know that real human neurons don't work at all like artificial neural networks, but that isn't a proof: phonons aren't bosons or fermions (their spin isn't well-defined), yet a sonic black hole is a useful model for Hawking radiation.
So, I'll provide an example: humans can learn to do mathematics. LLMs cannot. This example is particularly galling because there are computer programs that can do (some, limited) mathematics: those operate largely by brute-force, yet can solve more mathematics problems using fewer resources than LLMs.
Humans can generalize.
LLMs just predict the statistically-most-likely token.
human brains are just chemical reactions and electrical transmission between neurons too. You're comparing completely different layers of abstraction in your arguments.