I'm pretty sure LLMs have to generate text left-to-right, but there was a screen recording floating around that had "multiple cursors" of text generation at once, where it looked like the LLM was filling in a grid of text, so to speak. I'll see if I can find it.

There are LLMs that are trained to fill-in-the-middle, particularly code editing LLMs. There is nothing inherently required about left-to-right.

At least OpenAI creates more noise than useful output if you try to make it tackle a problem this way.