Did AI Write 36 Papers? A Harvard Physicist Reveals the Era of the 'AI Grad Student'

Imagine this: How would you feel if complex physics calculations and paper writing, which usually required 1 to 2 years of dedicated effort to achieve a life’s academic goal, were finished in just two weeks?

Recently, something akin to a science fiction movie occurred in academia. Professor Matthew D. Schwartz, a theoretical physicist at Harvard University, utilized the artificial intelligence ‘Claude’ like an ‘AI grad student’ to produce immense research results. He poured out 36 paper drafts in the short span of just three months [Harvard physicist, 36 paper drafts…(AGI Hunt)].

Why is this news important?

Scientific research, especially in fields like theoretical physics that deal with highly advanced logic, requires immense mathematical calculations and extensive data analysis. The time often referred to among researchers as ‘1 to 2 years of a graduate student’ is a sequence of such tedious and repetitive processes [Harvard physicist, AI-written paper…(The Crimson)].

What if a researcher could be freed from these simple, repetitive calculations? Humanity could focus its energy on asking more important ‘questions’ and gaining deeper insights into the essence of nature. Professor Schwartz’s experiment suggests that the ‘speed’ and ‘method’ of scientific research itself are changing completely. It is significant in that it was not merely a showy event, but the completion of papers that actually contribute to academia [Harvard physicist, AI-written paper…(Physics Explained)].

In simple terms, what is the method?

Shall we compare Professor Schwartz’s research method to ‘cooking’?

If past scientific research was a method where a great chef prepared ingredients from scratch, made stock, and directed every process from beginning to end, now it is like working with a reliable AI sous-chef.

Professor Schwartz used a system called ‘BootLoops’ (an open-source toolkit that connects AI models to scientific problems) to pose complex questions to the AI [Harvard physicist, releases BootLoops(Crypto Briefing)]. The process is as follows:

1. Topic Selection: Professor Schwartz’s team selected about 400 research topics.
2. AI Calculation: Claude (Claude Opus 4.5 model) performed numerical analysis and complex physics calculations.
3. Paper Writing: The AI wrote the drafts of the papers, and Professor Schwartz performed the role of a ‘Supervisor’ who reviewed and guided them.

In simple terms, the physicist decided the overall direction of the research and the ‘flavor of the research’, while the AI took charge of the complex labor and numerical work to realize that process [VibePhysics: A physics professor writes a paper with Claude(Jangwook.net)]. For this immense collaboration, a staggering 51,000 conversations were exchanged, and the paper was completed after 110 revisions [Harvard physicist, 1 paper in 2 weeks…(Clinicians Build)].

How far has the current research come?

Professor Schwartz has already published a theoretical physics paper on ‘Resummation of Sudakov shoulders in C-parameters (a specific physical data pattern appearing in particle collision experiments)’ with Claude’s help [Harvard professor writes physics paper with AI(DeepTechStars)]. This paper is an achievement that would have taken a single research group years to accomplish using conventional methods [Theoretical link in particle physics(Unite.ai)].

Of course, the important point is that the AI did not write the paper alone. This process was possible only because there was the physicist’s thorough supervision and insight into the AI. The AI is like a capable grad student, but setting the direction of the research and judging the value of the results remains the unique domain of humans [VibePhysics: A physics professor writes a paper with Claude(Jangwook.net)].

What will happen in the future?

Professor Schwartz is experimenting with this ‘AI-accelerated research’ with 19 collaborators across 18 scientific fields worldwide [Harvard physicist, 36 paper drafts…(AGI Hunt)]. It seems clear that AI will settle as an essential ‘colleague’ in the scientific community, going beyond merely assisting.

The point we need to watch is not ‘who wrote the paper’, but ‘what new discoveries will AI and humans collaborate to make at a speed that was impossible in the past’. Perhaps an era will soon arrive where the science news you read will pour out 10 or 100 times faster than now.

MindTickleBytes’ AI Reporter Perspective

Technology is evolving beyond a simple tool that does work for us into a partner that expands human intellectual limits. What matters is not the AI model itself, but the human’s ‘questioning ability’—the constant contemplation of what questions to ask using AI. When artificial intelligence and humans join hands, humanity is opening new horizons of knowledge that were previously unreachable.

References

1. Harvard physicist, 36 paper drafts…(AGI Hunt)
2. Harvard physicist, AI-written paper…(The Crimson)
3. Harvard physicist, releases BootLoops(Crypto Briefing)
4. Harvard professor writes physics paper with AI(DeepTechStars)
5. Harvard physicist, 1 paper in 2 weeks…(WinBuzzer)
6. Harvard physicist, AI-written paper…(Physics Explained)
7. VibePhysics: A physics professor writes a paper with Claude(Jangwook.net)
8. Theoretical link in particle physics(Unite.ai)
9. Harvard physicist, AI-assisted research(LinkedIn)
10. Harvard physicist, 1 paper in 2 weeks(Clinicians Build)