These results are predicted by LLM Scaling Laws and the GPT authors knew it before they started.