Humans provide an untapped wealth of evidence about alignment — LessWrong