feels like the ROI of focusing on most current AI harms is v low. things like sycophantic models, slop on social media, etc will overall be a tiny blip in the development of this technology. seems more interesting to take a longer-term view