Scaling informal language models alone is an insufficient path to achieving reliable, general intelligence in mathematics; a grounding in formal, verifiable languages is essential.
Formal verification represents a severe economic and temporal bottleneck in critical industries like hardware and cloud computing, creating a large commercial opportunity for AI-driven solutions.
The ultimate goal is to create a self-improving, superhuman AI reasoner, with mathematics serving as the ideal initial domain to develop and prove this 'generation and verification' loop.
AI will augment, not replace, the highest level of human mathematical genius. It will function as a 'diligent grad student,' allowing top mathematicians to work at a higher level of abstraction by formalizing and proving their intuitions.
Objective, high-profile achievements against established human benchmarks like the Putnam exam and unsolved mathematical conjectures are the most effective way to demonstrate progress and capability in AI reasoning.
2019
A paper co-authored by future Axiom technical staff member Francois Charton demonstrates that Transformer models can outperform traditional computer algebra systems, signaling the potential of neural networks in symbolic math.
Pre-December 2023
Carina Hong reports that Axiom's AI system scored 9 out of 12 on the Putnam exam, a result that would have won the competition in the previous year against 4,000 human participants.
September 2023
Version 4 of the Lean programming language is released, which Hong notes is the first version to support industry-scale engineering, a key enabler for Axiom's technology.
December 2023
Hong announces a significant milestone: Axiom Prover achieved a perfect score of 120 out of 120 on the Putnam exam, becoming only the sixth perfect score in the exam's history.
Post-December 2023
Following its Putnam success, Hong reports that Axiom Prover solved four open mathematical conjectures from various universities and that Professor Ken Ono stepped down as Vice Provost of UVA to join the company.
▶Formal Verification as the Path to Superhuman AI
Hong argues that true AI reasoning, especially in mathematics, cannot be achieved by simply scaling informal language models. She posits that grounding AI in formal languages like Lean is essential to overcome the limitations of probabilistic systems and create verifiable, trustworthy, and ultimately superhuman intelligence.
This positions Axiom in the AI safety and reliability space, offering a potentially more defensible and enterprise-ready technology compared to general-purpose LLMs, which could be a key differentiator for investors concerned about the hallucination problem.
▶Benchmarking Against the Pinnacle of Human IntellectApr 2026
Axiom's strategy involves demonstrating its AI's capabilities by competing in arenas traditionally dominated by elite human minds. By achieving a perfect score on the Putnam exam and solving open mathematical conjectures, Hong creates clear, dramatic proof points of the system's superiority.
This high-profile benchmarking is a powerful marketing and narrative-building tool that makes the abstract concept of 'AI reasoning' tangible and impressive to investors, customers, and potential talent.
▶Bridging Elite Academia and Commercial Application
The company actively recruits top-tier academic talent, such as number theorist Ken Ono and Transformer pioneer Francois Charton. Hong highlights how these experts are drawn to Axiom by its ability to solve problems they couldn't, effectively turning academic breakthroughs into a commercial engine for solving industrial verification challenges.
Axiom's ability to attract and retain world-class academics serves as a strong validation of its technical approach and is a leading indicator of its potential to maintain a competitive edge.
▶The Economics of VerificationApr 2026
Hong repeatedly frames the market opportunity by quantifying the immense cost of verification in time, money, and personnel. She points to multi-year verification cycles for chips and the massive manual effort by AWS as evidence of a severe economic bottleneck that Axiom's technology is designed to solve.
By focusing on the tangible business pain point of verification lag, Hong translates a complex technological solution into a clear value proposition for enterprise customers, focusing on ROI rather than just technical novelty.