Translated data: Google DeepMind researchers conducted experiments and found that the Transformer model has limitations in its generalization capabilities. The study results indicate that data quality is crucial for model performance, especially when dealing with new problems and function categories. This research underscores the limited generalization ability of the Transformer model, which holds significant implications for the fields of machine learning and artificial intelligence.