OpenAI launches a general purpose agent in ChatGPT

OpenC is launching a new general purpose AI agent at ChatGPT, which the company says that users can complete various computer-based works for users. OpenAI says the agent can automatically navigate a user calendar, create editable presentations and slideshow and run the code.

This tool, known as the ChatzPT Agent, combines a number of capabilities with previous agent tools in Openai The ability to click on the operator’s websitesAs well as The ability to accumulate deep research data From dozens of websites in the short research report. Open says that users will be able to contact the agent by requesting the Chatzipi in natural language.

On Thursday, the Open is turning the ChatGP agent for its Pro, Plus and Team Plans customers. To activate the equipment, users can select “Agent Mode” in the dropdown menu of ChatzPT equipment.

The introduction of the ChatzPT agent is that the bold attempt of the Openai still represents the ChatGP to transform the Chatzipt into an agent product that can only take steps for users without just answering the questions and take the offload task. In recent years, Silicon Valley companies with Open, Google and confusion have unveiled dozens of AI agents that have only promised to do it. However, this initial version of AI agents has proven to fight complex tasks, and the final Vision Tech executives around AI agents seem to be less compulsory than the pitch of the Executive.

It was said that Open said that the ChatzPT agent is much more capable of his previous offers.

The new agent of OpenAI can access ChatzPT connectors, allow users to connect applications such as Gmail and Githab so that agents can get relevant information on your prompt. Furthermore, OpenAI says that the ChatzPT agent has access to a terminal and can use API to access specific apps.

According to the Openai, the underlying ChatzPT agent offers sophisticated performances to a number of criteria.

The agency says that the ChatzPT Agent Model scores 5..6% in the last test of humanity (pass@1), a difficult test composed of thousands of questions across more than one hundred subjects. Opening and 3 and 4-Minute tests scored fairly doubled.

In the Frontiermath, OpenAI, one of the most well -known mathematical mathematics criteria, says that the ChatzPT agent score 27.4% when it has access to its tools such as terminals for code execution. The previous sophisticated score came from and 4-minute, which just scored 6.3%.

OpenAI notes that it has developed the agent with the protection of security, because the product represents some new powers that can make it more dangerous in the hands of a bad actor. However, how much the ChatzPT agent is really capable of the agent.