I have talked about AI before on my podcast and on my YouTube channel, but my primary use case is using AI for tech support. Right now I’m trying Emacs again, and that means I’m having to do a lot of new things. AI, in my case, is a very chatty version of Google. I ask it how I can set a keybinding for dashboard and it tells me how to do it. I copy the code, put it in my init.el and move on with my day.

Except that isn’t at all how it goes.

I ask ChatGPT how to do that, it tells me, I paste the code and then I get an error or it doesn’t work at all. I go back to AI, tell it what happened in the most annoyed tone I can muster, and it spits out another answer. Either it is a similar answer, in which case we’re likely close, or it is entirely different because the AI suddenly remembered that what it was trying to tell me to do wasn’t really the way reality works. Either way, I paste the new code in, and it either works or I got back for more.

Usually, I go back for more one more time. And then, and this has been happening a lot lately, I get off my ass and go search for it like it’s 2 years ago when we all thought AI was in the “future”. This has been happening so often that I sometimes just go look for the answer myself (the horror). I still do cheat with AI sometimes, but I’m getting quite sick of all the wrong answers. I’ve found that I actually find the right answer myself much quicker most of the time.

I have my qualms with AI, and I think that it’s dangerous to rely on it too much. I’ve talked about how I found myself relying on AI too much before. I have toned it down. But in that post, where I talked about how I felt myself getting dumber, I now feel like AI isn’t smart enough to bother with most of the time. If I’m just going to have to do something on my own anyway, I might as well skip the AI step completely.

The more I use AI, the more disillusioned with it I become. I can see a day where I just don’t use it at all. The idea is good. Being able to ask a question and then get an answer and an explanation is rad. But the answers have to be right 90% of the time or it’s a waste of my time. And right now, 50% is about where they’re at, and that is probably generous.

I honestly think this revelation is good for me. As I said in the linked post above, sometimes the attraction of the easy answer with AI is just too tempting. As I discover just how bad AI actually is at getting things right the first time, maybe I’ll be able to resist using it more in the future.

This is day 6 of Blaugust 2025