The ChatGPT Agent From OpenAI Can Do Intricate, Multi-Step Tasks.

(CTN News) – Recently, OpenAI unveiled a highly advanced ChatGPT artificial intelligence program that can do complex, multi-step tasks.

To incorporate Operator and Deep Research technologies, the organization has created a unique model for the new agent. It can perform tasks like creating spreadsheets and PowerPoint presentations because it has its own virtual computer installed.

OpenAI designs it to operate in the background, obtaining permission before initiating “actions of consequence,” such as sending an email. By simply choosing “agent mode” in the composer at any time during the conversation, ChatGPT users with Pro, Plus, or Team subscriptions can activate the new agentic features.

Businesses’ communication, automation, and operational processes have undergone significant changes as a result of the deployment of AI agents.

OpenAI has made available an agent named ChatGPT.

The purpose of this agent’s development was to streamline iterative and collaborative processes. According to OpenAI, the ChatGPT agent outperforms earlier models in terms of its ability to offer more flexibility and interactivity, making it more suitable for iterative and collaborative workflows.

You can pause ChatGPT whenever you want to make changes to the assignment, make it more clear, or guide it toward the intended results. It will continue where it left off, keeping all of the work that has already been done.

ChatGPT may ask for more info to ensure the task meets your goals. If a task appears to be stalled or takes longer than anticipated, you have the option to pause it, obtain a progress report, or cancel it utterly.

On the other hand, you will only receive incomplete results. If you have installed the ChatGPT app on your mobile device, it will alert you when your task is complete.

ChatGPT’s ability to adjust and change is mainly due to its special features, making it much more useful for both personal and work situations, according to OpenAI.

“In the workplace, you can automate repetitive tasks, including the conversion of screenshots or dashboards into presentations with editable vector elements, the reorganization of meetings, the planning and booking of offsites, and the addition of new financial data to spreadsheets while maintaining the same formatting.”

There are unique risks connected to new additions.

For the first time, users will be able to suggest that ChatGPT performs a particular online activity.

In addition, OpenAI said, “This introduces new risks, particularly because the ChatGPT agent can interact directly with your data, whether it is information accessed through connectors or websites that you have logged into via takeover mode.”

To solve issues including managing sensitive information on the live web, growing its user base, and limiting access to the terminal network, the company has put in place stronger security measures.

Additionally, the organization has said that it has strengthened the strict limitations associated with the operator’s research preview. “The risk is significantly reduced by these mitigations;

However, the ChatGPT agent’s overall risk profile is higher due to its ability to reach a greater number of individuals and possess a greater number of tools.”

Additionally, OpenAI has made significant efforts to safeguard the ChatGPT agent against the threat of malicious manipulation through quick insertion, which is present in all agentic systems. “Prompt injections are the result of incorrect instructions that a ChatGPT agent may encounter on the internet while performing a task, with the intention of altering its functionality.”

In addition to being taught and tested on how to detect and prevent prompt injection attacks, the agent has been trained to employ monitoring to quickly identify and prevent them.

By requiring users to verify their activities before execution, we can further reduce the risk of damage from these attacks. Additionally, users can pause or resume tasks as needed. When choosing the information to provide to the agent, users should take these trade-offs into account.

They should also implement strategies to reduce the probability of encountering them. For example, they must make sure that connectors are turned off when they are not in use.

SOURCE: PEN