This raises an operational question: should we tell LLM to use Latex, markdown, or either to get better proofs? Does the answer depend on whether the field is math or theoretical computer science?
That, and also Latex is pretty much the only way to communicate math over email, so humans had to develop an ability to read plain latex off a page too in the past couple of decades. An alternative would be using some buggy Chrome plugin like Latex for Gmail (of which there are many), but much easier to just parse it in your head. Although LLM probably also read arxiv papers in latex source. Humans probably still prefer PDF.
This raises an operational question: should we tell LLM to use Latex, markdown, or either to get better proofs? Does the answer depend on whether the field is math or theoretical computer science?