OpenAI's Deep Research feature, launched for ChatGPT Pro users, offers a glimpse into the future of AI technology. This update goes beyond simple text generation, providing in-depth analysis and information for complex questions. Below is a summary of this feature's key information.
1. Key Features of the Deep Research Function
-
Multi-Stage Investigation System
Deep Research gathers, analyzes data, and provides results in a multi-stage process for user-inputted questions. For example, instead of simply searching for information, the process involves planning and execution stages, enabling more structured answers.
-
Real-time Situation Handling
The AI can adjust its plans or expand its research in response to real-time information changes as it gathers data. This is a significant difference from the simple text responses of existing GPT models.
-
Process Visualization and Citation
The process of answering is shown through a sidebar in the form of a research steps summary and citations. This increases user confidence in the results and allows for immediate review of reference information.
-
Support for Various Input Methods
In addition to text, users can utilize images, PDFs, and spreadsheets to provide additional information. This feature allows for visual data review and consideration of structural context.
2. How to Use
-
Basic Procedure
- Users provide information in various forms, including text, images, and files.
- The Deep Research system sets analysis stages appropriate to the question and gathers data.
- If the solution to the given problem is not immediately clear, it supports an iterative searching process with focused investigation and result reflection.
- Finally, the answer is output in a structured format, including tables, charts, and lists.
-
Response Time
On average, it takes 5 to 30 minutes to generate a response. This can vary depending on task complexity and data requirements.
3. Limitations and Precautions
-
"Hallucination"
The AI may generate non-existent information or draw incorrect conclusions based on unreliable data.
- To prevent this, always review the provided summaries and source information.
-
Accuracy Issues
Deep Research may have difficulty distinguishing between authoritative information and rumors. Verification by an expert is necessary when used for important projects.
-
Confidence Level Indication
The model attempts to indicate its confidence in the results, but this is still in its early stages and has clear limitations.
4. Competition and Comparison
Companies like Google are also developing research-focused AI tools (e.g., Project Mariner). However, OpenAI's Deep Research is already deployed to subscribers and boasts high accuracy and speed.
- Humanity’s Last Exam AI Benchmark Results
- Deep Research model: 26.6% accuracy
- GPT-4o model: 3.3% accuracy
This demonstrates the power of OpenAI's new model for research and analysis.
5. Pricing and Access Conditions
- Deep Research is available to ChatGPT Pro ($200/month) users, with 100 queries allowed per month.
- Phased expansion of limited access is planned for ChatGPT Plus, Team, and Enterprise users.
- The high computational resources consumed in processing responses have been mentioned as a reason for potential future cost increases.
6. Business and Personal Use Cases
-
Business Sector
- Market change analysis: Example: understanding retail market changes over the past three years.
- Complex data analysis and report generation.
-
Personal Sector
- Academic research: Gathering information for a thesis topic.
- Hobby-based exploration: In-depth learning about areas of interest.
OpenAI's Deep Research demonstrates that AI technology can perform comprehensive thinking and analysis similar to humans, going beyond simply retrieving information easily. If this feature stabilizes completely, it is expected to provide innovative value to individuals and large organizations alike.
*Source URL:
https://www.theverge.com/news/604902/chagpt-deep-research-ai-agent
Leave a Reply