Poisoning needs a roughly fixed number of documents, not a percentage of training data, so attacks on web-scraped corpora are more practical than assumed. Anyone training or fine-tuningTaking a trained model and training it a bit more on your own examples so it gets better at one specific job.Full definition → on public text should weigh data provenance.
Terms in this piece · Glossary
fine-tuning — Taking a trained model and training it a bit more on your own examples so it gets better at one specific job.
