Thanks so much, this is reasonable and insightful. So I guess it's safe to say that ChatGPT/turbo (not GPT-2) passed the threshold for "acceptable" performance

Thanks so much, this is reasonable and insightful. So I guess it's safe to say that ChatGPT/turbo (not GPT-2) passed the threshold for "acceptable" performance for summarization, Q/A, etc., which means that we can use GPT-4/5/... for tricky tasks that turbo was not able to do.