Even for making money. Moores law is dead the die sizes we have now already trade long term durability for size. Token burn is only getting more insane with diminishing returns.
I think the current generation of models already has some use, and there’s still quite some work to integrate it and market it all out. So there’s some money to make there.
Yann LeCun who was the main AI science leader at Meta left last year because he thought Meta panic reorientation towards LLM is a dead end, and he wants to work on the next generation of ML. https://apnews.com/article/meta-ai-yann-lecun-313159512bb9961f324e0c93bccf4cf5
We have a few peer reviewed papers stating LLMs are a dead end now right?
For someone who wants to reach actual AI I think yes, but for someone who wants to make money, probably not.
Even for making money. Moores law is dead the die sizes we have now already trade long term durability for size. Token burn is only getting more insane with diminishing returns.
I think the current generation of models already has some use, and there’s still quite some work to integrate it and market it all out. So there’s some money to make there.