x
Rigged reward learning — LessWrong