Raw LLM Responses

I really love you explaination about how AI work but, you say AI can do 80% of white collar work... I'm a software architect (26 years in the biz), I use the big four AI for my work (although mostly Claude and Codex) because I cannot do any proper work with just one, they are all "Artificial Idiots" if you ask me. They are very usefull none the less if you know your craft, they show me things that I didn't know I didn't know, or make for me an initial scaffolding of things (which I always have to correct anyway) but more than 50% of the time their proposed solutions when they have to "think" are shitty, seriously. Some times they give you solutions that work out of the box but seem to be done by a Junior or they give you over-engineered solutions that seem to come from a just graduated engineer that wants to apply all his acquired knowledge in every pieco of thing he does. And I'm not even talking about every day life questions, they hallucinate like hell. I do not agree with you about what they can do today. But well, I don't pay 100/200 USD AI services, just 20/30 USD. Probably all what you say they can do works when you expend all the money it takes to let them work for hours??? But I don't understand how you would let them work for hours.... For instance, I have a 300K LOC project, if I take Opus or Codex High and I make them "understand" the whole codebase and then ask them questions, it just takes minutes to do that. And they give me 50% of the time shitty answers (believe me it's not my prompts). It's not that I can select an option for them to run 5 hours to givem a "correct" answer.... So, at least in the "normal" way I use them, the are still idiots...