This is where 'bigger reliably means better' got its evidence, and where the curve's shape — smooth and predictable across orders of magnitude — first let labs forecast a model's performance before training it. It also set the priorities Chinchilla later corrected.
