OpenAI, a leading AI firm, has recently unveiled a new AI agent, designed to assist users in conducting comprehensive, intricate research through a platform known as ChatGPT.
This new AI tool, aptly named deep research, is specifically tailored for those involved in intensive knowledge-based professions such as finance, science, engineering, and policy, requiring meticulous, accurate, and reliable research. It’s also a handy tool for individuals making significant purchases, like automobiles, home appliances, or furniture, that demand thorough research.
In contrast to providing quick answers or summaries, the deep research feature of ChatGPT is created for situations that require careful evaluation of information from various websites and other sources.
As of now, OpenAI has made this feature available to ChatGPT Pro users with a limit of 100 queries per month. It’s expected to extend this service to Plus and Team users, and later to Enterprise users. The release timeline for ChatGPT users in the UK, Switzerland, and the European Economic Area remains undisclosed.
Utilizing ChatGPT’s deep research is as simple as selecting the “deep research” option in the composer and inputting a query. Users can also attach files or spreadsheets. The research may take anywhere from 5 to 30 minutes, and users will be notified once the search is completed.
While the outputs of ChatGPT deep research are currently text-only, OpenAI plans to incorporate embedded images, data visualizations, and other analytical outputs. The AI giant also aims to connect more specialized data sources, including subscription-based and internal resources.
However, as with any AI, ChatGPT deep research is not flawless. Given that AI is prone to errors, every output of ChatGPT deep research will be fully documented with clear citations and a summary of the thinking process to ensure easy referencing and verification of the information.
In terms of enhancing deep research’s accuracy, OpenAI is employing a unique version of its recently launched o3 “reasoning” AI model, which was trained using reinforcement learning on real-world tasks. This version of the OpenAI o3 model is optimized for web browsing and data analysis.
ChatGPT deep research was tested using Humanity’s Last Exam, a challenging evaluation comprising over 3000 expert-level questions across various academic fields. Despite its limitations, the deep research o3 model outperformed other models, achieving an accuracy of 26.6%.
However, OpenAI acknowledges that ChatGPT deep research has its limitations, occasionally making errors and incorrect inferences. Notwithstanding the potential challenges, this type of in-depth, well-cited output is likely more appealing to users seeking comprehensive and accurate information online.
Interestingly, Google announced a similar AI feature with the same name just a few weeks ago.
