Inducing self-other overlap with SFT reduces deception at scale, but generaliza

(lesswrong.com) by joozio | Aug 8, 2026 | 0 comments on HN