ChatGPT Introduces Deep Research Feature for Enhanced Data Gathering

OpenAI has introduced an exciting new feature for ChatGPT called deep research. This innovative capability allows the AI to function autonomously, meaning it can now plan and execute a series of steps to gather necessary data. According to OpenAI, the AI can even backtrack and adjust its approach based on real-time information when needed.

Rather than merely generating text responses, the deep research feature highlights its thought process by providing a summary in a sidebar. This includes citations for the information presented and an outline of the research methodology, which users can reference.

Users have the flexibility to pose questions using text, images, or even various documents such as PDFs and spreadsheets to provide additional context. The response time for queries generally ranges from 5 to 30 minutes, with the promise of future updates that may allow for embedded images and charts within responses.

OpenAI has outlined some limitations of the deep research feature. The AI might occasionally hallucinate, creating incorrect facts, or have difficulty distinguishing between reliable information and mere rumors. Additionally, the AI can struggle to assess how confident it should be in its responses.

The development of tools like deep research signifies OpenAI's commitment to enhancing the practical applications of generative AI technologies and making them more valuable. This feature is designed to operate at a level comparable to that of a research analyst. One demonstration video showcases the AI tackling a question about changes in the retail industry over the past three years, providing a detailed response complete with bullet points and tables.

This release follows another recent OpenAI feature called Operator, which allows the AI to utilize a web browser for task completion, echoing the design of the Project Mariner prototype that Google previewed last December. Google’s tool is not available for public use yet, whereas the deep research feature has been launched specifically for Pro users.

For those subscribing to the Pro plan, OpenAI is providing up to 100 queries per month for a fee of $200. There is a promise of limited access for Plus, Team, and Enterprise users to this feature, which is noted to be demanding in terms of computational resources. As responses take longer to research, more computational power is necessary. The company also mentions that all paid users can expect higher rate limits in the future, especially when a faster and more cost-efficient version of the feature is released.

A recent press release from OpenAI states that the deep research model achieved a new benchmark in accuracy on what they have termed Humanity’s Last Exam. This test evaluates the model's responses to expert-level questions. The deep research model reached an accuracy score of 26.6% when equipped with browsing and Python capabilities, significantly outperforming GPT-4o, which scored just 3.3%. Even the next highest-scoring model, the o3-mini (high), which was evaluated solely on text, accrued a score of 13%.

AI, Research, OpenAI