AI is good at a great many things, but so far, its utility has largely been centered on analyzing large amounts of data and creating content. Now, a new startup plans on taking the tech to a less cerebral and more utilitarian place with JACE.
AI’s usefulness in the commercial and science sectors seems to grow daily. In the last few months alone we’ve seen it identify a new rare-earth-free magnet, analyze the emotions of crisis-line callers, help with weather and climate forecasting, and even attempt to make airline food better.
But in the consumer sector, the technology has thus far been limited to answering questions, explaining concepts, and creating various types of content. In short, it’s a great “thinker,” but not really much of a doer. Startup Zeta Labs plans to change that by linking its proprietary web-interaction AI engine, AWA-1 (which stands for autonomous web agent-1) to its JACE interface. The result is a system that can execute a series of online tasks with just a few simple prompts.
For example, if you’re looking to take a trip, you should be able to ask the system to book an Airbnb for your specific dates, in a certain neighborhood, and at a specific price point.
In a video on the Zeta Labs website, company co-founder Frydery Wiatrowski explains that in testing the system, the Zeta Labs team was even able to have it successfully form a math-tutoring company. According to a press release from the company, the system not only walked the user through creating the company, but it also found the first client and made actual revenue within two weeks.
“At Zeta Labs, we aim to enable AI models to interact with the digital world around them,” says Peter Albert, another Zeta Labs co-founder. “The development of our AWA-1 model allows JACE the ability to control a browser, essentially giving the AI assistant the digital equivalent of arms and legs. You can’t ask any current chatbot on the market to book a trip, pay an invoice or set up a job post. With JACE, you can.”
The company says that in further testing – that included ordering a pizza and looking for two apartments in London meeting specific criteria – when JACE operated using the AWA-1 model, it demonstrated an 89% task-completion model. That’s opposed to a 68% success rate when JACE was plugged into GPT-4o or a measly 25% success rate when using “an open-source web agent using GPT-4o.”
Still, Wiatrowski says the system has its challenges.
“JACE still has its limitations,” he says in the company video. “It can struggle with complicated tasks, and the current browsing speed is somewhat slow. We are working very hard to make it faster and more reliable.”
Of course, despite JACE’s impaired performance with GPT-4o, it’s easy to imagine that OpenAI could easily match or exceed AWA-1’s performance with its own web-interface system specifically designed to work with its own AI engine. And we already know that Google has a mighty AI-powered task-oriented system in the works called Assistant with Bard, so it remains to be seen how JACE will compete against the bigger players in the field.
Still, Zeta Labs was recently able to snag US$2.9 million in pre-seed funding led by Daniel Gross, who was the former head of AI at Y Combinator and former Github CEO Nat Friedman, so it’s got its own steam behind it.
“Web browsing was invented 34 years ago and forever changed the global economy, yet AI models don’t understand how to use it,” says Gross. “The next paradigm shift in AI is unlocking that behavior, and Zeta Labs is poised to make that happen. I am pleased to be an investor in this grand mission!”
If you want to be among the first to try out JACE, you can sign up to the waitlist on Zeta Labs’ website.
Source: Zeta Labs
