External analysts: Deep Research in OpenAI Log

Institutional companies need a note Deep Openai Research. It provides a strong product that depends on new capabilities, which is very good so that it can get many people out of jobs.

Deep search on the edge of bleeding from the increasing direction: combining the LLMS models with search engines and other tools to greatly expand their capabilities. (Just as this article was reported, for example, Elon Musk revealed Xai Grok 3, which claims similar AbilitiesIncluding a deep search product. However, it is too early to assess the realistic performance of Grok 3, because most subscribers have not actually received their hands yet.)

Openai’s deep search, which was released on February 3, requires a professional account with Openai, at a cost of $ 200 per month, and is currently available only for users in the United States. To date, this restriction may have limited early reactions from the global developer community, which is usually fast in dissecting new developments of artificial intelligence.

With a deep search mode, users can ask OPenai’s leading O3 model on any question. The result? The report often outperforms what human analysts produce, they surrender faster and in a small part of the cost.

How deep research works

While deep research has been discussed widely, its broader effects have not been fully recorded yet. Initial reactions praised her impressive research capabilities, despite the accidental hallucinations in her martyrdom. There was The man who said he used to help his wife Those suffering from breast cancer. He said he provided a deeper analysis than what oncologists gave about how radiotherapy. Consensus, Summarized by Professor Wharton Ai Ethan MalikIt is that its advantages are far outperforming the incomplete accuracy, as examining facts takes less time than artificial intelligence in general. This is something I agree with, based on my own use.

Financial institutions are already exploring requests. BNY Mellon, for example, sees potential to use deep research to assess credit risks. Its impact will extend through industries, from health care to retail, manufacturing and supply chain management – almost any field dependent on knowledge works.

A more intelligent research agent

Unlike traditional artificial intelligence models that try to one answers, deep research first requires clarification of questions. He may ask four or more questions to make sure they understand exactly what you want. Then it develops an organized research plan, performs multiple inspections, revishes its plan based on new visions, and is repeated in a ring until it collects a good and well -founded report. This can take between a few and a half hours. Reports range from 1500 to 20,000 words, usually cited from 15 to 30 sources with URL delicate addresses, at least, according to my use during the past week and half.

Technology behind Deep Research: Thinking LLMS and Agentic Rag

Deep research does this by integrating two technologies in a way that we have not seen before in the comprehensive market product.

Thinking llmsThe first is the advanced model of Openai, O3, which leads to logical thinking and extended thought processes. When it was announced in December 2024, O3 recorded 87.5 % unprecedented on the super-standard arc-AGI standard Designed to test new problem solving capabilities. The interesting thing is that O3 has not been released as an independent model for developers for its use. In fact, CEO of Openai Sam Altman Announced last week Instead, the model will be wrapped in a “unified intelligence” system, which will unite models with agent tools such as research, coding and more. Deep search is an example of such a product. While competitors such as Deepseeek-R1 approached the O3 capabilities (one of the reasons behind a lot of excitement a few weeks ago), Openai is It is still widely considered to be a little bit.

Agent’s rag: the second , Agent’s ragIt is a technique that has been around for about a year. Agents are used to search for information and context independently from other sources, including the Internet search. This can include other tools to find information other than the Internet through applications interfaces; Coding factors that can complete complex serials more efficiently; And search for the database. Starting, Deep search in Openai is mainly looking for the open webBut the company’s leaders suggested that he be able to search for more sources over time.

Openai competitive feature (and its limits)

Although these technologies are not completely new, Openai’s improvements that have been enabled through things like a work leap on these technologies, massive financing, and closed source development model-have taken a deep search to a new level. It can work behind closed doors, and take advantage of the comments from more than 300 million active users for the famous Chatgpt product from Openai. Openai led in these areas, for example, how to do it Step -step check for better results. It has clearly implemented the research in an interesting way, and perhaps borrowing from Microsoft’s Bing and other technologies.

Although it still insults some results from the searches, it does it less than competitors, perhaps partially because the main O3 model itself Determine the industry low of this hallucinations by 8 %. There are ways to reduce mistakes more, using mechanisms such as Thresholds of trust, quotation requirements And other Developed credibility checks.

