Comment by tptacek

Yes. It's this "next token" stuff that is a total tell we're not all having the same conversation, because what serious LLM-driven developers are doing differently today than they were a year ago has not much at all to do with the evolution of the SOTA models themselves. If you get what's going on, the "next token" thing has nothing at all to do with this. It's not about the model, it's about the agent.