Europe has successfully tested AI fighter jets

Welcome to this edition of Loop!

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

‏‏‎ ‎ HIGHLIGHTS ‏‏‎ ‎

• Uber and Wayve’s plan to launch robotaxis in London

• Why fibre optic drones are a problem for us all

• How Multiverse Computing is significantly reducing AI costs

• … and much more

Let's jump in!

1. Uber and Wayve plan to launch robotaxis in London

The company is partnering with Wayve on the initiative, which is one of the UK’s leading self-driving startups. Wayve has close ties with the UK Government and hopes to deploy their vehicles within 12 months.

Just last year, the startup worked with the Government on policy and their input was used to draft new regulations on autonomous vehicles.

Under the UK Government’s new plans, self-driving cars will be operating on the roads by spring 2026. Previously, the Government was aiming for late 2027.

Heidi Alexander, who is the UK’s Transport Secretary, confirmed the new timeline and said that it will lead to more investment in the country.

This follows the wider AI investment strategy that was set by Peter Kyle, which I described in detail last week.

While there are few details about how many robotaxis are planned, I expect that Wayve will start with only 3-5 vehicles and expand from there.

The UK Government is more tolerant of risk than it was before, but I think it’s fair to say that wider British society is slightly more cautious than their US counterparts.

This has led to fewer investment opportunities and new startups, which the new Labour Government is keen to fix. Alongside the initiatives I described last week, this is just one way that they hope to do that.

2. We now have drones that can’t be electronically jammed

Quite frankly, this should alarm us all. The Ukraine-Russia war has led to new advancements in drone technology and how it’s deployed on the battlefield.

Based on what UK national security officials have said recently, there are real concerns that the West is not keeping up and innovating fast enough.

This all started earlier in the year, when Russian forces deployed fibre-optic drones to evade electronic jamming. Since they use physical cables to send data, not radio frequencies, it’s difficult for the other side to stop these drones.

Both sides are now deploying the drones and developing new techniques to slow them down, like netting on critical roads - but that won’t completely stop them.

Ukraine’s only other options are to cut the wire, down the drone with weapons, or to hit it with a friendly drone.

This raises new challenges for the general public and how we manage large events. At Pope Francis’ funeral, the Italian military had electronic countermeasures to down any drone.

But this new advancement could make that more difficult to do.

3. Corporate AI adoption may be levelling off

According to the payments firm Ramp, business appetite for AI tools is starting to plateau - following years of enthusiastic adoption.

The company's AI Index, which tracks business spending on AI products across roughly 30,000 firms, shows adoption stalled at 41% in May. This follows on from ten months of steady growth.

As you’d expect, larger enterprises lead the way with 49% adoption. While medium-sized companies are at 44%, and smaller businesses at 37%.

Although, it’s worth noting that Ramp’s methodology isn’t perfect. It’s entirely possible that some AI spending has been missed, since they heavily rely on merchant names and billing data.

Regardless, it does point to a broader trend and shows that companies are facing a reality check with productivity gains. In short, it’s harder to achieve than we were promised with the launch of ChatGPT in late 2022.

Klarna has recently discovered this the hard way. After they announced plans to replace hundreds of support staff with AI, the company quietly reversed course when service quality took a nosedive.

It seems businesses are learning what technologists have long known: there's often a chasm between demo magic and real-world deployment.

That doesn’t mean these productivity gains will never happen, but companies are starting to separate today’s real-world capabilities from the relentless Silicon Valley hype.

4. Meta creates a new world model for advanced robotics

V-JEPA 2 is the latest version of their “world model”, which has a better grasp of the physical world.

This builds on last year's V-JEPA, which was trained on over a million hours of video footage. With the new version, Meta hopes that robots will become more capable and have a bit more “common sense”.

This is a different approach to today’s LLMs, which are able to make links between words and phrases - but fundamentally struggle to understand the world around us. As we continue to scale and create larger models, this problem remains an issue.

In Meta’s examples, they showcased a robot that was holding a plate and spatula. It then approached a stove, which was being used to cook eggs.

V-JEPA 2 was then able to predict the logical next move: using the spatula to transfer the eggs to the plate.

You might be wondering, “but isn’t this already possible with LLMs?” Yes, but it’s really limited.

Today’s LLMs can analyse images and some other files, but not videos. Once we start to give them more complex images, they fall down and don’t perform well.

This is what Meta’s team is trying to tackle. They want an advanced model that can analyse videos and have a more concrete understanding of the real world - with minimal hallucinations.

Once that’s possible, it will become easier to build robots that can perform tasks. Although, there are a lot more challenges around form factor and safety that need to be figured out. This is just one step of many.

5. Midjourney is sued by Hollywood for stealing copyrighted content

The AI startup is facing legal action from both Disney and Universal, who say that the company is a "bottomless pit of plagiarism" that profits from their intellectual property.

They claim that Midjourney's image generator is being used to create unauthorised copies of their characters - including Darth Vader, Spider-Man, and the Minions.

It’s an interesting move. A few weeks ago, Disney allowed gamers to speak with an AI version of Darth Vader.

