When we briefed people on 🍓 before o1-preview's release, seeing the CoT live was usually the "aha" moment for them that made it clear this was going to be a big deal. These aren't the raw CoTs but it's a big step closer and I'm glad we can share that experience with the world. https://t.co/72ZpPrHmfK
— Noam Brown (@polynoamial) February 6, 2025
Thinking models, such as o3-mini, thoroughly check the facts before providing a result. Although this requires more time to prepare the answer, it significantly increases their accuracy. Along with the answer, the model also provides a brief summary of its thought process.
This prompted OpenAI to provide users with more information about the model's reasoning. The company still won't show the full process of o3-mini's "reasoning", but said it has "found a balance": o3-mini can "think freely" and then organize its "thoughts" into more detailed summaries.
"To improve clarity and safety, we’ve added an additional post-processing step where the model reviews the raw chain of thought, removing any unsafe content, and then simplifies any complex ideas, explains OpenAI. "Additionally, this post-processing step enables non-English users to receive the chain of thought in their native language, creating a more accessible and friendly experience."