1 paper
Nikolaus Howe, Ian McKenzie, Oskar Hollinsworth +5
Increasing model size has unlocked a dazzling array of capabilities in modern language models. At the same time, even frontier models remain vulnerable to jailbreaks and prompt inj…