Description
LLMs can perform tasks that seem to require reasoning, especially with techniques like "chain-of-thought" prompting (showing the model step-by-step thinking). However, the depth and reliability of this reasoning are unclear, and they often struggle with problems that require robust, out-of-distribution reasoning. It's an open question whether their limitations are fundamental or will disappear with more scale or better training.