/ai
When posed with a logical puzzle that demands reasoning about the knowledge of others and about
counterfactuals, large language models (LLMs) display a "distinctive and revealing pattern of failure," according to a bulletin from the Bank for International Settlements.