It would probably struggle to see the larger picture. I can see it being used to add comments in self-contained functions though without too much difficulty.

Honest question: would an LLM be able to write useful comments in code like this?

It can be better han nothing, but not really. The LLM faces the same challenge that any competent coder does: neither were present to learn the human, business and organization context when the code was first written.

100% I use them a lot to ingest and understand shitty code for me. Of course it's not perfect, it's like having a colleague who's not super strong but has infinite patience for bullshit

use the LLM to generate regression tests for the large file, then start refactoring it