4 ms·

Poison Fountain: https://rnsaffn.com/poison3/ https://rnsaffn.com/poison3/

• dandersch 9mo ago
  > Small quantities of poisoned training data can significantly damage a language model. Is this still accurate?
  • embedding-shape 9mo 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 9mo ago
      How would they put guards against poisoned data ? How would they identify poisoned data if there are a lot/obfuscated ?