Learning to use local AI is exciting, overwhelming, and frustrating
The part about local AI being frustrating rings true to me. Anyone who has spent an evening getting a model to run on their own hardware knows the drill: a download, a command line, an out-of-memory error, a search for the right flag, and suddenly it's midnight. The excitement is real too, because once it works, it's genuinely impressive that a laptop can carry a conversation or write code without phoning home.
The privacy motivation is the most interesting part of this. The author was held back not by skepticism about AI's usefulness but by where their data would end up. That's a legitimate concern, and local models are a real answer to it. But the trade-off is that you become your own sysadmin, and that's a steep price for people who just want to ask a question and get an answer.
I'd want to know how long the author stuck with it after the initial setup. The tools have gotten better, and there are friendly frontends now, but the gap between "runs locally" and "works as smoothly as a cloud product" is still wide. My take is that local AI is worth the trouble for specific use cases, not as a general replacement. The frustration is the cost of control.