Letter from the Director, August 2026

Venkat Guruswami, photographed from the shoulders up against a gray background. He's wearing glasses and a dark shirt.

Dear friends,

Greetings from Berkeley! This has been a dizzying summer, one in which we witnessed an inflection point in AI’s capability to autonomously solve major open problems in math and theoretical CS. Like many of you, I feel a mix of amazement at this rapid acceleration and uncertainty about what it means for the future of TCS research as a profession. Scientifically, though, there is much to celebrate: the field has advanced by leaps and bounds this summer, whether done the old-school way, AI-assisted, or AI-driven. Even within the corners of theory close to my own interests, we now have a deterministic NC algorithm for bipartite matching, polynomial-factor inapproximability of finding close vectors in lattices, a one-shot algebraic PCP sans composition, a powerful approach to sparsification rooted in convex geometry, efficient unclonable encryption, quasipolynomial cryptanalysis of the McEliece cryptosystem, and improvements — in not one but two ways — on rate upper bounds for codes with prescribed distance. (The next time I teach coding theory, I’m glad I won’t have to say that these bounds haven’t budged in my lifetime.) And with humans and AI working together, ever greater discoveries in TCS may be on the horizon. Perhaps, just as the brilliance of my students and postdocs has always inspired me, I can learn to take joy in understanding AI’s contributions as well. 

For the TCS research community, this is a moment for reflection and exploration — about how to collaborate with these models and avail ourselves of their transformative powers, while honoring the uniquely human capacities of understanding and judgment. We are already seeing disruptions to the way results are discovered, shared, and credited, and these will have downstream implications for how we train students, evaluate research, and build careers. I know this uncertainty weighs most heavily on our students, postdocs, and junior colleagues, and their future must be at the center of these conversations. Informed by the results of a widely circulated survey, the Simons Institute will convene a working group next month with the goal of producing some concrete, near-term suggested actions for the community around attribution, conferences, and hiring. We recognize the significance of this moment, and hope to provide a forum for discussion and an evolving shared vision on these important topics.

Last week, we welcomed the participants in our Fall 2026 research programs on Spectral Theory Beyond Graphs and Pseudorandomness & High-Dimensional Expansion. Spectral graph theory is a central topic in TCS belonging to a broader mathematical tradition that studies spectra of operators on more general spaces; the Spectral Theory Beyond Graphs program brings the TCS and mathematical communities together to think about spectra of graphs, groups, and manifolds. Pseudorandomness is the theory of efficiently generating random-like objects using little or no randomness; high-dimensional expansion is a generalization of expansion in graphs to higher dimensions. The Pseudorandomness & High-Dimensional Expansion program seeks to strengthen the ties between these two research communities, and advance applications to and connections with other areas (including spectral theory, as represented in the concurrent program). These two interconnected programs kicked off with a joint workshop last week: the Simons Institute’s ICM Satellite Conference on Spectral Theory, High-Dimensional Expansion, and Pseudorandomness. Both programs boast an exceptionally strong roster of research fellows. 

Since I last wrote to you, we’ve wrapped up a highly successful Summer Cluster on Quantum Computing and held workshops on Topics in Intelligence: World Models and Social Reasoning (June 8 – 12) and Diffusion Generative Modeling: Progress and Next Steps (August 3 – 7), as well as a reunion workshop (July 13 – 17) for our 2025 Cryptography program. We share some highlights from these, and from our May workshop on The Role of TCS in Modern Machine Learning, in our SimonsTV corner this month, with talks on deep data mining, the impact of LLMs on the human activity of research, the Komlós conjecture, self-correcting quantum memory, sampling for diffusion models, and fault tolerance against adversarial errors and PCPs.

Last month, we introduced Berkeley machine learning researcher Jason Lee as our new Senior Scientist. To help you get to know Jason, we’re delighted to share a Q&A with him about his research, how LLMs are influencing CS theory, and vice versa. We round off this month’s newsletter with highlights from press coverage of our field. 

Here’s to a fall of great theorems — whatever their source,
Venkat

Venkatesan Guruswami
Director, Simons Institute for the Theory of Computing

,