Kaj_Sotala comments on Is GPT-3 already sample-efficient?