Some thoughts on AI development

Just thinking out loud here

If say a system needs to get RSI, this pressumes that it has ways to test its hypothesis from theory to reality, which LLM's current state certainly cannot except for obscure math problems that can be proven strictly on paper or computer simulation. I do think text based LLM can go further in terms of intelligence but without either hiring actual top level researchers and thinkers to input new data or a recursive loop of proving its own thinking over time within the realms of reality (either by a super insanely good simulation or machines to do the tests in reality), claims of RSI is mostly an illusion.

Another thing is that there is a real and expanding market of novelty toys which i only learnt of this year. Can be eink productivity tools, to wearables, and other fun toys that are cute and functional. Maybe its always been there but Im too deep into software to realize. Seems to be growing market in any case.

And something I dont really get as much is how these new startups can buy and sell data at such insane prices. Highly doubt this can be sustainable.