OpenAI is allowing some users to try out the new ChatGPT feature artificial intelligence Use your web browser to book trips, buy groceries, hunt for bargains, and do more online.
The new tool, called Operator, is an AI agent: It relies on an AI model trained on both text and images to interpret commands and figure out how to use a web browser to execute them. OpenAI claims it is capable of automating many daily tasks and errands during the workday.
OpenAI operator follows rival releases of both Google and Anthropic, yes proven ones Able to use the web. AI agents are widely considered the next stage of evolution for AI tracking chatbots, and many companies have jumped on the hype bandwagon by touting them. In most cases, these capabilities are very limited and simply use the language model to automate things that are typically done with conventional software.
“AI is evolving from being a tool that can answer your questions to a tool that can act in the world, executing workflows,” said Peter Welinder, Vice President of Product at OpenAI. complex, many steps”. “We're going to see a lot of impact on people's productivity—as well as the quality of work that people can get done.”
OpenAI acknowledges that granting ChatGPT access to web browsers introduces new risks and says the Operator may sometimes misbehave. It said it has implemented many new safeguards and plans to gradually expand the Operator's capabilities.
Welinder and Yash Kumar, product and engineering lead for OpenAI's Computational Agent, said the plan is to learn from how people use the tool. They admit that the tool can make unwanted bookings or purchases but add that they have to do a lot of work to ensure that the tool asks before doing anything risk. “It will come back to me and ask for confirmation before taking steps that may be irreversible,” Kumar said.
OpenAI today also released a new “system card” highlighting possible issues with the Operator. These include the possibility of it misinterpreting commands or deviating from what the user requested; abused by users; or become a target of cybercriminals.
“It also poses numerous safety challenges,” Kumar said. “Because your attack vector area and your risk vector area increase quite significantly.”
Initially, the operator will be available as a “research preview” for ChatGPT users with Pro accounts, at a whopping cost of $200 per month. The company said it plans to expand access while rolling out the tool slowly, as there will inevitably be some mistakes along the way.
In several demonstrations, Operator has shown the potential for AI to take on a more active role as a web helper. This tool has a remote web browser and chat window for communicating with users.
At WIRED's request, the Operator was asked to book an Amtrak train from New Haven, Connecticut to Washington, DC. It goes to the correct website and enters the correct information needed to display the timetable and then request further instructions. If the user is logged into the Amtrak website or to a browser profile with credit card information stored, the Operator will be able to go ahead and book a ticket — although it is designed to require prior authorization.
Kumar asked the Operator to reserve a table at Beretta, a restaurant in San Francisco. The program visits the OpenTable website, finds the right restaurant, and looks up availability before asking what to do next. OpenAI says it has partnered with several popular websites, including OpenTable, to ensure that the Operator operates smoothly on them.
The new tool is based on OpenAI's GPT-4o AI model, which can recognize browsers and websites as well as chat using typed text. The tool incorporates additional training designed to help it understand how to perform online tasks. OpenAI will also provide a Compute Agent through its API.