Its customers want their sheeple base to only use computers via full MITM of a networked (or at least with undecipherable functionality, like a local LLM) agent, that predicts what you want from what you say, so that god forbid you'd never do direct computation.

I mean, an LLM model is literally a program whose logic is hidden in weights. A way to thoroughly obfuscate functionality.

And those customers you might consider smart, with such power, but in fact just like with everything else they are not, just in the right place in the structures of power to have their wishes tried first.

I'll repeat, they are not as smart as one would expect. But if your asshole gets torn while playing superhero, it's your problem and of those who did it. If their asshole gets torn while playing superhero, it's a problem of everyone in the street, town, district, country, continent, ethnicity, maybe even statistical bucket of those who did it.

And they do think they are some sort of superheroes.

Though when the AI bubble bursts, and we'll have plenty of cheap hardware suited for this technology, who says there won't be plenty of more specific applications and even toolkits based on LLMs? And then they'll get their wish, not in the sense of agents, but in the sense of programs far more resistant to reverse-engineering than normal binaries being popularized.

Not even talking about the scenario where all that cheap hardware is bought by parties which can use it to their normal goals, unlike most real commercial activity. That is, by nation states with their surveillance needs.

So perhaps those people are smart enough.

OK, maybe it's just another BAD psychosis.