Two Training Paths, One Smarter AI Strategy

RLSD blends verifiable rewards with self-distillation to train models more stably and avoid the collapse seen in naive self-supervision.