OpenAI introduces new, super-talented AI models that look humans in the eye

The AI ​​race is absolutely frenetic and any pause, even a small one, can potentially be very costly for the players who operate in this vibrant field of activity. OpenAI knows this and that is why it has decided to take a run-up with the presentation of o1, a new series of AI models that pride themselves on being able to reason and are specifically designed to solve complex problems and work on tasks fraught with complexity.

OpenAI’s o1 models are trained to take longer than other AI systems to complete the tasks assigned to them. and thinking about the problems they face in a similar way to how a human would.

According to OpenAI, its new AI models “They can refine their thinking process, try different strategies and identify their own mistakes.”

The performance of OpenAI o1 is equivalent to that of the PhD students when working on problems in physics, chemistry or biology.

Not in vain, OpenAI’s o1 models have scored 83% on a qualifying exam for the International Mathematical OlympiadBy contrast, GPT-4o, OpenAI’s previous AI model, scored just 13% on the same test, the company led by Sam Altman said.

At the moment, only two OpenAI o1 previews are available.

According to OpenAI, its new o1 models could be used, for example, by medical researchers in cell sequencing data, by the physicists to generate the complex mathematical formulas needed in quantum optics, and for the developers to create and execute multi-step workflows. OpenAI models also demonstrate exceptional performance in mathematics and code generation.

Within OpenAI, the models that the company is now presenting had first been codified under the name Q* and then under the name Strawberry.

OpenAI has taken a cautious approach when launching its new AI models. AndFor the moment, it will be limited to launching a couple of “previews” of the two models that make up the o1 series.

Those with current ChatGPT Plus or Teams accounts will be able to access “o1-preview” and “o1-mini” (which is a faster version of the model) in the OpenAI chatbot navigation menu.

For its part, Developers and researchers will have access to the new models from ChatGPT and the application programming interface.

OpenAI’s new o1 models will not be able to connect to the internet initially. Users will also not be able to upload files and images to the models. OpenAI insists that it has strengthened the security measures of its new models and has also informed the US federal authorities of their complexity.