Welcome to Weekly News Roundup Issue #455. This week, we will take a closer look at what Apple is cooking with their upcoming new products and services powered by generative AI.

In other news, Elon Musk said that the first patient with a Neuralink implant can move a computer mouse by thinking. Over in the AI world, Nvidia reported record revenue figures while Google released new open-source models and had some problems with images generated by Gemini.

Apple is famous for joining late to major tech trends. iPod, iPhone, iPad, Apple Watch and recently released Apple Vision Pro were not the first products in their respective categories. The same is happening now with generative AI. Apple hasn’t shown any generative AI products yet but that could change soon, according to the recent leaks coming from Cupertino.

Almost every tech giant has joined the generative AI game. Microsoft is reinventing itself around chatbots and copilots. After some hiccups, Alphabet finally released Gemini, its competitor to OpenAI’s GPT-4. Meta embraced open-source with Llama 2 and is now training Llama 3. Amazon has Q, a generative AI-powered assistant, and offers a spectrum of generative AI services on AWS.

Apple, just like everyone else in the tech industry, was caught by surprise by the meteoric rise of ChatGPT. While others were showcasing their newest AI products throughout 2023, Apple remained suspiciously quiet. However, behind the scenes, Apple has been working on a new suite of products powered by its very own large language model. And we can see some of them later this year, Tim Cook said in a recent earnings call.

Apple might be quiet but it is not sitting still and doing nothing. Quite the opposite. In 2023, Apple quietly acquired 32 AI startups, the most of all tech giants. For comparison, in the same year, Google acquired 21 AI startups, Meta 18 and Microsoft 17. I wouldn’t be surprised if most if not all of those acquisitions were acqui-hires to get more AI talent in the company rather than to get specific AI products or technologies. Apple has also released an open-source machine learning framework called MLX that makes it easier to train AI models on Apple Silicon chips, a text-to-image editing AI model MGIE, and an AI animator called Keyframer.

According to a report by Mark Gurman, Apple is planning to announce a suite of new generative AI tools and features to its products at the WWDC conference around the first week of June. Some of them are going to be part of iOS 18 which, according to leaks, could be the biggest update in iOS’s history.

Productivity apps such as Pages and Keynote are rumoured to get auto-summarising and auto-complete features as well as other tools assisting in writing and creating slides. The release of MGIE suggests users could be able to edit the photos using text prompts. Apple Music might get an AI to generate playlists, similar to Spotify’s AI DJ. Messages are getting AI features, too. Meanwhile, developers building apps in Xcode will benefit from an AI-powered coding assistant, akin to Github’s Copilot. Apple is also apparently testing the use of generative AI for internal customer service apps within AppleCare.

But the biggest star of the show is rumoured to be the new Siri. Siri has fallen behind other digital assistants even before ChatGPT was released. The new generation of chatbots powered by large language models have made Siri feel even more stale and outdated. As part of Apple’s new push into generative AI, Siri is expected to get a complete overhaul in iOS 18. New Siri is rumoured to be powered by Apple’s very own large language model codenamed Apple GPT. The first leaks about this new language model emerged mid-2023 and it was reported that Apple is spending millions of dollars per day training the new model.

Details about Apple’s large language mode are scarce but Mark Gurman reveals that Apple GPT uses Apple’s LLM framework called Ajax. Apple GPT is running on Google Cloud and built with Google JAX, a framework created to accelerate machine learning research.

It is almost certain Apple GPT will be running locally on iPhones and other Apple devices. Apple prefers to run as many things as possible on the device in the name of protecting privacy. There is no information about how big the Apple GPT running on an iPhone will be. Early leaks mentioned that AppleGPT is a 200-billion parameter model but that’s way too big a model to fit inside an iPhone and sounds more like a model designed to be deployed on a server, not to run on a small device.

It is possible Apple could take a similar approach that Google took with Gemini and release a family of models where the smaller models are distilled from the biggest model. However, Apple might have some clever tricks in their sleeve here.

Recently, Apple engineers and researchers published a paper describing a new method of running large language models on devices with limited memory by using the device's flash memory, which sounds like something that could be very useful on a device like an iPhone. This new method not only enables running models up to twice the size of the available memory but also increases the inference speed by 4-5 times when the model is run on the CPU and by 20-25 times when run on GPU. Add to that rumours about a significantly upgraded Apple Neural Engine for the upcoming iPhone 16 and we might be getting the most powerful language model running on any mobile device.

Apple GPT is currently being tested internally by Apple employees who recently were banned from using competitors’ tools like ChatGPT or Github Copilot to minimise the chances of accidental leaks.

As I mentioned earlier, it is expected that Apple will announce its new AI products and services during the WWDC conference in June. In that case, Apple’s new AI features will be available as part of Apple’s beta program shortly after the conference. Everyone else will get these updates in autumn when the newest versions of iOS, iPadOS and macOS are usually released to the public. However, not all AI products will be released at once as Apple plans to take a gradual approach to AI development with some features not coming for years.

It’s going to be interesting to see what Apple is going to bring to the generative AI game. Apple is behind the competition and many of their AI-powered products will be aiming to achieve parity with what Microsoft and Google offer. When Apple releases their AI products in autumn, they will join the game almost two years after ChatGPT entered the stage and probably just weeks before GPT-5 is out. Microsoft, Google and Amazon will have year to year and half head start. Apple will have to catch up very quickly.

Let’s not forget that Apple likes to do things differently. Sometimes, Apple’s insistence on doing things their own way fails miserably. But sometimes we get ideas with Apple’s magic touch that later become industry standards.

If you enjoy this post, please click the ❤️ button or share it.

