Openi has added a new ‘Deep Research’ mode to Chat GPT, which allows users to do a multi -step research on the web for more complicated tasks. In particular, Deep Research is the second AI agent of Openi, which was after the launch of the operator AI for browser -related works last month.

What is deep research? How does it work?

The deep research is powered by the latest Openi’s latest O3 reasoning model, which is better for web browsing and data analysis. Openi claims that the AI ​​agent finds, translates and analyzes a large amount of text, photos and PDF on the latest web to prepare a comprehensive report that is close to the research analyst level.

Unlike the usual search on Chat GPT, deep research questions will take between 5 and 30 minutes to return, and chat boot will send users a notification when their research is completed.

Although Deep Research currently only supports the text output, Open says he plans to add embedded images, data vegetables and other analytics outputs in the coming weeks.

Why does a deep research do not mean much?

Deep research, such as other AI tolls Open IIs still suffering from deception, and the company confesses in its blog post. In addition, Openi also says that deep research can “struggle to distinguish authentic information from rumors, and currently shows weakness in confidence calibration, which often fails to properly reveal uncertainty. Lives “.

All of this raises the question of how useful the ‘detailed report’ is really useful, if you can’t believe whether AI has made some facts or if you are getting the production of real news sources Based on. Certainly, you have the option to pass through the detailed references to confirm the information in your deep search, but it negates the purpose of having AI agent at least in the first place.

Its price Open, Open, says the deep search will be able to connect with “special data sources”, which will include either subscription or internal resources, so that “its output even stronger And be made of personal nature. “



Source link