The alignment problem in artificial intelligence is becoming increasingly complex. OpenAI's research suggests that supervising large models with smaller ones can significantly enhance generalization performance in natural language processing tasks. Traditional human supervision may be insufficient for super AI models, but the weak-to-strong generalization method holds promise for significantly improving performance. This research encourages more empirical studies, provides open-source code, and initiates a funding program. Substantial progress in the field of super AI alignment may be on the horizon in the future.