I’ve been seeing a lot of chatter about how the real bottleneck in AI might not be compute or model size… but the fact that we’re running out of usable training data. Google DeepMind just shared something called “Generative Data Refinement” basically, …