12 Days of OpenAI
Revisit the highlights below.
Deliberative alignment: reasoning enables safer language models
Introducing our new alignment strategy for o-series models, which are directly taught safety specifications and how to reason over them.
Introducing our new alignment strategy for o-series models, which are directly taught safety specifications and how to reason over them.
We're offering safety and security researchers early access to our next frontier models.
Introducing OpenAI o1, Realtime API improvements, a new fine-tuning method and more for developers.
Explore the potential of ChatGPT search, a speedy and timely way to get answers from relevant web sources that debuted in October 2024.
Discover all the possibilities of canvas, an interface for writing and coding in collaboration with ChatGPT that we launched this past October.
Create videos in various formats, generate new content from text, or enhance, remix, and blend your own assets.
Sora builds on learnings from DALL-E and GPT models, and is designed to give people expanded tools for storytelling and creative expression.
This fine-tuning technique is designed to improve the performance of our reasoning models on verifiable domain-specific tasks. We find it's most useful on tasks that can be easily graded against the “correct” answer and where accuracy is important—in industries like math, science, legal, healthcare, and financial services.
The updated OpenAI o1 system card builds on prior safety work, highlighting robustness evals, red teaming insights, and safety improvements.
