A master’s degree course that normally takes one week took one University of Pennsylvania professor 12 hours with the help of AI.
Jesús Fernández Villaverde, an economics professor at the University of Pennsylvania, said he used Anthropic’s chatbot Claude to design and complete a personalized course on sociologist Erving Goffman and discuss the content in real time.
After about 12 hours of study, he said he reached a level of understanding comparable to covering the total study time of a week-long master’s course.
“Goffman typically takes one week of classes, which equates to about nine to 12 hours of total study time for students,” he told Business Insider.
“The time I invested is comparable to the time spent by students in a well-structured course,” he said. “The ability to learn at this level at almost zero marginal cost is extraordinary.”
AI as a personal tutor
According to Fernández Villaverde, there were three stages in the process.
First, Claude created a customized syllabus based on existing knowledge, including readings, major themes, and connections to other thinkers.
Next, he read books by sociologists, such as “The Presentation of the Self in Everyday Life” and excerpts from “Exile” and “Stigma.” Claude selected the readings and assembled the order, he said, “but I did the readings myself.”
Third, he used AI as an interactive partner, seeking explanations, connections, and parallels to what he was already familiar with in economics.
“There’s a difference between reading a book alone and reading with a knowledgeable colleague who has infinite patience and no office hours,” he said of his experience with AI-generated courses.
He said Claude was “excellent” at managing what to do, in what order, and what to emphasize, taking into account students’ special backgrounds, calling it “one of the most difficult things a professor does.”
“Claude did this at a level that I would say is above the 90th percentile of real professors, at least for this type of work,” he added.
Not a perfect replacement
However, Fernandez-Villaverde said he fell short of Claude on several levels.
Like a good teacher, they don’t challenge their students.
“It answers the questions you ask, not the questions you should ask,” he said.
Nor can they recreate the peer experience in the classroom. Still, he said, human professors aren’t perfect either, adding that they don’t always answer their students’ questions accurately or accurately.
“You should compare Claude to a real professor, not an ideal professor,” he said.
university calculations
Fernández-Villaverde believes that AI is a huge positive for learning, but also a serious challenge for certain educational institutions.
Several professors and prominent academics, including University of Texas history professor Stephen Mintz and economist Tyler Cowen, expressed similar concerns, saying AI is exposing long-standing weaknesses in higher education, such as standardized assignments and outdated teaching models.
For Fernández Villaverde, the impact is more economic, with AI forcing universities to justify student payments.
University programs built primarily around delivering lectures are particularly at risk, he added.
“If your primary value proposition is to transfer existing knowledge in the classroom, and students can get an equivalent or better version of it for $20 a month, your business model is under severe pressure,” he said.
But that doesn’t apply equally to all institutions, he said, adding that top universities still offer advantages that cannot be easily replicated with AI, such as proximity to research frontiers, strong peer networks, and valuable credentials.
He said he doesn’t think AI will eliminate traditional higher education, but believes it will force a rethink about what it offers.
“Universities that thrive are those that offer things that AI cannot,” he said, citing research mentorship, access to labs, a true peer community and trusted credentials.
“Companies that are essentially selling access to lectures and diplomas will face the most difficult questions,” he added.