At the recent conference of Intel Innovation, Intel General Director Pat Gelsinger announced the speedy “II -PC” – personal computers optimized to accelerate the wide range of artificial intelligence tasks without connecting to the Internet.
Intel is not alone in her aspirations. The largest IT companies, such as Apple and Qualcomm, are actively developing hardware and software solutions for the local performance of neural networks on user devices instead of remote servers in the cloud. The goal is personalized AI, so naturally integrated in our lives that we will stop noticing “artificiality” in it.
However, in the way of embodiment of this dream there were obstacles. Today, the most powerful AI models require expensive specialized iron and high-speed Internet connection. Even with access to remote computing capacities, the response of systems such as ChatGPT can slow down due to server overload.
Pallavi Mahajan , Vice President of Intel, emphasized that about 50% of the tasks of the AI are fulfilled today On user devices, mainly due to the processing of a natural language and computer vision. Nevertheless, the most powerful AI models currently depend on the data centers, which limits their speed and accessibility.
oliver Lemon , a professor of computer sciences at the University of Edinburgh, talked about his experience with large language models , noting their insufficient speed for real tasks. Instead, his team used the smaller model vicuna-13b from Meta.
companies are now actively developing tools for local AI, such as Openvino from Intel and Core ML from Apple. This not only accelerates the operation of AI, but also improves data confidentiality, since all processing occurs on the user’s device.
An example of this approach is the REWIND application for Mac and PC, which allows you to perform complex AI tasks, including the search for remote letters and files, locally on the user’s device.
New CALCOMM Snapdragon chips will allow the use of powerful AI models, such as Llama 2 from META, directly on smartphones without the need to connect to the Internet.
These innovations not only improve performance and confidentiality, but also open up new opportunities for smart assistants and household devices, providing intellectual opportunities even with limited resources.
It seems that the industry moves to the new paradigm – when the powerful AI “on the edge of the network” will become a familiar part of our devices.