Apple’s visual search is changing how we use the iPhone

Welcome to this edition of Loop!

To kick off your week, we’ve rounded-up the most important technology and AI updates that you should know about.

‏‏‎ ‎ HIGHLIGHTS ‏‏‎ ‎

• Why the EU has ordered Apple to return €13 billion

• OpenAI’s new model and why it’s disappointing

• Microsoft’s ambitions to provide reliable quantum computing

• … and much more

Let's jump in!

1. Astronauts complete the first-ever commercial spacewalk

In a historic mission, four astronauts have completed the first-ever commercial spacewalk.

Since their space-craft doesn’t have an airlock, all crew members had to wear the SpaceX-designed suits.

Both Jared Isaacman and Sarah Gillis spent less than 10 minutes outside, with each testing the new spacesuits.

It’s a significant achievement for SpaceX, who are planning even more private spaceflights - including missions to both the Moon and Mars.

The Polaris Dawn crew have since splashed down in the Gulf of Mexico, after they completed their mission.

2. Apple must pay Ireland €13 billion in unpaid taxes

The European Court of Justice has ruled that Apple must pay Ireland a staggering €13bn in unpaid taxes, which brings the 8-year dispute to a close.

In 2016, the EU launched the case against Apple - which accused it of collaborating with Ireland to gain an illegal tax advantage.

Ireland’s corporation tax is just 12.5%, which is far lower than other Western nations - such as the US (21%) and UK (25%).

When combined with Ireland’s skilled workforce and access to the EU single market, it’s a very attractive country for foreign investment. Apple, Google, Meta, and many others have since setup offices in the country.

Given how heavily Ireland relies on foreign investment, their Government has opposed this move by the EU.

Regardless, this decision is unlikely to have an impact on whether corporations setup offices in Dublin. But it shows that the EU is cracking down on what it sees as tax avoidance.

3. US, China and other nations attend summit on AI use in military

South Korea has hosted an international summit, which will outline how militaries should responsibly use AI.

Representatives from over 90 countries attended the event, including the United States and China.

In Ukraine, AI-enabled drones are already being used to repel the Russian forces. Ukraine is already producing 1 million drones per year, with the aim of growing this to over 2 million.

The summit has led to a “blueprint for action” being created, which around 60 countries have signed up to. This includes the US and its allies.

The blueprint outlines steps that countries can take to minimise risks, such as AI models being used by individuals to create Weapons of Mass Destruction (WMDs).

While the document is not legally binding, China and 30 other countries have not signed up.

4. Microsoft could become the first to provide reliable Quantum computing

The company has announced that it will create the first platform for reliable quantum computing.

This follows research they conducted in April, which was a major breakthrough and saw them run 14,000 experiments - without encountering a single error.

Microsoft is partnering with Atom Computing to build “the world’s most powerful quantum machine”, which will eventually be available to enterprise customers on Azure.

Atom Computing is an impressive startup. Just last year, they announced that they had developed the first 1,225-Qubit computer.

While there’s no timeline for when the new quantum computer will be available, it’s a clear commitment from Microsoft in the technology.

If properly achieved, quantum computers could solve complex problems much faster than the computers we have today - leading to significant advances in areas like drug discovery and cryptography.

5. New AI tool can spot warning signs for over 1,000 diseases

AstraZeneca has developed a new AI tool called MILTON, which can detect early warning signs for over 1,000 diseases - even years before any symptoms show.

The tool works by analysing the patient’s data, such as their blood or urine tests, and can then suggest whether the patient may have a specific disease.

To create the AI model, the researchers used data from 500,000 people in the UK Biobank.

This is a huge medical database, which patients are able to opt into and provide their data to researchers - with many of them developing new ways to treat cancer.

The model has performed very well and was “exceptional” at identifying over 120 diseases. For the other 1,091 diseases, MILTON’s still did a good job and was seen to be “highly predictive”.

There’s no doubt that this is an incredible achievement. Since the model has “exceptional” performance for over 120 diseases, doctors could reliably use this to intervene much earlier.

If this is properly implemented in the UK, with safeguards around it, we could see a huge improvement in patient outcomes.

That’s badly needed, as the UK Government is struggling to fund their National Health Service. Advances like this could help reduce those costs.

Why Apple’s Visual Search is a big move

At their latest iPhone event, the company released several new products - including the iPhone 16, Apple Watch 10, and updated AirPods.

However, it was Apple Intelligence (nice branding) that stole the show. The new iPhone will be able to categorise information and show you the most important messages.

For example, if you ask Siri about a specific appointment that’s buried in your emails, then it will be able to search through your messages and find that information.

Previously, you had to manually add the event to your calendar, before Siri could “see” that information.

