Introduction to Deep Research

Summary

Key Points

• OpenAI believes agents will transform knowledge work by streamlining processes and increasing productivity for both enterprises and consumers. 0:37
• Deep Research is a new agentic offering designed to perform multi-step research on the internet, discovering, synthesizing, and reasoning about content. 1:31
• Unlike traditional models, Deep Research can take extended periods, from five to thirty minutes, to complete tasks, which is considered a positive attribute for autonomous, unsupervised work. 2:08
• The output of Deep Research is a comprehensive, fully cited research paper, similar to what an analyst or expert would produce. 2:28
• The model can handle complex queries, such as market research for new product launches or detailed product comparisons for personal purchases, often requiring clarifying questions upfront. 2:36
• Deep Research is accessible starting today on ChatGPT Pro and will soon be rolled out to Plus, Team, Education, and Enterprise tiers. 3:17
• Deep Research is powered by a fine-tuned version of the O3 reasoning model and trained using reinforcement learning on browsing and reasoning tasks, enabling it to plan, execute, and backtrack as needed. 9:17
• Deep Research can browse user-uploaded files, use Python for calculations and plots, embed images, and cite specific sentences and passages from its sources. 9:43
• The model demonstrates advanced capabilities, achieving new highs on benchmarks like Humanity's Last Exam and HELM, and performing complex tasks that would take humans many hours. 10:01
• Future developments for Deep Research include connecting to custom contexts and enterprise data storages, furthering its role in OpenAI's AGI roadmap for agents that can think longer and solve more difficult tasks autonomously. 19:42