Nicolas Saunier
Thomas FuchsAnd yet another reply, “But the human brain works just like an LLM!”
Yes, famously we fully understand how human brains work, maybe just some minor details left to figure out.
Natalie 🪗@thomasfuchs what, you don’t use high-dimensional matrix multiplication to think? (sarcasm)
Raptor's Nest (He/Him)@thomasfuchs I find it incredibly depressing that people think our thoughts are composed of next word prediction…
eclexic@drmambobob @thomasfuchs I'm convinced they are right... but only about their own cognition.
Neo Ehproque@drmambobob @thomasfuchs it would explain a lot though (about those people)
acid@thomasfuchs I think we know well enough about the human brain to know that this claim isn’t true at all.
Stefan Böhringer@thomasfuchs Therefore I highly recommend to read another „Thomas Fuchs“ (but sadly only German):
„Das Gehirn - ein Beziehungsorgan“
https://elibrary.kohlhammer.de/book/10.17433/978-3-17-045784-3
@datenschauer looks like another Thomas Fuchs with good takes
@thomasfuchs in fact for deaf-mutes it doesn't work at all until they learn how to read!
Blåhaj@thomasfuchs Sometimes the argument is "well, we don't know how brains work, so you can't prove they *don't* work like an LLM"
Albert CardonaAs the book “The Idea of the Brain” by Matthew Cobb says, we always explain how the brain works by whatever is the latest technology: clocks, steam engine, hydraulics, computers, now LLMs… and all explanations fall very, very short, revealing more about those who propose them than about the brain.
https://www.theguardian.com/books/2020/nov/19/the-idea-of-the-brain-by-matthew-cobb-review
v@thomasfuchs they replicate our best understanding of what neurons do when we recognize a pattern or association, right? That feels very different from cognition
@NocturnalNessa they don’t replicate biological neurons, but a highly abstract model of what they thought of neurons in the 1950s
@thomasfuchs my basic understanding that this is an abstracted reconstruction of pattern recognition and very much not anything that could be called cognition - is that correct enough?