Training large language models on narrow tasks can lead to broad misalignment(nature.com)5 points by thebeardisred 215 days ago | 1 comment