Openai is updating the power of AI model OperatorIts AI agent who can browse the web independently and use some software in the cloud hostess virtual machine to meet users’ requests.
Soon, the operator -based model will use O3One of the latest in the Openi’s Openi series of “reasoning” models. Earlier, the operator relied on the customs version GPT-4O.
According to many standards, O3 is a much more modern model, especially on mathematics and works involved.
“We’re taking the current GPT-4O-based model for the operator with an Open AO3-based version.” Is written In a blog post. “API version [of Operator] Will be based on 4O.
The operator is one of the many agent tools released by AI companies in recent months. Companies are running to make a very sophisticated agent who can do more or less work reliably without supervision.
Google offers a “Computer use“The agent through his Gemini API, who can browse the web and take steps by consumers, is also called a more consumer -based offer. Marine. Anthropic models are also worth Perform computer workOpening files and naving on the web pages.
According to the Open, the new operator model, called the O3 operator, was “properly developed with additional additional additional protective data,” which included datases that were designed to teach the model. [OpenAI’s] Decision limits on verification and denial.
Open has released a technical report showing the performance of the O3 operator on a specific safety diagnosis. Compared to the GPT -4 operator model, the O3 operator is less likely to perform “illegal” activities and refuse to find sensitive personal data, and AI is less likely to be an attack form.
Openai wrote in his blog post, “O3 operator uses the same multi -layered approach to safety we use for the 4O version of the operator.” “Although the O3 operators inherit O3’s coding capabilities, it does not have a coding environment or local access to the terminal.”