Mathematical Superintelligence: Harmonic's Vlad Tenev & Tudor Achim on IMO Gold & Theories of Everything - Cognitive Revolution Recap
Podcast: Cognitive Revolution
Published: 2026-02-18
Duration: 1 hr 31 min
Guests: Vlad Tenev, Tudor Achim
Summary
Vlad Tenev and Tudor Achim discuss the creation of Aristotle, an AI system that achieves gold medal performance at the International Mathematical Olympiad by using Lean for formally verified proofs. They explore how such systems could transform software engineering, mathematical practice, and lead to trustworthy AI by 2030.
What Happened
Vlad Tenev and Tudor Achim, co-founders of Harmonic, share their journey in developing Aristotle, an AI system that achieved gold medal performance at the 2025 International Mathematical Olympiad. They explain how Aristotle leverages the Lean programming language to generate candidate proofs, ensuring each step is formally verified. Aristotle's architecture is built on a large transformer model, Monte Carlo tree search strategy, lemma guessing, and a specialized geometry module.
Lean, a dependently typed programming language, plays a central role, enabling the expression of complex properties and logical concepts. This language is changing mathematical practice by fostering digital collaboration and reducing peer review needs. It represents a significant shift towards formal verification in mathematics, aiming to make reasoning more reliable and collaborative.
The discussion highlights how Lean's kernel checks each move against a concise rule book, ensuring accuracy and trustworthiness. The Mathlib project, mentioned as the largest digital repository of mathematical knowledge, supports this effort by providing a vast resource for Lean users. Through Mathlib, complex proofs can be formalized and shared, contributing to a new era of mathematical practice.
Aristotle's capabilities extend beyond theoretical math problems. It is utilized for reasoning in software verification, such as cryptography implementations, highlighting its potential to power critical real-world applications. By expanding its boundary to formalize more complex real-world events, Aristotle represents a leap towards superintelligent systems.
Harmonic's development process involved reinforcement learning from verifiable rewards and synthetic data generation, ensuring Aristotle's outputs are both correct and usable. Community input through API submissions helps direct Aristotle's focus, underlining the collaborative nature of this AI system.
The episode concludes with a vision for 2030, where theoretical explanations for everything might be accessible, akin to historical scientific breakthroughs. As AI capabilities grow, formal reasoning is set to become mainstream, accelerating scientific progress and potentially reshaping our understanding of the universe.
Key Insights
- Aristotle, an AI developed by Harmonic, achieved a gold medal at the 2025 International Mathematical Olympiad by using Lean, a dependently typed programming language, to formally verify each step of its proofs. This ensures a level of accuracy and trust that traditional mathematical approaches often lack.
- Lean's kernel functions like a strict referee in a sports game, checking each mathematical move against a concise rule book. This meticulous process reduces the need for peer review and facilitates digital collaboration, transforming how mathematicians can work together.
- The Mathlib project acts as a vast digital library, enabling Lean users to formalize and share complex proofs. This repository not only supports Aristotle's mathematical capabilities but also signals a new era of collaborative and reliable mathematical practice.
- Aristotle's success in theoretical math extends to practical applications like cryptography. By using reinforcement learning from verifiable rewards and synthetic data, Aristotle is positioned to tackle real-world problems, pushing AI towards superintelligent systems.
Key Questions Answered
What is Harmonic's Aristotle AI system?
Aristotle is an AI system developed by Harmonic that achieved gold medal performance at the 2025 International Mathematical Olympiad. It uses the Lean programming language to formalize and verify mathematical proofs with high accuracy.
How does Lean programming language impact mathematics?
Lean is changing mathematical practice by enabling digital collaboration and reducing the need for peer review, fostering a more reliable and accessible approach to formal verification and proof development.
What role does Mathlib play in mathematical reasoning?
Mathlib is the largest digital repository of mathematical knowledge, providing resources for Lean users to develop, formalize, and share complex mathematical proofs, thus advancing the field of mathematics.