OpenAI recently unveiled a tool called Deep Research which can autonomously research topics and create reports without human assistance. Deep Research can complete tasks which would take humans 30 days of research in minutes.
This research tool, powered by OpenAI o3 model, allows for searching, analyzing and synthesizing text, images and PDFs. Initially released to ChatGPT Pro users for limited preview, plans are underway to expand access for Plus/Team users as well as Enterprise customers in due course.
What is Deep Research?
Deep Research is an agent that creates in-depth reports from multi-step thorough and precise internet research, to assist people "do intensive knowledge work in fields like science, engineering and finance".
The new feature is powered by an unreleased version of OpenAI's reasoning model o3 that autonomously searches the web for relevant information, including texts, images, PDFs and videos, while adapting its approach as more data becomes available. According to OpenAI's announcements, this system can cite or paraphrase what it finds as well.
Users can take advantage of ChatGPT's new research capability by selecting it from its message composer and entering their query. Once the tool starts to work, a sidebar will show its steps and sources as it constructs its report; when complete, users are notified via chat notification when it has arrived. At present it only searches public web content and uploaded files but will eventually have access to subscription-based databases, according to ChatGPT.
How does Deep Research work?
Deep Research autonomously plans and executes multi-step web searches upon receiving a prompt, using advanced reasoning algorithms from its forthcoming o3 model to access different sources and synthesize information into an in-depth report, complete with clear citations and an explanation of its thought process.
If an engineer needs to troubleshoot a complex technical issue, an agent can gather case studies and technical documents from online sources in minutes instead of hours; freeing up time for professionals to focus on higher value tasks according to the company.
Deep Research marks an important step toward OpenAI's aim of developing artificial general intelligence - defined as any system capable of performing any intellectual task better than humans - as defined by OpenAI itself. Last month, Operator was unveiled by OpenAI; an AI agent which can book restaurant reservations or conduct grocery shopping trips by scanning images of your shopping list and making necessary reservations on your behalf.
How long does each research task take?
OpenAI's latest ChatGPT feature -- Deep Research -- offers a more extensive research tool. Utilizing a variant of the upcoming o3 reasoning model, Deep Research can autonomously search online databases and interpret information to produce reports suitable for research analysts.
Users provide a prompt to Deep Research, then step away while it works its magic. After completion, an interactive report appears within chat and includes detailed explanations of its thought process and citations from its sources.
Each query typically takes five to 30 minutes for processing and delivery; users can leave the computer alone while it does its work. Due to its compute-intensive nature, Pro ChatGPT users will initially have access to this service with up to 100 queries each month; it will eventually expand with Plus and Enterprise subscribers joining soon after.
What is the final output?
OpenAI claims its tool can do in 10 compute minutes what would take human research analysts hours to accomplish. It will find, analyze and synthesize hundreds of online sources into a comprehensive report for those doing intensive knowledge work in finance, science or policy fields. Furthermore, OpenAI offers clear citations as well as summaries of their thought processes so users can verify information themselves.
Relying on an unreleased version of their reasoning model o3, the agent can search the internet and interpret what it finds, searching through massive amounts of text, images and PDFs for pertinent data. Furthermore, its agent was capable of adapting quickly in response to new information -- an ability demonstrated when it beat out GPT-4o on Humanity's Last Exam(opens in new window), an assessment designed to test expert-level knowledge across various fields.
Current Deep Research offerings are only available to ChatGPT Pro subscribers (at $200 per month), via a new tab within their chat interface, with 100 queries being permitted per month due to its computational intensity. OpenAI promises "significantly higher" query rates in future by developing smaller models more efficient of processing information more rapidly.
0 Comments