#8 future of AI (videos that appeared on my YouTube feed that seem believable)

Basically, there's a new surge of opinions about AI being a bubble that is going to burst and that the AI development is plateauing. Now the counter argument to that is every time we've said that it is going to or has plateaued, it didn't, there was a newer model that was marginally better (4.5 -> 4.6) or a breakthrough (deepseek-v4). The exact question to be asked is "What exactly caused this breakthrough?" Like, was it JUST the research part of finding an optimal solution? OR did the optimal solution already exist and the reason we hadn't found it yet was because the lack of resources being spent on looking at that particular problem? Which one is it? Is it a cycle where the more AI proved to have breakthroughs, the more we spent money on making the models bigger than the last? It is obvious that spending more compute and human resources at a problem would solve it sufficiently faster than if it was purely based on academic research. So then, was the ceiling for what's possible with AI raised JUST BECAUSE more money was allocated to it? And now that we have money to allocate but can't meet the demands for it PHYSICALLY, is why there is going to be a deceleration in the development causing a chain of dominos to fall one by one rendering AI useless / unreachable at the pace companies are aiming for? This is a very raw thought I was just directly typing into my WhatsApp, but I genuinely want to look into this a bit more because when you say the sentence you can't get it out of your head that "All of AI breakthroughs have happened because more money was thrown at the problem". Like, what that means is, and I still don't feel satisfied in my explanation even though I've repeated the same thing I think 3 times now, what it means is that... okay what does it mean to throw money? Buying more compute, right? So just because more compute was allocated to the model, the model becomes bigger and bigger, and it's obvious that previous minima(s) that have been reached by older models are bound to be surpassed by the newer, bigger ones. So, the actual like... takeaway is that, if we kept the money same, if we kept the compute same, would that breakthrough have happened? Because that's the exact problem we're gonna run into, not lack of money but lack of compute. And then the house of cards starts folding in on itself. Because there's not gonna be more compute allocated than what is now, the "breakthroughs" won't happen, if they won't happen, less money will be allocated, less money means no breakthroughs, and less less less, and gone. AI would become a highly compressed language generator that helps you generate a little slop so you don't have to start with a blank slate. Because it's a closed system, the weights don't change after it's done training. The next training, if you want to keep the wheel of breakthroughs spinning, requires more compute than the last, which now it doesn't have. So... What? What's gonna happen next?