x
Sufficiently optimized agents appear coherent — LessWrong