Investigating an insurance-for-AI startup — LessWrong