OpenAI has unveiled a groundbreaking feature, ChatGPT Deep Research, designed to facilitate complex and in-depth research using artificial intelligence. This new feature is powered by a specialized version of OpenAI's o3 "reasoning" AI model, meticulously trained through reinforcement learning on real-world tasks that require extensive use of web browsers and Python tools. Available today to ChatGPT Pro users, the feature permits up to 100 queries per month, with plans to extend its availability to Plus and Team users in the near future.
The o3 model, at the heart of ChatGPT Deep Research, achieved an accuracy rate of 26.6% on Humanity's Last Exam—a benchmark devised to evaluate expert-level questions across various academic disciplines. OpenAI refined this model to optimize web browsing and data analysis, enabling users to conduct multi-step research, backtrack, and respond to real-time information efficiently.
“[This] version of the upcoming OpenAI o3 model [is] optimized for web browsing and data analysis,” – OpenAI
ChatGPT Deep Research is designed to handle complex research queries with precision. It can generate graphs, embed images from websites, browse user-uploaded files, and specifically cite hundreds of sources and passages. This functionality ensures a well-documented approach to research, enhancing reliability and depth.
“[It] leverages reasoning to search, interpret, and analyze massive amounts of text, images, and PDFs on the internet, pivoting as needed in reaction to information it encounters […]" – OpenAI
OpenAI rigorously tested the feature using Humanity's Last Exam, comprising over 3,000 expert-level questions across diverse academic fields. Despite its potential, the feature has limitations and may sometimes produce mistakes or incorrect inferences. To address this, all outputs undergo review and correction by OpenAI to maintain accuracy.
ChatGPT Deep Research is currently limited to a web-only experience but will soon integrate with mobile and desktop applications later this month. However, geo-targeting restrictions apply; the release timeline for users in the U.K., Switzerland, and the European Economic Area remains unspecified. OpenAI plans to roll out support for Plus users within approximately a month and Enterprise users subsequently.
“[The] model is also able to browse over user uploaded files, plot and iterate on graphs using the python tool, embed both generated graphs and images from websites in its responses, and cite specific sentences or passages from its sources.” – OpenAI
While Google introduced a similar AI feature named ChatGPT Search less than two months ago, OpenAI asserts its version surpasses in accuracy. Nonetheless, like Google's feature, ChatGPT Search encounters limitations and occasionally provides incorrect answers.
“[This] type of in-depth, well-cited output probably sounds more appealing than a deceptively simple chatbot summary with no citations.” – Anthony Ha
Leave a Reply