Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

That seems like a reductive way to consider it. What percent of music was created by Led Zeppelin? What percent of art was painted by Monet? What percent of films by Alfred Hitchcock? It may be a small percentage objectively but they are hugely influential.


I don't think back propagation care whose text it is back propagating.


The data sets aren't naively fed into the training runs.

Instead, training attempts to sample more heavily from higher quality sources, with, I'm sure, a mix of manual and heuristic labeling.


fwiw, no llm ive ever used generated in the writing style newspapers and -sites use - hence i honestly doubt they've been given a meaningful boost in relevancy.

their idioms would leak occasionally otherwise




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: