logoalt Hacker News

CompleteSkeptictoday at 8:27 PM0 repliesview on HN

we have played with this! the fascinating thing we've found so far is that adversarial examples for our model are quite different from that of LLMs so that they work even better together