If that's their strategy, it's pretty dumb since fine tuned LLAMA2 is already almost as good as GPT-4 now and will likely surpass it by the end of the year.