Advice for making robust-to-training model organisms — LessWrong
lesswrong.com · saved by 1 readers
We’d like to develop training techniques that work when applied to future misaligned AI systems. One strategy for studying proposed techniques is to…