OpenAI Unveils ChatGPT: This AI Can Take Control of Your PC!

Since 2022, artificial intelligence has seen several key advancements, including web search capabilities, canvases, and image generator integrations. The current trend is the development of agent-based AI tools that can navigate the web on our behalf. Naturally, OpenAI aims to be a leader in this area. To maintain its pioneering status, the company has launched the ChatGPT agent. Let’s delve into the details!

Operator’s Capabilities in ChatGPT

“ChatGPT can now work for you using its own virtual computer, handling complex tasks from start to finish by seamlessly transitioning between reasoning and action,” OpenAI states in its announcement. To achieve this, the chatbot now incorporates the capabilities of Operator, the agent-based AI technology introduced in January, combined with the strengths of ChatGPT.

This means that you can now turn your ideas and questions into concrete actions, whether it’s conducting in-depth financial research, filling out online forms, or creating polished presentations, highlights OpenAI.

In practical terms, you can input a request in natural language to ChatGPT (for example, scheduling a meeting in a calendar or looking up flights for a trip), after which ChatGPT will proceed to check dates online or compare prices. In various benchmarks, OpenAI notes that the ChatGPT agent has set “new records” in the latest human exams as well as in spreadsheet modification capabilities.

However, for now, the ChatGPT agent seems designed for simple and easily automatable tasks. OpenAI does not fail to mention that “results may appear basic in terms of formatting and finishing, especially when starting without an existing document”.

How the ChatGPT Agent Works

For the user, everything will happen directly within ChatGPT. The agent operates on a built-in virtual computing environment, capable of managing tasks from start to finish. It utilizes a set of specialized tools depending on the task at hand. Three main channels are employed:

• A visual browser, which allows ChatGPT to interact with websites like a human: clicking, scrolling, and filling out forms.
• A textual browser, optimized for simpler queries where only the content is relevant, such as extracting specific data or summarizing an article.
• The agent can also use direct API access to instantly fetch structured data (financial, sports, etc.).

Visually, the web search will open directly beneath your query, resembling an in-house browser. The user retains total control over access – Operator will ask for permission before “performing significant actions” – and can take over navigation at any time. Moreover, certain “high-risk” operations, such as financial transactions, will not be possible.

Availability of the ChatGPT Agent

Currently, the ChatGPT agent is available to users of the Pro, Plus, and Team versions, with expansion to the Enterprise and Education plans expected this month. Except for the Pro plan, which offers almost unlimited use, a limit of 50 tasks per month will be set. However, French users will need to wait a bit longer: OpenAI is “currently finalizing the launch date of the ChatGPT agent in France,” according to a press release.

Activation of the feature will be simple: just select the “agent mode” from the dropdown tool menu. This is the menu under the prompt bar, where tools like web search and image generation are found. The site displayed in the Operator preview will be functional for 30 days, after which it will be deleted.