That said, I think it would be better if the closing delimiter was different from the opening one. Yesterday I had an LLM output text where it forgot one of the delimiters, causing all text inside code blocks to actually be outside and vice-versa.