ORC Seminar: Scaling AI Capabilities in Mathematical Reasoning

Theophane Weber Research Scientist Google Abstract: TBD Bio: Théophane Weber is a Research Scientist and Team Lead in Gemini at Google DeepMind, where he focuses on RL science, LLM reasoning and AI-based scientific discovery. His research interests span sequential decision-making, reinforcement learning, generative models, probabilistic inference, inference-time scaling, and planning. At DeepMind, he was a key contributor to the IMO and DeepThink efforts, co-inventing the training technique behind the Gemini model that achieved a gold medal at the 2025 International Mathematical Olympiad. He is also a member of Project Big Sleep, which discovered the first real-world software vulnerability using AI (late 2024), pre-empted an in-the-wild zero-day exploit (July 2025), and demonstrated AI’s viability for cybersecurity by finding over 20 vulnerabilities across open-source software. Prior to DeepMind, he was at Lyric Labs (acquired by Analog Devices). He earned his PhD and MS from the MIT Operations Research Center (ORC), where he was co-advised by David Gamarnik and John Tsitsiklis for his PhD and Jérémie Gallien for his MS. His doctoral research explored the intersection of statistical physics, machine learning and computer science, following earlier work on dynamic control of queueing systems.