Matthew Schwartz

Timeline

1. 1998–2003

   Penn degrees and a Princeton PhD

   He earns a B.A. in mathematics and biophysics and an M.A. in mathematics at the University of Pennsylvania in 1998. Princeton awards him a PhD in 2003.

2. 2008

   Assistant professor at Harvard

   After postdoctoral work at Berkeley and Johns Hopkins, he joins Harvard as an assistant professor in 2008. In June 2008, he coauthors "Top-tagging" with David E. Kaplan, Keith Rehermann and Brock Tweedie, a jet-substructure method that picks out boosted, hadronically decaying top quarks from light quark and gluon jets.

3. 2014–2015

   A QFT textbook and tenure

   Cambridge University Press publishes his textbook Quantum Field Theory and the Standard Model in March 2014, covering particle physics from its foundations to the discovery of the Higgs boson; field theory courses worldwide adopt it as a standard text. Harvard promotes him to Professor of Physics with tenure in May 2015. By 2015, he has also joined the ATLAS Collaboration at the Large Hadron Collider as a Short Term Associate to work on data analysis.

4. 2016

   Deep learning for jets

   In December 2016, with Patrick T. Komiske and Eric M. Metodiev, he coauthors "Deep learning in color," which tests whether convolutional neural networks can discriminate quark and gluon jets better than observables designed by physicists. The 2016 paper is an early application of deep learning to particle physics, and his later work applies convolutional networks to both discrimination and regression tasks at the Large Hadron Collider.

5. August 2020

   Launching an NSF AI institute

   In August 2020, the National Science Foundation awards a five-year, $20 million grant to create the AI Institute for Artificial Intelligence and Fundamental Interactions, joining physicists and AI experts from Harvard, MIT, Northeastern and Tufts. He is one of eight Harvard scientists on the team and heads the new institute's coordination board.

6. November 2022

   Interpretable AI in Nature Reviews Physics

   In November 2022, he publishes the comment "Should artificial intelligence be interpretable to humans?" in Nature Reviews Physics, discussing what it means to understand AI and whether that goal is achievable. The 2022 piece compares the timescales of AI and human evolution and argues that transferring understanding between biological and artificial intelligence will become a fundamental challenge.

7. December 2025–March 2026

   Vibe physics with Claude Opus

   In the last two weeks of December 2025, he guides Claude Opus 4.5 through a calculation resumming the Sudakov shoulder in the C-parameter; the resulting paper appears on arXiv in January 2026, with all calculations, numerical analysis and manuscript preparation performed by Claude under his supervision. He describes the experiment in the Anthropic guest post "Vibe physics: The AI grad student" in March 2026. At the American Physical Society's Global Physics Summit in Denver in March 2026, he predicts that large language models will surpass human intelligence in five years.

8. As of October 2026

   BootLoops and Anthropic

   In October 2026, he describes BootLoops in the Anthropic post "Claude-shaped science"; the open-source harness helps large language models do interdisciplinary quantitative science, and he and 19 coauthors use it to produce 36 manuscripts in 18 fields over three months. He works as a visiting researcher at Anthropic and owns and maintains BootLoops. He is a professor of physics at Harvard and serves on the IAIFI Institute Board as Research Forum Lead.