OpenAI’s ChatGPT, powered by the GPT-4 model, has recently demonstrated its ability to pass a typical freshman year at Harvard University, achieving a GPA of 3.34. The experiment, conducted by Maya Bodnick, an intern at Slow Boring, aimed to assess the chatbot’s performance at an Ivy League college.
To evaluate ChatGPT, Bodnick asked eight professors and teaching assistants to grade the essays generated by the AI in response to real Harvard prompts. In an effort to reduce bias, she did not disclose whether the essays were written by her or the AI.
The results were promising, with the AI achieving mostly As and Bs, along with a solitary C, across various social science and humanities subjects during a typical freshman year. The average GPA amounted to 3.34, as reported by the newsletter.
Many of the academics praised the writing skills of the chatbot, except for one who suggested simplifying the writing style. However, the professors expressed concerns about the content and arguments presented in the essays. According to Bodnick, one grader found the essay arguments to be consistently extensive and unclear.
The advent of generative AI, exemplified by OpenAI’s ChatGPT, has caused significant disruption in the higher education sector. The AI’s capability to complete complex assignments has raised accusations of academic dishonesty, though investigations have debunked some of these claims.
While previous experiments involving chatbots and professors have been conducted, Bodnick’s experiment stands out due to its scale. Christian Terwiesch, a professor at Wharton, tested an earlier version of ChatGPT powered by GPT-3.5 by asking it questions from a final exam, but the AI only achieved a B or B- grade.
Colleges and universities have struggled to navigate the implications of this new technology, leading some professors to take matters into their own hands. The rise of generative AI has engendered a sense of distrust between students and lecturers. Consequently, the higher education sector is now introducing guidelines and policies to effectively manage the proliferation of generative AI.
In conclusion, the experiment conducted by Maya Bodnick illustrates that ChatGPT, powered by GPT-4, is capable of achieving a satisfactory performance at a prestigious institution like Harvard University. However, concerns remain regarding the quality and clarity of the content generated by the AI. The higher education sector continues to grapple with the challenges posed by generative AI, necessitating the implementation of regulations and measures to navigate this rapidly evolving landscape.
