Harvard DSR
Language Models
Humans are terrible at spotting when AI's lying — and it's making us worse
You ask a confident friend for directions and they sound so sure you trust them, even though they have no idea where they're going. The more convincingly wrong they sound, the more you believe them next time.
This means the more we rely on AI suggestions (in spreadsheets, code editors, or dashboards), the worse our own judgment becomes — a hidden cost of the AI we're already using.
Bug reported: No