We are releasing the course materials of the Iliad Intensive, a new month-long, full-time AI alignment course that runs in-person every second month. The course targets students with strong backgrounds in mathematics, physics, or theoretical computer science, and the materials reflect that: they include mathematical exercises with solutions, self-contained lecture notes on topics like singular learning theory and data attribution, and coding problems, at a depth that is unmatched for many of the topics we cover.
Around 20 contributors were involved in developing these materials for the April 2026 cohort of the Iliad Intensive. By sharing the materials, we hope to:
- create more common knowledge about what the Iliad Intensive is;
- invite feedback on the materials;
- and allow others to learn via independent study.
We are developing the materials further and we will add, remove, and modify modules going forward to improve and expand the course over time.
Structure
The Iliad Intensive is structured into clusters — loose collections of related topics — which decompose into modules taught within one day. Each module consists of learning outcomes, prerequisites, and the content itself, including a fast-track and pointers for how to learn more. Some modules include a teaching guide that explains how the content was taught during the April Intensive.
- 0 Prerequisites. This module collects what is useful to know before engaging with the materials. It points to writings that inform a background worldview — why AI matters, and safety risks — and then lists the technical prerequisites: deep learning, linear algebra, calculus, probability & statistics, information theory, and some theoretical computer science. General mathematical maturity is also very valuable.
- Cluster A — Alignment. This cluster is on AI alignment, the problem of how to align AI systems or collections of such systems with a vision for how they should behave.
- Cluster B — Learning. The most powerful modern AI systems are based on deep learning, and it is increasingly likely that the first AGI systems will be too. Yet we lack a satisfying theoretical understanding of why deep learning works at all, let alone how to ensure its safety properties hold in new settings. This cluster asks what a rigorous understanding of deep learning would look like and why it matters for safety. We begin with the theoretical foundations of learning in general and the specific empirical mysteries of deep learning, then spend three days on concrete research directions — singular learning theory, training dynamics, and data attribution — that each attack a piece of the puzzle.
- Cluster C — Abstractions, Representations, and Interpretability. This cluster studies the internal representations and mechanisms of cognitive systems: how can we reverse-engineer the features and circuits of trained neural networks (module on mechanistic interpretability), what is the structure underlying optimal prediction in context, and do transformers recover it (module on computational mechanics), and how can we define a notion of “natural” abstractions on which different agents modelling the same world converge (module on abstractions and natural latents). This cluster also contains a module on practical ML engineering foundation.
- Cluster D — Agency. The alignment problem is believed to be particularly difficult if AI systems act as agents with long-term goals in an unbounded environment, thus displaying agency, which we study from a largely theoretical perspective in this cluster.
- Cluster E — Safety Guarantees and their Limits. This cluster discusses debate, an approach to directly align AI systems with human overseers and a proposal for how its safety could be established, and several limits to ensuring the transparency and safety of AI systems in general.
Feedback
Please provide feedback on the materials in the comments of the LessWrong post, or as an email to [email protected].