(San Francisco) OpenAI, the creator of ChatGPT, on Thursday launched Operator, an AI (artificial intelligence) agent capable of performing online tasks for the user, such as planning a vacation, reserving a restaurant or running errands, an important step in the race for ever more efficient AI assistants.
Posted yesterday at 8:32 p.m.
Operator “uses its own browser”, it can “look at a web page, scroll through it, click on buttons” and “fill in text fields like people do on a daily basis”, the Californian company said in a press release .
The new function is currently only available to professional ChatGPT subscribers, in order to improve it thanks to feedback.
“Operator is one of our first agents, that is to say AI capable of performing tasks for you autonomously: you give it a task and it executes it,” summarizes OpenAI.
The explosion of generative AI with the success of ChatGPT since the end of 2022 has launched a frantic race for AI assistants between technology giants, who are rapidly deploying tools capable of writing messages, answering questions, generating pictures, etc.
The Holy Grail of Silicon Valley is AI agents, when the machine becomes a sort of omniscient secretary, available at any time and capable of carrying out numerous tasks, from sending messages to shopping on the Internet.
-In this area, OpenAI is not the fastest, at least in terms of deployment.
Operator resembles “Computer Use,” a feature launched in October by Anthropic, a start-up rival.
Computer Use allows Claude, Anthropic’s generative AI interface, to use computers like a human, from selecting buttons to entering text and handling different software.
Google, which presented Gemini 2.0, its new family of generative AI models, in December, is also moving forward with more complex interactions with technology, so that AI agents navigate the Internet autonomously, seek additional information in line or in a document, etc.
All the companies specify that AI assistants act under the supervision of humans: while they can select products to purchase on an e-commerce site, they cannot (yet) click the payment button.
OpenAI’s release includes a video showing how Operator works. An engineer asks him to find a recipe and add the necessary ingredients to his basket on an online ordering service: the AI agent goes to the cooking site, asks the user additional questions and asks him to connect when necessary.
