yes in many cases the token count is in fact less it is almost impossible to get GPT-4 to regenerate the original text word for word when decompressing the compressed text but usually, the gist of it remains pretty intact

→ View original post on X — @alexalbert__,