What's this about?
A paper tests AI models on real MBA teaching cases — open-ended, messy business problems with no single right answer — and finds models already perform very well across fields, improving fast generation over generation.
What exactly is a "case study" benchmark?
Business schools teach via cases: multi-page messes of finance, strategy, and people problems. Unlike math benchmarks, there's no answer key — grading requires expert judgment, making this a test of unbounded reasoning rather than lookup.
Why is it worth noting?
Most benchmarks are verifiable puzzles; this one targets exactly the fuzzy judgment work managers and consultants are paid for. Mollick (who teaches these cases at Wharton) flagging it gives it weight.