Recently, OpenAI announced the launch of a new feature called "Deep Research," which adds a powerful information querying and analysis tool to its AI chat platform, ChatGPT. This feature is primarily aimed at users who require in-depth research, such as professionals in finance, science, policy, and engineering, as well as consumers looking for more detailed information when choosing to purchase cars, appliances, or furniture.
Unlike ChatGPT's traditional quick response method, the "Deep Research" feature is better suited for complex queries that require comprehensive analysis from multiple websites and information sources. Users simply need to select the "Deep Research" mode in the input box, enter their relevant questions, and can also attach related documents or spreadsheets to provide the system with more contextual information.
It is worth noting that this feature is currently only available on the web version, with integration into mobile and desktop applications expected to roll out later this month. After a user submits a query, the system's response time typically ranges from 5 to 30 minutes, and users will receive a notification once it is complete. This feature currently supports text output only, but OpenAI has stated that it plans to add images, data visualizations, and other analytical outputs in the future to further enrich the user's research experience.
To enhance the accuracy of "Deep Research," OpenAI has employed a special version of the newly released o3 reasoning model. This model is trained through reinforcement learning on real tasks and has capabilities in web browsing, data analysis, and processing user-uploaded files. In a test called "The Ultimate Human Exam," this model achieved an accuracy rate of 26.6%, significantly outperforming other similar models. However, OpenAI also acknowledges that this feature has limitations, as it may confuse authoritative information with rumors, and formatting errors may appear in reports.
To mitigate these issues, the outputs of "Deep Research" will include detailed citations and summaries of thought processes, allowing users to verify the authenticity of the information themselves. Currently, this feature is available to ChatGPT Pro users, with a limit of 100 queries per month. Support for ChatGPT Plus and Team users will be rolled out in the future, and an enterprise version will follow later.
It is important to note that OpenAI's new feature is similar to a namesake AI function launched by Google two months ago, aimed at providing users with comprehensive information querying services. Although OpenAI has promised to launch this feature for Plus users within a month, the release date in the UK, Switzerland, and the European Economic Area remains uncertain.
