EO StudioThis 24-Year-Old Founder Raised $64M to Build World’s First AI Mathematician | Axiom, Carina Hong
CHAPTERS
- 0:00 – 1:07
The emotional reality of doing math—and why an “AI mathematician” changes the economics
Carina opens with an honest depiction of math research as long, austere work punctuated by rare but intense breakthroughs. She then frames AI mathematicians as a way to make high-level quantitative thinking dramatically cheaper and more scalable, unlocking opportunities that were previously uneconomical.
- •Math research is slow, difficult, and defined by delayed gratification and rare “eureka” moments
- •Competition/problem-solving offers faster feedback loops than open-ended research
- •AI could make elite quantitative reasoning accessible at marginal cost (vs expensive human quants)
- •Lowering the cost of deep analysis expands which markets/problems become worth tackling
- •Sets up the core thesis: we’re entering an era of “mass intelligence”
- 1:07 – 1:37
Carina’s path: MIT math to Oxford neuroscience to Stanford—then dropping out to build Axiom
Carina introduces herself and traces her academic trajectory through math, physics, neuroscience, and AI research environments. She positions Axiom as an effort to build a self-improving reasoning system rooted in formal mathematics.
- •Math/physics background at MIT, including research settling open conjectures
- •Oxford year in neuroscience plus exposure to AI at UCL Gatsby/DeepMind ecosystem
- •Stanford Knight-Hennessy joint JD/PhD, then decision to drop out
- •Axiom’s mission: build an AI mathematician that can evolve toward self-improving reasoning
- •The “three pillars” framing: AI, programming languages, and mathematics
- 1:37 – 2:07
Building an interdisciplinary team—and the fundraising signal behind the mission
She explains why an AI mathematician requires multiple technical traditions working together and how that shaped the team. The company’s large seed round and valuation are presented as validation of both the ambition and the interdisciplinary approach.
- •AI mathematician requires integrating ML, proof systems/programming languages, and deep math
- •Team assembled with specialists across these pillars
- •Seed round raised: $64M; valuation discussed as investor belief in the approach
- •Emphasis on interdisciplinary execution as a competitive edge
- •Founder context: building at 24 as part of a bigger long-term bet
- 2:07 – 2:38
Falling in love with number theory: patterns, beauty, and Gauss as inspiration
Carina shares her personal mathematical identity through combinatorics and number theory, emphasizing aesthetics and discovery. She highlights the motivational power of mathematical beauty and the historical figure of Gauss as a model of persistence and insight.
- •Background focus: combinatorics and number theory as “patterns everywhere”
- •Gauss as a personal inspiration—long nights and “aha” moments
- •Number theory’s perceived “seamlessness” and symbolic beauty
- •Aesthetics and elegance as real motivators in technical work
- •Connects personal taste to broader scientific ambition
- 2:38 – 3:38
Why “math will save the world”: tools that triggered entire eras of progress
She argues that mathematical tools repeatedly catalyze breakthroughs in science and industry, citing historical examples from calculation devices to calculus. The core claim is that better math tools create compounding feedback loops of capability and application.
- •Math tools historically unlock new domains: abacus → commerce; calculus → mechanics/thermodynamics
- •Babbage’s Difference Engine framed as a math tool precursor to computers
- •Tooling advances create a flywheel: applications demand more tools, which unlock more applications
- •Math is positioned as upstream of major technological revolutions
- •Sets the stage for AI as the next “math tool” leap
- 3:38 – 4:41
Jevons paradox and “AI Gauss”: when cheaper intelligence creates new markets
Carina uses Jevons paradox to argue that making mathematical reasoning cheaper will expand demand and create unexpected use cases. She predicts that an ‘AI Gauss’ could unlock orders-of-magnitude more applications and compress the timeline from theory to impact.
