Openai uses the company’s AI -based chatbot platform, CHATGPT, to announce a new AI “Agent” designed to help people perform in -depth complex research.
Properly, it is called a deep study.
Openai said in a blog post published on Sunday, “This new feature is designed for those who perform intensive knowledge in areas such as finance, science, policy and engineering, and needs a thorough, accurate and reliable study.” The company can also be useful for people who need prudent research such as cars, home appliances and furniture.
Basically, Chatgpt Deep Research is a case where you need to uncomfortable consideration of information on multiple websites and other sources without having to answer quick answers or summaries.
Openai is limited to 100 queries a month by CHATGPT PRO users, supporting plus and team users and enterprise. (Openai is aiming for a plus roll out for about a month from now on, and the query restrictions on paid users will soon say that it should be “very high.”) Openai did not have a release time line to share for CHATGPT customers in the UK, Switzerland and European economies.
To use Chatgpt Deep Research, select “Deep Research” from the composer and enter a query with an option to attach a file or spreadsheet. (This is a web -only experience that includes mobile and desktop app integration at the end of this month.) Deep Research can take 5 to 30 minutes to answer questions and get a notification when the search is completed. .
Currently, the output of CHATGPT Deep Research is only for text. But Openai said it will add built -in image, data visualization and other “analysis” output soon. The roadmap also has the ability to connect “subscription base” and “more specialized data source” including internal resources.
The biggest problem is how accurate the Chatgpt Deep Research is? AI is eventually incomplete. It is easy to generate other types of errors that can be particularly harmful in hallucinations and “in -depth research” scenarios. For this reason, Openai said that all Chatgpt Deep Research production will be completely documented to make it easier to refer to and confirm information in a summary of clear quotations and ().
The jury explains whether this mitigation is enough to fight the AI mistake. In the Chatgpt and Chatgpt search, Openai’s AI -based web search functions rarely provide answers to questions without creating origins. TechCrunch’s test shows that the Chatgpt search has less useful results than Google searches for certain queries.
To enhance the accuracy of Deep Research, OpenAI uses an AI model that has been trained by using a special version of the recently released O3 “Inference” AI model to “real tasks that need to be used for browsers and python tools.” there is. Reinforcement learning “teaches” the model through trial and error to inherently achieve certain goals. As the model gets closer to the goal, it is ideal to receive a virtual “reward” to improve the work in the future.
This version of the Openai O3 model added, “It is optimized for web browsing and data analysis.” Responding to the information (…) (…) This model can also explore the user uploaded files and plot and repeat the graph using the Python tool, and include both the graphs and images generated on the website. And you can quote certain sentences or phrases. source.”
The company said it tested Chatgpt Deep Research using the last test of mankind. The O3 Model Powering Deep study achieved 26.6%accuracy, which may seem like a failed grade, but the last test of mankind is designed to be ahead of the development of other benchmarks. According to Openai, the Deep Research O3 model was ahead of Gemini Thinking (6.2%), Grok-2 (3.8%) and Openai’s GPT-4O (3.3%).
Nevertheless, Openai pointed out that Chatgpt Deep Research has limitations and sometimes causes mistakes and wrong reasoning. In -depth studies may have difficulty in distinguishing authoritative information from rumors, and they can not convey them when they are uncertain and make formal errors in reports and citation, the company said.
If you are worried about those who are trying to find information online or online, this type of in -depth quoted output will sound more attractive than a simple chatbot summary without citation. However, most users can actually apply the output to actual analysis and double confirmation or simply treat them as more professional text to copy paste.
If all of this sounds familiar, Google has actually announced exactly similar AI features of the same name two months ago.
