Twenty-three AI alignment research project definitions — LessWrong