Loading: Researchers propose bias fix for GPT-3 and other language models