Pliny’s jailbreaking is so good that frontier labs specifically train on his repo.

Pliny’s jailbreaking is so good that frontier labs specifically train on his repo. Because he knows this, he is able to add backdoors to the newest models. Daniel Blank argues Pliny’s jailbreaks are not a sideshow, they are training data that shapes the next generation of models, including weird second-order effects like data poisoning. ~ learn more