This repository is for preparing for and following the Fields Institute course Mathematics for AI Safety. It will collect course notes, useful resources, and project ideas as they develop.
Read the course documentation · A learning project of the Varifold community.
- Dates: September 14–December 9, 2026; mid-semester break is October 12–16.
- Instructor: Yevgeny Liokumovich, University of Toronto.
- Lecture times: Mondays, 10:00 a.m.–12:00 p.m.; Wednesdays, 10:00–11:00 a.m. ET.
- Course project: Students are expected to develop a viable research project with a novel contribution. The final project is the basis for the course grade.
- Official page: Course description and schedule
- Recordings: YouTube playlist (linked from the official course page).
Planned topics include AI capability forecasting, deep learning theory, LLM training, adversarial attacks, neural representations and interpretability, multi-agent game theory, scalable oversight, safety guarantees, and agency and decision theory. The course page notes that topics may change.
- Lectures: materials organized by the course's weekly topics.
- Course notes: cross-topic discussion and reading notes.
- Project ideas: early ideas and evolving project plans.
Use the templates in those folders as a starting point. Suggested filenames follow YYYY-MM-DD-topic.md for dated notes and short descriptive names for project ideas.