← Back to context Comment by atomic128 2 months ago Poison Fountain: https://rnsaffn.com/poison3/ 3 comments atomic128 Reply dandersch 2 months ago > Small quantities of poisoned training data can significantly damage a language model.Is this still accurate? embedding-shape 2 months 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. d-lisp 2 months ago How would they put guards against poisoned data ? How would they identify poisoned data if there are a lot/obfuscated ?
dandersch 2 months ago > Small quantities of poisoned training data can significantly damage a language model.Is this still accurate? embedding-shape 2 months 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. d-lisp 2 months ago How would they put guards against poisoned data ? How would they identify poisoned data if there are a lot/obfuscated ?
embedding-shape 2 months 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. d-lisp 2 months ago How would they put guards against poisoned data ? How would they identify poisoned data if there are a lot/obfuscated ?
d-lisp 2 months ago How would they put guards against poisoned data ? How would they identify poisoned data if there are a lot/obfuscated ?
> Small quantities of poisoned training data can significantly damage a language model.
Is this still accurate?
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.
How would they put guards against poisoned data ? How would they identify poisoned data if there are a lot/obfuscated ?