Why can't ChatGPT tell time? (www.theverge.com)

Despite being sold as a capable personal assistant, ChatGPT still can’t reliably tell the current time. Users report inconsistent answers—sometimes it says it can’t access the device clock, sometimes it guesses a local time, and occasionally it’s correct until a few minutes later when it’s wrong again. The inconsistency highlights a real-world shortcoming that undermines trust in chatbots: they can’t be assumed to have up-to-the-second awareness unless explicitly given a path to it. The reason is architectural. Large language models are trained to predict text from their training data and don’t have built-in access to real-time system state (like a clock) unless an app or plugin supplies that information via the chat context or by running a web/system search. Adding live timestamps or frequent clock reads consumes context-window space and can clutter or confuse the model, so designers trade off timeliness against context management and privacy. Some clients enable search or local-system hooks and can therefore report correct time; others don’t for safety (prompt-injection risks) or resource reasons. The issue also extends to related tasks—reading analog clocks and calendars remain error-prone. OpenAI acknowledges the limitation and says it’s working to make the model more consistent about when and how it fetches up-to-date facts.