Microsoft researchers have successfully conducted studies on selective forgetting in large models, allowing models to forget specific information rather than the entire content through fine-tuning and reinforcement learning. The research method has certain limitations and requires further validation.