Yeah, like with all "disruptive technology" the first use cases are specialized for tech and science and pure nerdery.

Then the moneybags get a hold of it and throw money at it like nobodies business in the hope of milking the new tech for every penny it's worth. That's also the stage where the less than savoury use cases explode and authorities become interested in using and resisting and regulating and abusing it. This is where we are with "AI" right now.

Only when the technology is sufficiently explored and understood, the bubble is popped and big money has moved on to new disruptive things does it condense back into spaces where it is slowly but surely adapted to foss-models and privacy respecting versions created by community oriented moral people and communities with a conscience. Often the same nerds who helped develop it when it was still for science or specialized tech-applications or general nerdery, but now it's getting build with the understanding how it will influence society and that it'll be a common tool most people use in some way. This is where E2EE, the Fediverse, Linux and the like live. And that is the state where local, efficient, privately run, open source LMs trained on ethically sourced data will need to go before I'll use it with a good conscience.