Results from the interpretability hackathon — AI Alignment Forum