>>106219906
>unironically linking to the ai-2027 meme site
Bwhahahahahahaha
I see you drank the kool aid entirely.
Read/watch what Yan LeCunn (the man who has been trying to solve AGI for decades) has to say about LLMs:
Language alone cannot provide a world model for the emergence of intelligence. You cannot understand the world through language alone. LLMs have no understanding of the concepts behind the words and no theory of mind. To create true artificial intelligence, you would need to come up with a whole different structure altogether, more similar to a Human brain. Which is hard to do, because we do not fully understand how the Human brain works to begin with.
You are abscribing qualities to the LLMs that they do not posess. (as does that site)
They are simple neural networks that make relations between words and predict the next token based on heuristic probability algos. Probability which in turn comes from the training datasets they were given.
They've been trying to improve the LLMs through scaling up (bigger datacenters, more training data) but there are hard limits to both things.
Datacenters already consume a massive amount of electricity and water and have insane operating costs, all these AI companies are losing money.
And training data eventually runs out, they've been trying to cope by giving it "synthetic data" but using AI slop to train AI is only going to create more AI slop, there is no knowledge being generated there, only a degraded output.
The grift is about to end pretty soon, that website's predictions are already off.
