Blog

Interpretability borrowed the term model organism from biology and kept the toy. Biological foundation models are the real thing: small enough to trace exhaustively, with ground truth that exists outside the model, and a wet lab at the end of the argument.

A field guide to 109 open research problems spanning evidence standards, model internals, causal circuits, biological discovery, model design, replication, biosecurity, and general mechanistic interpretability.

A framework for translating mechanistic interpretability into concrete discovery outputs: biomarkers, regulatory hypotheses, cell-state taxonomies, drug target prioritization, compact algorithm extraction, artifact detection, and manifold-guided experimental design.

An argument for treating biological foundation models as compressed maps of biological reality, and for using mechanistic interpretability to extract validated knowledge, causal structure, and compact biological algorithms from their internals.