"These models are good in automating tasks, not workflows or jobs."
Sam Altman on Joe Rogan.
Think about the scenario of a home, in case the homeowner is going on a vacation:
i) The homeowner talks to an A.I. assistant, and orders the agent, to open the windows every day automatically, and feed the dog. The A.I. agent may hallucinate, and do none of the tasks.
ii) The homeowner with the help of the A.I. assistant, will write a small program, to open the windows in the exact time he wishes, and with the right interval of hours, and feed the dog exactly that many times. That program will run correctly and deterministically till the end of the universe.
Complicated programming is not getting automated any time soon, maybe never. Simple programming tasks are going to be automated by humans, who lack formal C.S. training and use the statistical engines to complete some not so obvious tasks to them.
Think about the scenario of a home, in case the homeowner is going on a vacation:
i) The homeowner talks to an A.I. assistant, and orders the agent, to open the windows every day automatically, and feed the dog. The A.I. agent may hallucinate, and do none of the tasks.
ii) The homeowner with the help of the A.I. assistant, will write a small program, to open the windows in the exact time he wishes, and with the right interval of hours, and feed the dog exactly that many times. That program will run correctly and deterministically till the end of the universe.
Complicated programming is not getting automated any time soon, maybe never. Simple programming tasks are going to be automated by humans, who lack formal C.S. training and use the statistical engines to complete some not so obvious tasks to them.