Post #4335015
2026-06-16 18:39 UTC
RE: https://mastodon.cloud/@aarontay/116761154439972422
I enjoyed this thread, and have a few notes of my own to add:
1. distinguishing between clerical errors and dealbreakers is kind of a big deal and why experts exist
2. we still have a mostly insurmountable gap between what LLMs can do with statistical knowledge from training data; and the tasks that require semantic / referential knowledge (see point 1)
Replies (1)
-
@aarontay@mastodon.cloud 2026-06-16 18:49
@hye@glammr.us you are an expert in your field. But is #1 really an issue? When i ask llms to critique my work it can definitely tell me which are serious issues which are nitpicks. Granted at your level it might not be there yet in domain expertiss but from what I hear from many researchers it's almost there , and I don't see why progress will just stop now