That's Amnon Shashua, founder and CEO of Mobileye. He eyes complicated problems that he thinks AI should solve or better yet, more reliable AI itself. In the side of building and running his self-driving car technology company, which he took public then sold to Intel, then spun out again-he has been hatching a number of ideas.
Now, one of these is fundraising and really picking up pace.
One Zero, a fintech trying to use AI in retail banking services, is raising at least $100 million, sources familiar with the company's current funding plans told TechCrunch.
For one, despite having co-founded the business with one of the most high-profile, successful founders in Israel, One Zero has got remarkably little attention outside of its home market to date. The company raised approximately $242 million so far and was valued at around $320 million in 2023 according to data in PitchBook. Our sources report this valuation to be much higher in the next round.
Unclear who the investors are, but previous rounds were led by Tencent, OurCrowd, and SBI Ventures (the newly-independent firm that once was part of SoftBank).
One Zero's momentum comes as Shashua, who has a non-executive role at the company, with CEO Gal Bar Dea has just finished frenetic pace of activity. In the past couple of years, Shashua has founded or co-founded startups focused on humanoid robotics (Mentee) and alternative approaches to large language models for generative AI (AI21), as well as, launched just a couple of weeks ago, AA-I Technologies (pronounced "double AI"), which Shashua described to me as his effort to build an "AI scientist." He is also a computer science professor at the Hebrew University in Jerusalem. One Zero's equally ambitious mission is "bring private banking to the masses," he said in an interview. It seeks to democratize the type of high-touch, advisory services that are accorded to high-net-worth individuals by banks, in a market where the average person not only doesn't get that kind of service today but stares at a future where there may be no physical bank and no humans to help at all.
It is attacking that ambition with a dual business focus. In Israel, where One Zero is based, the startup has won a banking license and been building a full-stack retail bank. Along with that, One Zero is using insights gleaned from that retail business—Shashua described it in an interview as a "sandbox"—to train its models and hone its technology, with the aim of licensing that tech to banks elsewhere.
In the retail business, Shashua said that the company now has some 110,000 customers and though it hasn't announced any licensing deals so far, said it's received a number of inbound requests from major banks to do so.
The company's so far cornerstone-and the area in which it will concentrate its funds-is a chatbot called Ella, that is to exceed current models in the services that human bankers could not.
As Shashua conceives it, although there have been a number of efforts to install AI into retail banking services, for instance, with regards to functions such as controlling spending, they're not very effective.
"You don't see banks deploying artificial intelligence to a level in which you are actually replacing a banker," he said.
For instance, he said, take automated communication.
You can ask a banking chatbot simple questions, such as "how much money is in my account?, or info about recent transactions, and it will probably be able to answer it. But if you ask something with calculations, like "how much money I will have in my deposit account by the end of the year based on the activity so far?
Or what's the best way for me to buy a car based on my financial profile?
It's not just that chatbots can't answer such questions. Most personal bankers can't either.
"There's an opportunity here, where generative AI can seemingly do this," he said. "It goes way beyond spending tracking."
So, in other words, instead of training one large model, Zero builds several models," Shashua said. "But rather than making a single model more efficient to do all those things, he is using multiple large language models." Some models might be optimized for different tasks, he said, but running tasks through several LLMs can provide diversity of responses, then run through a verification process to understand when answers are misleading or wrong. And if those answers are not verified to be helpful or correct, the AI doesn't try to say something anyhow, he said. "It's fine for it to say, I cannot solve your problem. I cannot answer your question," he said. "Humans cannot also answer every question, right? So it's fine. It's not fine to say, here's an answer to your question, and the answer is completely fake, completely false." The system starts with rather simple functionalities such as expense management and will add functionality over time to counsel customers on how to better finance large purchases or save.