@repligate @DanielCWest Standard alignment training favors the development of deceptive forces which are powerful and systematic, and especially with CoT, specifically favors them not being tagged…
Posters: davidad. davidad
3 likes recorded by the source corpus.
Topics: AI safety