1 of 2
Researchers Find Decline
- Researchers from Carnegie Mellon, MIT, Oxford and UCLA found performance fell after an OpenAI GPT-5 chatbot session ended.
- Performance on the last three math questions fell 20 % versus controls while skip rates nearly doubled.
- Users who received AI hints performed like participants who never used AI.
- The decline showed in SAT reading tasks, matching findings from Microsoft and Polish doctor studies that warn AI reliance may erode critical thinking.
1 / 2
