Analysis of AI Safety surveys for field-building insights — LessWrong