[NooneAtAll3]

I thought "data efficiency" meant same quality with less parameters

instead it's more parameters with less training data... but I don't really see any quality control?

[naasking]

Seems like they're relying on the loss as a measure, at least for now.