How to make System 1 LLMs show System 2 competencies?
Summary
This post discusses how System 1 LLMs might exhibit System 2-level competencies through formalization work in mathematics. It weighs the benefits and risks of large-scale AI-assisted formalization, highlights exemplars like Mathlib, Tau Ceti, Formal Frontier, and Palomar, and argues for high-quality, community-governed libraries over indiscriminate automation. The piece outlines a roadmap for scaling formal mathematics with AI while focusing on alignment, provenance, and practical utility.