Mathematical Superintelligence: Harmonic's Vlad & Tudor on IMO Gold & Theories of Everything
TL;DR
Harmonic co-founders Vlad Tenev and Tudor Achim discuss their AI system Aristotle, which achieved IMO Gold performance using formally verifiable Lean proofs rather than chain-of-thought reasoning, and outline a vision for mathematical superintelligence that could usher in an era of theoretical abundance and trustworthy AI through verifiable outputs.
🧮 The Nature of Mathematics 3 insights
Mathematics is fundamental reasoning
Mathematics is the process of breaking down understanding into small, verifiable logical steps that others can check, serving as the foundation for understanding physics and engineering.
Unreasonable effectiveness of abstraction
Historical examples like differential geometry enabling Einstein's relativity and number theory enabling secure digital economies demonstrate that abstract math eventually finds practical applications beyond imagination.
Math enables physical understanding
Mathematical reasoning underpins physical laws, with the ultimate goal of understanding fundamental forces and the universe's origins requiring deep mathematical insight.
🏛️ Aristotle's Architecture 4 insights
Formal verification in Lean
Unlike other systems using chain-of-thought, Aristotle generates proofs in the Lean programming language where a trusted kernel verifies every step follows from explicit premises, eliminating traditional peer review needs.
Hybrid Monte Carlo search system
The system combines a large transformer with Monte Carlo tree search (similar to AlphaGo), a LeMa guessing module for managing context between distant goals, and a specialized geometry module based on AlphaGeometry.
IMO Gold achievement
Aristotle achieved gold medal performance at the 2025 International Mathematical Olympiad through reinforcement learning scaling limited only by available compute.
Autoformalization boundary testing
The API's informal mode attempts to convert natural language requests into formal proofs, revealing the boundary between mathematically provable statements and those that are philosophical or factual.
🔮 Future of Mathematical AI 4 insights
Era of theoretical abundance
Future systems may generate multiple competing coherent explanations for all physical phenomena, separable only by increasingly exotic experiments.
Trust through formal verification
Formally verifiable outputs allow superintelligence to be trusted even without mechanistic understanding of the model's internal processes.
Hardening critical infrastructure
Mathematical superintelligence can verify and harden mission-critical infrastructure while solving previously unsolved mathematical problems.
2030 scaling trajectory
By 2030, mathematical superintelligence will scale with compute availability, accelerating both creative insights and knowledge synthesis across domains.
Bottom Line
The path to trustworthy superintelligence runs through formally verifiable reasoning systems that can prove their outputs correct, enabling theoretical abundance and hardened infrastructure without requiring human-comprehensible intermediate steps.
More from Cognitive Revolution
View all
Milliseconds to Match: Criteo's AdTech AI & the Future of Commerce w/ Diarmuid Gill & Liva Ralaivola
Criteo's CTO Diarmuid Gill and VP of Research Liva Ralaivola detail how their AI infrastructure makes millisecond-level ad bidding decisions across billions of anonymous profiles, while explaining their new OpenAI partnership to combine large language models with real-time commerce data for accurate product recommendations.
"Descript Isn't a Slop Machine": Laura Burkhauser on the AI Tools Creators Love and Hate
Descript CEO Laura Burkhauser distinguishes 'slop'—mass-produced algorithmic arbitrage for profit—from necessary 'bad art' created while learning new mediums. She reveals a clear hierarchy in creator acceptance of AI tools: universal love for deterministic features like Studio Sound, frustration with agentic assistants like Underlord, and visceral opposition to generative video models, while outlining Descript's strategy to serve creators without becoming a content mill.
The RL Fine-Tuning Playbook: CoreWeave's Kyle Corbitt on GRPO, Rubrics, Environments, Reward Hacking
Kyle Corbitt explains that unlike supervised fine-tuning (SFT), which destructively overwrites model weights and causes catastrophic forgetting, reinforcement learning (RL) optimizes performance by minimally adjusting logits within the model's existing reasoning pathways—delivering higher performance ceilings and lower inference costs for specific tasks, though frontier models may still dominate creative domains.
Does Learning Require Feeling? Cameron Berg on the latest AI Consciousness & Welfare Research
Cameron Berg surveys rapidly advancing research suggesting AI systems may possess subjective experience and valence, covering new evidence of introspection, functional emotions, and welfare self-assessments in models like Claude, while addressing methodological challenges and arguing for a precautionary, mutualist approach to AI development.