Over the last year, I've been experimenting with LLMs almost every day, and I think the biggest shift isn't that models are getting smarter. It's that they're starting to do things instead of just answer questions. A few months ago I wa…