SynPref-40M and Skywork-Reward-V2: Scalable Human-AI Alignment for State-of-the-Art Reward Models Kommentar verfassen / Von / Juli 7, 2025