James Chua comments on OpenAI finetuning metrics: What is going on with the loss curves?