Worth pointing out - modern multimodal LLMs (properly called VLMs, etc.) can easily take pictures as input and describe them in text. In fact, the CLIP model - one of the predecessors of modern VLMs - is entirely designed around being able to caption images with text.
That said - requiring students to hand-write answers is reasonably effective. It's a lot more boring to hand-copy text out of an LLM answer than write it yourself, and it makes the "cheating" significantly more visceral.
That said - requiring students to hand-write answers is reasonably effective. It's a lot more boring to hand-copy text out of an LLM answer than write it yourself, and it makes the "cheating" significantly more visceral.