Iliad Intensive
A four-week, full-time taught course on foundational alignment research for researchers with backgrounds in math, theoretical physics, computer science, or related disciplines.
The Iliad Intensive provides a broad overview of alignment research in areas like deep learning theory, agent foundations, AI interpretability, and more. It runs in-person monthly, alternating between London and the Bay Area.
Apply nowOverview
Duration
Four weeks, with occasional three-week iterations.
Schedule
Five days a week, 10am–6pm.
Locations
London and Berkeley.
Cost & support
No tuition. Participants receive a $5,000 USD travel-and-housing allowance, office space, and lunch and dinner five days a week.
What You’ll Learn
The curriculum is organized into five clusters, each composed of one-day modules. There is more material than fits in a single month, so the exact content varies between iterations.
You can browse the curriculum here.
Alignment
The alignment problem and how to decompose it into subproblems; how frontier labs align models in practice in pretraining, post-training, and deployment; and reward learning theory, including an analysis of various forms of misspecification.
Learning
The mystery of how deep learning works, and components of a theory: singular learning theory, training dynamics, and data attribution.
Abstractions, Representations, and Interpretability
Mechanistic interpretability, computational mechanics and belief-state geometry, natural latents, and an ML engineering introduction.
Agency
Agents with long-term goals and how to model them theoretically: reinforcement learning, idealized agency and AIXI, coherence and expected utility, agent foundations, decision theory, and world models.
Safety Guarantees and their Limits
Debate and its complexity-theoretic foundations, steganography and backdoors, and worst-case interpretability and heuristic arguments.
Who Should Apply?
You should apply if you have a strong mathematical background—typically pursuing or having completed a graduate degree in mathematics, physics, or theoretical computer science—and want to work on foundational AI alignment research.
We also consider research experience, general competence, and motivation for pursuing the program.
Apply nowTalks & Lectures
The Mathematics of Concepts: How AIs See The World
Deep Learning Is Singular - Here’s What That Means
The Dumbest Neural Network Worth Studying
Uncovering the Hidden Biases of Deep Learning
The Five Rules of Rationality Behind Reinforcement Learning
When Bad Goals ALSO Fit Your Training Data
FAQ
What is the Iliad Intensive?
The Iliad Intensive is a four-week, full-time, in-person course in technical AI alignment, run by Iliad. It targets people with strong backgrounds in mathematics, physics, or theoretical computer science, and aims to give you both a coherent map of the alignment research landscape and genuine technical depth.
How is the Intensive different from ARENA or a research fellowship?
The Intensive has similarities to ARENA, but it focuses on foundational AI alignment research rather than alignment research engineering; expect more math and less coding. It is also a taught course, not a research program.
How much coding is involved?
There is a dedicated ML engineering day, and several modules include hands-on coding exercises. Comfort with Python isn’t strictly necessary, but will make these sessions more productive.
Do I need a PhD?
No. The requirement is mathematical maturity, not a specific credential.