Guessing The AI ​​Capabilities of Apple’s iPhone

Morrissey Technology – Apple has tended to hide the generative AI they are working on for the iPhone. But now they have released OpenELM, a lightweight AI model that will probably run locally on their devices. OpenELM stands for Open-source Efficient Language Models. And, there are four language models released by Apple researchers, all based on Hugging Face.

According to Apple, Hugging Face operates very efficiently for text-based tasks, for example writing emails. This model is open source and available for use by developers. OpenELM is divided into four types based on the number of parameters. For your information, parameters refer to how many variables this language model understands to determine “decisions” from its training dataset. Namely 270 million, 450 million, 1.1 billion, and 3 billion parameters FOR4D.

Microsoft and Google have also previously showcased other language models for devices such as laptops and cellphones, namely Phi-3 (supports 3.8 billion parameters) and Gemma (supports 2 billion parameters).

Previously, Apple CEO Tim Cook also promised that generative AI would be featured in Apple’s newest devices. He said that Apple had spent enormous time and effort developing generative AI. They have previously released other AI models, but they have not released an AI model that can be a foundation for commercial purposes like those released by their competitors.

Last December, Apple also released MLX, a machine learning framework that makes AI models easier to run on Apple Silicon. They also released an AI model for image processing called MGIE, whose function is to process photos using text commands. There is also Ferret-UI FOR4D, which can be used to navigate on the cellphone.