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

(nature.com) by petemetefete | Jan 16, 2026 | 0 comments on HN