top of page

OpenAI launches Deep Research to cut academic manhours

Vishal Narayan


OpenAI has launched 'Deep Research,' a ChatGPT feature that will allow users to prepare expert-level reports across fields, with proper citations and sources. 


In a blog post on Sunday, OpenAI said the feature was designed for "people who do intensive knowledge work in areas like finance, science, policy, and engineering and need thorough, precise, and reliable research." 


The feature could also be used to make discerning shopping decisions such as those which require extensive research, it said.


Deep Research is powered by the company's o1 reasoning model "that’s optimized for web browsing and data analysis." 


OpenAI says its o1 "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."


To use the feature, users may select ‘Deep Research' in the composer on ChatGPT and enter their query. The agent may take 5-30 minutes to come up with the response, which will arrive in a report-like form with citations and sources appearing in a sidebar. 


OpenAI said that Deep Research has been trained using "end-to-end reinforcement learning on hard browsing and reasoning tasks across a range of domains."


The model can browse uploaded files, plot graphs using the python tool, embed both generated graphs and images from websites, and cite specific sentences or passages from its sources, it said. 


To test its robustness, OpenAI pitted Deep Research against Humanity's Last Exam, a cutting-edge multi-modal benchmark to test LLMs, and that uses a dataset of 3,000 questions. 


According to OpenAI, Deep Search scored 26 per cent in the test, which may sound like a failure, till we see others' results.


As compared to Deep Research, the accuracy level of Gemini Thinking was 6.2 per cent, Grok-2 3.8 per cent, and OpenAI's own GPT-4o 3.3 per cent. Deepseek R1, the raging Chinese startup, which upended OpenAI's hegemony recently, scored 9.1 per cent in the same test. 


Flattering results notwithstanding, OpenAI conceded Deep Research's limitations, saying the model may "struggle with distinguishing authoritative information from rumors, and currently shows weakness in confidence calibration…"


Deep Search is available to ChatGPT Pro users at present, with up to 100 queries per month. Plus and Team users will get access next, followed by Enterprise, it said. 


The feature is not yet available in the UK, Switzerland, and the European Economic Area. 


Image Source: Unsplash




 


Commentaires


Stay up-to-date with the latest news in science, technology, and artificial intelligence by subscribing to Voltaire News.

Thank You for Subscribing!

  • Instagram
  • Facebook
  • Twitter

© 2023 by Voltaire News Developed & Designed by Intertoons

bottom of page