Openai announces a new AI “Agent” designed to help people carry out in -depth, complex research using the Chatgpt, the company’s Chatbot platform.
Quite appropriate, it is called deep research.
Openai said a blog post He published Sunday that this new opportunity was designed for “people who are doing intensive knowledge in areas such as funding, science, politics and engineering and need thorough, accurate and reliable research”. It could also be useful, the company added, for those who make “purchases that usually require careful research, such as cars, devices and furniture”.
Basically, Chatgpt Deep Research is intended for cases where you don’t just want a quick answer or summary, but you should consider information from multiple websites and other sources.
Openai said it is deeply researching available to Chatgpt Pro users today, limited to 100 questions a month, with support for Plus and the users of the group coming next, followed by the business. (Openai aims at a habitual association in about a month from now, the company said and the quest limits for paid users should be “significantly higher” soon.) It is a geographical targeted launch. Openai had no release schedule to share Chatgpt customers in the United Kingdom, Switzerland and European Economic Area.
To use the Deep Research Chatgpt, you will simply select “Deep Research” to the composer, and then enter a query, with the option to attach files or computing sheets. (It’s an experience only on web for now, with the integration of mobile and desktop applications to come later this month.) Deep research could then take anywhere from 5 to 30 minutes to answer the question and you will receive a notice when the search is complete.
Currently, the results of Chatgpt Deep Research are only for text. But Openai said he intends to add built -in images, data illustrations and other “detailed” outputs soon. Also on the course map is the ability to connect “more specialized data sources”, including “subscriptions” and internal resources, Openai added.
The big question is, how accurate is Chatgpt Deep Research? AI is incomplete, after all. They are prone to hallucinations and other types of errors that could be particularly harmful to a “deep research” scenario. This is perhaps the reason why Openai said that every chatgpt deep research production will be “fully documented, with clear references and a summary of [the] Thought, making it easy to report and verify the information. ”
The jury is out of whether these mitigation will be sufficient to fight AI errors. Openai’s tissue search capability in chatgpt, chatgpt search, not rarely makes gaffes and gives incorrect answers to questions. The TechCrunch test found that the Chatgpt search produced less useful results than Google search for some questions.
To enhance the accuracy of Deep Research, Openai uses a special version of the recently announced O3 “logic” model AI trained by learning a reinforcement of “real world duties that require use of the browser and Python”. Learning aid essentially “teaches” a model through testing and error to achieve a particular goal. As the model approaches the target, it receives virtual “rewards” which, ideally, make it better in the work it goes on.
He said this version of the Openai O3 model is “optimized for browsing the internet and data analysis”, adding that “utilizing reasoning for searching, interpreting and analyzing huge quantities of text, images and pdfs on the internet, rotating as Reaction to information is required […] The model is also able to browse files that upload the user, design and repeat in charts using the Python tool, incorporate both created graphs and images from the websites in its answers and report specific suggestions or excerpts From its sources ”.
The company said it examined the use of Deep Research Chatgpt Deep The last examination of humanityAn evaluation that includes more than 3,000 questions at the level of experts in various academic areas. The O3 model, which feeds deeply, has achieved a precision of 26.6%, which may look like a failed score – but the latest examination of humanity was designed to be tougher than other reference points to stay in front of the model’s advances. According to Openai, the Deep Research O3 model was nominated in front of Gemini’s thought (6.2%), Grok-2 (3.8%) and Openai’s GPT-4O (3.3%).
Still, Openai notes that Chatgpt’s Deep Research has limitations, sometimes making mistakes and incorrect conclusions. Deep research can fight to distinguish valid information from rumors, the company said and often fails to carry when it is uncertain of something – and can also make formatting errors in reports and references.
For those who are worried about the impact of the genetic AI on students or anyone trying to find information on the internet, this guy in depth, well -mentioned exit probably sounds more attractive than a deceptively simple chatbot summary without references. But we will see if most users will actually submit the exit to actual resolution and double control, or if they are simply treated as a more professional text to copy the copy.
And if all of this sounds familiar, Google has actually announced a similar AI function with exactly the same name two months ago.
