The new Deep Research agent is integrated into chatgpt to synthesize knowledge and conduct complex research

Openai He has announced Deep Research, an artificial intelligence agent (AI) who uses reasoning to synthesize large amounts of information and execute research tasks, which is already available for users of Chatgpt for and will be expanded to Plus and Team subscribers.

The technology company has presented this tool, which “achieves in Dechedas of minutes what a human being would take many hours” only two days after announcing the deployment of the O3-mini model, which has also reached free users and offers An “exceptional” performance in Stem disciplines.

In this case, Deep Research, which is translated as’ deep research ‘, is an agent who can carry out research that requires several steps, simplified and can work independently to find, analyze and synthesize hundreds of sources’ Online ‘, in order to Create a full report at the level of a research analyst.

This generative AI tool is driven by a version of the OPENAI O3 model optimized for web navigation and data analysis and take advantage of the reasoning to search, interpret and analyze large amounts of text, already rchivas images in PDF format on the Internet.

Deep Research, which “changes their attitude as necessary in response to the information it finds,” Openai explained in a shared publication in his blog, is an important step for the technology towards his goal of developing an intelligence “capable of producing new scientific research. “

This solution is designed for ersonas that “perform intensive knowledge” in areas such as finance, science, politics or engineering, so that it serves as a complement to Conduct an “exhaustive, precise and reliable” investigation.

Each result is documented, by appointment and a summary of its reasoning, thanks to which the researchers can verify information that this agent provides, which “is particularly effective in finding niche information, not intuitive,” that is, the one that requires navigating numerous web pages.

OpenAI has advanced that this agent has trained with “real world” assignments that require the use of a browser and a python tool, as well as using the same reinforcement learning methods that are behind OpenAi O1.

To use it, the subscribers of Chatgpt Pro -Proximously will reach Users of Plus and Team- they must select the option ‘Deep Research’ in the ‘PROMPT’ and introduce the consultation. To add context to the application, you can attach spreads or leaves.

Deep Research takes between 5 and 30 minutes to offer the result and, once it reaches it, will notify users through a notification, to tell them that the investigation has been completed. This will be presented as a report within the chat, which will soon have support for integrated images and other analytical results, which will allow “greater clarity and context” to research.

Openai has finally acknowledged that Deep Research still has limitations and that it can hallucinate facts in the answers or make incorrect inferences, “although at a tas remarkably less than the existing chatpt models,” according to internal evaluations it has carried out.

It also hopes to launch a faster and more profitable version driven by a smaller model that continues to provide high quality results, as well as its corresponding mobile versions and desktop next month. Currently you can only access it through the Chatgpt website.

It should be remembered that Google presented a tool very similar to this last December, which also has the same name. Deep Research is now available in more than 45 languages, including Spanish, and more than 150 countries through subscription to its Gemini Advanced model.

This other agent has also been designed to expedite the collection of information and with the objective of being in charge of “hard work” for searching and exploration of documents under the supervision of a human, the technological explained then.