Edward Hopper

I'm currently playing around a bit with ChatGPT. It's fascinating what's possible. The following picture was created from a prompt à la ‘Paint a picture of a longboarder doing a pirouette on a longboard in the style of painter Edward Hopper’. That doesn't look bad at all. The longboard is more like a skateboard and the proportions aren't quite right, but the picture comes pretty close. This also applies to the posture and arm position. However, when I consult the large language model in my main field of research, the modelling of diseases, I only get a good result at first glance. In the end, all queries produced recognisable errors or dangerous misunderstandings, at least for me. There is probably not enough good knowledge in circulation for artificial intelligence to produce good patterns or correlations. The problem is that the AI always answers very deliberately. And that is quite a difference. I can easily reject the image or I don't like it. Incorrect scientific answers are usually not recognisable to the layperson.