2 ms·
Imagine trying to write a better take on GPTs than https://generative.ink/posts/simulators/ https://generative.ink/posts/simulators/ it will never happen. Anywa
by ftxbro 4y ago
Imagine trying to write a better take on GPTs than https://generative.ink/posts/simulators/ https://generative.ink/posts/simulators/ it will never happen. Anyway, Yud's main point is obviously true and many people either don't understand it or don't believe it:
"GPT-4 is still not as smart as a human in many ways, but it's naked mathematical truth that the task GPTs are being trained on is harder than being an actual human."
"And since the task that GPTs are being trained on is different from and harder than the task of being a human, it would be surprising - even leaving aside all the ways that gradient descent differs from natural selection - if GPTs ended up thinking the way humans do, in order to solve that problem."
- user_named 4y agoIf being human is less hard, then how come GPT fails at it?
- camel-cdr 4y agoBecause it's not trained to be a human.
- catchnear4321 4y agoyud celebrates yud the comparison of gradient descent to natural selection is just silly of course such things won’t think the same but they might simulate it