Openai reveals a new Chatgpt agent for “Deep Search”
New Openai is announced Amnesty International “Agent” Designed to help people conduct complex, in -depth research ChatgptChatbot platform on behalf of artificial intelligence.
Suitable enough, it is called deep research.
Openai said in Blog post It was published on Sunday that this new ability was designed for “people who do extensive knowledge in areas such as financing, science, politics and engineering and need comprehensive, accurate and reliable research.” The company added that the company may also be useful for anyone who “making purchases that usually require accurate search, such as cars, devices and furniture.”
Basically, Chatgpt Deep Research aims at cases where you don’t just want an answer or a quick summary, but instead you need to think hard in information from multiple web sites and other sources.
Openai said it provides a deep search for Chatgpt Pro users today, limited to 100 quotes per month, with Plus users and team users after that, followed by Enterprise. (Openai aims to operate an over operation within almost a month from now, as the company said, and the limits of inquiries for paid users must be “much higher” soon). It is the launch of a geographical target; Openai had a timetable for the release of its participation in the Chatgpt customers in the United Kingdom, Switzerland and the European Economic Zone.
To use the Ratgpt Deep Research, you will only determine a “deep search” in the composer and then enter query, with the option to attach files or data schedules. (It is an online experience only at the present time, with the integration of mobile applications and desktop later this month).
Currently, Chatgpt Deep Research is the text only. But Openai said that he intends to add included images, data perceptions and other “analytical” outputs soon. Openai added that on the road map is also the ability to connect “the most specialized data sources”, including “subscription based” and internal resources.
The big question is, how deep is the deep search search? Artificial intelligence is incomplete, after all. It’s vulnerable to Halosa and other types of errors This can be particularly harmful in the “deep search” scenario. Perhaps for this reason Openai said that all the deep Chatgpt search outputs will be “fully documented, with clear and summary categories of [the] Thinking, which makes it easy to return and verify information. “
The jury is outside whether this reduction will be sufficient to combat artificial intelligence errors. The ISIS search feature in ChatGPT, ChatGPT search, not recurrent Makes gaffes and gives wrong answers to questions. The techcrunch test found that SHATGPT is searching I produced less useful results From Google search for certain inquiries.
To enhance the accuracy of deep search, Openai is used A special version of the “logical” O3 model artificial intelligence model This has been trained by learning reinforcement on “real tasks that require the use of the browser and the use of the Python tool.” Essential learning reinforcement “teaches” a model through experience and error to achieve a specific goal. As the model approaches the target, it receives virtual “bonuses”, making it better in the task to move forward.
He said that this version of the Openai O3 model is “improved to browse the web and analyze data,” adding that “it benefits from thinking about research, interpretation and analysis of huge quantities of text, pictures and PDF on the Internet, and declined according to the need in a reaction to the information it faces […] The model also can browse the files uploaded to the user, plan and repeat the graphs using the Python tool, and include both the created graphics and pictures of web sites in their responses, and indicate specific sentences or clips from their sources. “
The company said it tested Chatgpt deep research using The last humanity examEvaluation includes more than 3000 questions at the level of experts in a variety of academic fields. The O3 model that works to operate deep research has achieved 26.6 % resolution, which may seem to be a failed degree – but the last humanity exam is designed to be stricter than other criteria for survival before typical developments. According to Openai, the Deep search form came Gemini thinking (6.2 %), Grok-2 (3.8 %), and Openai’s GPT-4O (3.3 %).
However, Openai notes that the deep search chains have restrictions, and sometimes it makes incorrect errors and conclusions. The company said that deep research may be struggled to distinguish between reliable information about rumors, and often fails to transport when it is not sure of something – and it can also make coordination errors in reports and martyrdom.
Anyone concerned about the impact of the artificial intelligence on students, or anyone trying to find online information, this type of in -depth and in -depth output may seem more attractive than a very simple Chatbot summary without martyrdom. But we will see whether most users will be already undergoing real analysis and double verification, or if they simply treat it as a more professional text for copying.
And if all this looks familiar, then Google is already A similar AI feature announced With exactly the same name for less than two months.
