When pre-training data is limited, it is better to train a smaller model for multiple epochs than to train a larger model on unique data once.

What these subjects mean

simplicity Less, but better: reduction as the way to make something good, and the argument about when it goes too far.

scaling laws Whether more compute keeps buying more capability, and where the curve now bends -- pre-training, post-training, test-time.

What Sebastian Ruder actually said

Do you hold this korrent?Do you also believe this?

Shared subject first, then similar wording. They may agree or disagree.