Humans built codebases many millions of lines long, well before LLMs existed. Human memory has not been a restriction on us in a long time.
replyLook at all the libraries full of books we've built. It's useful for more than mere training sets.
I think the trick here is plural; I guarantee no single human knows all 1 million lines. Note this is different than knowing how to orient yourself in a million line codebase quickly.
replyThe limit here I think the ancestor comments are getting at is cognitive load, which is real and measured. We only have so much memory to devote to a "stack" when executing, and it's usually quite constrained.