Comment by supermatt

thats a real shame. Any idea why GPT-3 generations differ so much? Maybe eleuther/whatever could look at refining their model in a similar way.