GPT-4.5 by Openai, who can use the new model and how to exploit its refined “personality”

GPT-4.5 is distinguished by greater competence in writing, a wider knowledge of the world and a more refined “personality” than its predecessors. The interaction with the model is designed to be more natural, thanks to its improved ability to recognize schemes and establish connections. These characteristics make it particularly suitable for writing, programming and resolution of practical problems. The model was trained using a combination of new supervision techniques and consolidated methods, such as supervised optimization (SFT) and learning by reinforcement from human feedback (RLHF), already used for GPT-4O. A significant aspect is the reduction of “hallucinations” (ie the generation of incorrect or misleading information) compared to GPT-4O and, to a lesser extent, compared to the “O1” model. Openii has placed particular emphasis on the alignment of GPT-4.5 to make it a more effective collaborator, with conversations that appear more engaging, intuitive and emotionally faded. Test conducted with human evaluators have confirmed an improvement in almost all categories compared to GPT-4O.

The CEO of Openai, Sam Altman, recognized that GPT-4.5 is a “large and expensive” model which, however, does not aim to overcome existing performance benchmark. After the initial launch for Pro, GPT-4.5 users will be made available to Plus and Team users the following week, and following Enterprise and Ed users. The launch of GPT-4.5 is part of a larger context of development in Openai, with the expected release of GPT-5 by the end of May. GPT-5, described by Altman as a “system that integrates many of our technologies”, will include the new “O3” reasoning model by Openai. The approach reflects the long -term goal of the company to combine its large linguistic models to create a more powerful system, potentially approaching what is called general artificial intelligence (agi).