This 24-Year-Old Founder Raised $64M to Build World’s First AI Mathematician | Axiom, Carina Hong
In a Nutshell
A 24-year-old founder raised $64M at $300M valuation to build an AI mathematician that combines AI, programming languages, and mathematics to create self-improving reasoning systems. The core mission is to compress centuries-long math-to-application timelines by enabling AI mathematicians to work alongside human researchers, solving complex theoretical problems through deductive logic in Lean. Math serves as both the foundational "sandbox of reality" and a transferable skill that enhances performance across coding and scientific domains, with the goal of making mathematical research more collaborative and less isolating.
These notes were generated by AI and may contain inaccuracies.
Math research is a process of almost a monk praying in the temple day after day, keeping pushing harder and harder. The dopamine hits feel natural and effortless, similar to how one feels looking at progress in competitions where the clock is ticking. The feeling is exhilarating. A hedge fund can afford quant researchers typically paid at $14 million each year starting, but an AI mathematician at $5 each hour enables tackling a market of total trading volume of $8 million that likely has not been deeply studied previously.
We are entering an era of mass intelligence and an AI mathematician is a crucial part of this future.
Karina is founder and CEO of Axiom. She was a mathematician for most of her life, doing math and physics double major at MIT where she worked on research projects that settled some open conjectures. She was a Rhodes scholar at Oxford pursuing a degree in neuroscience and stumbled into AI at the UCL Gatsby Institute, which is the home of DeepMind. She was at Stanford as a Hennessy scholar pursuing a joint JD PhD, then dropped out to start Axiom.
Axiom is building an AI mathematician that is the first model that will eventually evolve to be a self-improving super intelligent reasoner.
To build an AI mathematician, you need three pillars: AI, programming languages, and mathematics. Because of this vision, Axiom assembled a world-class team of experts from each of these three pillars, taking a very interdisciplinary approach. Axiom raised a seed round of $64 million with investors valuing the company at $300 million valuation. Karina is 24 years old.
Karina's background is in combinatorics and number theory. Number theory is about patterns everywhere in sets and numbers, with correspondences that are unexpected. She loves everything number theorist. She finds reading about Gauss's life very inspiring, particularly all the aha moments and long nights when he was pushing for Eureka moments. Number theory is the crown jewel of mathematics, as G.H. Hardy put it. It is quite beautiful and feels timeless compared to other fields of math. It involves symbolic expressions where the symbols look beautiful on sketch paper.
In history, every mathematical tool after its invention has led to terrific breakthroughs not just in fundamental science but also in real world applications. The invention of the abacus led to the bloom of trade and commerce. Integrals and calculus led to mechanics and thermodynamics, which led to the industrial revolution. The Babbage engine or difference engine was a math tool to calculate log tables faster and was the prototype of the computer. Mathematical tools spark the flywheel in real world applications and in turn require more computational tools.
Jevons paradox states that when the price of a tool becomes elastic, unexpected use cases and applications emerge, requiring more tools and making these tools more valuable. By building AI mathematicians at your fingertips, many magnitudes of use cases and markets will be unlocked. Pragmatically, the time span from mass invention to real world application has historically taken centuries, but AI compresses this timeline.
AI mathematicians can work together with applied scientists, which human mathematicians seldom do. AI mathematicians can go into applied fields and solve complex systems that have never been theoretically understood, shortening the century-long time span to much shorter durations.
Karina grew up loving math. Every time you solve a math problem, you get instant reinforcement and a dopamine hit. In elementary school at fourth grade, about 1,000 really bright kids were assembled into 24 classrooms to compete. After each exam, students were ranked, and only those performing outstandingly could go to the next level. It was eye-opening to read the proof of quadratic reciprocity and learn names of French and German mathematicians, motivating her to wonder about visiting French and German research institutes.
