3 ms·Training large language models on narrow tasks can lead to broad misalignment1 points by Anon84 9mo ago