[trc001]

Well, there sort of is a linear relationship between the ammount of training data and the quality of the model [1]

[1]: https://arxiv.org/abs/2203.15556