We think of LLMs as not math-proficient (because they aren't yet) but what about those multimodal models?

I wonder if anyone has tried getting them to "imagine" math in a way to "visually compute".