She went through additional stages of advanced training and security testing, similar to those used for GPT-4. GPT-4.5 combines traditional large-scale training with an approach in which the model learns intermediate steps of reasoning. This increases her "associative" abilities and reduces the likelihood of errors known as hallucinations.
In addition, the model has become more user-oriented, demonstrating the accuracy of responses and the ability to conduct a dialogue. Compared to GPT-4, its computational efficiency has increased by more than 10 times.
Photo: ru.freepik.com
However, despite the expanded knowledge, improved writing skills and more subtle personalization, the model does not offer fundamentally new features that would significantly distinguish it from previous versions.
After the official release of the new model, Sam Altman, CEO of OpenAI, announced that the company was faced with a shortage of graphics processors. In this regard, access to the new model will first be granted to users of the Pro plan, and Plus subscribers will be able to connect later.
It is expected that next week, by March 9, 2025, OpenAI will add "tens of thousands" of new processors, after which access to GPT-4.5 will be provided to all subscribers.
Earlier, we reported that Alibaba, a Chinese holding company, is investing $53 billion in AI development.
