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 some errors -- but not all.
https://bit.ly/3hrH9wH