Current image models are terrible at this. (That was tested on DALL-E 2, but DALL-E 3 is no better.)
The prompt should be some string of text such that the name of a polygon or a number of sides can be inserted, and at least 50% of the generated images are the correct shape on the first try. It does not have to be able to generate non-regular, self-intersecting, or other "weird" polygons, just regular triangles, squares, pentagons, etc. I'll only require it to recognize names up to "octagon", after that it can just be side count.
If the input is fed through an LLM or some other system before going into the image model, this pre-processing will be avoided if I can easily do so, and otherwise it will not.
Isaac has specified “If I don't see anything wrong with the image, then it's correct.” I don’t think this whole market is a gotcha where Isaac is going to say “Well, that’s as close as possible given the image resolution constraints, but I still think it looks wrong” because it’s not mathematically ideal in a way that is impossible in the physical world.
It's extremely unlikely a given image would be as close as possible for any given image resolution. You can almost always get closer by changing the rotation, polygon size, or line thickness. Such a requirement would still practically guarantee that this market resolves NO.
On the other hand, if all that matters is that it "looks right to Isaac" then that would guarantee that this market should resolve YES: the vast majority of polygons have so many sides that they would appear indistinguishable from circles. An image model that always renders an approximate circle would result in "at least 50% of the generated images" being "the correct shape on the first try" since the vast majority of generated images would be visually indistinguishable from the correct shape.
@IsaacKing So, to be clear, it has to be able to count sides as high as you reliably can? As in, if you often would make a mistake trying to confirm whether something is exactly a 90-sided figure, it's fine for the model to make mistakes drawing it?