the ML word for learning — and where the analogy breaks in ML, training is one-directional: data goes in, model weights come out. training ends, then inference begins. in cyber, every cyberlink is a weight update to the cybergraph, and learning and inference run continuously, interleaved. the graph…