ChatGPT's New Deep Research Feature: A Step Forward in Autonomous AI

OpenAI has recently introduced an exciting new feature for ChatGPT known as deep research. This capability allows the AI to operate independently, enabling it to plan and execute a detailed sequence of steps to gather the necessary data. When facing challenges, it can backtrack and make adjustments based on real-time information.

Unlike the previous versions that merely generated text, the deep research feature provides users with a sidebar that summarizes the AI's research process. This sidebar includes citations and an overview of the methods used, making it easier for users to understand how the information was obtained.

With this upgrade, users can pose questions in various formats, including text, images, and even additional documents like PDFs or spreadsheets. This added context helps the AI provide more informed responses, which typically take between 5 to 30 minutes to develop. Future improvements are also anticipated, with plans to include embedded images and charts in the responses.

However, OpenAI has also highlighted some limitations of the deep research feature. The AI can sometimes generate inaccurate information, often referred to as 'hallucinations,' and may struggle to accurately differentiate between credible facts and rumors. It may also have difficulty assessing the level of certainty in its responses.

The introduction of deep research aligns with OpenAI's broader goal to make generative AI tools more useful and justifiable for users who are willing to pay for enhanced functionalities. The company claims that this feature operates at a level comparable to that of a research analyst. In a demonstration video, the AI was tasked with providing insights into changes in the retail industry over the past three years, resulting in a response that included both bullet points and tables.

This announcement follows closely after OpenAI's launch of another tool called Operator. This new Operator tool can utilize web browsing to complete tasks for users. The deep research feature is somewhat similar to Google's Project Mariner, which was briefly showcased in December but is still not publicly available.

OpenAI is offering this deep research feature to Pro users, with subscriptions costing $200 per month for up to 100 queries monthly. There is also a mention of "limited access" for Plus, Team, and eventually, Enterprise users. The deep research functions are described as being highly demanding in terms of computational resources, which means the longer a research task takes, the more computational power it requires. All paid users will benefit from increased rate limits in the future when a faster and more efficient version of the system is released.

A press release also cited that the model behind the deep research feature achieved a new high score for accuracy on an AI benchmark known as "Humanity's Last Exam," which is designed to assess responses to expert-level questions. The deep research model reached an accuracy of 26.6 percent when both browsing and Python tools were enabled, far surpassing the previous GPT-4o model's score of 3.3 percent, as well as the next highest model, which scored 13 percent while evaluated solely on text.

AI, Research, Technology