I just tried a couple of cases that ChatGPT is bad at (reproducing certain scenes/setpieces from classic tabletop RPG adventures, like the weird pyramid from cl

I just tried a couple of cases that ChatGPT is bad at (reproducing certain scenes/setpieces from classic tabletop RPG adventures, like the weird pyramid from classic D&D B4 The Lost City), and Gemini fails in just about the same way of getting architectural proportions and scenery details wrong even when given simple, broad rules about them. Adding more detail seems kind of pointless when it can't even get basics like "creature X is about as tall as the building around it" or "the pyramid is surrounded by ruined buildings" right.