Recently, the company revealed a new reasoning model called “o3” the successor of the popular “o1” reasoning AI model which was released a couple of months ago. This new OpenAI “o3” reasoning model promises to have higher accuracy compared with the previous model.
In the Friday announcement, stated that they are working to launch soon two versions of the “o3” new reasoning model, the OpenAI “o3” and “o3-mini”. These two models are expected to be launched by the end of January if the company develops all the impressive capabilities. If the “o3” is planned to be a reasoning AI model that can handle more complex tasks in order to offer more accurate responses, the “o3-mini” is created especially to handle more particular tasks.
Currently, these two OpenAI new reasoning models are not available for public use due to the fact that they are still part of internal safety testing rounds, but the company opened a list where all tech researchers can join in order to test a preview of the reasoning AImodel before the official launch.
Most people are probably wondering why OpenAI decided to name it “o3” instead of the obvious “o2” since the first OpenAI reasoning model is called “o1”. Well, according to different sources, the company avoided naming the model “o2” in order to prevent a potential conflict with the British telecom provider named O2.
According to OpenAI, the new OpenAI “o3” reasoning AI model is capable in some cases of reaching the AGI capabilities but with some minor limitations.
Compared with most AI models, all reasoning models including the OpenAI new reasoning model “o3” are designed to fact-check themselves which is a good capability to prevent errors in responses. But it’s important to mention that the fact-checking capability usually comes with latency, taking seconds to minutes longer for models to come up with an accurate answer. But the advantage here is that all reasoning models can be more trustworthy in industries such as science, physics, and mathematics.
The “o3” model was trained using reinforcement learning in order to “think” before offering an answer, described by the company as a “private chain of thought”. This offers the option for the model to reason through different tasks, plan ahead, and even execute a series of actions over time and even execute a series of actions over time to come up with a solution.
It’s important to mention that a new capability for the “o3” reasoning AI tool compared with the previous model is the option to adjust the thinking time or the reasoning time. Now, the two models “o3” and “o3-mini” have the option to change the reasoning time to low, medium, or high, where the higher the computing means a better performance on different tasks.
It remains to be seen when these two reasoning AI models will be officially released, and what other impressive capabilities will OpenAI develop.
Stay tuned for more!
.webp)
