Towards Reliable Misinformation Mitigation: Generalization, Uncertainty, and GPT-4