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

(nature.com)

5 points | by thebeardisred 9 hours ago ago

1 comments