In deep learning models, is it 'data quality' or 'model size'?

15 Tem 14:00

In recent times, especially with large language models, accessing data has become easier, but issues regarding data quality and cleanliness are increasing. On one hand, there are models working with massive parameters, while on the other, there are smaller approaches focused on improving data. Which do you think will dominate in the future? Or is a balance between the two important? Will models that achieve the same performance with less data drive progress, or will massive systems fueled by more data lead the way?