OpenAI with its new AI agent, Deep Research, will serve its users in conducting complex and in-depth research. Deep Research is much nerdier than it sounds, it is designed to assist professionals in fields such as, finance, science, engineering, and policy making, as it is an instrument for time consuming and intensive knowledge work. Deep Research is perfect for users who seek comprehensive, accurate, and reliable research and not just quick summaries or overviews.
For people who need to cross-source data analysis, along with ensuring a more rigorous approach to gathering information from multiple resources, the new AI agent is well equipped for that. OpenAI mentioned that, “Deep Research was aimed for people who do intensive knowledge work in areas like finance, science, policy, and engineering and need thorough, precise, and reliable research. It could also be useful for anyone making purchases that typically require careful research, like cars, appliances, and furniture.”
Availability and Access:
The features can be accessed by selecting “Deep Research” in the ChatGPT composer, entering a question and optionally adding files or spreadsheets. In contrast to typical chatbots, Deep Research takes a considerable time before being fully tested, it takes from 5 to 30 minutes to complete its queries, and users are alerted once the results are available. Although OpenAI was initially restricted to the web, it is set to be released soon in both mobile and desktop versions.
The Deep Research feature is currently accessible to ChatGPT Pro users, with a monthly limit of 100 queries. Enterprise customers will have additional access to Plus and Team, as well as other levels of functionality provided by OpenAI. The access for Deep Research is currently restricted to few areas, and OpenAI has not made it available to users in the U.K, Switzerland, or the European Economic Area.
Enhanced Capabilities and Future Upgrades:
Initially, Deep Research only provides text-based outputs, but OpenAI has announced that it will soon provide embedded images, data visualizations and analytic output in upcoming versions. Moreover, the company is striving to integrate the tool with specialized data sources such as subscriptions and internal resources.
One of the primary challenges with AI-driven research is its accuracy, to overcome that OpenAI has arranged for all Deep Research outputs to be fully documented with citations and a summary of reasoning, it will be easier for users to verify information. OpenAI admits that AI-generated content is plagued by hallucinations, misinterpretations and citation errors and to facilitate user verification of information, OpenAI said, “Every ChatGPT Deep Research output will be fully documented, with clear citations and a summary of [the] thinking, making it easy to reference and verify the information.”
AI Model and Performance:
A specialized version of OpenAI’s o3 “reasoning” AI model is being used for Deep Research. This model has been trained in strengthening the learning for real-world tasks, such as searching the web and analyzing data in Python. OpenAI claims that o3 is designed to comprehend and scrutinize vast amounts of online material along with adapting as it processes information. OpenAI stated, “This version of o3 is optimized for web browsing and data analysis, it leverages reasoning to search, interpret, and analyze massive amounts of text, images, and PDFs on the internet, pivoting as needed in reaction to information it encounters.
The model is also able to browse over user-uploaded files, and plot and iterate on graphs using [a Python] tool, embed both generated graphs and images from websites in its responses, and cite specific sentences or passages from its sources.” Deep Research scored an accuracy of 26.6% in the Humanity’s Last Exam benchmark, which included 3,000 expert-level questions and was a high level difficulty test, while it outperformed other competitors such as Gemini Thinking (6.2%), Grok-2 (3.8%), and OpenAI’S GPT-4o (3.99%). Despite the scores, OpenAI asserts that Humanity’s Last Exam is designed to be more challenging than a typical AI test.
Competitive Landscape:
Google is not the only one in this race to achieve AI research, as it recently unveiled an artificial intelligence tool bearing the same name. This is just one example of the growing competition, as AI-led research assistants are increasingly being sought after as they go beyond basic chatbot functionality. Deep Research offers an exciting prospect for researchers, students, and professionals to explore knowledge facilitated by AI and is a promising area, but it’s not entirely free of obstacles.
The AI limitations in this particular agent are set out to be a caution for the users, as the tool can occasionally misunderstand authoritative sources, fail to indicate uncertainty and produce formatting errors in reports and citations. It is crucial for users to be willing to rely on the validity and critique of AI-generated content before accepting it. As AI progresses, only time can give an insight about whether the presence of AI-controlled research personnel will improve human knowledge or will they simply make it more accessible to copy and paste.
Read More: OpenAI Launches o3-Mini Reasoning Model with Free ChatGPT Access