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

(nature.com) by Anon84 | Jan 15, 2026 | 0 comments on HN