How to Scale a Model to Get Better Results

Chinchilla is the model and paper by Google. And they were showing that it's both parameters and data. So as you scale parameter two times, you need to scale data two times to get tracing data. It was showing a lot of models that people were doing were training inefficiently. They had way less data. But just for like some context, like how much data is a trillion? So it's about like one word per like data is what I'm how it's defined.