OpenAI has begun previewing a new tool called Operator that can navigate within a web browser. According to a blog post published Thursday, the software is powered by what the company calls a Computer-Using Agent. “CUA is trained to interact with graphical user interfaces (GUIs) — the buttons, menus, and text fields people see on a screen — just as humans do,” says OpenAI of the model. “This gives it the flexibility to perform digital tasks without using OS- or web-specific APIs.“
The current release of Operator builds on OpenAI’s GPT-4o model. It combines the vision capabilities of that algorithm with “advanced reasoning” trained through reinforcement learning. Operator has the ability to “break tasks into multi-step plans and adaptively self-correct when challenges arise.” According to OpenAI, that capability represents the next stage in AI development.
As with past research previews, OpenAI warns that Operator is “still early and has limitations,” and that it won’t “perform reliably in all scenarios just yet.” For instance, depending on the complexity of the task and interface involved, the agent greatly benefits from the user taking a few extra moments to write a more detailed prompt. Per The Verge, Operator will give the user control if it ever gets stuck on a task. It will also hand control over whenever a website asks for sensitive information, including login credentials. The company says it designed the tool to “refuse harmful requests and block disallowed content.”
OpenAI is making Operator first available to users of its $200 per month ChatGPT Pro subscription. It is also partnering with companies like Instacart to offer the agent on their platforms, though there again you’ll need a ChatGPT Pro subscription to test the integration.
Operator joins a growing list of AI agents that can either navigate a web browser or an entire operating system. Anthropic was the first to offer the capability with the release of its Claude 3.5 Sonnet model in October, followed more recently by Google with its Gemini 2.0 model and Project Mariner.
If you buy something through a link in this article, we may earn commission.
Credit: Source link