At the same time, there are limits to the leadership of Openai and its capabilities. Within two days of Deep Research, he presented Hugingface Open source research agent called Open Deep Research That got the results were not Very far from Openai’s Merging the leading models similarly and the abolitional capabilities available freely. There are a few trenches. It seems that open competitors like Deepseek remain close to The field of thinking formsAnd microsoft’s MAGENIC-One offersFor example only two other examples.

Moreover, deep research has restrictions. The product is really effective in searching for mysterious information that can be found on the web. But in the areas where there is not much Internet and where the field experience is specially especially – whether in the heads of peoples or in private databases – it does not work at all. Therefore, this will not threaten the jobs of high hedge boxes, for example, who are pushed to speak with real experts in an industry to find out the information that is very difficult to remove, such as Bin Tumson argued in a modern publication (See the drawing below). In most cases, the deep research in Openai will affect the functions of low -skills analysts.

The most intelligent product so far

When you combine first -class thinking with the agent’s retrieval, it is not really surprising that you get such a strong product. The deep OPENAI research achieved 26.6 % on The last humanity examIt can be said that the best criterion for intelligence. This is a relatively new standard of artificial intelligence designed to be the most difficult for any Amnesty International model to complete it, and covers 3000 questions across 100 different topics. In this criterion, the deep research of Openai greatly outperforms deep research (20.5 %) and previous models such as O3-MINI (13 %) and Deepseek-R1 (9.4 %) that were not deficient with the lever of the agent. But early reviews indicate the leadership of Openai in both quality and depth. The deep search of Google has not yet been tested against this standard, but early reviews indicate Openai performs in both quality and depth.

How is it different: The first Amnesty International to replace jobs

What is different with this product is its ability to eliminate jobs. Sam Whithen, founder of Red Dragon and developer of artificial intelligence factors, was observed in a deep diving Discussion of the video With me that many people will say: “Safe foolishness, I can get these reports for $ 200 I can get from a four -way consulting company that will cost me $ 20,000.” He said this would lead to some real changes, including taking people out of jobs.

This brings me back to my interview last week with Sarthak Pattanaik, President of Engineering and Amnesty International at Bny Mellon, a major American bank.

Patanek certainly did not say anything about the repercussions of the product to get actual functions in its bank. This will be a particularly sensitive topic that any institution is likely to be ashamed to publicly address it. But he said that he could see Openai’s deep research in credit and other credit reports on the strategy (and) research, such as the comparison seller management, comparison A product against the product B. He added: “So I think everything is more on system thinking – More exploring, as it may not have a correct answer, because the correct answer can be installed once this scenario definition is obtained – I think this is an opportunity. “

Historical perspective: job loss and job creation

Technological revolutions have led to the transplantation of workers historically in the short term with the creation of new long -term industries. One of the cars that replaces horse -drawn vehicles to computers that are automated by religious work, job markets are evolving. The new opportunities created by sabotage techniques tend to spoil the new employment. Companies that fail to embrace these developments will fail to their competitors.

Openai’s altman acknowledged the link, even if it is indirect, between deep search and work. At the top of artificial intelligence in Paris last week, he was asked about his vision of artificial general intelligence (AGI), or the stage in which artificial intelligence can lead to any task that a person can. As he answered, his first reference was to the deep research: “It is a model that I think is able to do as a low percentage in all tasks in the economy in the world at the present time, and it is a crazy statement, and a year ago, I do not think anything that people think He will come. ” (See the three minute from This video). He continued: “For 50 cents of the account, you can do like $ 500 or $ 5,000 of work. Companies do this to be more efficient.”

Ready -made meals: a new era for knowledge works

Deep research represents the moment of water gatherings for Amnesty International in knowledge -based industries. By combining advanced thinking with independent research capabilities, Openai created an intelligent, faster and more effective tool than human analysts.

Wide effects, from financial services to health care to decisions. Organizations that benefit from this technology effectively will get a great competitive advantage. Those who ignore this do this in danger.

For a deeper discussion on how to work deep research for Openai, and how knowledge work is reshaped, check my in -depth conversation with Sam Witteven in our latest video:

Daily visions about business use cases with VB daily

If you want to persuade your boss at work, you have covered VB Daily. We give you the internal journalistic precedence over what companies do with obstetric artificial intelligence, from organizational transformations to practical publishing operations, so that you can share visions of the maximum return on investment.

Read with us privacy policy

Thanks for subscribing. Check more VB news bulletins here.

An error occurred.