Be My Eyes taps GPT-4 to give 250M blind users a virtual eye

After a decade connecting sighted volunteers to people who are blind or have low vision, Danish startup Be My Eyes is betting its future on a synthetic volunteer. The company is integrating OpenAI’s GPT-4, with its new visual input capability, directly into its app to create a Virtual Volunteer that can analyze images and hold a real conversation about them.

CEO Michael Buckley doesn’t mince words about the tech, calling the performance ‘unparalleled’ compared to any existing image-to-text tool. The leap here isn’t just identifying objects—it’s contextual reasoning. Where a basic recognition app spots groceries, GPT-4 can parse the contents of your fridge and suggest a recipe based on what’s available. CTO Jesper Hvirring Henriksen draws a sharp distinction: older models tell you what’s in front of you, but GPT-4 can discuss whether those noodles have the right ingredients or warn that an object on the ground isn’t just a ball but a tripping hazard.

Beta testing started in early February with employees and select users, including blind broadcaster Lucy Edwards, and the feedback has pushed the company to accelerate. Buckley says the feature will roll out to users ‘in weeks.’ One early tester already used the system to navigate a railway system solo—getting not just map details but point-by-point walking directions to their destination.

Screen navigation might be the bigger unlock. Henriksen explains that GPT-4 can now scan a cluttered webpage and summarize it the way a sighted person naturally would, bouncing between important data points instead of reading every line sequentially. That makes e-commerce and news sites actually usable in real-time. Buckley frames it bluntly: a profound development for humanity that also represents an enormous commercial opportunity.