Meet the Fellow: Jaume de Dios Pont

This entry is a part of our Meet the Fellow blog series, which introduces and highlights Faculty Fellows who have recently joined CDS.

Meet CDS Faculty Fellow Jaume de Dios Pont, who will join us this January after completing a postdoc at ETH Zurich. Pont brings a unique perspective to the intersection of mathematics and artificial intelligence, driven by what he describes as both excitement and apprehension about AI’s growing capabilities in mathematical research.

“I kind of joke that if someone has to take my job away, I’d rather it be me,” said Pont, explaining his motivation for transitioning from pure mathematics to data science. “So that’s part of the motivation to move to a data science department.”

De Dios earned his PhD in mathematics from UCLA under the supervision of Terence Tao. He previously completed a master’s degree in mathematics at ETH Zurich in 2018 and dual bachelor’s degrees in mathematics and physics from Universitat Autonoma de Barcelona in 2017. Between completing his PhD and starting his postdoc, he worked as a research intern at the Machine Learning Foundations group at Microsoft Research. His research is funded by the Simons Collaboration on Localization of Waves.

At CDS, de Dios plans to explore two complementary research directions. The first involves applying mathematical insights to understand and create more efficient AI systems, drawing on his background in harmonic analysis and spectral theory to tackle questions about diffusion models and scaling limits of neural networks. This work connects to broader mathematical questions about how these systems behave at scale and what theoretical guarantees can be made about their performance.

The second direction de Dios plans to explore examines how AI can revolutionize mathematical practice itself. He believes that within two to three years, mathematicians who effectively use AI-based systems will become “radically more efficient” than those who don’t. Current AI systems for mathematics, however, fall short of what mathematicians actually need, according to de Dios. Through conversations with colleagues and his own experience, he’s discovered that the mathematical community wants tools that can operate across multiple levels of abstraction — from high-level intuitive reasoning down to formal verification. “We want to see these layers in the system,” he explained, describing a vision where mathematicians can steer AI assistance at different levels of precision and have the system fill in the gaps.

De Dios is particularly interested in formal proof verification systems like Lean, which force AI models to communicate in a programming language that computers can verify for correctness. This approach addresses a fundamental problem with current large language models: their tendency to generate plausible-sounding but incorrect mathematical statements. “If it lies to us [in the language of Lean], it’s very clear and easy to catch,” Pont said, explaining the benefits of systems that use formal verification.

One challenge De Dios identifies with current AI development is the industry’s focus on measurable benchmarks. Companies optimize for tasks with clear numerical scores, but the problems that interest him most — like improving the user experience of mathematical AI tools — resist simple quantification. “I find it really hard to attach a number to the tasks that I’m interested in,” he noted.

De Dios is excited to collaborate with CDS Professor of Computer Science and Data Science Joan Bruna, who introduced him to the Center, as well as other theory-focused faculty. He’s also looking forward to New York’s vibrant scientific ecosystem, which includes not only CDS but also Columbia, the Flatiron Institute, and Princeton.

“There’s so much going on, so much happening,” de Dios said of New York. “It’s so easy to meet people doing interesting, exciting things.”

To view all our current faculty fellows, please visit the CDS Faculty Fellow page on our website.

By Stephen Thomas

Have feedback on our content? Help us improve our blog by completing our (super quick) survey.