OpenAI has released a new functionality named ๐๐๐๐ โ๐๐๐๐ถ๐๐ธ๐ฝ, that is the first of its type designed to execute multistep research tasks with ease. Thanks to this innovation, ChatGPT is capable of a thorough online investigation that is possible only within a few minutes, while a human researcher would need a couple of hours, if not days.
OpenAI sees ๐๐๐๐ โ๐๐๐๐ถ๐๐ธ๐ฝ as a major step towards the development of artificial general intelligence (AGI).
"The ability to synthesize knowledge is a precursor to generating new insights," OpenAI said. "Deep Research is a major achievement in our longer goal of AGI development."
Enabling AI with Expertise in Research:
๐๐๐๐
โ๐๐๐๐ถ๐๐ธ๐ฝ is the feature that allows ChatGPT to obtain self-motivating capabilities. The facility to find, analyze, and synthesize information from hundreds of sources is to be found in ๐๐๐๐
โ๐๐๐๐ถ๐๐ธ๐ฝ. Apparently, with the assistance of a user, one can receive a report, which is almost as detailed and accurate as that compiled by a human analyst.
It is operated by an advanced version of OpenAI's new "o3" model which is leaving out the manual data-gathering job. ๐๐๐๐ โ๐๐๐๐ถ๐๐ธ๐ฝ guarantees persistent and complete reports about different types of data including competitive analysis, policy review, and personalized recommendations of a new commuter bike.
One of the brightest spots of the product has been the full integration of transparencyโeach report includes all citations, thus, users can verify sources and trust the findings in a hassle-free manner.
It is operated by an advanced version of OpenAI's new "o3" model which is leaving out the manual data-gathering job. ๐๐๐๐ โ๐๐๐๐ถ๐๐ธ๐ฝ guarantees persistent and complete reports about different types of data including competitive analysis, policy review, and personalized recommendations of a new commuter bike.
One of the brightest spots of the product has been the full integration of transparencyโeach report includes all citations, thus, users can verify sources and trust the findings in a hassle-free manner.
Finding Secret Insights Across Several Industries:
๐๐๐๐ โ๐๐๐๐ถ๐๐ธ๐ฝ is highly efficient in finding niche and non-trivial insights. Thus, it may be used not only in the fields such as finance, science, politics, and engineering but also by the ordinary person who, for instance, would like to get the most personalized, appropriate product purchase advice or would like to hunt for a rare collectible.
A Practical Example of the Product:
๐ฎ๐ถ๐ ๐๐๐๐๐ถ๐, the CEO of OpenAI, narrated his experience with ๐๐๐๐
โ๐๐๐๐ถ๐๐ธ๐ฝ, saying that no other tool could do what ๐๐๐๐
โ๐๐๐๐ถ๐๐ธ๐ฝ did:
"I was in Japan searching for a classic NSX. After hours of fruitless searching, I was about to give up. Then, Deep Research found it instantly."
How It Works
๐๐๐๐ โ๐๐๐๐ถ๐๐ธ๐ฝ is the most effortless operating element among the features in the ChatGPT user interface. If the "Deep Research" option is just selected from the composer and the query input, the AI does the rest. Users also have the option of submitting files or spreadsheets for context.
Once the search is started, the AI starts off with the procedure, which ranges from 5 to 30 minutes long. A sidebar area will be accessible for live updates on its progress and users are alerted once the whole and well-documented report is done.
During the next few weeks, OpenAI intends to revise this feature by introducing pictures, data visualization, and interactive graphs for deeper insights.
Designed for In-Depth Comprehension Over Speed
Unlikely to GPT-4o, which mainly focuses on real-time, multimodal conversations, ๐๐๐๐ โ๐๐๐๐ถ๐๐ธ๐ฝ rates high on the depth and coverage of the information. Owning the robust citation system and the careful analysis, the AI draws the public focus from the brief summaries so typical in other sources of AI analysis to a more profound analysis of high-quality, research-grade insights.
Built for Real Problems of the Real World
This ๐๐๐๐ โ๐๐๐๐ถ๐๐ธ๐ฝ was made possible by the application of modernized training procedures like human browsing and reasoning quests. By means of reinforcement learning, progress is assured as the AI plans, backtracks, and re-refines its approach with the acquisition of new knowledge.
This powerful tool can:
โ Scan user-uploaded files for details
โ Create and modify data visualizations based on Python
โ Include media elements e.g. images, interactive web graphics
โ Use excerpts from sources as citations
Results Beyond Expected
OpenAI has tested ๐๐๐๐ โ๐๐๐๐ถ๐๐ธ๐ฝ against "Humanityโs Last Exam," the tough 3000+-question AI exam that requires real expertsโ knowledge in areas from rocket science to linguistics and ecology.
The results achieved were as follows:
โ GPT-4o: 3.3%
โ Grok-2: 3.8%
โ Claude 3.5 Sonnet: 4.3%
โ OpenAI o1: 9.1%
โ DeepSeek-R1: 9.4%
โ Deep Research: 26.6% (with browsing + Python tools)
Moreover, ๐๐๐๐ โ๐๐๐๐ถ๐๐ธ๐ฝ outperformed all other AI under mock testing scenario, at the end of the GAIA exam. It was definitely at the top of the leaderboard with a score of 72.57%.
