OpenAI unveils a new ChatGPT agent for ‘deep research’

OpenAI is a new announcement You have a “agent” Designed to help people using deepest, complex research to people ChatziptOrganization’s AI-powered chatbot platform.

Properly sufficient, it is called deep research.

Opena said Post a blog On Sunday, it was published that this new power was “intense knowledge in territories such as money, science, principles and engineering, and was designed for people who work for their thorough, definite and reliable research.” It may also be effective, the company has added, “Why” why to buy a purchase that usually requires careful research like cars, equipment and furniture. “

Basically, the chatzPT Deep Research is created for examples where you do not want to just answer or summarize, but instead you need to consider information from multiple websites and other sources.

OpenAI says that it is currently available in deep research for pro users for pro users, with 100 questions per month, with the support of the plus and team users, followed by the enterprise. (Open is targeting a plus rollout within about a month from now, the company says) This is a geo-launched launch; There was no release period for sharing the ChatzP for customers in the UK, Switzerland and European Economic Zone of Open.

To use ChatzPT Deep Research, you simply select “Deep Research” in the composer and then enter a query with an option to attach files or spreadsheets. (It is a web-experience experience with the Mobile and Desktop App integration later this month).

Currently, the output of the chatzPT Deep Research is only text. However, Open is said that it wants to add embedded images, data visualization and other “analytical” outputs soon. Openai added, “Subscription-based” and “more specialized data sources”, including internal resources and also have the ability to connect.

The big question is, how precise is the deeper research of the ChatzPi? AI is incomplete, above all. It is prone to Hallucinations and other types of errors It can be particularly harmful in a “deep study” scene. This is why Openai said each chatzipti deep research output “would be fully documented, including clean quotes and summaries [the] Thoughts, the information makes it easier to mention and verify ”” “

The jury has come out on whether these mites will be enough to deal with the AI ​​mistakes. OpenAI’s AI-powered Web Search feature is ChatzPT, ChatzPT Search, rarely Gaff makes and answers the wrong answer to the questionThe In search of the ChatzPT that has been shown in the TechCrunch exam Produced less useful results Than Google search for specific questions.

To enhance the accuracy of deep research, using OpenAI Its recently announced and a special version of the AI ​​Model of AI “Argument” It was trained by learning reinforcement on “Browser and Python Equipment Real-World Works”. Learning reinforcement basically “teaches” a model through trial and error to achieve a specific goal. As the model approaches the target, it receives virtual “prizes” that ideally, makes it better in order to move forward.

It says that this version of OpenAI and 3 models is “optimized for web browsing and data analysis”, adding “It presents a large amount of text, image and PDF search, explanation and analysis on the Internet, pivoting it as necessary as necessary Response […] The model is able to embed both graphs and images generated from websites to the users uploaded files, plots and repetitions in graphs using Python tools, and quoting specific sentences or passages from its source. “

The company has said that it has tested the ChatzPT deep research using it The last test of humanityAn evaluation that includes more than 3,000 expert-level questions in various academic cases. O 3 model Powering Deep Research achieved 26.6%of which could look like a failed grade – but the last test of humanity was designed to be stronger than other criteria to remain ahead of model progress. According to Openai, deep research and 3 models came forward Gemini (6.2%), Grock -2 (3.8%), and Openai’s own GPT -4O (3.3%).

Nevertheless, Openi note that the ChatzPT has a limit to Deep Research, sometimes incorrect and incorrect references. Deep research can struggle to separate authoritative information from rumors, the company said and often fails to reveal when it is uncertain about something – and it can also make formatting errors in reports and quotes.

For anyone who is concerned about the impact of the generator AI on students, or for anyone who is trying online, this type of deeper, well-covered output probably seems to be more attractive than a chatter without a quote. However, we will see that most users are actually subject to real analysis and double-cheeking outputs or if they only consider it as a more professional-led text in copy-paste.

And if all these seem to be familiar then Google is actually Similar AI features have announced Two months ago with exactly the same name.