Openai lets some users try a new chatgpt function that uses its artificial intelligence Operate a browser to reserve travel, buy food, hunt discountes and do many other online tasks.
The new tool, called operator, is an AI agent: it depends on AI model trained on both text and images to interpret commands and find out how to use a browser to carry out them. Openai claims it has the opportunity to automate many daily tasks and working tasks.
Openai’s operator follows rival editions of both Google and Anthropic, who have proven able to use the network. AI agents are widely seen as the next development stage To follow talks, and many companies jumped on the Hype train by propagating them. In most cases, these are very limited in their capabilities and simply use a language model to automate things usually done with regular software.
“AI develops from this tool that could answer your questions to one who is also capable of acting in the world, carrying out complex, multi -step workflows,” says Peter Welinder, VP about a product at Openai. “We will see a lot of impact on the productivity of people – but also the quality of work that people are able to perform.”
Openai admits that giving ChatGPT access to a browser does indeed introduce new risks, and it says that an operator can sometimes misbehave. It says it has implemented various new safeguards and plans to extend the operator’s capabilities gradually.
Welinder and Yash Kumar, product and engineering guide for a computer-using Openai agent, says the plan is to learn from how people use the tool. They acknowledge that the tool could make unwanted reserves or purchases, but adds that a lot of work has been to make sure it asks before doing something risky. “It will return to me and ask for confirmations before taking steps that could be irreplaceable,” says Kumar.
Openai today also released a new “system card” outlining the problems that could arrive with an operator. These include the possibility for it to misunderstand commands or divert from what a user requests; to be misused by users; or to be targeted by cybercriminals.
“It also presents an incredible amount of security challenges,” says Kumar. “Because your attacking vector area and your risky vector area increase quite significantly.”
An operator will initially be available as a “research preview” for ChatGPT users with a professional account that costs a lot of $ 200 a month. The company says it plans to expand access while slowly slowing the tool, as it will inevitably make a few mistakes on the way.
In several demonstrations, an operator has shown the opportunity to take a more active role as an online assistant. The tool has a remote browser and chat to communicate with a user.
At Wired’s request, an operator was asked to reserve Amtrak train trip from New Haven, Connecticut, to Washington, DC. It went to the correct site and entered the necessary information correctly to bring the schedule, later requested further instructions. If a user had been logged in to the Amtrak website or a browser profile with stored credit card information, operator could continue and book a ticket – though it is designed to request permission first.
Kumar asked operator to order a table at Beretta, a restaurant in San Francisco. The program went to OpenTable’s website, found the right restaurant and sought availability before asking what to do next. Openai says it has partnered with some popular sites, including OpenTable, to make sure Operator works smoothly on them.
The new tool is based on Openai’s GPT-4o AI AI, which can perceive a browser and web page and converse in a typed text. The tool includes further training designed to help it understand how to perform tasks online. Openai will also provide his computer agent with his API.
