You assume it will improve. There is an asymptote from what I can see.
Surely you're not seriously that we've hit the peak (or even the top of an S curve) of what AI is capable of right? Even if you think LLMs have a limit to what they can do and current AI is overhyped (both of which I would agree with) I find it difficult to believe you really think what we have today is the best AI will ever be able to do.
The problem is there isn’t a viable revenue model. The moment you ask money for it the customers disappear. That means there isn’t a revenue stream to support the extreme cost of model training.
And this is all paid for by people who expect a return. In the middle of a very volatile market.
It’s dead even from a non technical perspective.
From a technical perspective every reducing gain requires more money than the last step. That isn’t something that ever works.