Updated
Updated · OpenAI · Aug 27
Study of 1,000 Students Finds ChatGPT Lifts Scores by Nearly 1 Point as Training Boosts Originality
Updated
Updated · OpenAI · Aug 27

Study of 1,000 Students Finds ChatGPT Lifts Scores by Nearly 1 Point as Training Boosts Originality

1 articles · Updated · OpenAI · Aug 27

Summary

  • More than 1,000 first-year Bocconi students in a randomized experiment used GPT-4o, causal-reasoning training, both, or neither on a real marketing case, letting researchers isolate each intervention’s effect.
  • ChatGPT access raised assignment scores by almost a full point on a five-point rubric, producing more ideas, clearer logic and answers closer to expert recommendations.
  • Causal-reasoning training did not lift rubric scores, but text analysis showed students generated a wider range of more distinct ideas and better explained why proposals might work or fail.
  • Students given both interventions combined those gains—their originality matched the training-only group, while their scores and idea counts resembled the ChatGPT-only group.
  • The findings suggest schools may need assessments that capture originality and reasoning, not just polished final answers, as AI helps novices produce expert-like work.

Insights

When AI provides perfect polish, how can educators redesign assignments to prove a student actually understands the underlying cause and effect?
Why did traditional university grading miss the most original ideas, and what does this mean for evaluating human intelligence?