OpenAI's Operator is an AI agent that automates web tasks using a new model trained on screenshots and text commands. It can book concert tickets, reserve restaurants, and shop for groceries. The tool costs about $200 monthly in Research Preview and could help people with disabilities or limited computer skills access the web.
Chapters
OpenAI just announced Operator, an AI agent designed to perform web tasks, and it's pretty crazy. Powering Operator is a new model called Computer Using Agent, or CUA for short. CUA has been trained on graphical user interfaces and in a nutshell it takes in as input a text command, a screenshot of the interface that it's on, and then as output it performs actions on said interface. The model has state of the art performance on web browsing and computer use. Benchmarks and examples that were shown were getting concert tickets, booking reservations for restaurants, and buying groceries. But what is exciting about these models is their potential to help those with little to no computer skills perform complex digital web tasks and even more importantly, those with disabilities or visual impairments have access to computers, especially when it's combined with voice. Operator is only in Research Preview at the moment and it's only available in the US for pro users. So if you're in the US and you have $200 a month to spare, let us know what you think of it.