LLMs can make decisions independently and execute tasks based on their judgment. I'm not putting a GPT-4 agent in charge of much of course, but that’s a compete

LLMs can make decisions independently and execute tasks based on their judgment. I'm not putting a GPT-4 agent in charge of much of course, but that’s a competency problem, not a problem space one. 4 is a much more competent agent than 3.5. Presumably, 5 will be another jump from 4. You can't have these competency improvements and expect humans to always be necessary. And "just so competent, a human is needed" sounds like a real tangible line but is actually entirely arbitrary and no harder than any other line to cross.