How important do you think AI safety filter research is for the security community? Critical — understanding weaknesses is essential for building better defences Useful but should be carefully controlled Too risky — this research helps attackers more than defenders I haven't thought much about it Every AI application that filters content is making a bet. The bet is that the categories of harmful outputs the developers anticipated at deployment time cover all the categories attackers will try at runtime.…
No comments:
Post a Comment