I agree. I have run a few models on my phone and server and they are interesting but on such weak hardware they are slow and less capable. That said, a project like KDE has a certain amount of funding which could be put to use in part finding bugs using AI tools. They could be run on reasonable hardware with all the ethical safeguards we could come up with and have it be the least awful version possible. Some would argue it would still be unethical and I can see their perspective. That said, for me this is the line we should not cross. Keep the development by humans and for humans, and if we are to use AI models then we should be doing so to find problems to work on, not to solve them.