Andrej Karpathy: "AI gets better by learning from its mistakes. The catch is someone has to find them first." One sentence that explains why AI progress is slower than it looks from the outside. The models improve when errors are identified, labeled, and fed back into training. The bottleneck is not compute. It is not architecture. It is the human judgment required to find the mistakes that the model doesn't know it's making. The most important job in AI right now is not building the model. It's finding what it gets wrong.(CyrilXBT)