OPENAI announces a new AI “agent” designed to help people conduct in-depth complex research using Chatgpt, the Chatbot platform supplied by the company’s AI.
Appropriately, this is called Deep Research.
Openai said in A blog article Published on Sunday that these new capacities were designed for “people who make intensive knowledge in fields such as finance, science, politicians and engineering and need in -depth, precise and reliable research”. It could also be useful, added the company, for all those who make “purchases that generally require meticulous research, such as cars, household appliances and furniture”.
Basically, the deep research of chatgpt is intended for cases where you do not only want a quick response or summary, but you should rather consider information assiduously from several websites and other sources.
OPENAI said that this made in -depth research available to Chatgpt Pro users today, limited to 100 queries per month, with support for more and the users of the team then, followed by the company. (Openai targets a more deployment in about a month, said the company, and the requests for the request for paid users should soon be “much higher”.) This is a geo-target launch; Openai had no liberation calendar to share for Chatgpt customers in the United Kingdom, Switzerland and in the European economic field.
To use Chatgpt Deep Research, you simply select the “deep search” in the composer, then enter a request, with the possibility of joining files or calculation sheets. (It is a web experience only for the moment, with the integration of mobile and office applications later this month.
Currently, the results of Chatgpt Deep Research are only in text. But Openai said he intended to add on -board images, data visualizations and other “analytical” outputs soon. The roadmap is also the possibility of connecting “more specialized data sources”, including “subscription -based” and internal resources, added Openai.
The big question is, how precise the deep research is? AI is imperfect, after all. It is subject to hallucinations and other types of errors which could be particularly harmful in a scenario of “deep research”. This may be why Openai said that each profound chatgpt research outing will be “fully documented, with clear quotes and a summary of [the] Think, facilitate the reference and easily check the information. »»
The jury is explained if these attenuations will be sufficient to fight against AI errors. Openai AI web search function, chatpt search, not rarely made blunders and gives bad answers to the questions. Techcrunch tests have revealed that chatgpt research has produced less useful results than Google research for certain requests.
To strengthen the accuracy of Deep Research, Openai uses a special version of its recently announced O3 IA “reasoning” model which was formed by learning to strengthen “real tasks requiring use of the browser and the python”. The strengthening of learning “teaches” essentially a model via tests and errors to achieve a specific objective. As the model is getting closer to the objective, it receives virtual “rewards” which, ideally, improve it to the task in the future.
He said that this version of the OpenAi O3 model is “optimized for web browsing and data analysis”, adding that “draws a reason for reasoning to search, interpret and analyze massive quantities of text, D ‘Images and PDF on the Internet, swivel according to the needs in reaction to the information he meets […] The model is also able to browse the files downloaded by the user, draw and iterate on graphics using the Python tool, integrate the graphics and images generated from websites in web his answers and cite specific sentences or passages from his sources. »»
The company said that it had tested deep research from Chatgpt using The last examination of humanityAn assessment that includes more than 3,000 expert questions in a variety of academic fields. The O3 model fueling deep research reached 26.6%precision, which could look like a failing note – but the last examination of humanity was designed to be more difficult than other benchmarks to stay in advance on The progress of the model. According to Openai, the deep research model O3 was ahead of Gemini (6.2%), Grok-2 (3.8%) and OPENAI (3.3%).
However, Openai notes that the deep research of Chatgpt has limits, sometimes making incorrect errors and inferences. In -depth research may find it difficult to distinguish the authority of rumors, said society, and often does not manage to transmit when it is uncertain of something – and it can also make formatting errors in reports and quotes.
For anyone who is concerned about the impact of the generative AI on students, or on anyone who tries to find online information, this type of in -depth and well cited output probably seems more attractive than a trumper chatbot summary without Quotes. But we will see if most users will really submit the output to a real analysis and double verification, or if they simply treat it as a more professional text to copy and paste.
And if all this seems familiar, Google has actually announced a similar AI feature with exactly the same name less than two months ago.