The elementary school math Olympiad problems were formulated in real world scenarios, like two trucks going across each other with questions about how long each truck is and how long it takes for them to pass each other. Converting that to mathematical formulation into equations that feel trackable is a magical process, and plugging the solution back into the real world scenario is beautiful. Mathematical thinking is very transferable to other fields.
At one point, Karina was introduced to research math, which involves much more delayed gratification. Research problems are really hard and take a long time to figure out. Unlike being a math Olympiad student where you can solve a dozen problems each day, research math often means a dozen months pass with nothing accomplished when encountering bottlenecks. She wanted to be a better mathematician, changing from a problem solver to a theory builder, owing a lot to her mentors who taught her how to do research, be patient, look around the corners, and look for unexpected connections from other fields to the current problem. This led to fruitful collaboration with Professor Ono and other collaborators.
The feeling of developing new mathematical theories based on the flow of past theories is fascinating. You invent definitions in the natural way, link definitions together to formulate interesting conjectures, prove them elegantly, and understand what natural, interesting, and elegant mean. These are questions about taste. By learning many branches of math, she formed her taste about math. In an era where AI is prevalent and can do a lot, taste becomes quite important, distinguishing between a good scientist and a mediocre one. The team wants to understand the problem of taste and intuition better using modern-day machine learning techniques, which will be a very difficult technical challenge.
At the Ross mathematics program when she was 15, professors taught how to think deeply about simple things. On the first day, they asked if you can prove that zero multiplied by everything is zero. This seemed obvious, but the task was to derive that statement strictly from a set of five axioms such as zero plus everything is that thing itself, and one multiplied by everything is that thing itself. This axiomatic and deductive reasoning was inspiring, teaching rigorous reasoning to reach the final destination.
The name Axiom comes from this inspiration to build out the knowledge graph and expand the frontier of mathematics through deductive logic using the programming language of proofs Lean. Axiom's mission is very short and concise: think very deeply about simple things. An AI mathematician requires various techniques joining together.
Colleague Hugh Leather and his team have been working on applying deep learning to code generation since 2017. Chart has been pioneering the field of AI for math discovery using transformers to solve symbolic integration in 2019, showing it worked better than computer algebra. CTO Shashank Gupta worked at FAIR for many years and at Facebook AI Research pioneered work such as large reinforcement learning models like OpenGo. The team believes solving the hardest problem is the best way to win, and this mission is incredibly attractive to talent.
Karina loves the fast-paced environment of startups, executing with the team, unblocking others, and asking for help. The instant reward signal is similar to childhood experiences solving math problems. Being in the moment and immersed in how research engineering is done on a day-to-day basis is a once-in-a-lifetime experience.
Math is the fundamental of lots of branches of sciences and the sandbox for reality where you can put real world objects into mathematical variables and formulate problems purely theoretically. Understanding math allows generalization to other domains. In machine learning, transferability of math reasoning is striking - a model that does well on math can likely do better in coding. Math concepts are unexpectedly present in many applied scientific fields.
Math is the sandbox of reality where you can understand real world problems theoretically through solving them mathematically. Another meaning is that modern machine learning models gather real world data to experiment and generate new knowledge. Math is a digital version of that - you can solve reasoning in the digital world without relying on real world data. It is the playground for trying out what we understand about reality and the universe.
Math research is a process like a monk praying in the temple day after day, hoping the storm being studied will have a flower grow out of it. The feeling of being stuck can feel depressing because the boundary between work and life blurs, with identity merging with work. This is a struggle many mathematicians face. With AI, the process will hopefully become more enjoyable. The ideal state is having a spec or lemma to prove, where instead of the human banging their head against the wall for days, weeks, or months, the AI mathematician can help prove it while the human mathematician guides the collaboration to the next lemma, making the journey more exhilarating. This is a collaborative endgame rather than replacing humans.
Five or 10 years is hard to imagine, but the team fully believes this is fundamental technology that will turn Axiom into a generational company.
Keep EO in your library
Save the videos and channels worth coming back to, and find them again in one place.
