Mike Austin shares hundreds of hours using ChatGPT and Copilot, but warns AI models mishandle licensing rules. They miss constraints like SQL's four core minimums. For complex, high-risk questions, he says, they're not ready yet.
Chapters
I've played quite a bit with Chat gbt, you know, probably a few hundred hours now, you know, playing with the various, you know, chatgpt and Copilot and a couple others, trying to understand how to use it to answer licensing questions and things like that. And the thing that I the reason I say be weary is it misinterprets the rules quite a bit. I mean, I've put in, you know, SQL licensing briefs, for instance, asked it very simple questions, and it's come back and ignored things like four core minimums and things like that. So I have seen people that are trying to use it. I'd be cautioned when it comes to answering questions about licensing. I think there's better use cases for it today. I think it's just a complex topic, that it's going to take a while for the models to be trained to actually do that sort of thing.