A new paper from Microsoft Research and Salesforce finds that even the most capable Large Language Models (LLMs) fall apart ...