Deepfake calls are now happening on Zoom and almost tricked a US Senator.

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 European defence startups are lagging behind the US

• Microsoft’s new tool that can find AI hallucinations and fix them

• Google’s tool that converts your notes into a podcast

• … and much more

Let's jump in!

1. Investment in European defence startups increases by five-fold

We start with a report from Dealroom, which shows that venture capital investment in European defence technology is set to reach $1 billion in 2024. That’s a five-fold increase when compared to 2018.

Of course, that’s a tiny amount when compared with the US, which invests significantly more than other countries in their defence sector.

There are several reasons for this huge gulf in spending. The US is much more willing to give grants to small startups, which allows them to actually validate their ideas. Europe is much more cautious on what it funds.

The US also attracts the world’s best talent and has a very strong venture capital ecosystem, which means it’s easier for startups to succeed there.

Regardless, Europe has started to take more gambles on startups. This is due to the ongoing wars in Ukraine and the Middle East, which are focusing minds on global threats.

Germany, the UK, and France have secured 87% of Europe’s spending in this sector. Munich is the city which has seen the most investment, although this is due to the AI company Helsing - which is based there and is adding AI capabilities to Germany’s fighter jets.

2. Microsoft reveals a new tool that can find AI hallucinations and fix them

Correction is a new tool that can be used to automatically spot mistakes in an AI model’s message and fix them.

It does this by scanning the text, identifies claims that could be wrong, then uses a source of truth to fact-check it.

The tool is part of their Content Safety API on Azure and can be used with any Large Language Model (LLM).

This is a step in the right direction, but it’s not a silver bullet that can remove all the mistakes. It’s entirely possible that the tool will itself make errors and add in new mistakes.

As I’ve said before, this all stems from how these LLMs work. They’re great at being creative and generating text, but they don’t “understand” what they’re responding with.

That causes them to make those mistakes in the first place.

I’m really hopeful that this tool will be able to remove a lot of AI hallucinations, but you shouldn’t expect it to work perfectly.

3. Mercedes wants their self-driving cars to travel at 59mph

If their plans are approved by German regulators, Mercedes’ semi-autonomous vehicles will be able to travel at up to 95km/h (59mph) on motorways.

It’s a clear sign of how confident Mercedes are in the system, following two years of use in production vehicles.

Drive Pilot is their Level 3 self-driving software, which means that drivers can take their hands off the wheel and look away from the road.

Back in 2021, Mercedes was one of the first manufacturers to release Level 3 systems onto the road.

When Drive Pilot is enabled, the company agrees to take on full liability for any crashes or damage that occurs - rather than the person behind the wheel.

4. Deepfake Zoom caller pretended to be a Ukrainian official

The head of the Senate Foreign Relations Committee, Ben Cardin, was targeted by a deepfake impersonator.

Senator Cardin was sent an email that asked to meet with Dmytro Kuleba, Ukraine’s former foreign minister, on Zoom.

On the video call, the deepfake looked exactly like the foreign minister and sounded like him.

The US Senator was asked about his position foreign policy, including the use of long-range missiles against Russia.

Cardin grew suspicious during the call and it was later confirmed that he did not speak with the real Kuleba.

This incident is alarming and shows just how advanced deepfake technology is becoming. It is moving away from ordinary videos on YouTube to more sophisticated attacks on Zoom.

There are companies out there who can detect these types of attacks, such as Reality Defender.

For businesses, this is a growing risk. It’s entirely possible for executives to suffer from the same kind of attack, with scammers pretending to be the CEO and ordering staff to send a payment to another company.

5. Marvel Fusion raises $70 million for laser-powered fusion

The startup is using laser technology to bring fusion power to market more quickly and cheaply.

Their approach involves firing femtosecond lasers at a nanostructured target, which ignites a fusion reaction.

The company has raised €62.8 million in a Series B round and are in-line for grants from the European Innovation Council.

Marvel’s ambition is to have a prototype by 2032 and this will contain hundreds of lasers that can fire 10 times per second.

That’s a long time from now, but it’s in-line with most expert’s expectations - many of whom believe that fusion power is at least a decade away.

Google can now convert your notes into a podcast

The company’s AI note-taking tool has gotten a lot of attention recently, as users have converted long write-ups into podcasts that explore the subject.

