Liam Fedus

@liamfedus · Twitter · 2025-02-03 00:05 UTC

@OpenAI's deep research is a significant milestone: AI can now directly complete economically valuable work. https://openai.com/index/introducing-deep-research/ This model also chips away at Humanity’s Last Exam with a score of ~27% (for comparison, DeepSeek-R1 at 9%, o3-mini-high is only 13%). Not sure our “last exam” will survive optimization pressure of 2025. Very proud of the team! Next we’ll work to bringing this capability to more and more of our users.

OpenAI

@openai · Twitter · 2025-02-02 23:51 UTC

Introduction to deep research 東京からのライブ🗼 https://openai.com/live/