Axiom says AxiomProver agent swarm completed Lean formalization of the Classification of Finite Simple Groups theorem in 24 days

Damn! The greatest mathematical proof in the history of human science has just been thoroughly verified by AI!! A 15,000-page magnum opus written over decades by more than 100 mathematicians has been reconstructed into 50 million lines of Lean code in 24 days by the AxiomProver agent swarm. The era in which computers take over the verification of truth may have officially entered the stage of industrialized mass production today! It feels like from now on we won't need to use hundreds of lines of Python scripts to judge AI's programming ability anymore. Axiom has just publicly unveiled a god-tier engineering feat by multiple agents: In 24 days and 50 million lines of code, the AI agent swarm directly conquered the largest mathematical theorem that humanity spent half a century and 15,000 pages of paper to pile up, and conveniently fixed all the loopholes in the original work.

If previously various large models writing code were still at the toy stage of a few dozen lines of script, the achievement Axiom has just announced is the first god-tier answer sheet delivered by multi-agent collaboration in the history of human engineering. Using an AI agent swarm called AxiomProver, they wrote about 50 million lines of flawlessly coherent Lean code in just 24 days. What they accomplished completely stunned the mathematics community and the entire software engineering community: They completely formally verified the largest and most complex "Classification of Finite Simple Groups" theorem in the history of modern mathematics. To help you understand how exaggerated this number is, you must first understand the weight of this theorem in human civilization. This mathematical monument known as the "Enormous Theorem" originally had its proof scattered across hundreds of academic papers, spanning a full 15,000 pages, pieced together by more than 100 top mathematicians worldwide at the cost of their physical and mental efforts over nearly half a century. Because it is so long and the derivations are so dense, over the past decades, all of humanity could not find even one person who could pat their chest and guarantee that there was not a single fatal logical flaw in these 15,000 pages of papers. Yet Axiom did not take the dead end of humans fighting alone, but instead turned this most bloated proof in human history into a multi-agent assembly-line assault: They forcibly translated 15,000 pages of natural language into formal mathematical code. In Lean, an extremely rigorous system where the compiler will instantly report an error and go on strike if even one line is logically inconsistent, this swarm of agents not only chewed through 50 million lines of code line by line, but even in the process of running the code, they turned around and caught logical loopholes buried for decades in the original works of the old human mathematicians, and wrote patches themselves to fix them all. Many people see this news and think it is just a closed-door celebration within the circle of mathematicians.

But if you look at it from the perspective of modern software engineering, this is a violent baptism of all human civilization in the capitalization of knowledge: First, the engineering ceiling of multiple agents chained together to write code has been directly raised to the tens-of-millions-of-lines level. It proves that as long as the boundaries and the syntax checker (Compiler) are hardcore enough, hundreds or thousands of agents in a high-concurrency collaboration lasting as long as 24 days will not only fail to collapse or go off track, but can instead complete ultra-large-scale code refactoring at a speed unattainable by humans. Second, the entire knowledge system that humanity has built over the past thousands of years on paper, intuition, and natural language finally has the ultimate tool to be "fully computerized." Previously we always felt that AI writing long code is prone to hallucination and forgetting. Axiom has taught the whole world a harsh lesson with these 50 million lines of code: When the reasoning ability of large models is connected to an absolutely rational compiler sandbox, agents are no longer chat assistants full of hot air, but a special engineering force capable of reconstructing the entire edifice of human science line by line.