OpenAI has changed the way it communicates about its latest AI model, o3-mini, revealing more details about its step-by-step "thought process." As reported by The Verge, the decision was made amid growing pressure from competitors, including Chinese startup DeepSeek.
According to statement by OpenAI, users of ChatGPT - both free and paid versions - will receive an updated chain of reasoning. Premium plan subscribers using o3-mini in "high level reasoning" mode will also see this update.
- Stepwise reasoning models thoroughly check their answers before producing a result. This reduces the risk of error, but slows down the decision-making process. In some cases, from a few seconds to several minutes.
- In addition to providing a deeper understanding of how the model thinks and analyzing AI performance, it also improves the quality of the user experience by allowing you to understand when the model is on the right or wrong track.
- OpenAI's competitor, DeepSeek, immediately implemented a similar approach in its R1 model.
Previously, OpenAI did not disclose the full inference process in o3-mini and its predecessors, o1 and o1-mini, in part for competitive reasons. Instead, users saw only brief summaries that sometimes contained inaccuracies. Although OpenAI does not yet demonstrate the full process, the company says it has found a balance: o3-mini was able to "think freely" and then structure its "thoughts" into detailed explanations.
It is worth recalling that Chinese DeepSeek quickly gained popularity after the introduction of the R1 model, and the startup's chatbot surpassed ChatGPT in the App Store ranking of free apps worldwide.
OpenAI and Microsoft, which is the largest investor in ChatGPT, are investigating the possible unauthorized acquisition of data by a group associated with DeepSeek.
