Comment by shikon7

Also, using an OpenAI model to judge the performance of an OpenAI model seems prone to all kinds of biases.