Weeks where decades happen at home:

If the AI were really as effective as the AI companies are saying, they'd use it for themselves instead of selling it at a loss

I'm starting to think he's actually stupid enough to think LLMs are intelligent

What kind of dweeb says "a few thousand days away" instead of "in a couple of years"?

This definitely 100% means that OpenAI has nothing they haven't shown the world yet and that they don't expect the models to get better.

So that's why we wacked that other guy that AI totally didn't put an automated bitcoin black market hit order on.

"AI will make it so that decades happen, every week" - Vladimir Linux

Basically admitting that so-called “AI” is nothing and will always be nothing, but doing it in a way that smooth-brained investors still think it’s smart to get on the hype train.

I’m shocked that a guy whose wealth is based on investor hype for AI makes a statement that AI is going to be exponentially more valuable.

The guy is such a con artist. All the statements about how they fear the AI will be so smart and powerful are coming from from the AI owners. It’s bullshit, If they actually thought their AI could kill everyone, they wouldn’t be bragging about it in the press.

I genuinely hope that this garbage technology falls apart at the seams and soon. Not only is it destroying the environment, but it's becoming a large factor into brain drain amongst students and adults.

never mind that the tendency of the rate of profit to fall also applies to AI training

This shit is gonna fail soon and take this sham of an economy with it. the more outlandish the claims, the more guaranteed that is.

Sam Altman should be tortured

there are weaks where seconds happen

and there are seconds where you go back for thirds, because the crismas turkey was real good

I highly highly doubt it. It is really good for the main thing it is designed for: language and translation. And because of it's function it can hallucinate for you decent code or answers to basic issues.

Try something, I've done this personally. Go to any of the big "AI" models right now. In another tab, open up the NYT games, letter boxed. Now try and teach the "AI" to play letter boxed. It isn't a complicated game but the few restrictions you do have are hard restrictions. For a language model coming up with words that fit the restriction and giving you a solution should be pretty straightforward if it is truly smart. Just a list of a few words that meet the criteria of the game that will solve today's puzzle is all it needs to do. Do it. See what happens.

EDIT: Been a few weeks since I last tried. Just tested it on 3 big models. All failed. The answers almost look right for some of them. But they fail. Here was my input:

You will solve a word puzzle. All rules must be followed all the time.

There will be four arrays of letters labelled 1, 2, 3, 4.
The first word may start with any letter from any of the arrays.
A word must be in the English language.
A word can contain more than one letter from any array.
A word can not contain two letters next to each other from the same array.
A word may not contain the same two letters next to each other.
After the first word is chosen the next word must contain the last letter of the previous word.
All letters from the arrays must be used at least once.
A solution should contain less than 5 words.

1: E, T, N;
2: O, S, B;
3: W, H, U;
4: V, A, L;

Solve this word puzzle using the rules and letter arrays as stated.