x
A Plan ‘B’ for AI safety — LessWrong