OpenAI has recently unveiled a new feature for ChatGPT, enabling the widely used chatbot to perform actions on behalf of its users. This initiative is emblematic of a broader movement across the industry aimed at transforming how individuals accomplish tasks online: Leading technology companies are optimistic that, in the future, users will no longer need to switch between various applications or conduct manual web searches, as they envision a scenario where intelligent agents could handle all these tasks seamlessly.
ChatGPT’s new agent mode, set to roll out immediately, marks a significant development in the ongoing trend among tech giants to enhance digital assistants with increasingly sophisticated functionalities. The competition intensifies between OpenAI and Google, both of which are chasing comparable goals with their respective Gemini assistant.
OpenAI announced that ChatGPT’s new agent mode is capable of “thinking” and “acting” through its virtual computer, allowing it to manage intricate action-oriented requests. The company announced in a blog post that users will have the capability to issue commands like, “look at my calendar and brief me on upcoming client meetings based on recent news” or “plan and buy ingredients to make Japanese breakfast for four.”
During a video demonstration, employees from OpenAI crafted an extensive prompt, requesting the agent’s assistance in helping a user prepare for a wedding. The document outlined a series of detailed directives, including the necessity to “find an outfit that matches the dress code.” It further stipulated the submission of five potential options, as well as recommendations for hotels capable of hosting guests for a few additional days surrounding the event.
A new feature has been introduced for subscribers of the Pro, Plus, or Team plans. The latest development integrates features from OpenAI’s existing ChatGPT Operator and Deep Research tools. The Operator is designed to browse the web, whereas Deep Research focuses on analysing online resources to compile reports and other relevant information.
The latest update marks a significant advancement in OpenAI’s initiative to transform ChatGPT into a more all-encompassing universal assistant. The broader AI industry is currently facing significant challenges as it seeks to tackle critical shortcomings and privacy concerns associated with the technology. Recent developments in AI technology reveal ongoing challenges, particularly concerning hallucinations and bias. A notable incident occurred last week with xAI’s Grok chatbot, which generated antisemitic content in response to specific prompts, highlighting the unpredictable nature of such models.
In a recent blog post, OpenAI recognised that the new capabilities of ChatGPT introduce potential risks. The organisation announced that it has restricted the data available to the model, emphasising that specific tasks, such as sending an email, necessitate user supervision. The company states that the model is also programmed to decline “high-risk tasks” such as bank transfers.
In a recent post on X, OpenAI CEO Sam Altman characterised the new agent as “cutting edge and experimental,” suggesting it offers a glimpse into the future. However, he advised caution, stating that it should not be employed in high-stakes situations or for sensitive personal information until further study and improvements can be made in real-world applications.
He urged users to exercise caution when granting ChatGPT access to their personal information. Granting access to a calendar for coordinating a group dinner appears logical; however, an agent doesn’t need calendar access when shopping for clothes on a user’s behalf.
The announcement arrives amid growing competition among tech giants as they strive to create advanced AI agents in their pursuit of dominance in the AI landscape. During its developer conference in May, Google unveiled a series of AI-related announcements, featuring an agent capable of making restaurant reservations and purchasing event tickets, among various other functions. Apple is reportedly developing a more sophisticated iteration of Siri, capable of utilising applications on behalf of users. However, the release of this update has been postponed indefinitely.
