Lessons learned from talking to >100 academics about AI safety — AI Alignment Forum