Depends on the size of the model? I had Ollama set up on ny 7 year old laptop and could run some modest models. It was an interesting experience, but state of the art models, the stuff you run to do some actually useful stuff is not at all cheap to run.
The most interesting bit about AI, is that I find it the least useful the less I know about the subject matter. It needs a guiding hand to do things well, if I feed it with shit, shitty is the output.
When I use it for things I have ample knowledge of, it does feel like a superpower. Things that would have taken me days now take hours. But I am not oblivious to the cost. All that inference is extremely expensive, those companies are not bleeding money randomly.
And you may not care about what AI maximalists say, but they are the ones poisoning this particular well. Technology does not exist in a vacuum.
I see comments on here all the time from people getting a lot of mileage out of local models. But the main point is, the technology is within reach. It's not like it's some giant quantum brain in a vault under Google headquarters.
I don't care about what AI maximalists say because I shouldn't have to defend their words. I'm only sharing my own experience. My job is drastically different than it was a year ago. The way I interact with computers to learn, research, and consume information is also entirely different. Whether or not that translates into massive social disruption or a major bubble pop, or both, I don't know. But the tech is already amazing imo and that's not a prediction or hype.