Axiom Releases a 16 Million Line Formalization of the Stacks Project Built With AI Coding
Most think vibecoding at scale is slop. Here鈥檚 a counterexample. Today we release a 16M-line formalization of the entire Stacks Project, critical infra underlying modern math, with optimized code quality and its highly interconnected structure preserved. A mathematical city 馃寙
Context
A full machine checked formalization of a core math library shows AI coding can tackle huge, highly linked scientific corpora, not only greenfield apps.
Axiom (@axiommathai) works on AI for mathematics. The Stacks Project is a large open encyclopedia of algebraic geometry and related math, maintained as a collaborative LaTeX corpus. Formalization means rewriting statements and proofs in a proof assistant so a computer can check them.
Deep dive
Axiom (@axiommathai) posted on Nov 4 that it is releasing a 16 million line formalization of the entire Stacks Project, calling the corpus critical infrastructure for modern math and saying code quality and the project's highly interconnected structure were preserved.
The company framed the release as a counterexample to the idea that vibecoding at scale is only slop. The post itself does not name the proof assistant, the exact method, or how much was human reviewed versus model generated, so those details need the linked materials.
A complete formalization of Stacks would be a major milestone for AI assisted mathematics. The project is dense, cross linked and written for human mathematicians, so preserving structure while producing machine checked code is harder than generating isolated lemmas. At about 5.9 million views, the post was one of the day's most watched AI research items.
The claim should still be read carefully. Line count alone does not prove every theorem is checked or that the formalization is usable by outside mathematicians. Independent inspection of coverage, proof assistant choice and remaining gaps will decide how much this moves formal math practice.
The release sits beside other Nov 4 research posts that test how far models go in science, including biology benchmarks and lab work with Claude.
What to expect next
Watch for a public repo, proof assistant details and coverage stats from Axiom or outside mathematicians. If the formalization loads and checks cleanly, other encyclopedic math projects may try the same approach. Gaps or broken links would temper the vibecoding claim. Tool builders may also study how Axiom kept interconnections intact at this scale.
Independent proof checking and a clear map of what remains informal will decide whether this becomes a default reference for formal algebraic geometry work.