Training large language models on narrow tasks can lead to broad misalignment

1 points | by Anon84 5 hours ago