Apple announced a breakthrough in using artificial intelligence (AI) based on large language models (LLM) on devices with limited hardware capabilities. In the future, this will help run ChatGPT and other formidable AI on smartphones and tablets. The study was published on: To collect scientific articles Arxiv.
The engineers tested their approach on the open-source Falcon 7B LLM neural network developed by the Abu Dhabi Technology Innovation Institute. According to the document, experts managed to make the AI work with a volume of available memory as low as half of the required one and accelerate the data throughput by 4-5 times and 20-25 times compared to the usual method of using a central processor. Times when graphics accelerator is used.
Today’s intelligent chatbots, such as OpenAI’s ChatGPT and Google’s Bard, run in data centers using massive computing power and consuming large amounts of energy.
Porting these systems to mobile devices is a very complex technical challenge. However, if this can be done, portable AI will be able to provide privacy to users as all information will be stored on the device and not in a cloud database on the company’s servers.
“Our work not only provides a solution to a current computing problem, but also sets a precedent for future research,” Apple said in its paper.
Previous scientists I learnedthat complex AIs independently learn to create simpler neural networks.