OpenAI has launched a new feature called “Deep Research” for ChatGPT, aimed at assisting users in conducting detailed investigations across various fields, including finance, science, engineering, and policy.
Unlike traditional chatbot interactions that provide quick responses, this tool focuses on delivering comprehensive analyses by sourcing and evaluating information from multiple platforms.
How It Works
Users can access Deep Research by selecting it within the ChatGPT composer, entering a query, and optionally attaching relevant files or spreadsheets.
The processing time ranges from 5 to 30 minutes, with notifications sent when the results are ready. Currently, the tool generates text-based reports, though OpenAI is developing features to incorporate visual elements such as graphs and data visualizations.
Key Features
- Data Analysis and Citations: Deep Research provides well-documented results, including citations and a transparent summary of the AI’s reasoning process, making it easier for users to verify findings.
- Enhanced Decision-Making: Suitable for evaluating investment strategies, comparing products, or examining policy options, the feature helps users navigate complex decision-making scenarios.
- User Control: Pro users can make up to 100 queries per month, with plans to expand access to Plus, Team, and Enterprise accounts in the future.
Accuracy and Model Updates
To improve data interpretation, OpenAI has optimized its o3 reasoning model for browsing large datasets, including internet text, images, and user-uploaded content.
Despite advancements, the AI may occasionally misinterpret information or fail to indicate uncertainty in its conclusions.
The model’s performance was tested using a rigorous evaluation involving over 3,000 expert-level questions, achieving an accuracy rate of 26.6%.
While this surpassed other models, OpenAI acknowledges that challenges persist, such as differentiating credible sources from rumors and formatting errors in reports.
Current Limitations
- Processing Time: Queries take significantly longer to process (5 to 30 minutes) compared to standard chatbot responses.
- Misinformation Risks: Despite reinforcement learning techniques, the AI may present false information or misjudge source credibility.
- Limited Availability: The rollout is geo-restricted, excluding regions such as the U.K., Switzerland, and the European Economic Area.
Future Enhancements
OpenAI is working on integrating visual elements like charts and data visualizations to improve the presentation of complex information. Plans are also underway to connect the tool with subscription-based databases and internal resources for more precise and tailored results.
Potential Use Cases
- Professional Research: Assisting experts in finance, science, and engineering with comprehensive data analysis.
- Consumer Applications: Supporting detailed product comparisons, such as evaluating vehicles or appliances.
- Academic Support: Helping students gather verified information for research projects.
As OpenAI continues to develop Deep Research, its potential applications for professionals and everyday users alike are set to expand.
This feature represents a step toward more sophisticated AI-driven solutions for information gathering and analysis.