OpenAI has introduced its first AI agent named Operator, designed to assist with various tasks. Operator can independently browse the internet to reserve a table at a restaurant, compile a shopping list, or create memes. In a demonstration video, Sam Altman and his colleagues showcased the skills Operator currently possesses.
Operator is capable of handling numerous tasks, though not all. The interface resembles ChatGPT, making it familiar to users. A new feature is the integration of services like Uber, OpenTable, and eBay, which can be selected to instruct Operator to complete tasks using these services.
For example, Operator can book a restaurant table. Users simply choose OpenTable and specify the restaurant and time. Operator then opens a cloud-based browser to access OpenTable’s website, searching for the desired restaurant and time. If a mistake occurs, such as selecting the wrong location, Operator can correct itself. In one instance, it initially selected the wrong restaurant location but adjusted it based on the user’s location in San Francisco. If the requested time is unavailable, Operator suggests alternatives.
In another demonstration, OpenAI staff uploaded a photo of a shopping list, allowing Operator to purchase the items independently. Users can monitor each step and intervene if necessary, or let the AI handle the task, freeing up their time for other activities. However, for logins, captchas, or payment methods during shopping, users must intervene, as the AI doesn’t handle payments with personal data.
Operator uses screenshots from websites to navigate with a virtual mouse and keyboard. OpenAI has developed a new AI model named Computer-Using Agent (CUA) to facilitate this. CUA utilizes GPT-4o to process visual elements and has been trained on user interfaces to recognize menus, text fields, and buttons. Despite being in early development, CUA has achieved excellent results in WebArena and WebVoyager benchmarks.
Currently, Operator is in a preview phase, meaning issues may still arise. OpenAI has limited access to a small group of users, specifically Pro users in the United States. This subscription costs $200 per month. OpenAI plans to eventually offer Operator to Plus, Team, and Enterprise users and integrate some features into ChatGPT.
OpenAI emphasizes that Operator is still being tested and refined. Future plans include expanding access and incorporating additional features into ChatGPT. Users interested in AI tools can explore local options using various available tools.