Ammon
Microsoft CEO Mustafa Suleiman said that the upcoming GPT-5 artificial intelligence model will bring performance improvements, but this development will not reach the level of AI agents.
During an interview with AI investor Seth Rosenberg, Suleiman explained that it is still difficult to make smart models follow users' instructions accurately, and to achieve this, it will require further development, as well as greater processing power in training processes, which will not be available before GPT-6, almost two years from now.
Microsoft's AI CEO added: "First, I don't think we're on a path to full autonomy for AI models, it's a highly undesirable reality. I think full autonomy is a serious risk, if a digital customer has autonomy and can make their own plans, set their own goals, and have access to their own resources, that would be a big risk."
Suleiman believes that what we are moving towards is "limited paths of independence", as smart customers can be directed towards dealing with a specific task that requires some degrees of thinking and planning, but it will be limited by limits, stressing that this future will provide a healthy environment of development and use in the event of legal regulation, for the use and development of artificial intelligence.
Suleiman pointed out that he and his team are currently working on developing the Microsoft Copilot smart assistant, by adjusting OpenAI models to work optimally with Copailot users, with a greater focus on memory capabilities, retrieving information from previous usage sessions, as well as the possibility of using it to perform specific tasks.
It is noteworthy that Suleiman's statements regarding GPT-5 may contradict somewhat with what Mira Murati, director of OpenAI technology, revealed that the next generation of artificial intelligence models for the company will not arrive before next year, especially as the company is still training it more deeply, in order to achieve safety and security.
Moratti described the evolution that GPT-5 will offer upon arrival: "The rate of evolution of AI models has become steady, the GPT-3 model was closer to the intelligence level of a young child, while the level of GPT-4 is close to that of a high school student, and the next generation will approach the level of a scientific doctorate."