'Scaling law' is the prediction that increasing the training data and model size will cause the result to be better. It's not a law, but it was the prediction that OpenAI staked their fledgling startup on to get to the ChatGPT moment.