Harmonic, an AI startup co-founded by Robinhood CEO Vlad Tenev, introduced Monday the beta launch of an iOS and Android chatbot app the place customers can entry its AI mannequin, Aristotle.
With this launch, the corporate goals to broaden entry to Aristotle, which Harmonic claims to supply “hallucination-free” solutions for questions involving mathematical reasoning — a daring declare given the reliability issues of as we speak’s AI fashions. Harmonic is targeted on creating “mathematical superintelligence” or MSI; the startup ultimately desires to assist customers with all fields that depend on math, together with physics, statistics, and laptop science.
“[Aristotle] is the primary product obtainable to people who does reasoning and formally verifies the output,” mentioned Harmonic CEO and co-founder Tudor Achim in an interview with iinfoai. “Inside the domains that Aristotle helps, that are quantitative reasoning domains, we truly do assure that there’s no hallucinations.”
Ultimately, Harmonic says it additionally plans to launch an API to let enterprises entry Aristotle, in addition to an internet app for customers.
The beta launch comes only a few weeks since Harmonic raised $100 million in a Collection B spherical led by Kleiner Perkins at an $875 million valuation. Achim claims that Harmonic is “advancing very quickly alongside” its path to attaining MSI and that traders believed that was a good valuation given the scope of his startup’s ambition.
A number of main tech corporations are centered on coaching their AI fashions to resolve math issues. AI that may do math is efficacious in its personal proper, however math can be thought of a uniquely verifiable area that calls for core reasoning expertise. Techniques that develop these capabilities could show helpful in different domains as nicely.
Achim says that Harmonic achieves its hyper-accurate options by having Aristotle produce responses within the open supply programming language Lean. Earlier than Aristotle offers a solution to customers, he says the mannequin double-checks that the answer is appropriate via an algorithmic course of that doesn’t contain AI. Harmonic’s CEO notes that related know-how is used to confirm outputs in high-stakes fields similar to medical gadgets and aviation.
Techcrunch occasion
San Francisco
|
October 27-29, 2025
Even in a slim area, attaining hallucination-free efficiency from an AI mannequin is an extremely troublesome job. Research have discovered that even main AI fashions hallucinate so much, and the issue doesn’t look like getting higher. OpenAI’s newest AI reasoning fashions hallucinate greater than its older ones.
Harmonic says Aristotle achieved gold medal efficiency on the 2025 Worldwide Math Olympiad (IMO) via a proper check (which means the issues have been translated right into a machine‑readable format). Google and OpenAI additionally developed AI fashions that achieved gold medal efficiency on this 12 months’s IMO, however via casual assessments taken in pure language.
