Researchers detail blind spots of large language models

Researchers at the Allen Institute for AI find that scaling up language models like OpenAI's GPT-3 can fix certain errors.