OpenAI has launched a new general-purpose AI agent within ChatGPT, designed to perform a wide range of computer-based tasks on behalf of users. This "ChatGPT agent" integrates functionaliti…
OpenAI has launched a new general-purpose AI agent within ChatGPT, designed to perform a wide range of computer-based tasks on behalf of users. This "ChatGPT agent" integrates functionalities from previous OpenAI agentic tools, such as the ability to navigate websites and synthesize information from multiple sources.
Users can interact with the agent using natural language prompts, enabling it to manage calendars, create presentations, and run code. The tool is currently available to subscribers of OpenAI's Pro, Plus, and Team plans, accessible through an "agent mode" selection in ChatGPT's menu.
The launch marks a significant step in OpenAI's efforts to transform ChatGPT into an action-taking agent, moving beyond mere question-answering. The agent can connect to apps like Gmail and GitHub via ChatGPT connectors, allowing it to access relevant information. OpenAI suggests users can utilize the agent for complex tasks such as planning and purchasing ingredients for a meal or analyzing competitors to create a slide deck.
The underlying model demonstrates state-of-the-art performance on benchmarks, including Humanity's Last Exam and FrontierMath, particularly when equipped with tools like a terminal for code execution. Recognizing the potential safety concerns associated with such advanced capabilities, OpenAI has prioritized safety measures in the development of ChatGPT agent.
The company has classified the model as "high capability" in biological and chemical weapon domains. New safeguards include real-time monitoring of user interactions, with classifiers analyzing prompts for biological relevance and responses for potential threats. Additionally, the agent's memory feature has been disabled to prevent potential misuse, such as data exfiltration through prompt injection attacks.
While the capabilities of the ChatGPT agent appear promising, its real-world effectiveness remains to be seen, as previous agent technologies have faced challenges in practical applications. OpenAI, however, claims to have developed a more capable model that can deliver on the promise of AI agents.
The company is taking a cautious approach to safety, implementing robust safeguards to mitigate potential risks associated with the agent's advanced functionalities.