Concrete Problems
Here are sources where one can quickly find projects to sink their teeth into! Many of these are difficult and probably require further reading before one can make progress, but they are hopefully easy to understand and get oriented around.
AISI Alignment Project: Computational Complexity Theory
This research area collects open problems that use computational complexity theory to study scalable oversight. It focuses on formal guarantees, impossibility results, and error bounds for methods such as debate and prover-verifier games.
Iliad Ecosystem
This page introduces organizations in the Iliad ecosystem, including groups working on interpretability, theoretical models, agency, and alignment. It explains what each organization studies and how Iliad supports or incubates technical research bets.
MAIS Open Problems
This directory collects mathematical open problems related to AI safety, with each problem stored in its own short document. It is useful for browsing concrete questions and finding a narrowly scoped problem to investigate.
Timaeus Project Ideas
This page collects project ideas for developmental interpretability and singular learning theory. It includes starter notebooks, active projects, and unstarted research questions with practical guidance about difficulty and status.