While they did this by copying James Earl Jones’ voice, and on the surface it might seem similar to Midjourney’s actions, Disney did ask for his family’s permission.

It appears that Midjourney were less careful, leading to the lawsuit. However, Midjourney is an easy target as they don’t offer any other services.

Whereas, companies like OpenAI have not faced the same action - as they offer more products that Disney could use to save money.

That diversification could save these AI labs from costly legal battles. But the downside is that smaller competitors are more exposed and likely to be picked off.

Europe successfully tests an AI fighter jet

Helsing’s latest AI system was able to successfully fly a fighter jet over the Baltic Sea, which is a remarkable achievement.

The startup partnered with Saab on the project, with a Gripen E fighter jet used for the test run.

But this wasn't just a simple autopilot demo. Instead, the AI took full control of the aircraft and ran simulated combat scenarios against another jet.

What’s fascinating is just how quickly this was achieved. In under six months, Helsing went from the initial concept to actual flight tests.

That's incredibly fast for the aerospace industry, where timelines usually stretch into years.

During the tests, the AI handled complex manoeuvres and even made firing decisions in beyond-visual-range combat situations.

It’s worth noting that a human pilot was onboard as a safety backup, ready to take control if needed.

European nations aren’t the only ones experimenting with this technology. The US was the first to successfully test AI fighter jets and simulate dog fights, with these tests happening last year. You can read my write-up here.

How drones could create 3D virtual worlds

The wider tech industry is very focused on Generative AI. When I was at SXSW London, there was only one talk that didn’t mention the topic.

But I believe we’re missing some advances that are happening right in front of us. Enterprise drones are one of them.

DJI is the most popular drone company in the world, with their technology used by both hobbyists and professionals. Recently, the Chinese company announced the Matrice 400 drone and it has some very impressive features.

The drone can fly for almost one hour and carry payloads up to 6kg. That gives you a lot of flexibility, as you can attach advanced camera systems to the drone - or even loudspeakers to broadcast safety messages.

It’s equipped with LiDAR, millimetre-wave radar, and low-light sensors that can detect objects - like power lines - and automatically prevent the drone from hitting them. The drone itself can even travel for over 40km (25 miles).

That all sounds great for search and rescue teams. However, I see this as being very valuable for tech companies - especially those who are interested in collecting data about the real-world.

In one of their examples, DJI used the drone to create a 3D scan of the environment around them. The drone was able to fly around the valley, scan the environment, and then create an incredibly realistic simulation.

Earlier on, I outlined how Meta is working to create new AI models for robotics. But we’re running out of data, since there’s a limit to how many text and video files can be collected for model training.

But if these companies can use drones to re-create the virtual world, they could merge these 3D worlds with static images - giving depth to those images that isn’t currently possible.

Either way, DJI is doing some really impressive work and it’s clear that they’re significantly ahead of the competition.

🍎 Apple introduces instant language translation for phone calls

🤖 Meta invests $14 billion in Scale AI, starts new "superintelligence" lab

💰 Qualcomm acquires semiconductor firm for $2.4 billion

📺 Warner Bros will split their cable and streaming businesses

🛑 Anthropic ends their AI-generated blog after one week

🚀 OpenAI hits $10 billion in annual revenue, releases o3-pro model

📉 News websites see their traffic fall, due to Google's AI search features

⚛️ Proxima Fusion raises $148 million to develop nuclear fusion

🆘 ChatGPT will avoid being shut down in some life-threatening scenarios, says former researcher

⏸️ Wikipedia pauses AI-generated summaries pilot, after editors protest

⚡ Fervo Energy raises $206 million to build a huge geothermal power plant

⚖️ Tesla sues former Optimus engineer for stealing trade secrets

Multiverse Computing

This Spanish startup is aiming to dramatically slash AI operating costs, with over $215 million secured in Series B funding.

Essentially, the company uses a compression system that’s inspired by some of the principles in quantum computing.

They claim that their technology can reduce the size of LLMs by 95%, without impacting the model’s performance.

If that claim stacks up, it could completely change how businesses deploy AI models and reduce overall costs.

They already offer compressed versions of Llama 4 Scout, Llama 3.3 70B, and Mistral Small 3.1 - with plans to support DeepSeek R1 soon.

Of course, these are open source models that are easy to access and customise. The company hasn’t been able to support proprietary models, like those from OpenAI and Anthropic as they don’t have that direct access.

You can use AWS Bedrock to access these models, which are 4x to 12x faster than the uncompressed versions. For inference tasks, that translates to cost savings of 50% to 80%.

Overall, the company has raised about $250 million to date and is certainly one to watch out for. It’s entirely possible that a company like OpenAI or Nvidia tries to acquire it.

This Week’s Art

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

• Uber and Wayve’s plan to launch robotaxis in London

• Why fibre optic drones are a problem for us all

• Ramp’s claim that corporate AI adoption is levelling off

• Meta’s new world model for advanced robotics

• Why Midjourney is being sued by Hollywood

• Europe’s successful test of an AI fighter jet

• How drones could be used to create 3D virtual worlds

• And how Multiverse Computing is significantly reducing AI costs

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

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.