OpenAI, Chatgpt developed for agent modeNU in Türkiye Plus He opened it to subscribers. According to the commands received from the user, the new system can perform tasks on the computer and receives user approval during operations.
OpenAI said in a statement, the agent mode is working via a virtual computer and the browser can carry out operations such as navigation, document preparation, reservation. The system can take a screenshot by accessing a browser from a remote browser, navigate between pages and perform operations on behalf of the user. However, a clear approval from the user must be obtained before each process. The agent mode consists of a combination of systems such as “Operator” and “Advanced Data Analysis” previously introduced. Thanks to this structure, all more comprehensive and multi -stage tasks can be maintained with user interaction.
OpenAI CEO Sam Altman said that this new system is an important turning point for him. According to Altman, the ability of agent mode to plan and transactions on computer, general artificial intelligence It offers an experience close to its vision. This structure is not limited to providing information, but can complete the tasks by taking action directly. Examples include a flight ticket reservation, calendar planning or e-mail sending. The agent is both transactions in all these processes and requires permission from the user if necessary. The system, which is opened for use in Türkiye, is offered with a limited quota of 40 messages per month during the initial stage.
Chatgpt agent mode tasks can complete step by step
The virtual computer environment, where the agent mode works, is managed by OpenAI and all tasks are carried out in this environment. The system can provide limited access to services such as Gmail, Google Drive or Github through certain connections. These connections mostly work with the permission of purely-reading, and the user does not require any additional verification. For example, if an e-mail sending or the authority to write to the document is required, the system only progresses when the user permits. It is also stated that the system consists of different layers in terms of security. In particular, there are various security measures against scenarios such as harmful commands, misleading ingredients or risk of infiltration of user data.
OpenAI offers both technical limiting and user information screens before use to ensure the safety of agent mode. In addition, the user can take over full control over the virtual browser at any time. This feature is activated, especially when sensitive data such as passwords are entered. The screenshot is stopped when scanner control is taken and the operations continue only with the user guidance. Company, agent mode first Pro And Team Limited its plans. Now, including Türkiye, including more in the region Plus It has also opened its users to access.
According to information from OpenAI, the system gives more successful results in complex tasks than previous models. For example, with multi -step mathematics problems Frontiermath The test was explained that 27.4 %was correct. Another measurement Humanity’s Last Exam In the test, the success rate was reported to be 41.6 %. These data reveal the competence of the system, especially in areas such as multi -layered thinking and processing. However, there are also warnings that the system is still experimental and that user data should be processed carefully. In any case, this feature will remain open only in a limited number of countries and certain user groups for now.