Early empirical results I consider evidence this framing makes sense:https://www.lesswrong.com/posts/wYZMmdWEt5QLM3m3e/steering-towards-automated-grading-degrades-alignment
Early empirical results I consider evidence this framing makes sense:
https://www.lesswrong.com/posts/wYZMmdWEt5QLM3m3e/steering-towards-automated-grading-degrades-alignment