It's hard to believe how much we've gained with AI and LLMs in the past couple of years, and now how much more it's expected to happen.
We've become so accustomed to continuous progress that at this point it's very easy to disappoint expectations with anything less than another major advancement that is deployed to the masses.
I do understand that a lot of cheesy hype has built-up, mostly because it's how business works. All sort of businesses are rushing to capitalize on the new buzzword, but this doesn't mean that we don't already have revolutionary technology that is already changing the world.
If all LLMs' development stopped today, we'd still have models like Llama 3.1 8B runnable on a laptop, and GPT-4-level models sometimes Open Source running externally with costs that are dropping rapidly (if we can predict one thing is that hardware keeps getting faster and cheaper).
What we can do today already is nothing short of amazing. There is so much untapped potential with the existing models, Open Source and not, and the only reason why this potential hasn't materialized is that we simply didn't have the time to work with it.
This is a good sign, because it means that things are evolving rapidly and there's little time to settle on an AI model, but if the rate of improvement of LLMs is truly slowing down or possibly reaching a plateau, then we, the developers/hackers/engineers, would gladly start to work towards making the best of what we've got.
Parallels with game console development
An example of this are game consoles, at least how they used to be, when every console had some new esoteric and quirky hardware developed perhaps on a hunch and under some heavy constraints by the hardware engineers and let loose to the programmers to make the best of it.
New games for a new console were usually the worst games in terms of technology, as the programmers had to struggle between learning the new hardware and getting the game out of the door.
Then, with the first releases out of the way, the developers would start to really hack the hardware, going lower level and starting to pull performance where possible.
As a result, games would start to look better and better, while the consoles would become less expensive to produce.
Count on hackers
There are many engineers/hackers out there that can and will make the best of what AI has to offer, even if progress in the field were to suddenly stop today. In fact, some of us may feel more at ease if everything slowed down a bit and we had a chance to build something on top of a relatively stable platform with not so many moving parts.
I don't think that AI (LLMs or whatever comes next) is about to slow down, but if it comes to that, we already have plenty to work with for years to come.