Apple’s AI can also write text messages for you, proof-read emails, or generate AI images. While it sounds like a small step, this ensures you don’t have to constantly switch to third-party apps - such as ChatGPT - and keeps you more engaged in the Apple ecosystem.

But the real advancement is “visual intelligence”. When you take a picture of a store, Apple will use your location data to figure out what you’re looking at.

Then it determines what action you probably want to do. In this case, you probably want to read reviews from others or check the menu.

Or, if you’re looking at a poster that’s promoting an upcoming event, you probably want to add the event to your calendar.

Actions are what separate Apple’s implementation from ChatGPT.

By having complete access to the user’s apps and data - such as their location, calendar, emails, text messages, and photos - Apple can pick the relevant information and give it to the AI model.

Of course, this data is stored securely on your device and is not accessible to other companies.

But Apple is moving towards a future where their AIs will suggest actions for us. That helps to simplify our lives, but also reinforces Apple’s strong ecosystem.

OpenAI’s new model has fallen short

The AI startup has released a preview version of o1, which can “think” before answering.

Their new model will spend more compute time on a task, with step-by-step “working out” used to better solve the problem.

By using smaller steps, it’s more likely that the model will spot when it has made a mistake and try to correct it.

Since it needs to use a lot more computing power, o1 is four times more expensive than GPT-4o and doesn’t support multi-modal queries - e.g. when you upload an image and then ask it a question.

For this reason, OpenAI admits that GPT-4o will be a better option for most prompts.

Based on my experience with it, and feedback from others online, this has been a disappointing launch.

It’s very common for the model to spend 20-30 seconds “thinking” about an answer and get it completely wrong, even for simple tasks.

This is to be expected, due to the inherent nature of how Large Language Models work.

These models are fantastic at generating text and spotting patterns. But they don’t actually understand what they’re saying. They have no idea if their answer is correct or completely made-up.

There is no internal reasoning within the model. And that’s a problem for complex tasks.

If a model spends 30 seconds or 2 minutes thinking about a task, and then returns the wrong answer, it’s a complete waste of time and money for businesses.

Strangely, OpenAI has showcased some very weak examples of how o1 could be used. You can already get similar responses from Claude 3.5 in under 5 seconds, while o1 takes over 20 seconds.

It’s not yet clear as to why businesses would use o1, rather than other models. We need more time to see where the value is in this model, but to-date it has been quite disappointing.

📱 Most people in UK now get their news from online sources, rather than TV

🎙️ Audible recruits voice actors to train AI models

👂 Apple says AirPods Pro 2 can work as 'clinical-grade' hearing aids

💰 OpenAI are in talks to raise funding at $150 billion valuation

🔍 Your public Facebook and Instagram posts were used to train Meta's AI models

🖼️ Mistral releases its first multimodal model, Pixtral 12B

🤖 DeepMind teaches a robot to tie its shoes and fix fellow robots

🤝 Amazon joins C2PA group to prevent deepfakes

Inversion Space

This startup, based in Los Angeles, are working on a space capsule that can re-enter our atmosphere and land anywhere on Earth.

They recently secured a $71 million contract from SpaceWERX, which is the development arm of the US Space Force.

Under that project, Inversion Space hopes to demonstrate that their re-entry vehicle is feasible for future missions.

In particular, this could be of huge benefit for the US Military - since it could allow them to deliver cargo in remote areas or battlefields.

Essentially, they are moving towards a model where Governments can have “warehouses in space”. When troops need urgent supplies, they can be dropped from space in under one hour.

It could also benefit commercial users - such as oil companies that need to urgently supply their oil rigs, or F1 teams that need to quickly retrieve spare parts.

The startup currently has around 25 people working on this, with the aim of bringing costs down and reliably dropping supplies to remote locations.

If you want to learn more, I’ve added a link below.

This Week’s Art

Apple certainly stole the show last week, even with OpenAI revealing o1 as their latest AI model.

We’ve covered quite a lot this week, including:

• The astronauts that completed the first-ever commercial spacewalk

• Why the EU has ordered Apple to pay €13 billion back to Ireland

• The South Korean summit that developed blueprints for how militaries should use AI

• Microsoft’s recent progress with Quantum computing

• AstraZeneca’s new AI tool that can spot warning signs for over 1,000 diseases

• Why Apple’s Visual Search is a big move

• OpenAI’s new o1 model and why it has fallen short

• And how Inversion Space are creating “warehouses in space”

Have a good week!

Liam

Share with Others

If you found something interesting in this week’s edition, feel free to share this newsletter with your colleagues.

About the Author

Liam McCormick is a Senior AI Engineer and works within Kainos' Innovation team. He identifies business value in emerging technologies, implements them, and then shares these insights with others.