In the year or so since large language models hit the big time, researchers have demonstrated numerous ways of tricking ...