@nileane I’ve recently concluded for all I don’t much like the LLM situation, or maybe specifically because I don’t like it, I need to educate myself on them early to face the future on my own terms.

I want to avoid the subscriptions for stolen data. Give a tool any responsibility, and you get the Post Office scandal.

But they’re getting more efficient. Give it a decade. Ethically trained models, running on-device, mostly delegating to scripts, user confirmation for destructive edits & IO?