Simons Institute Courses

AI for Mathematics and Theoretical Computer Science

Discover insights from Sophie Morel's talk at the Simons Institute and SLMath Joint Workshop on AI for Mathematics and Theoretical Computer Science.

• YouTube
• 1 hour 16 minutes
• Self-Paced
• Free Video

Formal Reasoning Meets LLMs: Toward AI for Mathematics and Verification

Explore AI for formal mathematical reasoning, focusing on theorem proving and autoformalization. Learn about challenges through projects on inequality problems and Euclidean geometry formalization.

• YouTube
• 1 hour 5 minutes
• Self-Paced
• Free Video

Challenges in State-of-the-Art Bit-Precise Reasoning

Explore the challenges and state-of-the-art techniques in bit-precise reasoning for Satisfiability Modulo Theories (SMT), focusing on bit-blasting approaches and recent procedures to improve scalability for increasing bit widths.

• YouTube
• 1 hour
• Self-Paced
• Free Video

Testing Artificial Mathematical Intelligence

Explore how to evaluate AI's ability to contribute meaningfully to mathematical processes through a series of tests inspired by Turing's work and Thurston's perspectives.

• YouTube
• 1 hour 5 minutes
• Self-Paced
• Free Video

How Can Machine Learning Help Mathematicians?

Explore how AI models can assist mathematicians in solving open problems, providing insights across various mathematical domains, and supporting automated theorem proving.

• YouTube
• 59 minutes
• Self-Paced
• Free Video

Computer-Assisted Intuition: SAT Solvers in Mathematical Discovery

Discover how SAT solvers and automated reasoning tools can generate insightful constructions in discrete mathematics, challenging intuition and guiding mathematical discovery.

• YouTube
• 1 hour 8 minutes
• Self-Paced
• Free Video

Solving the Kaplansky Unit Conjecture Using SAT Solvers

Discover how SAT solvers helped solve the 80-year-old Kaplansky unit conjecture for group rings by treating it as a Boolean satisfiability problem, making a theoretically undecidable problem practically solvable.

• YouTube
• 1 hour 3 minutes
• Self-Paced
• Free Video

Learning-Theoretic Foundations of Algorithm Design

Discover insights into AI applications for mathematics and theoretical computer science in this talk by Maria-Florina Balcan from Carnegie Mellon University.

• YouTube
• 1 hour 4 minutes
• Self-Paced
• Free Video

Out Of Distribution, Out Of Control? Understanding Safety Challenges In AI

Explore the safety challenges in AI, focusing on out-of-distribution issues and safety guarantees for large language models with Aditi Raghunathan.

• YouTube
• 59 minutes
• Self-Paced
• Free Video

Antidistillation Sampling for Safety-Guaranteed LLMs

Explore the concept of Antidistillation Sampling in the context of Safety-Guaranteed LLMs with Zico Kolter from Carnegie Mellon University.

• YouTube
• 1 hour 5 minutes
• Self-Paced
• Free Video

Adversarial Training for LLMs' Safety Robustness

Explore techniques for improving the safety robustness of Large Language Models through adversarial training methods with researcher Gauthier Gidel from IVADO-Mila.

• YouTube
• 1 hour 1 minute
• Self-Paced
• Free Video

Scalably Understanding AI with AI - Safety-Guaranteed LLMs

Explore Jacob Steinhardt's insights on using AI to understand AI at scale, focusing on safety-guaranteed LLMs and their implications for AI development.

• YouTube
• 59 minutes
• Self-Paced
• Free Video

Causal Representation Learning: A Natural Fit for Mechanistic Interpretability

Explore how sparse shift autoencoders (SSAEs) enable accurate steering of language models by mapping embedding differences to sparse representations that capture concept shifts, offering an unsupervised approach to LLM alignment.

• YouTube
• 59 minutes
• Self-Paced
• Free Video

LLM Negotiations and Social Dilemmas

Explore game-theoretic perspectives on LLM agents interacting with humans and other agents, focusing on Advantage Alignment algorithms for guiding policy learning toward cooperative behaviors in general sum games.

• YouTube
• 1 hour
• Self-Paced
• Free Video

Controlling Untrusted AIs With Monitors

Explore how to control untrusted AI systems through monitoring mechanisms, with insights from Anthropic's research on safety-guaranteed language models.

• YouTube
• 1 hour 1 minute
• Self-Paced
• Free Video