Grok is cooked
(twitter.com)
You're viewing a single comment thread. View all comments, or full comment thread.
Comments (17)
sorted by:
This is largely true, but only to the degree that most people don't think for themselves either.
This ain't how it works.
It is literally just a whole bunch of math to predict the next token in a sequence of tokens.
It literally isn't. You're more than a decade behind.
Grok (and others) can do math problems. And not just 2+2=4. It can execute algorithms. It can do spatial reasoning. If you have the mental capacity yourself, try asking it some difficult math problems and check the results. Try doing that with one of the statistics-based chatbots that you're thinking of and see what happens.
If it worked the way you claimed, all of that would be impossible. Neural nets aren't just "a whole bunch of math." Training them? Yes, that's a lot of math. Using them? No, that's not how they work at all.
Source: have designed and trained simple neural networks before. Have used Grok to analyze and help design a novel triangle mesh generation algorithm.
Does it have knowledge of the tokens or does it just have an improved ability to predict the next token in the sequence?
I don't think you even know you're asking, you're just repeating a received opinion that you think will make you "win". Whatever that means.