Data Flywheel / Feedback Loop

A mechanism in which outputs generated by a model are reused to train the next generation of models. Within an RL loop, this can compound with knowledge distillation to produce amplification effects.