Good empirical safety research is heavily bottlenecked on thinking more clearly. This post shares a few frames and tips I found useful, illustrated with concrete examples. I spent the better part of 2026 in empirical AI safety research apprenticeships. I was mentored by Alex Turner and Alex Cloud as part...
(see full author list at the end) About a year ago, METR showed that the length of tasks frontier models can reliably complete doubles every few months. A related safety-relevant question is this: what length of tasks can models complete without any chain of thought (CoT)? We investigate in our...
TL;DR: A couple months ago, we (Jo and Noah) wrote the first Wikipedia article on Mechanistic Interpretability. It was oddly missing despite Mech Interp’s visibility in alignment circles. We think Wikipedia is a top-of-funnel resource for journalists, policy staffers, and curious students, so filling that gap is cheap field-building. Seeing...
TL;DR: undergrad AI safety conference in Chicago on world-modelling & thinking about the cruxes of the future of TAI. Takes place March 29-30, apply by Feb 20. The UChicago AI Safety Team at XLab is excited to announce the Chicago Symposium on Transformative AI, an undergraduate AI Safety conference taking...
The University of Chicago Existential Risk Laboratory (XLab) is now accepting applications for the 2025 Summer Research Fellowship! We invite motivated undergraduate and graduate students interested in producing impactful, solution-oriented research on emerging threats that imperil global security (such as those from artificial intelligence, natural and engineered pathogens, nuclear weapons,...
Summary: Reflections and practice notes on a metacognitive technique aimed at refining the process of thinking, rather than the thoughts themselves. Epistemic Status: Experimental and observational, based on personal practice and reflections over a brief period. Introduction While doing a simple math problem, I realized that my faculty of thinking...