RT @chris_j_paxton: Its genuinely hard to believe a 70B model is up there with the 1.8T GPT4? I guess training data really is everything