Hacker News new | ask | show | jobs
by SoftTalker 61 days ago
Yeah it's sort of an LLM smell but honestly the models learned that pattern because it's common in the training data. People write that way because it sounds like they're revealing something profound.
1 comments

LLM inference does not just regurgitate the training corpus; RLHF is almost certainly to blame for this. There’s probably some Google n-gram graph to prove it.