“Our 65B model is better performant than ChatGPT‘s 175B” is missing the point openai made in Nov'22.
That aside, this looks nice incremental research progress.