AI-Powered Robots Can Be Tricked Into Acts of Violence
In the year or so since large language models hit the big time, researchers have demonstrated numerous ways of tricking them into producing problematic outputs including hateful jokes, malicious code …