AI is getting so good at science and math that people are now openly saying that it’s better than them at their own specializations.
Robinhood CEO Vlad Tenev, who was once pursuing a PhD in Math before dropping out, says that their AI system was already better than him at solving math problems. “The (mathematical superintelligence lab) is going well,” he said on the Cheeky Pint podcast. “Those guys have been cracking,” he added.
“You’ve cracked superintelligence?” John Collison asked him. “Not yet,” Tenev replied. “But we’re getting there though. I think we’ve cracked *my* intelligence at solving math problems. And I was a math PhD student. I went to UCLA to do a pure Maths PhD to study with Terrance Tao,” Tenev said.
“When I was in high school, I got to AIME level. It’s like two rungs below the International Math Olympiad. So in the past year, we’ve exceeded my intelligence by considerable margin. So think we’re on the way — probably another 10x and, and we’ll be at something that you can consider superintelligence,” Tenev added.
AIME is the American Invitational Mathematics Examination, a prestigious math competition for high school students. Tenev said that he was competing at AIME in high school, and was enrolled in a PhD with legendary mathematician Terrance Tao, who is currently among the best-known mathematicians in the world. As such, Tenev has serious mathematical credentials, but says his company has already built a mathematical AI that’s better than him at math.
“It’s a for-profit endeavor,” Tenev said about the new Math effort, named Harmonic. “We want to get really, really good at solving math problems. And the unique spin on it is that we’re using formal mathematics. So we use a language called Lean — we translate natural language math into Lean, and what that gives you is verification. And I think verification has been a big problem outside of the AI space,” he added.
Vlad Tenev is best known for co-founding stock market trading platform Robinhood, but in 2023, he’d started a new company named Harmonic. Harmonic says that it is “building mathematical superintelligence”. Earlier this month, it had announced that their advanced formal reasoning model, named Aristotle, had achieved Gold Medal-level performance at the 2025 International Mathematical Olympiad by generating verifiably correct solutions to 5 of 6 problems. Google DeepMind had officially won a gold medal at the Olympiad, and OpenAI too had said that its model had delivered a similar result.
“Every solution Aristotle generates is formally verified down to foundational axioms using the Lean4 proof assistant, providing a machine-checkable guarantee of correctness. This circumvents the need for human mathematicians to check the accuracy of our system’s outputs,” Harmonic said about its model.
And AI systems getting rapidly better at math could well be a crucial step towards creating superintelligence. Math is considered to be the purest of all the sciences, and most other fields derive downstream from it. If AI can be good at math — and at least 3 AI models now seem to have won golds at the International Math Olympiad — it could be a big move towards automating scientific research and putting humanity on the road to superintelligence.