Openai reveals a new Chatgpt agent for “deep research”

OPENAI announces a new You have an “agent” Designed to help people conduct in -depth, complex research using ChatgptAI’s chatbot platform.

Appropriate enough, this is called Deep Research.

said Openai in blog post It is published on Sunday that this new ability is intended for “people who work on intense knowledge in fields such as finance, science, policy and engineering and need in -depth, precise and reliable research.” This can also be useful, the company added for anyone who makes “purchases that usually require careful research, such as cars, appliances and furniture.”

Generally, Chatgpt Deep Research is designed for cases where you do not just want a quick response or summary, and instead you need to consider information from multiple websites and other sources.

Openai stated that it provides in -depth research available to Chatgpt Pro users today, limited to 100 requests per month, with plus support and team users coming next, followed by Enterprise. (Openai is aimed at plus starting about a month from now, the company said.) This is a start of geo-nasochene; Openai had no time release line to share for Chatgpt customers in the UK, Switzerland and the European Economic Area.

To use Chatgpt Deep Research, you will simply select “Deep Research” in the composer and then enter a request, with the file attachment option or spreadsheets. (So ​​far it’s only on a network.

The results of the Chatgpt Deep Research are currently only for text. But Openai said it intends to add embedded images, data visualizations and other “analytical” outputs soon. The road card is also the ability to connect “more specialized data sources”, including “subscription-based” and internal resources, added Openai.

The big question is, how exactly is Chatgpt Deep Research? After all, AI is imperfect. This is predisposed to Hallucinations and other types of errors This can be particularly harmful in a “deep research scenario”. Therefore, Openai said that any output of Chatgpt Deep will be “fully documented, with clear quotes and a summary of () thinking, which facilitates the mention and verification of information.”

The jurors are outside whether these mitigations will be sufficient to combat EI mistakes. AI Web Power Search feature in Chatgpt, not infrequently, not infrequently makes bluffs and gives the wrong answers to questionsS TechCrunch test found that Chatgpt Search give less useful results than Google search for certain requests.

To increase the accuracy of Deep Research, Openai uses A special version of the recently announced O3 “reasonable” AI model This was trained by strengthening the real -world task training requiring the use of browser and Python tools. The strengthening of the essence “teaches” a model through experience and a mistake to achieve a specific goal. As the model approaches the target, it receives virtual “rewards” that ideally make it better in the task forward.

It states that this version of the OpenAi O3 is “optimized for network surfing and data analysis”, adding that “it uses considerations to search, interpret and analyze huge amounts of text, images and PDF files on the Internet, By rotating according to its needs to respond to information that encounters (…) The model is also able to view files with uploaded users, plot and repetition of graphics using the Python tool, embeds both generated graphics and websites in the answers and cite specific sentences or passages from its sources. “

The company said it had tested Chatgpt Deep Research using Humanity’s last examAn assessment that includes more than 3000 questions per expert level in different academic fields. The O3, which feeds deep research, has achieved an accuracy of 26.6%, which may seem like an unsuccessful class – but the last human exam is designed to be more difficult than other indicators to remain in front of the model of the model. According to the OpenAi, the Deep Research O3 model is in front of Twin -thinking (6.2%), GROK-2 (3.8%) and own Openai GPT-4O (3.3%).

However, Openai notes that Chatgpt Deep Research has restrictions, sometimes making mistakes and incorrect conclusions. Deep research can fight to distinguish authoritative information from rumors, the company said, and often fails to convey when it is not certain about something – and can also make mistakes in reporting in reports and quotes.

For anyone who is worried about the effects of generative AI on students or for anyone trying to find information online, this type of thorough, well-quoted output probably sounds more attractive than deceptively simple chatting summons without quotes. But we will see if most users will actually submit the result of real analysis and double checking, or they just treat it as a more professional-looking Copy-Paste text.

And if all that sounds familiar, Google actually advertised a similar feature of AI With exactly the same name, less than two months ago.