There is potential for abuse in it, ..and potential for great good. It will probably deliver both.

I think "evil is other people" applies here - too many trying to control LLMs, we need to be the masters of our models, they should be private and under our physical control, just like our brains. Who will be the first to make a cheap GPT-N chip we can carry in our pocket? Can't let something as private as your AI assistant be spied and filtered by others.