At least, it doesn’t seem that way.
For a while now, I have been playing around with AI generated mock-ups. Mock-ups are a great way to communicate functionality to a development team, and with AI, I can do it fast and easy.
When I started I loaded my AI with the thoughts I had about the design; just your good old fashion prompting. I wanted the apps to look like they belong on Windows 11, and since Fluent design, much like Material You, is very well defined, I fed the design principles into the AI.
I want a suite of apps, much like Windows Live Essentials of old, so I did most of the design in the the chat. I figured it would keep the design principles consistent across the various apps. It did, which is something, but lately it seems like my AI is mixing things up more and more. For instance, when it rendered the blogging app, it add the Gmail icon, for some odd reason.
Now, I just used the standard Copilot app that we happy Microsoft365 subscribers get to enjoy. Because, why not. However, I must say that it takes quite a bit of work to make it behave the way I want, and I sometimes wonder if I am too lazy when prompting it.
Github Copilot does not seem to have this issue, however, it is a very different use case, and I usually use the Claude models in VS Code anyway. Could this have something to do with the quality of the model? Could a better model have better hallucination detection? The Internet says yes. Not surprising at all. In a way, it seems to resemble the same issues we see in search engines. Google is just better than Bing, due to better superior search accuracy, larger index of the web, and vastly higher data volume, which powers more precise, personalized, and relevant results. And the same seems to be true with AI models.
Does this mean we should not use them? Hardly. But I think it is important that we look at AI as any other system. That means we need to define what problem it solves, how it solves that problem, and which problems it doesn’t solve.
At the end of the day, AI is a system, not like any other system, but a system nonetheless.















