Comment by kubb

The wall is training data. Yes, we can make more and more of post training examples. No, we can never make enough. And there are diminishing returns to that process.