From the blog's footnote: "In addition, we’d like to note that engineers have worked to optimize prompts and few-shot samples for evaluations and reported high

From the blog's footnote: "In addition, we’d like to note that engineers have worked to optimize prompts and few-shot samples for evaluations and reported higher scores for a newer GPT-4T model"