New AI agent from OpenAI will be able to book tickets, order food, and perform other tasks on websites instead of users
OpenAI presented a new AI agent Operator that can perform tasks on the Internet for users.
Here's What We Know
The agent uses its own browser to browse the web, click on buttons, enter text, and scroll through content. This allows it to perform tasks on the Internet like a person who clicks buttons, scrolls through pages and enters text on websites. Initially, the new product will be available only to ChatGPT Pro subscribers in the United States.
The operator runs on a special model that combines the capabilities of GPT-4o with in-depth training. This allows it not only to "see" pages through screenshots but also to interact with interfaces as we are used to doing with a mouse and keyboard.
The most interesting thing is that the agent not only performs tasks, but can also correct itself. If something goes wrong, it will give you control over the process. It will also ask for permission if the site requires sensitive information (such as passwords) or ask for your consent before sending an email.
OpenAI has partnered with popular companies like Uber, DoorDash, Instacart, and others to ensure that the agent performs real-world tasks while adhering to security and ethical standards. However, not everything works perfectly - complex interfaces such as creating a slideshow are still difficult for it.
In the near future, Operator will be available for users of Plus, Team, and Enterprise plans, and OpenAI plans to integrate this technology directly into ChatGPT.
Source: OpenAI
