Huge AI Updates Are Coming To Apple’s iOS 18

After years of being silent, Apple is finally joining the generative AI hype. Two weeks ago, they released a paper on MM1, a multimodal large language model that can both understand text and images. This move indicates that Apple is ready to compete with companies like Microsoft, Google, Anthropic, and OpenAI which have had a head start in the AI race.

Recent rumors suggest that Apple is set to make significant AI advancements with the upcoming release of iOS 18 and the iPhone 16. The new operating system is expected to bring some of the biggest updates in the company’s history, with a strong focus on AI integration.

Here are some of the key features that are expected:

• A revamped version of Siri
• AI integration in the Messages app
• Auto-generated playlists in Apple Music
• The new iWork AI platform
• AI in Xcode for developers

Let’s delve into the details.

Biggest Update To Siri

At the center of iOS 18’s AI updates is a revamped version of Siri, Apple’s controversial virtual assistant. According to Bloomberg's Mark Gurman,

“the new operating system is seen within the company as one of the biggest iOS updates — if not the biggest — in the company’s history.”

There’s a high possibility that Apple will utilize large language models to power Siri. The company is said to be in talks with Google, OpenAI, and Anthropic to license their AI models for Siri in iOS 18.

I would love it if Apple picked Anthropic’s Claude language model because of its safety and quality of response.

Some users also found that the beta for iOS 17.4 included signs of AI changes to Siri, with evidence that four different AI models were being tested. Apple’s in-house LLM, codenamed Ajax, is a multimodal large language model with over 200 billion training parameters, possibly named Ferret.

The new Siri is expected to gain deeper integration with the Shortcuts app, allowing users to automate complex tasks with simple voice commands. For example, one might be able to take a series of five photos, turn them into a GIF, and send the resulting GIF to someone with just one command.

Imagine being able to show Siri a picture of your dog and ask to expand the background or remove unwanted elements from the image. The possibilities are endless, and it’s clear that Apple is aiming to make Siri the most advanced and capable virtual assistant on the market.

AI on Other Apple Apps

Aside from Siri, other Apple applications are also getting some AI love, including:

• iWork.ai—this is a new domain name that Apple acquired recently. This suggests that the company is working on a brand-new platform that relies heavily on AI.
• Apple Music—Apple’s music app will also gain more AI elements, such as automated playlist generation.
• XCode — Developers will be pleased to know that it’s possible that generative AI features like autocompletion are coming to XCode. Something similar to GitHub Copilot.

Other apps are also rumored to get AI upgrades, like Mail and Notes, but the specific features to be added are yet to be determined. Presentation apps like Pages and Keynote are also said to offer AI-assisted writing and slide deck creation.

Which of these rumored AI features are you most excited about?

Let’s talk about on-device AI

One of the most significant challenges in implementing AI features on mobile devices is the computational power required to run large language models.

AI models like GPT-4, Claude Opus, and Gemini Ultra have hundreds of billions or even trillions of parameters, requiring them to run on external servers and be accessed via the internet. This approach can be slow and prone to privacy and security risks.

Apple, however, is said to be working on an in-house LLM that can run locally on the device. This model, codenamed Ferret, is rumored to have 200 billion parameters, making it one of the largest models designed for on-device use.

In comparison, Google’s Gemini Nano has 3.28 billion parameters, and Anthropic’s Claude 3 Haiku has around 20 billion parameters.

Running such a large model on a mobile device is extremely difficult, as every 10 billion parameters require 1GB of memory to run. This means that Ferret’s size makes it practically impossible to run on today’s mobile devices with only 8–16 GB of built-in memory.

So, how does Apple plan to overcome this challenge?

The answer lies in a combination of hardware and software optimizations. Apple plans to run Ferret locally using flash storage, i.e., the storage space used for photos and apps on iPhones. Additionally, a report claimed that the A18 chip intended for the 2024 iPhone refresh will use a Neural Engine with more cores, allowing it to more easily process machine learning tasks.

If Apple succeeds in running Ferret locally on iPhones, it would be a game-changer in Apple’s software ecosystem.

Having an AI model run locally on the device would significantly improve privacy and security, as data would not need to be sent to external servers. Samsung has done it with their latest flagship S24.

Final Thoughts

Given the track record of Apple holding back on immature technology and the generative AI’s current state in terms of privacy concerns and hallucinations, I don’t expect iOS 18 to live up to the AI hype.

I do hope, though, that Apple proves me wrong.

If they figure out how to maximize the benefits of integrating AI into iOS 18, I think it will be a game-changer.

Only time will tell if iOS 18 will mark the beginning of a new era for mobile AI, but one thing is certain: the world will be watching closely as Apple unveils its latest innovations at WWDC 2024 on June 10th at Apple Park.

This story is published on Generative AI. Connect with us on LinkedIn and follow Zeniteq to stay in the loop with the latest AI stories. Let’s shape the future of AI together!