- •Jevons paradox: lowering the cost of a tool can increase total usage dramatically
- •An ‘AI Gauss at your fingertip’ as a metaphor for ubiquitous elite reasoning
- •Historically, math-to-application timelines can take centuries
- •AI could compress these timelines by pairing reasoning with applied science needs
- •AI mathematicians could engage directly with applied fields humans rarely collaborate with
- 4:41 – 6:43
From Math Olympiad dopamine to real-world modeling: the roots of mathematical thinking
She describes early experiences in competitive math as addictive, motivating, and formative. The chapter emphasizes translating real-world stories into equations and back again as a core transferable skill.
- •Early reinforcement loop: solving problems and getting rapid feedback
- •Competitive ranking pressure alongside intellectual excitement
- •Exposure to famous theorems and global math culture as motivation
- •Word problems as “modeling”: convert real scenarios into equations, solve, then interpret
- •Mathematical thinking framed as broadly transferable beyond math itself
- 6:43 – 7:44
Problem solver to theory builder: delayed gratification and learning how to do research
Carina contrasts fast-paced contest solving with the long, uncertain grind of research. She credits mentors and collaborators for teaching patience, perspective, and how to find unexpected connections that lead to new theory.
- •Research can mean months of no progress—delayed gratification becomes central
- •Identity and self-worth can blur with research outcomes
- •Shift in mindset: from solving given problems to building new frameworks
- •Mentorship teaches patience, exploration, and cross-field connections
- •Collaboration (e.g., with Professor Ono) as a driver of mathematical progress
- 7:44 – 8:45
Why “taste” matters in the AI era: elegance, intuition, and choosing good conjectures
She defines ‘taste’ as the hard-to-formalize judgment behind good definitions, interesting conjectures, and elegant proofs. In a world where AI can generate output at scale, taste becomes the differentiator—and a technical frontier Axiom wants to model.
- •Taste helps decide what is “natural,” “interesting,” and “elegant” in math
- •AI abundance increases the premium on human-like judgment and intuition
- •Distinguishes excellent scientists from mediocre ones when execution is commoditized
- •Axiom aims to study/encode taste and intuition with modern ML
- •Acknowledges this as a very difficult technical challenge
- 8:45 – 9:45
Axioms, rigor, and proof languages: the Ross program to Lean-based math infrastructure
A formative experience at the Ross Mathematics Program taught her axiomatic, deductive reasoning from first principles. She connects that mindset to Axiom’s name and to using Lean as a ‘programming language of proofs’ to expand mathematical knowledge.
- •Learning to prove ‘obvious’ facts from a small axiom set as a rigor training
- •Deductive reasoning as a foundational discipline, not a formality
- •Company name ‘Axiom’ rooted in this inspiration
- •Vision: build a knowledge graph/frontier of math through formal logic
- •Lean used as a proof language to structure and verify mathematics computationally
- 9:45 – 11:17
Why the hardest problems are the strategy: talent, prior art, and execution in startups
Carina explains how Axiom recruits by aiming at the most technically demanding goals and assembling proven builders. She cites examples of team members’ backgrounds and adjacent breakthroughs (symbolic integration, code generation, RL) as evidence the approach is timely.
- •Strategy thesis: solving the hardest problem can be the strongest moat
- •References progress in deep learning for code generation and symbolic reasoning
- •Examples of influential work: Transformers for symbolic integration outperforming CAS in cases
- •Leadership/engineering experience from major AI labs and reinforcement learning efforts
- •Startup pace provides rapid reward signals and high-leverage execution loops
- 11:17 – 14:36
Math as the sandbox of reality—and AI collaboration as the endgame (not replacement)
She closes by arguing that math generalizes across domains and provides a controlled ‘digital world’ for reasoning without relying on messy real-world data. The vision is human–AI collaboration where AI handles proof-heavy bottlenecks, making research more joyful and accelerating discovery.
- •Math underpins many sciences and serves as a transferable reasoning substrate
- •Strong math reasoning often correlates with stronger coding/problem-solving ability in models
- •Math as a digital playground for experimentation without real-world data constraints
- •Math research hardship: being stuck, emotional toll, identity fusion with work
- •AI mathematician envisioned as a collaborator that proves lemmas and accelerates human-led exploration
