Welcome to this Tuesday edition of
The AI Reverie 

In this newsletter I will go break down the AI-announcements from Apple, and translate what it actually means (for us mere mortals). I spent quite some time going through comments and initial reactions for this edition.

Yesterday Apple released “Apple Intelligence (read about it here) which means that every Iphone and device will get an AI that is trained on your OWN data. It’s supposedly only going to be run locally - meaning no one else will get YOUR data, but I guess we’ll have to wait and see how that plays out.

They also released a large whitepaper explaining HOW the AI model will be trained, and I’ve gone through it and will try to explain under. I went through feedback from tech-savvy people on Hacker News and other sources, and I present a summary of the feedback in this newsletter.

Let me know what you think about today’s email, as it’s a new format compared to before (longer and much more detailed). You can let me know through the poll at the bottom of the email, or by replying to this email (i read every reply).

Have a great day,

Espen Opdahl
Founder of The AI Reverie

Tuesday 11th of June
Today we’ll cover

• TOGETHER WITH GAMMA

• 🍎 Apple's New AI Technology Explained: What Does On-Device Models Mean For You

TOGETHER WITH GAMMA

An entirely new way to present ideas

Gamma’s AI creates beautiful presentations, websites, and more. No design or coding skills required. Try it free today.

AI ROBOTICS
🍎 Apple's New AI Technology Explained: Everything You Need to Know

Apple has unveiled its latest AI advancements, integrating smart adapters and robust privacy measures into everyday devices.

I went through the announcement from Apple, feedback and the debate on this announcement on the Hacker News. The discussion revolves around the novelty of these features, the focus on user experience and security.

The Magic of Adapters

At their 2024 Worldwide Developers Conference, Apple rolled out "Apple Intelligence," an AI system embedded in iPhones, iPads, and Macs. Central to this innovation are "adapters," small modules that plug into the main AI model to perform specific tasks. Think of the main AI model as a versatile machine that can handle many general jobs, but sometimes it needs a special tool to do a particular task really well.

For example, imagine you want your device to summarize a long email or create a fun image for a chat. Instead of the main AI model trying to do everything on its own, it uses an adapter specifically trained for summarizing text or creating images. These adapters act like special helpers that fine-tune the AI's abilities for specific jobs, making it more efficient and accurate. Similar to Agents, we teach you how to use in the community.

For the ordinary person, this means that their devices can perform tasks more quickly and accurately. Whether it's getting a clear and concise summary of your messages, automatically sorting your photos, or generating creative content, these adapters ensure the AI works smarter and more effectively, providing a better overall experience.

Some tech enthusiasts on Hacker News were quick to point out that Apple's "adapters" are similar to an existing technique called LoRA (Low-Rank Adaptation). It looks like, Apple has repackaged a known method, presenting it as something new and shiny.

While the tech-savvy might see this as old news with a new name, for everyday users, these adapters could offer significant practical benefits, making their devices smarter and more versatile.

Apple's Tech Philosophy

Apple has always been about taking existing technology and refining to make it user-friendly and beautiful. Their new AI models are no different, designed to integrate seamlessly with their devices, enhancing everyday tasks without compromising performance or privacy.

A commenter noted that Apple’s strength lies in perfecting technology for mainstream use. Just like how Apple wasn’t the first to create smartphones but revolutionized the market with the iPhone, they're now aiming to do the same with AI.

Apple is not always first, but they often make the best, and most well-designed, version of existing tech. They focus on user experience and reliability, which has always set them apart.

Apples focus: Privacy First

Apple is serious about privacy. Their AI processes most data on your device, keeping your personal information safe. They’ve even set up a system for web publishers to control how their content is used for AI training.

Some users appreciated Apple’s emphasis on privacy, noting it as a strong point compared to other tech giants. However, there's some skepticism about the effectiveness and transparency of these measures.

Apple's privacy-first approach is commendable, but users want more assurance and transparency to fully trust these claims.

Powering Apple AI with Apple Silicon

Apple’s AI models run on their own silicon chips rather than relying on NVIDIA's GPUs, which are common in other AI applications.

So, what exactly is "Apple Silicon"? Apple Silicon refers to the custom-built processors that Apple designs and manufactures for their devices. These chips, like the M1 and A15, are specifically optimized for performance and efficiency. Unlike general-purpose processors, Apple Silicon is designed to handle specific tasks like AI computations very efficiently. This means that tasks like voice recognition, image processing, and running AI models can be done faster and use less power.

Using their own chips allows Apple to tightly integrate hardware and software, leading to better performance, longer battery life, and potentially lower costs in the long run, as they don't have to rely on third-party chips from companies like NVIDIA. This move enhances the overall user experience by ensuring that Apple devices are not only powerful but also efficient and cost-effective.

One commenter highlighted that using Apple’s silicon could challenge NVIDIA’s dominance in the AI hardware market. However, there's a debate on whether Apple’s pricing will make this a viable alternative for everyone.

Apple’s move to use its own chips for AI tasks is bold and could shake up the hardware market, but it remains to be seen if they can deliver on performance and cost-effectiveness.

Making AI Accessible

Apple’s AI enhancements are designed to work seamlessly across all their devices, ensuring that even regular users can benefit from powerful AI features without needing special hardware.

A Hacker News user mentioned that this standardization is a big win for developers. Unlike Windows and Linux, where hardware can vary widely, Mac developers can assume all users have capable processors.

Apple’s approach simplifies things for developers and ensures that consumers get consistent and reliable AI capabilities across all devices.

Benchmark Comparisons

Apple compared their AI models with both open-source models and commercial ones like GPT-3.5 and GPT-4. They claim their models perform well, especially in terms of reducing harmful content and improving user satisfaction.

While some were intrigued by benchmarks showing Apple’s models as safer and more helpful, others are curious about the specifics and fairness of these comparisons.

The tech community is interested but cautious. They appreciate Apple’s focus on safety and quality but want more transparent comparisons to fully trust the claims. It’s also interesting to note that GPT4o is conveniently not included.

Why Should You Care?

Apple's introduction of on-device and server foundation models is very intriguing and seems like a step in the right direction for AI mass-adoption.

While some see it as a repackaging of existing concepts, others appreciate the practical improvements and strong focus on privacy. Apple's knack for refining technology to make it user-friendly and reliable continues to impress, even if they're not always on the cutting edge of innovation.

I will be watching closely to see how these models perform in the real world with real-world. What’s so exciting, is that this is rolled out to EVERYONE with an Iphone or Apple device, making it possible to get user feedback quickly.

Recommended reading
If we had to recommend other newsletters

Agent. AI
Written by Dharmesh Shah. Dharmesh Shah is co-founder and CTO of HubSpot, and writes in-depth, technical (data-science background) insights in how AI works. This is a great supplement to The AI Reverie:

AI Minds Newsletter
“Navigating the Intersection of Human Minds and AI”. This newsletters dives deeper into usecases, and features research papers and tools that help you become smarter about AI. Highly recommended reading.

✨TOGETHER WITH Connecting Dots

I really recommend this newsletter, so sign up under.

FEEDBACK

We're always on the lookout for ways to spice up our newsletter and make it something you're excited to open.

Got any cool ideas or thoughts on how we can do better? Just hit 'reply' and let us in on your thoughts.