Long before LLM and stuff, it was common practice to train one neural network to create something and a second one to decide which is AI and which isn’t. Once the second can’t detect the first anymore, the first is good enough