Share

Do you like my work? Consider becoming a paying subscriber to support it

Become a paid subscriber

For those who prefer to make a one-off donation, you can 'buy me a coffee' via Ko-fi. Every coffee bought is a generous support towards the work put into this newsletter.

Your support, in any form, is deeply appreciated and goes a long way in keeping this newsletter alive and thriving.

🦾 More than a human

Neuralink implant patient can move computer mouse by thinking, Musk says
The first human patient implanted with a brain-computer interface from Neuralink has reportedly fully recovered and is capable of controlling a computer mouse with their thoughts, Elon Musk said in a Spaces event on X. Musk said Neuralink was now trying to get as many mouse button clicks as possible from the patient. However, no evidence has been provided to support Musk’s claims.

New Wyss project aims to control exosuit with brain signals
The Synapsuit project aims to develop high-performance algorithms that decode complex brain signals to then control a lightweight, soft, wearable exosuit that supports arm and hand movement in real time. Combining the algorithms with a brain-controlled exosuit could accelerate neuro-rehab methods, Wyss says, by supporting the movement of those suffering from stroke and spinal cord injury.

🧠 Artificial Intelligence

Nvidia reports enormous revenue as AI hits a tipping point
Nvidia is still riding the AI wave. The company has reported fourth fiscal quarter earnings that beat Wall Street’s forecasts. The chip maker posted $22.1B in revenue on an expected $20.55B. Revenue was 22% higher than the previous quarter, up 265% from a year ago. The revenue from data centres was up more than 400% from the same period last year, reaching $18.4B.

Hundreds of AI luminaries sign letter calling for anti-deepfake legislation
Hundreds of people, including notable figures in the AI community, have signed an open letter demanding strict regulation of AI-generated deepfakes, emphasizing the need to criminalize deepfake child sexual abuse materials and penalize harmful deepfake distribution. This action reflects growing concerns over the societal threats posed by deepfakes and highlights the urgency for developers to prevent their misuse.

Google Gemma: Introducing new state-of-the-art open models
Google has released Gemma, a new family of open large language models. Gemma comes in two sizes - 7B and 2B. In terms of performance, Gemma 7B is roughly at the same level as Meta’s Llama 2 70B model while Gemma 2B tops the list for similarly sized models, according to benchmarks published by HuggingFace. Both models are available on HuggingFace as well as on Colab and Kaggle.

Google to pause Gemini AI model's image generation of people due to inaccuracies
Google paused its Gemini AI model's ability to generate images of people after users discovered inaccuracies in images generated by the AI depicting historical figures, such as German WWII soldiers, Vikings, popes and US founding fathers by portraying them in a variety of historically inaccurate ethnicities and genders. Google said it is working to fix the issue.

Air Canada ordered to pay customer who was misled by airline’s chatbot
Air Canada, Canada’s largest airline, has been ordered to pay compensation after its chatbot gave a customer inaccurate information, misleading him into buying a full-price ticket. Interestingly, Air Canada tried to put the blame on the chatbot arguing that despite the error, the chatbot was a “separate legal entity” and thus was responsible for its actions. “While a chatbot has an interactive component, it is still just a part of Air Canada’s website. It should be obvious to Air Canada that it is responsible for all the information on its website,” wrote the judge. “It makes no difference whether the information comes from a static page or a chatbot.”

🤖 Robotics

Drones could deliver medical supplies under UK travel watchdog plans
The UK's Civil Aviation Authority (CAA) is proposing guidelines to expand drone use for medical deliveries and infrastructure inspections by allowing flights beyond visual line of sight at low altitudes near buildings, aiming to safely integrate drones into the airspace. This move, part of the UK's future flight program, could significantly benefit industries and pave the way for drone-based delivery services in the UK.

▶️ Meet Apollo, the real-life robot who wants to give you more free time (17:40)

This video shares the story of how Apollo, Apptronik’s humanoid robot, came to be. This is over a decade-long story of hard work and overcoming difficult engineering challenges to build a humanoid robot that can safely and successfully operate in an unpredictable world. This is also a story of how motorcycle design inspired the looks of the robot to make people feel comfortable and safe around these robots.

Disney’s Newest Robot Demonstrates Collaborative Cuteness

In this article, Morgan Pope, a research scientist at Disney Research, shares how the team at Disney Research made two robots collaborate with each other to an engaging and dynamic performance at the Shanghai Disney Resort. Duke, an expressive bipedal robot, and the kiosk, a stable wheeled system, work together to deliver a performance that neither could achieve on their own. This project presented unique challenges, as most robots are designed in ways that make direct collaboration with other robots difficult.

🧬 Biotechnology

New CRISPR tomatoes need less watering
By removing a gene coding for a specific protein, researchers have created a tomato plant that can partially close the small holes on its leaves, called "stomata," at midday. This reduces the amount of water evaporating from the plant, allowing it to conserve water without compromising the quantity or quality of the tomatoes produced.

Japan startup creates pigs with organs suitable for human transplants
A Japanese startup has announced the production of three cloned piglets with organs suitable for human transplant with reduced risk of immune rejection. These piglets will be made available to medical institutions in Japan for pre-clinical research, according to the company.

Thanks for reading. If you enjoyed this post, please click the ❤️ button or share it.

Share

Humanity Redefined sheds light on the bleeding edge of technology and how advancements in AI, robotics, and biotech can usher in abundance, expand humanity's horizons, and redefine what it means to be human.

A big thank you to my paid subscribers, to my Patrons: whmr, Florian, dux, Eric, Preppikoma and Andrew, and to everyone who supports my work on Ko-Fi. Thank you for the support!