The name ai has been used to describe many things. We called Chess engines ai, we still call video game opponents ai. Even the guy who coined the phrase intended it to describe something that is simulating inteligence. I think LLMs are something that completely fit the bill. When we invent something better, we will probably stop calling LLMs ai, but I think it's completely fair to do it until then. It's a simulation, I think the general public understanding is that this is not a living intelligence we are conversing with.

Validate what it's telling you against actual sources, and it can indeed be helpful. Yes, but.... I don't have to do this anymore, because I ask it to explain each command to me and while I would still struggle without it, right now I call already understand the commands. It can't delete everything or something epic like that. I don't use agents for this stuff, that would be more risky.

However.... I often wonder when I do cli commands and paste them back to llms, how stupid that is. Maybe it's still necessary to keep control? But when I extrapolate into the future and let's say llms get just a bit more reliable, doing all the stuff I did would take no effort at all.

I think of it sort of like soup. You don't need soup to be extremely well defined to work. Sure you can mess it up, but it's not that likely and if that situation doesn't hold a big financial penalty then why not risk away?

Also, I'm not some ai evangelist. I'm more like extremely pragmatic.