news 
OpenAI teases new reasoning model—but don’t expect to try it soon
In a surprise move, OpenAI has announced the development of its latest “reasoning” model, dubbed o3 and o3-mini. However, users won’t be able to get their hands on the revolutionary AI just yet.
According to the company, o3 surpasses previous performance records across the board, outperforming its predecessor in coding tests by 22.8 percent and even besting OpenAI’s Chief Scientist in competitive programming. The model also came close to acing one of the most challenging math competitions, missing only one question, and achieved an impressive 87.7% on a benchmark for expert-level science problems.
But don’t get too excited just yet – it seems that OpenAI is still putting o3 through its paces before deeming it ready for public consumption. The company has announced that it will be accepting applications from the research community to test these systems ahead of their public release, which they have yet to schedule.
The new model’s capabilities go beyond simply spitting out answers; instead, it breaks down instructions into smaller tasks to produce stronger outcomes. This feature allows for transparency in the decision-making process, as users can see the reasoning behind an AI’s response.
In a related development, OpenAI has also announced research on deliberative alignment, which requires AI models to process safety decisions step-by-step rather than simply following yes/no rules. The company claims that this paradigm is much more effective at ensuring its AI models follow safety guidelines, as demonstrated by improved performance in tests using the previous model, o1.
While we wait for OpenAI’s latest innovation to become available, fans of the platform can take solace in other recent announcements. For instance, ChatGPT has now integrated with a few new desktop apps, and its AI search engine is rolling out to everyone.
Source: http://www.theverge.com
