Comment by m00nsome

Why do they not release the unnatural Variant of the model? According to the paper it beats all of the other variants and seems to be close to GPT-4.