OpenAIthe developer of ChatGPThe announced Operatoran agent of artificial intelligence able to browse the web and independently carry out activities on behalf of people. Availability is currently limited to United States. Operator is based on the model Computer-Using Agentwhich combines the visual capabilities of GPT-4o (called ‘multimodal’) with an advanced reasoning and learning system.
According to the official note from OpenAI, Operator is capable of interacting with graphical interfacesi.e. computers, starting from a few indications from users. An example is the possibility of book a table for a restaurant as soon as he gets free or a plane ticket when the price falls below a threshold.
After some internal tests, Operator successfully completed the87% of operations web-based and the 38.1% of those that require full use of the computer. The agent can also fill out forms on portals, order food delivery, buy products online And create images. In more complex cases, when human intervention is required, Operator sends a notification to leave control to the user. It happens, among other things, after theAI has written an email and is ready to send it or when sensitive data needs to be entered online, such as those of credit cards.
OpenAI has announced collaborations with several companies, including Uber, DoorDash And Instacartto integrate Operator in everyday life and simplify activities. Subscribers to ChatGPT Pro in the USA they can try the novelty while in the future Operator it will also come for users ChatGPT Plus, Team and Enterprise.
«By collecting feedback from the real world, we can refine the security measures and constantly improve them, as we prepare for a future that will see increasing use of digital agents», the words of OpenAI.