ChatGPT showed it could pass a freshman year at Harvard and achieve a 3.34 GPA, student says

Advertisement
ChatGPT showed it could pass a freshman year at Harvard and achieve a 3.34 GPA, student says
The entrance to Harvard Yard.Scott Eisen/Getty Images
  • ChatGPT is capable of achieving respectable grades at Harvard, an experiment found.
  • Per the experiment conducted by a Harvard student, the bot ended the year with a 3.34 GPA.
Advertisement

A version of ChatGPT powered by OpenAI's GPT-4 is capable of passing a typical freshman year at Harvard, a recent experiment found.

Maya Bodnick, an intern at Matthew Yglesias's newsletter "Slow Boring," decided to see how the chatbot would fare at an Ivy League College.

To conduct the experiment, Bodnick, a Harvard student herself, asked eight professors and teaching assistants to grade ChatGPT's essays generated in response to real Harvard prompts. In an attempt to lessen potential bias, she told the professors the essays would either be written by her or by an AI.

Complimentary Tech Event
Transform talent with learning that works
Capability development is critical for businesses who want to push the envelope of innovation.Discover how business leaders are strategizing around building talent capabilities and empowering employee transformation.Know More

The bot's performance was largely impressive, achieving mostly As and Bs, along with one C across Bodnick's social science and humanities-focused freshman year. The grades averaged out to a 3.34 GPA, the newsletter said.

Many of the academics heaped praise on the chatbot's writing skills, aside from one who urged the "student" to simplify their writing. However, the professors were less positive about the content and claims made in the essays. One grader called the essay's arguments "consistently large and unclear," according to Bodnick.

Advertisement

The rise and popularity of generative AI, prompted largely by the release of OpenAI's ChatGPT last year, has rocked the higher education sector. The bot's ability to complete students' take-home assignments at a relatively sophisticated level has sparked widespread accusations of cheating, some of which were proved to be false.

Professors at various colleges have tried similar experiments before, although perhaps not at the scale of Bodnick's. Wharton professor Christian Terwiesch tested an earlier version of ChatGPT powered by GPT-3.5 with questions from his final exam. The chatbot, however, only earned itself a B or B- grade on the exam.

Colleges have struggled to deal with the consequences of the new technology, prompting some professors to take matters into their own hands. An atmosphere of distrust has arisen between some students and lecturers as a result. Higher education is starting to bring in guidance and policies to manage the rise of generative AI.

After a shaky start that resulted in several outright bans being introduced, some colleges are now trying to incorporate and capitalize on generative AI tools.

{{}}