I often give ChatGPT simple tasks and it gets them wrong. Yesterday I took a picture of a video game and told it I was stuck, only to be given instructions that lead down a bad path in-game.

I also often retest new models with tasks old models failed, and see some improvements. I really liked “format this SQL generated by my ORM and explain it” last week.

I honestly have no insight on if the tasks it is failing to do are right around the corner or if they are decades away.

You need to have more blind faith in Sam and his machine.