The results are very, very impressive.

I tested the tool with last week’s write-up and it far exceeded my expectations. You can listen to it here.

Google’s AI voices sound incredibly realistic, with two “people” being mimicked by the AI as they discuss your topic.

It can also seamlessly flow from one topic to another, without making it sound jarring or sudden.

In one example, the AIs talked about a startup that was doing really well and then transitioned to 23andMe - who have seen their valuation fall off the cliff in recent years.

While the AI voices do have an American accent, the changes in tone were spot on. One of them even included a slight stammer to make it sound more authentic.

Quite frankly, I have been blown away by just how good this tool is.

I don’t see it replacing your favourite podcast hosts, as these AIs are quite shallow and obviously lack a genuine personality, but we are seeing a rise in audio-only media.

Both the New York Times and Wall Street Journal offer audio-versions of their articles, which are voiced by an AI bot.

NotebookLM now opens up the door for anyone to do this, including students. People often have different learning styles, whether it’s learning from writing, doing, or visuals.

This could really benefit those kids that learn best from audio recordings.

Meta reveals their impressive smart glasses

Meta has unveiled a prototype of its Orion smart glasses, which it believes could one day replace smartphones.

The Orion prototype combines augmented reality, eye and hand tracking, AI, and gesture detection. Although, the technology is still pretty young and incredibly expensive - it’s believed that the glasses cost around $10,000 to produce.

Meta is not alone in the race to create a sleek, functional pair of smart glasses, with companies like Snap, Google, Apple, and Jony Ive's LoveFrom also working on similar projects.

These devices aim to replicate and surpass today’s smartphones as they can offer a more natural, wearable form factor.

Mark Zuckerberg has been supporting AR glasses for a long time, as he sees it as Meta’s only hope for platform independence.

Currently, Meta has to play ball with Apple and Google - which means it has to follow their rules on data collection and user privacy.

While the glasses still look a little strange, especially considering it has cameras that are always active, it’s encouraging to see their progress.

For years we have seen videos that suggest what the future might look like - whether it’s looking at ingredients and getting a list of meal options, or instant translation as someone talks to you in Spanish - but it was always a distant reality.

That doesn’t seem so distant nowadays, but the huge costs and form factor need to be improved for this to become viable.

🔓 Meta is fined $100 million for storing passwords in plaintext

🛰️ Starlink now has over 4 million users

🛫 Pyka raises $40 million for electric autonomous planes

🤖 Jony Ive is working with OpenAI on a new AI device

📞 FCC fines a man $6 million for creating a deepfake robocall of President Biden

🐰 Only 5,000 people use the Rabbit R1 device every day

🚫 US moves to block all Chinese auto imports, due to security concerns

💰 OpenAI plans to double the price of ChatGPT in next 5 years, as losses mount

Radian Aerospace

This Seattle-based startup has taken a significant step towards realising the "holy grail" of spaceflight: a reusable space plane that can take off and land like a conventional aeroplane.

The company has recently completed ground tests in Abu Dhabi, which featured a prototype plane.

The tests gathered data about the vehicle's handling and this will now be compared to the company's simulations.

Over time, they will be able to refine the final design and improve their flight control system.

Radian's ultimate goal is to develop the Radian One space plane, which will take off from a rail sled, ignite engines on orbit, and return to Earth on a runway.

This reusable space plane concept is considered a holy grail because it eliminates the need for a launch vehicle, making space as accessible to space vehicles as the upper atmosphere is to aeroplanes.

Radian Aerospace has raised $32 million to date from investors and aims to begin full-scale flights of the space plane in 2028.

However, that seems very ambitious given how small the startup is. They’re certainly one to watch as they try to validate their design with further tests.

This Week’s Art

A lot has been covered this week, including:

• Why investment in European defence startups is increasing, but still lagging behind the US

• Microsoft’s new tool that can find AI hallucinations and fix them

• Mercedes’ progress with their self-driving cars

• How a deepfake Zoom caller almost tricked a US Senator

• Marvel Fusion’s work on laser-powered fusion

• Google’s NotebookLM, which allows you to convert notes into a podcast

• Meta’s impressive smart glasses

• And how Radian Aerospace are working on re-usable spaceflight

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.