General purpose intelligence: arguing the Orthogonality thesis — LessWrong