AI reinforcement learning works by modeling the past, but it falls apart when things change in real time. What to do? @blaiseaguera.bsky.social and his team have the answer: empathy.