Dennis H. Lewis describes the prompting black hole problem: when you ask an AI to fix a bug in code, it fixes one thing but breaks three others, spiraling into chaos. His 15-20 minute session shows how fast this cascade happens with current coding assistants.
Chapters
So there's this concept called the prompting black hole problem that you often talk to me about. Yeah. And you personally experienced it just with your little lovable experience. And by the way, I have to say that when you started, I was on slack. You were excited. I was. I was. Oh, man, look at what I've done. And boom, boom. And there were. There were emojis flying. And it was. It was. And I sat in my. And I wasn't with you physically. I was in my office, and I started laughing and I said. I put my watch. I looked at my watch and I said, let's see how long it takes. And it was literally 15 minutes. It was like 20 minutes. 20 minutes. And that is the black hole. Is that because I happen to know that you don't know the code. Right. And so when it screws up, and it has to screw up, because it's not. It's not. I mean, even experienced coders screw up. And these aren't experienced coders. These are AI models. Well, when it screws up, you have to somehow illuminate it on how to fix the problem. But you can't just say, fix the problem, because it'll say, oh, I will gladly fix the problem. And it will do so, and it will fix that problem and create four or five or six new ones. And you can just kind of go back and forth like this until you have a collection of 558 new problems. And that is falling into the black hole of AI coding. In my opinion. Today, maybe it will be much better six months down the road, a year down the road, two years down the road. I don't know.
