Bug in Original Scaling Laws Led to Suboptimal Model Training

An error in original scaling law calculations potentially caused widespread industry inefficiency by encouraging the training of oversized, undertrained models. This oversight failed to account for the relationship between parameter count, training duration, and inference costs.