Apple Develops Method of Running LLM on iPhone


The generative AI service works with a multi-language model (LLM) running on a server. Users can then use any chatbot either in a special application or in a website. Google has successfully run LLM on Pixel 8 Pro with Gemini Nano and Mozilla has introduced Llamafile to run LLM on personal computers.

Recently, Apple has successfully developed a method of running LLM on the iPhone which is a device with limited memory. This is possible because Apple uses a Windowing method that recycles data that has been processed so as not to consume memory each time new data is processed. The second is Row-Column Bundling which collects data and reads it at once instead of reading data one by one. This method is said to speed up the AI's understanding of things.

According to this Apple study, these two methods allow AI to be processed 4-5 times faster on the CPU and 20-25 times faster on the GPU. This means that, in the future, the iPhone could have artificial intelligence capabilities on Siri that are smarter, have better photography, can translate foreign languages in real time or even bring smoother augmented reality features.

According to sources, Apple is developing a competitor to ChatGPT, Copilot, Claude and other generative AIs with a project dubbed Ajax. It is expected to be launched together with iOS 18 next year.

Previous Post Next Post

Contact Form