OpenAI’s new feature is deep research for ChatGPT. It runs automatically when activated in deep research. The AI can plan and execute a sequence of steps to collect information.
Should it encounter obstacles on its search path, it can backtrack and tweak itself at that time. In order to interact with the system, users can ask questions via typed text, images, and other file types. The typical time taken for generating an answer is approximately anywhere from 5 to 30 minutes.
In its description of the functionality of deep research, OpenAI does acknowledge a few limitations. Sometimes the AI produces untruths-about what it thinks-in a so-called “hallucination.” Additionally, it is sometimes incapable of distinguishing legitimate news sources from non-verified rumors. The machine might also struggle with assessing how confident it is in its responses.
This new feature comes on the heels of the OpenAI Operator web-browsing tool’s launch. Operator is analogous to Google’s, which is a Project Mariner research prototype. Those who pay $200 for their monthly subscription may send up to 100 queries a month. To lesser extents, other users such as Plus, Team, and Enterprise members are also allowed to experience the service.
An AI model that supports deep research has gotten the highest score on accuracy ever reached. It’s performing well on a benchmark test called “Humanity’s Last Exam.” With activated browsing and Python tools in operation, it got an accuracy score of 26.6 percent, which is a big jump from Intel GPT-4o’s 3.3. Subsequently, the open3 mini scored 13 percent, having been evaluated only on the text part. This is a move that points to how deep research could improve the capability of the AI further.