Recursive training of generative models degenerates into reinforcement learning with random feedback. You need strong feedback signal in the loop to survive rec

Recursive training of generative models degenerates into reinforcement learning with random feedback. You need strong feedback signal in the loop to survive recursive training. The feedback does not have to come from humans though. You can use anything that grounds the model in reality.