A Princeton and UC Berkeley study says that AI chatbots could be relying on “partial truths” and “weasel words” as their ...