During the keynote of WWDC 2023, amidst the release of numerous new products such as the Apple Silicon Mac Pro and Apple Vision Pro, Apple's presenters never mentioned the term "AI," which is a notable omission considering its competitors like Microsoft and Google have been heavily focused on current generative AI. However, AI still played a part in Apple's demonstrations, albeit under a different name.
While "artificial intelligence" is now a very vague term, surrounded by both astonishing advancements and extreme hype, Apple chose to avoid this association and instead focused on terms like "machine learning." For example, during the iOS 17 demo, Senior Vice President of Software Engineering Craig Federighi discussed improvements in autocorrection and dictation:
"Autocorrection is powered by machine learning on the device, and we've been advancing these models for years. Now, the keyboard input leverages this transformer language model, which is state-of-the-art technology for word prediction, making autocorrection more accurate than ever. And with the power of Apple Silicon, the iPhone can run this model with every keystroke."
It is worth noting that Apple mentioned the term "transformer" in one of their keynote presentations. The company specifically referred to the "transformer language model," indicating that their AI models utilize the transformer architecture, which has driven many recent generative AI innovations such as the DALL-E image generator and ChatGPT chatbot.
The transformer model, introduced in 2017, is a neural network architecture used for natural language processing (NLP). It employs self-attention mechanisms that prioritize different words or elements in a sequence. Its ability to process inputs in parallel significantly improves efficiency and has led to breakthroughs in NLP tasks such as translation, summarization, and question answering.
Clearly, Apple's new transformer model in iOS 17 allows for sentence-level autocorrection, completing a word or even an entire sentence as you press the spacebar. It also learns from your writing style to guide its suggestions.
All of this AI processing on Apple devices is made relatively easy due to the presence of AI-specific hardware in Apple Silicon chips (starting with the A11 chip introduced in 2017). Apple also stated that voice input "employs a new transformer-based speech recognition model, utilizing the neural engine to make voice input more accurate."
During the keynote, Apple also mentioned "machine learning" multiple times: when describing the new iPad lock screen feature ("When you choose Live Photo, we use advanced machine learning models to synthesize additional frames"), iPadOS PDF functionality ("Thanks to new machine learning models, iPadOS can recognize fields in PDFs, so you can quickly fill in contact information like names, addresses, and emails using autofill"), adaptive audio feature for AirPods ("By personalizing the volume, we use machine learning to understand your listening preferences over time"), and the Apple Watch widget feature called Smart Stack ("Smart Stack uses machine learning to show you relevant information when you need it").
Apple also introduced a new app called Journal, which allows for locking and encrypting personal text and image logs on the iPhone (similar to an interactive diary). Apple stated that AI plays a role in this app but did not use the term "AI."
"With the machine learning capabilities on your device, your iPhone can create personalized moment suggestions to inspire your creativity," Apple said. "Suggestions will be intelligently curated based on information on your iPhone, such as your photos, location, music, workouts, and more. You can control what content is included when enabling suggestions and what content is saved to your journal."
Finally, during the demo of the new Apple Vision Pro, the company revealed that the moving image on the front of the goggles comes from a special 3D avatar created by scanning your face - you guessed it, using machine learning technology.
"With our state-of-the-art machine learning technology, we've created a novel solution," Apple said. "After a quick registration process using the front-facing sensors on Vision Pro, the system uses advanced encoder-decoder neural networks to create your digital character."
The encoder-decoder neural network is a type of neural network that first compresses the input into a compressed digital form called a "latent space representation" (encoder) and then reconstructs the data based on the representation (decoder). While we can only speculate, the encoder part might analyze facial data captured during the scanning process and compress it into a more manageable low-dimensional latent representation. The decoder part could then use this compressed information to generate a 3D model of your face.
M2 Ultra - The AI Powerhouse?
During the WWDC keynote, Apple introduced the most powerful Apple Silicon chip to date, the M2 Ultra, which boasts up to 24 CPU cores, 76 GPU cores, and a 32-core neural engine, reportedly capable of delivering 316 trillion operations per second. Apple claims it represents a 40% performance increase over the M1 Ultra.
Interestingly, Apple directly stated that this capability could come into play when training "large transformer models," which is the most prominent mention of AI in their keynote (albeit in passing):
"M2 Ultra supports up to 192GB of unified memory, 50% more than the M1 Ultra, enabling it to do things other chips can't. For example, within a single system, it can handle a large number of ML workloads, such as large transformers that even the most powerful discrete GPUs struggle with due to memory limitations."
This development has excited some AI experts. Perry E. Metzger, an AI expert, wrote on Twitter, "Whether intentional or not, Apple Silicon's unified memory architecture means that high-end Macs are now truly amazing machines for running large AI models and doing AI research. There really aren't many other systems at this price point with 192GB of GPU-accessible RAM."
Here, larger RAM means more powerful AI models can fit into memory. These systems are the new Mac Studio (starting at $1999) and the new Mac Pro (starting at $6999), which may bring AI training within reach for many newcomers in the form of desktop and tower machines. Apple seems to have openly entered the hardware field for generative AI training.
Source: https://arstechnica.com/information-technology/2023/06/at-apples-wwdc-keynote-ai-never-came-up-by-name-but-it-was-there/
Alphabet Q1 paid subscribers reached 350M, fueled by YouTube & Google One. YouTube ad revenue slightly missed at $9.88B amid Premium growth. Total Q1 revenue hit $109.9B, beating forecasts.
Anthropic releases Claude connectors for Adobe, Blender, and Ableton, enabling AI workflow automation and app control. It also sponsors the Blender Development Foundation.
Apple introduces an App Store model splitting 12-month plans into monthly installments. On iOS 26.4+, it lowers upfront costs without reducing total price, adding billing tracking.
Google launches Aadhaar digital IDs in Wallet for India, enabling secure QR/NFC verification. A phased partner rollout targets key sectors, while passport-based credentials expand globally.
YouTube tests "Ask YouTube," an AI tool blending text and video for complex queries. Initially for US Premium users, it supports interactive follow-ups, expanding Google's AI search.
Google grants US DoD classified AI access, joining OpenAI and xAI in defense deals amid Anthropic's dispute. Contractual safeguards against surveillance and lethal weapons remain unenforceable.
Mistral AI launches Workflows, a Temporal-powered orchestration platform for secure, Python-driven enterprise AI automation featuring human-in-the-loop approvals.