Comment by specked-citrus

I'm also quite surprised to see that LLMs can do this. I guess it is possible that "make an image in MS paint" is a type of RL environment used for image understanding. This is one of those areas where people inside the labs have a very different view into how much models are generalizing.