Challenges & Future Improvements
Notwithstanding all its splendid features, OpenAI cannot deny that ๐๐๐๐ โ๐๐๐๐ถ๐๐ธ๐ฝ still leaves room for improvement. Of the said drawbacks we may list here:
โ Instances of hallucinations (misinterpretations or incorrect data)
โ Difficulty in distinguishing between trustworthy and speculative sources
โ At times, confidence calibration appears troublesome as the machine overannounces the level of confidence in its findings
โ The minor issues of formatting in reports and citations
OpenAI has informed customers that continuous product amendment efforts will take care of the problems in the long run.
Step-by-Step Introduction & Future Objectives
Deep Research's rollout plan proceeds as follows:
๐ Pro users: Immediate access (100 queries per month)
๐ Plus & Team users: Arriving soon
๐ Enterprise users: Subsequent phase
The feature will, initially, not be available in the UK, Switzerland, and EEA, but OpenAI is currently busy with expansion projects.
The Road Ahead: AI that Works
AI, as it continues to evolve, is indeed a spectacular step forward, and the Deep Research is a remarkable proof of it.
Be, therefore, prepared for the time when intelligent and independent robots will handle problems not only in the industry but in everyday life, too.
"I was in Japan searching for a classic NSX. After hours of fruitless searching, I was about to give up. Then, Deep Research found it instantly."
How It Works
๐๐๐๐ โ๐๐๐๐ถ๐๐ธ๐ฝ is the most effortless operating element among the features in the ChatGPT user interface. If the "Deep Research" option is just selected from the composer and the query input, the AI does the rest. Users also have the option of submitting files or spreadsheets for context.
Once the search is started, the AI starts off with the procedure, which ranges from 5 to 30 minutes long. A sidebar area will be accessible for live updates on its progress and users are alerted once the whole and well-documented report is done.
During the next few weeks, OpenAI intends to revise this feature by introducing pictures, data visualization, and interactive graphs for deeper insights.
Designed for In-Depth Comprehension Over Speed
Unlikely to GPT-4o, which mainly focuses on real-time, multimodal conversations, ๐๐๐๐ โ๐๐๐๐ถ๐๐ธ๐ฝ rates high on the depth and coverage of the information. Owning the robust citation system and the careful analysis, the AI draws the public focus from the brief summaries so typical in other sources of AI analysis to a more profound analysis of high-quality, research-grade insights.
Built for Real Problems of the Real World
This ๐๐๐๐ โ๐๐๐๐ถ๐๐ธ๐ฝ was made possible by the application of modernized training procedures like human browsing and reasoning quests. By means of reinforcement learning, progress is assured as the AI plans, backtracks, and re-refines its approach with the acquisition of new knowledge.
This powerful tool can:
โ Scan user-uploaded files for details
โ Create and modify data visualizations based on Python
โ Include media elements e.g. images, interactive web graphics
โ Use excerpts from sources as citations
Results Beyond Expected
OpenAI has tested ๐๐๐๐ โ๐๐๐๐ถ๐๐ธ๐ฝ against "Humanityโs Last Exam," the tough 3000+-question AI exam that requires real expertsโ knowledge in areas from rocket science to linguistics and ecology.
The results achieved were as follows:
โ GPT-4o: 3.3%
โ Grok-2: 3.8%
โ Claude 3.5 Sonnet: 4.3%
โ OpenAI o1: 9.1%
โ DeepSeek-R1: 9.4%
โ Deep Research: 26.6% (with browsing + Python tools)
Moreover, ๐๐๐๐ โ๐๐๐๐ถ๐๐ธ๐ฝ outperformed all other AI under mock testing scenario, at the end of the GAIA exam. It was definitely at the top of the leaderboard with a score of 72.57%.
Challenges & Future Improvements
Notwithstanding all its splendid features, OpenAI cannot deny that ๐๐๐๐ โ๐๐๐๐ถ๐๐ธ๐ฝ still leaves room for improvement. Of the said drawbacks we may list here:
โ Instances of hallucinations (misinterpretations or incorrect data)
โ Difficulty in distinguishing between trustworthy and speculative sources
โ At times, confidence calibration appears troublesome as the machine overannounces the level of confidence in its findings
โ The minor issues of formatting in reports and citations
OpenAI has informed customers that continuous product amendment efforts will take care of the problems in the long run.
Step-by-Step Introduction & Future Objectives
Deep Research's rollout plan proceeds as follows:
๐ Pro users: Immediate access (100 queries per month)
๐ Plus & Team users: Arriving soon
๐ Enterprise users: Subsequent phase
The feature will, initially, not be available in the UK, Switzerland, and EEA, but OpenAI is currently busy with expansion projects.
The Road Ahead: AI that Works
AI, as it continues to evolve, is indeed a spectacular step forward, and the Deep Research is a remarkable proof of it.
Be, therefore, prepared for the time when intelligent and independent robots will handle problems not only in the industry but in everyday life, too.