Wharton School Professor Christian Terwiesch found OpenAI's ChatGPT was able to pass a final exam for the school's MBA program, scoring between B- and B
Christian Terwiesch's exam result landed two and a half weeks after GPTZero went viral as a ChatGPT essay detector — and it undercut the detection thesis at the source: the model wasn't just producing plausible prose, it earned a B- to B on an actual Wharton MBA final. Within a month, another Wharton professor had made ChatGPT mandatory for class essays, with students doing the accuracy and bias checking themselves.
The result also drew a boundary rather than a verdict: when GPT-4 failed a CFA Institute exam two months later despite answers being available online, it was clear exam performance depends heavily on format and domain. By September, ChatGPT was out-generating Wharton's own students, taking 35 of the top 40 spots in an idea-generation competition between the model and MBAs.
First-order effects
Standard written MBA assessments lose their signaling power immediately: if a model scores B- to B unaided, grades on conventional finals can no longer separate a student's knowledge from the tool's.
Wharton's own response within weeks — Terwiesch's colleague building assignments around required ChatGPT use — shifts the school from policing the tool to teaching with it, moving verification work onto students.
Second-order effects
Exam-proctoring and AI-detection vendors get their strongest demand case yet: GPTZero's viral moment preceded this result, and every documented model pass raises the stakes for institutions deciding whether to buy detection or redesign assessment.
Credentialing bodies gain a reason to benchmark rather than panic — the CFA Institute exam failure that followed shows a single MBA pass doesn't generalize, so professional testers can calibrate which formats remain discriminating.
Third-order effects
The trajectory visible across the corpus runs from containment to procurement: by late 2025 OpenAI had sold over 700K ChatGPT licenses to roughly 35 US public universities, meaning institutions that once feared the model now pay for it at scale.
If assessment signal keeps eroding while usage is institutionalized, the durable question for accreditors and employers shifts from 'did the student write this' to whether programs teach students to direct, verify, and correct AI output — the skill the Wharton essay assignments were already practicing.
The trend: Higher education is moving from detecting AI-written work to licensing and teaching with it, turning exam performance into a recurring benchmark contest rather than a gatekeeping barrier.
You probably shouldn't need a chatbot to understand that a lot of post-secondary exams are just pattern-matching and not a meaningful examination of understanding a topic. https://twitter.com/...
ChatGPT passed exam on a Wharton MBA course. And this platform has potential to further improve. Massively. The age of machines is near !! https://twitter.com/...
Rather than thinking this speaks well about the quality of ChatGPT, we should consider how it speaks poorly of the intellectual quality of business schools https://www.nbcnews.com/...
too lazy to set up a word filter to change “ChatGPT” to “a photocopier” but it does feel like it would work out like when people ask “can a photocopier replace therapy?” or when there's a headline like “a photocopier, when trained on answers, passed the MBA exam”
Just read the paper on ChatGPT passing a Wharton MBA exam. One part asked ChatGPT to design “queuing theory” questions as it relates to Homer Simpson's life. The answers are really good: https://twitter.com/...
I had @OpenAI Chat GPT3 take my @Wharton Operations Management MBA exam. It got a B/B-. Is that good or bad news? Can we replicate this at other schools / subjects? Some implications are in this @MackInstitute report https://mackinstitute.wharton.upenn.edu / ... https://twitter.c…
Stories like this remind me of “IBM Watson beats humans at Jeopardy” stories. These are publicity stunts and tell us nothing about the technology or the human activity it's supposedly mastered. https://www.nbcnews.com/...
Not yet. Can ChatGPT lever up a portfolio firm, lay off 60% of employees, degrade software quality, and sell the firm to its own secondary fund in a soft form of fraud? https://twitter.com/...
“ChatGPT Passes medical/law/mba/etc. Exam”. Yeah? I would too if I had petabytes of relevant or exact training data available to me (aka open book exam). Another day, another set of worthless metrics.
ChatGPT has now successfully passed a Wharton MBA exam. Sure, this isn't the hardest problem, but to execute it in 1 second is the breakthrough. https://twitter.com/...