Hacker News new | ask | show | jobs
by dandersch 165 days ago
> Small quantities of poisoned training data can significantly damage a language model.

Is this still accurate?
1 comments
embedding-shape 165 days ago
Probably always be true, but also probably not effective in the wild. Researchers will train a version, see results are off, put guards against poisoned data, re-train and no damage been done to whatever they release.
link
d-lisp 165 days ago
How would they put guards against poisoned data ? How would they identify poisoned data if there are a lot/obfuscated ?
link