GARDO: Reinforcing Diffusion Models without Reward Hacking Paper • 2512.24138 • Published 21 days ago • 28