Previously, yes, that's the behavior it relied on, why it was broken, and why the article had a story to write about. Currently, no, it replaced that behavior w

Previously, yes, that's the behavior it relied on, why it was broken, and why the article had a story to write about. Currently, no, it replaced that behavior with external information not sourced from the LLM and that's why it doesn't matter if the LLM can or can't say "I don't know": the data for this type of task is no longer coming from the LLM itself.