My personal test. We’re almost there. Maybe 3.5 will do it.

Since I've started using Gemini and ChatGPT, there was one test I always did. I teach German and there is a communication model by a man called Schulz von Thun.

And the prompt was always (in German): "Create an image that explains the Four-Ears-Model of Schulz von Thun".

In the beginning, the result was always complete nonsense. Just a whirl of words, ears and faces. With 2.5 the result was almost recognizable, but still made no sense and was full of mistakes.

Today I tried it again and the result is almost good. It still uses two English words instead of the German words and a few orthographical errors. But it's a giant leap ahead.

I think we're one model away from a perfect result. Which is almost frightening.