Apple, known for its closed approach to development and unwillingness to participate in digital hype, took a significant step forward in the field of artificial intelligence. Despite its usual laconicity regarding future products, Apple announced major achievements in AI research, presenting two new scientific works.
The first study of Apple concerns the creation of animated 3D authors. The company’s specialists have proposed the Hugs (Human Gaussian Splats) method, which allows you to generate animated 3D-dutars from short video format videos. This method automatically shares a static scene and creates a fully animated human avatar in just 30 minutes, which is a hundred times faster than previous methods. This discovery can lead to new possibilities in virtual fitting, technologies for television priority and creation of synthetic media.
The second work of Apple is focused on the key problem of unfolding large language models on limited memory devices such as smartphones. Modern language models (LLM), similar to GPT-4, contain hundreds of billions of parameters, which makes their use on consumer equipment close to the impossible. The proposed system significantly minimizes the transfer of data from flash memory to limited DRAM (dynamic RAM) during the output.
“Our method includes creating a model of the output cost agreed with the behavior of flash memory, which allows us to optimize two critical areas: a decrease in the volume of data transmitted from flash memory and reading data with large, more solid blocks”, – says in the study.
Both developments can soon allow complex AI helpers and chat boots to work directly on the iPhone, iPad and other mobile devices, efficiently expanding the functionality of the latter.
Thus, Apple has demonstrated significant progress in the field of artificial intelligence, which can have a long -term effect on consumer technologies and the experience of their use as a whole.