Apple is rumoured to be taking a different approach to deploying generative AI in iOS 18 and in next-gen iPhone models, by keeping all processing on the device rather than sending it to the cloud and back to yield answers.
Those reports appear well-grounded considering Apple’s robust approach to user privacy and past form. Keeping requests entirely local will likely be faster and more secure than sending the information into the stratosphere and back.
However, it’s unclear whether the on-device models will have access to the same wealth of knowledge as models that consult the cloud, like Google’s Gemini and OpenAI’s ChatGPT. Samsung, for example, uses a combination of on-device prowess and cloud processing for it’s Galaxy AI. Apple is rumoured to be mulling a deal with Google to fill in the gaps by bringing Gemini to iPhones.
It’s also unclear whether using an on-device model will limit the new features to the next-generations of iPhone hardware, rather than existing devices.
Get a ‘Like new’ iPhone 12 Pro Max for just £449
You can now buy a ‘Like new’ refurbished iPhone 12 Pro Max for just £449, which is a decidedly mid-range price.
- GiffGaff
- ‘Like new’ refurbished condition
- Now £449
Now there’s a little more evidence to suggest that’s precisely the route Apple will look to take. This week, Apple has released a number of open source large language models that are, you guessed it, built for on-device processing.
As MacRumors reports, the company has published a white paper on the launch of eight OpenELM (Open-source Efficient Language Models) within the AI community on the Hugging Face app.
Apple reckons the performance is on a par with other LLMs that do utilise help from the cloud, despite receiving less training. It hopes developers will get involved in to help move forward the trustworthiness and reliability of results.
The paper explains: “To this end, we release OpenELM, a state-of-the-art open language model. OpenELM uses a layer-wise scaling strategy to efficiently allocate parameters within each layer of the transformer model, leading to enhanced accuracy. For example, with a parameter budget of approximately one billion parameters, OpenELM exhibits a 2.36% improvement in accuracy compared to OLMo while requiring 2× fewer pre-training tokens.
“Diverging from prior practices that only provide model weights and inference code, and pre-train on private datasets, our release includes the complete framework for training and evaluation of the language model on publicly available datasets, including training logs, multiple checkpoints, and pre-training configurations. We also release code to convert models to MLX library for inference and fine-tuning on Apple devices. This comprehensive release aims to empower and strengthen the open research community, paving the way for future open research endeavors.”
Do you have high hopes for Apple’s dive into generative AI within iOS 18 and future iPhones? Let us know @trustedreviews on Twitter.
