ChatGPT Gains Agentic Capability for Complex Research
OpenAI has introduced a new feature called Deep Research that is revolutionizing AI-powered research. With this tool, online research tasks can be performed by ChatGPT in multiple steps, which otherwise could have taken hours or even days for human researchers to do in just minutes.
According to OpenAI, this is a significant step toward achieving artificial general intelligence (AGI). It clearly mentioned that the capability of abstracting knowledge synthesis is a significant milestone toward developing AGI.
Deep Research enables ChatGPT to independently search, evaluate, and summarize information from hundreds of sources. Whether doing competitive analysis, policy reviews, or personalized product recommendations, the tool delivers accurate, well-documented reports with full citations so users can easily verify findings.
The feature uses the upcoming o3 model of OpenAI. This will significantly reduce time and effort taken for deep research. It offers a "Deep Research" feature wherein one can trigger ChatGPT to perform deep investigations and receive an output in detail within 5-30 minutes.
Built for Real-World Challenges
Deep Research was trained with reinforcement learning on real-world browsing and reasoning tasks. Its capabilities include:
Browsing through user-uploaded files
Generating and refining graphs using Python
Embedding images, web pages, and citations
It performed exceptionally well in OpenAI’s “Humanity’s Last Exam,” a rigorous benchmark testing AI’s ability to solve complex problems. Deep Research achieved a remarkable 26.6% accuracy, outperforming leading models like GPT-4o and Claude 3.5 Sonnet. Additionally, it set a new standard on the GAIA benchmark, scoring 72.57% for real-world reasoning and multi-modal fluency.
Limitations and Future Prospects
Despite its advancements, Deep Research has some limitations:
Occasional hallucinations or incorrect inferences
Unable to tell the difference between authoritative and speculative content
Lack of calibration in confidence scores
Formatting issues in citations and reports
OpenAI is implementing this feature step by step starting with Pro users (100 queries/month), then Plus, Team, and finally Enterprise. Further, it's expanding into Europe, mobile, and desktop devices and integrating Deep Research with its Operator chatbot for real-world execution of tasks.
Frequently Asked Questions
What is Deep Research in ChatGPT?
Deep Research is an advanced agentic AI capability that enables ChatGPT to perform complex, multi-step research tasks autonomously, synthesizing knowledge from multiple online sources.
How does Deep Research differ from GPT-4o?
Unlike GPT-4o, which focuses on real-time multimodal conversations, Deep Research focuses on depth, detailed analysis, and rigorous citations for research-grade insights.
How long does it take for Deep Research to generate a report?
Depending on the complexity of the query, Deep Research takes between 5-30 minutes to complete an investigation and provide a comprehensive, well-documented report.
What are the current limitations of Deep Research?
Although highly capable, Deep Research can at times hallucinate facts, has trouble with confidence calibration, and occasionally sees minor formatting errors in citations.
Who has access to Deep Research?
Deep Research is now available to Pro users with up to 100 queries per month. The feature will be released to Team and Enterprise tiers as well, while the European region is already expanding.
Deep Research is a quantum leap in AI-based research and is promising its users a far more efficient and accurate way to perform research.