Harvard PhD Student: "ChatGPT Explains Lectures, but Research Still Needs People"
Nia Robles, a physics PhD student at Harvard, wrote a personal essay about studying and doing research with AI. The tool helped her make sense of lectures, she says, but research calls for human guidance to verify answers and discuss them.
ChatGPT helped her get through difficult lectures, but once she started doing research she needed a human being to go over the answers with. That is the account offered by Nia Robles, a physics PhD student at Harvard University, in a personal essay published on October 15 local time, describing how AI changed the way she studies and works. At the heart of the piece is the gap between getting an answer from AI and reviewing that answer with a person.
How much did ChatGPT help with understanding lectures?
Robles writes that at first she found it hard to trust ChatGPT’s answers on physics. Later, she found it helpful to ask ChatGPT to explain lecture material she had struggled to understand. It was useful, she says, because it gave her another way to understand what she had missed in class.
Once she moved into PhD research, what Robles emphasizes is human mentorship. She argues that she needs someone who can verify an answer and debate the substance of it with her. The help she got from having a lecture explained again, in other words, is not the same as the review and discussion that research requires. That distinction is relevant to readers who found AI’s explanations helpful for studying and are now considering how to use it in research.
The essay extends to the lab a problem Robles had previously taken up in the context of education. In an earlier piece she wrote about the competitive atmosphere of physics education at Oxford and students’ use of ChatGPT. This time she focuses on collaboration and mentorship in the lab. She also observed that colleagues spent less time working through problems together as they began using AI.
Who verified the equation GPT-5.2 Pro proposed?
In February 2026, OpenAI said GPT-5.2 Pro had proposed an equation in theoretical physics. The equation concerns gluon scattering amplitudes, quantities used to calculate the probability that particles interact in a particular way. OpenAI said that a separately configured internal version of GPT-5.2 proved the equation and that researchers verified the result.
The case OpenAI presented is one in which a model proposed an equation and humans reviewed it to turn it into a result. This example connects with the role of verification and discussion that Robles describes. While acknowledging how useful AI was for making sense of her coursework, Robles emphasizes that research requires someone who can check answers and discuss ideas with her.
For readers who study with AI, the question that lingers is who can help them check an answer after AI explains it. What remains to be seen is how classrooms and labs will provide opportunities to verify answers and find partners for discussion.