Invited talks and distinguished guests at The 2026 Universal Cup Finals.
Abstract
I will talk about my experiences in academia as a former competitive programmer. Along the way, I will discuss some research projects in algorithms that I have undertaken over the years, specifically about problems in graph theory, variants of the Traveling Salesperson Problem, and/or data clustering.
Abstract
The rapid rise of Large Language Models (LLMs) has raised a major question for our community: is AI about to replace human competitive programmers, or is it simply becoming another tool? This talk cuts through the hype by examining the real algorithmic abilities of state-of-the-art AI through three recent milestones, tracing its role from contestant, to problem setter, to heuristic optimizer.
First, we examine AI as a competitor through LiveCodeBench Pro (NeurIPS 2025). To avoid contamination from past online archives, we evaluate models in real time on fresh Codeforces, ICPC, and IOI problems before solutions are published. The results show a clear divide: models do well on knowledge-heavy implementation tasks, but struggle on ad-hoc, constructive, and observation-heavy problems. We discuss why AI still misses the creative "aha" moments that human experts often find naturally.
Next, we turn to AI as a problem setter with AutoCode (ICLR 2026). Writing strong statements, validators, and edge-case generators is one of the hardest parts of contest design. We present an automated system in which LLMs generate robust test suites that catch tricky "cheese" or hack solutions. More surprisingly, the system can also create novel contest-grade problems that the AI itself still cannot solve to Accepted.
Finally, we explore AI as a heuristic optimizer through FrontierCS (under review, ICML 2026). Unlike standard competitive programming, which relies on binary AC/WA verdicts, heuristic contests use continuous scores on open-ended optimization tasks. FrontierCS extends this setting to NP-hard problems and evaluates how well models improve solutions when the true optimum is unknown. We show how these challenges naturally connect competitive programming to real-world computer science research.
Associate Professor
Carnegie Mellon University
Richard Peng is an associate professor at Carnegie Mellon University. His research focuses on the design and analysis of efficient algorithms for fundamental computational problems, including graph algorithms, dynamic algorithms, and linear algebraic algorithms. His representative work includes linear system solvers, maximum flow and minimum cut algorithms, and time- and space-efficient data structures for problems involving matchings, electrical flows, and matrices. He has received best paper awards at SODA and FOCS. In high school, he represented Canada and won an IOI gold medal and an IMO silver medal. He later earned a bronze medal at the 2008 ICPC World Finals as a member of the University of Waterloo team.
Associate Professor
University of Alberta
Zachary Friggstad is an associate professor at the University of Alberta. His research interests primarily concern combinatorial optimization with emphasis on the design of approximation algorithms for NP-hard combinatorial optimization problems. This includes, but is not limited to, problems in data clustering, vehicle routing, network design.
Associate Professor
University of British Columbia
Robert Xiao is an Associate Professor in the Department of Computer Science at the University of British Columbia in Vancouver, Canada, and a core member of the interdisciplinary Designing for People research cluster. He uses his background in computer science and mathematics to craft novel interactive technologies driven by sensors, machine learning and computer vision. These technologies range from improving touchscreen input, to novel motion tracking systems, on-world projected interfaces, and much more. Alongside his research interests, he routinely competes in DEF CON CTF (1st place 2016, 2017, 2019, 2022, 2023, 2024, 2025) and other security contests.
Master's Student, Combinatorics & Optimization
University of Waterloo
Roger Fu is a master's student in Combinatorics and Optimization at the University of Waterloo, advised by Jim Geelen, with primary research interests in matroid theory. Beyond research, he has long been active in the competitive programming community as an open-source developer and maintainer of DMOJ, and has helped host more than 200 programming contests for various organizations, including several national-level events.