A joint team between the University of Edinburgh and Nvidia in the UK has developed a new method that can compress the memory that artificial intelligence models rely on to run, thereby improving the accuracy of the model processing complex tasks or significantly reducing its energy consumption while maintaining the same response speed. This also means that smaller memory will bring “stronger AI,” which is expected to break the performance bottleneck of large language models.

A joint team between the University of Edinburgh and Nvidia in the UK has developed a new method that can compress the memory that artificial intelligence models rely on to run, thereby improving the accuracy of the model processing complex tasks or significantly reducing its energy consumption while maintaining the same response speed. This also means that smaller memory will bring “stronger AI,” which is expected to break the performance bottleneck of large language models.