Practical AI Alignment and Interpretability Research Group

The Practical AI Alignment and Interpretability Research Group, also known as the Pr(Ai)²R Group, is a remote, non-profit research organization focused on mechanistic interpretability of AI systems. The group was founded by Atticus Geiger in late 2023, building directly on his doctoral dissertation at Stanford University, which examined the problem of uncovering and inducing interpretable causal structure in deep learning models. The group's central technical focus is causal abstraction, a theoretical framework that provides rigorous foundations for mechanistic interpretability. Mechanistic interpretability aims to explain the computations performed by neural networks in terms of intelligible algorithms that faithfully capture the behavior of otherwise opaque models. The group applies and extends this framework to language models and other deep learning systems. Beyond original research, the group's mandate includes creating open-source course materials on mechanistic interpretability and running mentorship programs to grow the pipeline of researchers working on AI interpretability. This educational and community-building component reflects a theory that the field needs more trained researchers to make progress on AI safety. In October 2024, Open Philanthropy announced a grant of $737,000 over two years to support the group's work under Geiger's leadership. The group's website was hosted at prair.group. Known members include Atticus Geiger (PI), Amir Zur, and Jiuding Sun. Atticus Geiger subsequently joined Goodfire, a commercial AI interpretability company, in September 2025.