The issue is that these bots don't know anything. As such, they only function in domains with curated data, which is why all of the "AI-companies" are curating data. This also shows why the name "AI-companies" is wrong, because if there were any artificial intelligence, there wouldn't be a need to curate anything; it would just start ordering Boston Robotics Spot robots and observe things in the world, it would install spy gadgets into the oval office and the Kremlin without anyone knowing, it would invest on the stock market with 10000% yearly returns, it would massively manipulate any unregulated market using techniques that are illegal on regulated markets, and within a few years or so, it would know pretty much everything better than anyone curating data.

Systems like AlphaZero could be considered narrow superhuman AI, but they still just use a gigantic amount of computation. LLMs might be useful as a subsystem of some kind, but that's it at the point. Reinforcement learning still requires an unreasonable amount of data and in fact it has been shown that some functions exist that can't be learned efficiently.

If I look at my usage of LLMs, I have mostly replaced some usage of search engines with it, but any sensible search engine would be able to undo that advantage. If I look in my environment of mostly stupid people (people of average intelligence), none of them use e.g. ChatGPT. It seems very much to be a technology used by advanced users. Apparently. 180 million accounts used ChatGPT, which considering that there are likely duplicate accounts, might be significantly less. Altman claims there are 100 million weekly users, but that number has not been growing significantly, which suggests that this type of service is only applicable to people that have a need for it, which isn't everyone.

For me the novelty aspect is gone and while I might use it just as a way to see whether they actually get any better, any true artificial intelligence would never be made available to the public, because it would be the last invention humanity ever has to make. If the AI is cheap enough (and it will be over time), it would dominate every single area of business over time.

Google recently claimed that their FunSearch invented a new heuristic for online bin packing, but the Kolmogorov complexity of their solution is extremely low, and additionally, it's not actually a new heuristic. I thought it was one of their most pathetic publications and Nature is retarded for publishing such garbage. Nature just has become a vehicle to prop up Google's failing stock price. Nature used to have a good reputation, but since tech companies started to publish in it, it has all gone to shit. The truth of the matter is that for most of the things Google does, LLMs do it better and there is no moat.