Harmonic, The Robinhood Ceo’s Ai Math Startup, Launches An Ai Chatbot App

Harmonic, an AI startup co-founded by Robinhood CEO Vlad Tenev, announced Monday nan beta motorboat of an iOS and Android chatbot app wherever users tin entree its AI model, Aristotle.

With this launch, nan institution intends to broaden entree to Aristotle, which Harmonic claims to connection “hallucination-free” answers for questions involving mathematical reasoning — a bold claims fixed nan reliability problems of today’s AI models. Harmonic is focused connected creating “mathematical superintelligence” aliases MSI; nan startup yet wants to thief users pinch each fields that trust connected math, including physics, statistics, and machine science.

“[Aristotle] is nan first merchandise disposable to group that does reasoning and formally verifies nan output,” said Harmonic CEO and co-founder Tudor Achim successful an question and reply pinch TechCrunch. “Within nan domains that Aristotle supports, which are quantitative reasoning domains, we really do guarantee that there’s nary hallucinations.”

Eventually, Harmonic says it besides plans to merchandise an API to fto enterprises entree Aristotle, arsenic good arsenic a web app for consumers.

The beta motorboat comes conscionable a fewer weeks since Harmonic raised $100 cardinal successful a Series B information led by Kleiner Perkins astatine an $875 cardinal valuation. Achim claims that Harmonic is “advancing very quickly along” its way to achieving MSI, and that investors believed that was a adjacent valuation fixed nan scope of his startup’s ambition.

Achim says that Harmonic achieves its hyper-accurate solutions by having Aristotle nutrient responses successful nan open-source programming connection Lean. Before Aristotle gives an reply to users, he says nan exemplary double checks that nan solution is correct done an algorithmic process that doesn’t impact AI. Harmonic’s CEO notes that akin exertion is utilized to verify outputs successful precocious stakes fields specified arsenic aesculapian devices and aviation.

Even successful a constrictive domain, achieving hallucination-free capacity from an AI exemplary is an incredibly difficult task. Studies person recovered that even starring AI models hallucinate a lot, and nan problem doesn’t look to beryllium getting better. OpenAI’s latest AI reasoning models hallucinate much than its older ones.

Harmonic says Aristotle achieved golden badge capacity connected nan 2025 International Math Olympiad done a general trial (meaning nan problems were translated into a machine‑readable format.) Google and OpenAI besides developed AI models that achieved golden badge capacity connected this year’s IMO, but done informal tests taken successful earthy language.