And you cannot Overfit 75B on a 3M because there's no capacity for it its forced to generalize, what can overfit is 1.6B on 1M parameters which is what photon uses
You have a textbook definition of overfitting. It doesn't work like that.
Both of your sides are using "scaling" differently. One side is exploring the small-model efficiency frontier, the other is defending the large-model scaling paradigm. You guys are almost talking past each other.
