OpenAI recently introduced its new AI agent, “Operator,” which is based on the CUA (Computer-Using Agent) model. This technology helps users automate various browser-based tasks. The CUA model is trained in such a way that it asks for confirmation from the user before finalizing any task, such as submitting an order or sending an email. This feature gives users a chance to double-check the work of the AI ​​​​so that any mistake is not permanent. OpenAI said in a statement to TechCrunch that this model has proved useful in many cases, and the company wants to expand it to other tasks as well.
However, OpenAI also admits that CUA is not completely perfect at the moment. The company has said that “CUA is not yet able to perform reliably in all circumstances.” For example, it is currently unable to handle complex or specialized tasks, such as creating detailed slideshows, managing complex calendar systems, or working with customized web interfaces.
Precautions and limitations
OpenAI requires user monitoring of the operator for highly sensitive tasks, such as banking transactions. Users must enter credit card information manually. The operator ensures that on sensitive websites, such as email or financial services, users can monitor it directly and correct any mistakes immediately. This precaution helps prevent AI from making bad decisions and potential risks, such as accidentally spending money in the wrong place.
Google has also adopted a similar approach. Their “Project Mariner” AI agent also avoided entering sensitive information such as credit card numbers. Such measures ensure that the technology remains useful as well as safe.
Limitations of the Operator
The operator has some significant limitations in its use. It has daily and task-based “rate limits.”. It can perform many tasks simultaneously, but its functionality at one time is limited. Currently, the operator refuses to perform certain tasks for security reasons, such as sending emails or deleting calendar events. OpenAI says this feature will be added in the future, but no timeline has been given.
In addition, the “Operator” can get “stuck” when it encounters obstacles such as a complex interface, password field, or CAPTCHA. In such cases, it asks the user to take control to complete the task.
The future of AI agents
OpenAI has taken slow steps in the development of AI agents, as this technology involves security risks. When AI systems can perform tasks on the web, it can open doors to dangerous use cases. Such as automating phishing scams or stealing sensitive information. OpenAI has released the “Operator” in its current form to mitigate these risks.
Tools have been added to the “Operator” that protect it from malicious commands and phishing attempts. A monitoring system can detect suspicious activities and stop the task. Also, automated and human reviews constantly update its security measures.
User control and data privacy
The “Operator” keeps users in control at all times. It asks for confirmation before finalizing important tasks and hands over the workload to the user when entering sensitive information. Furthermore, users can control the data privacy options of the “Operator.”. The “Improve the model for everyone” option can be turned off to ensure that the data is not used in model training.
Furthermore, users can delete their browsing data in a single click and log out of all sites. OpenAI has also ensured that the “Operator” can identify and avoid hidden commands or malicious code.
Using the “Operator”
The “Operator” is easy to use. Users can instruct it by simply describing a task. It can also handle custom instructions like setting preferences for airline bookings. Users can instruct it to perform multiple tasks simultaneously, like ordering a mug on Etsy and booking a campsite at the same time.
Further plans
OpenAI plans to bring the “CUA” model to the API so that developers can use it to build their own AI agents. Also, there are plans to improve the capabilities of the “Operator” and make it capable of handling long and complex processes. OpenAI will make it available to Plus, Team, and Enterprise users for wider use.
Conclusion
The “Operator” is a huge leap forward in the world of AI. While it still has room for improvement, it is an impressive step towards making daily tasks easier for users. OpenAI is committed to making it safe, reliable, and user-friendly.