OpenAI recently unveiled its latest innovation, named Deep Research, which significantly enhances the functionalities of ChatGPT. This new feature allows the AI to conduct extensive and complex research with minimal human interaction, marking a notable leap toward greater autonomy for artificial intelligence.
Announced during a live stream at 4 PM Pacific Time on November 1, 2023, Deep Research is now accessible to users of ChatGPT Pro at the cost of $200 per month, with a limit of 100 research queries. OpenAI plans to widen access to users on Plus and Teams tiers across other countries as they roll out this feature more broadly.
Deep Research is crafted for individuals engaged intensively with knowledge across various fields, such as finance, science, politics, and engineering. OpenAI states, 'Deep Research is intended for those who work intensively with knowledge, needing thorough, precise research.' The intention is to provide professionals with reliable, methodical inquiries now possible through this high-powered tool.
So how does Deep Research actually work? The AI operates autonomously, generating action plans and searching the internet for information. It not only aggregates and analyzes data but also processes complex queries through Python scripts. The actual research process may take anywhere from 5 to 30 minutes to yield results.
This new feature allows users to upload various file types—including images, PDFs, and spreadsheets—enhancing the analysis's accuracy. Nevertheless, OpenAI has cautioned about the limitations of Deep Research, pointing out the potential for 'hallucinations' where the AI might create inaccurate facts, or confuse credible sources with mere rumors.
Deep Research has already made waves with its performance metrics. According to OpenAI, it achieved a record-breaking score of 26.6% on the Humanity's Last Exam benchmark—a test devised for assessing AI's competencies across diverse scientific disciplines. To put this in perspective, prior year models like o3-mini registered only 13%, indicating substantial improvement.
Performance assessments of Deep Research have revealed intriguing correlations; for example, simpler tasks tend to produce more accurate results. OpenAI has highlighted, 'The correlation suggests the more valuable the task, the harder it is to achieve good results.' This is likely due to the inherent complexity involved with more significant challenges.
An analysis of how Deep Research interacts with tools—such as web browsing—shows dramatic results. The agent's performance significantly improves relative to the number of browser interactions. At least 20 inquiries are necessary for acceptable results, with quality peaking around 60 queries.
Although the results currently are strictly text-based, OpenAI intends to integrate additional features, including built-in images and data visualizations, to enrich the research produced by the AI. This ambitious roadmap indicates OpenAI's commitment to advancing AI research tools, setting the stage for more substantial capabilities.
With offerings like Deep Research and Operator introduced this year, it appears OpenAI is pivoting its focus from merely developing foundational models to creating practical AI applications. Users eagerly await the release of comprehensive and complex AI products integrating various functionalities.
Looking forward, the AI industry is watching closely to see how well Deep Research performs across different sectors and applications. The tool has generated excitement not only for its groundbreaking abilities but for its strategic potential across industries where research and data analysis are abundant. OpenAI's continuing evolution of its AI functionalities hints at both the revolutionary potential and the challenges still upcoming.