It's greed at the top pushing the market into insanity. "AI" vendors are dangling the carrot of "eliminate manpower" in front of the manpower-owning class and they are throwing everything they have at it. The wealth that's been thrown into this hole could have payed for a couple million work-years of developers.
Insanity in the market is drawing in profit-seekers. "Greed at the top pushing the market into insanity" isn't really something that can happen: greedy folks are adept at leveraging incentive structures that have already emerged, but very bad at trying to purposefully shift macro-level incentives at a grand scale (which is something that humans in general are quite bad at).
If AGI is the next evolutionary goal, its the difference between being stuck and not.
Wealth is also only relevant in a capitalistic system, which we invented. It could easily be that all other species never created capitalism and are therefore waiting for us somewere which we will not reach because we are stuck on 'wealth'.
And if we are stuck in a local minimum, it might seem that GPT-2 broke us out of there.
Or LLMs will end up being the local optimum we organize society around and sacrifice everything to, and end up not being the child dream of infinite wealth and benefits the AI boosters believe in. laissez-faire capitalism tends to result in power concentration and the creation of a distinct class of hyper-wealthy and powerful individuals with close to infinite power to dictate the future. That’s pretty much the situation we are in.
If the current AI bet turns out to not be the next revolution, what’s the plan? What will be the pivot? What happens to all the capex and commitments, the reputation of all the people who promised that was the journey to the holly land? The answer is that there is no plan, AI has to work to justify the system itself. It’s almost a natural result of the economical and ideological system we conceived.
It’s really not that different from blockchains, though at least LLMs have some actual use cases. But there is a complete disconnect between the actual ROI and the vision sold by the AI folks
All this massive massive massive compute can be used for different types of machine learning. Nuclear Fusion physics simulation, neural networks for every other use case etc.
I'm not sure if we needed AI/ML for breaking the memory wall.
But yeah we will see how the society will respond to more and more and more automatisation.
ML/AI/Robotics is for sure the next automatisation revolution.
