How Meta's new video generator could change movie production

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 ‏‏‎ ‎

• Microsoft’s big re-design for Copilot

• How students used Meta’s smart glasses to dox people in real-time

• US Space Force plans to identify new threats in space

• … and much more

Let's jump in!

1. Microsoft can create medical X-rays with Gen AI

The company’s researchers have developed RadEdit, which uses Generative AI to create medical images.

Essentially, it can generate images that look very similar to real X-rays. These are then used to test Microsoft’s AI models and determine how accurate they are.

Because you can use AI to generate these X-rays, you can also specify what diseases it should simulate.

Strangely, this research hasn’t got a lot of attention yet - but the potential is huge.

This could really reduce the time it takes to create new AI models for medicine, as companies can test them sooner and with millions of images.

Meta have already done similar work, but they’ve used diffusion models to train new robots.

Microsoft could take a similar path and use RadEdit to train new AI models - alongside testing the tools they already have.

2. Toyota invests heavily into an air taxi startup

I covered Joby Aviation several weeks ago, which you can read here. Toyota is now doubling its investment into the air taxi startup.

The company has just invested $500 million, which brings the total investment to $894 million. Certainly not a small bit of change.

Joby will use the funds to apply for certification with the US FAA, which is a very lengthy process.

The two companies have partnered for several years, with Toyota providing their manufacturing experts to help Joby with their eVTOL production.

3. Ukraine is using a “dragon drone” that can burn tanks

In a significant development in the Russia-Ukraine war, Ukrainian forces have begun deploying "dragon drones" that are equipped with thermite.

The drone is able to drop molten thermite that burns at over 2,200 degrees Celsius (4,000 degrees Fahrenheit).

So far, these drones have mostly been used against trenches that contain Russian soldiers - but they’re now being used against armoured vehicles as well.

The Ukrainian army's 30th Mechanised Brigade was able to immobilise a Russian tank, thanks to an anti-tank missile.

The dragon drone was then sent in to finish it off, with thermite sprayed over the tank and setting it ablaze.

While the drones' incendiary spray is pretty inaccurate, it’s a worrying glimpse into the future of warfare - with multiple drones used to rain fire down on troops.

4. Impulse Space secures contract to identify threats in space

The company has secured a $34.5 million contract for two spacecraft missions, which will be able to quickly respond to threats in space.

Since many more satellites are being launched into space, and there are growing fears that Russia has a weapon that can disable American satellites, it’s crucial that the US Space Force can respond quicker.

Impulse Space's vehicles will be used to monitor, track, and understand object behaviour in space.

This will help the agency to understand which objects pose a threat to US interests, alongside any satellite malfunctions, and shorten the response time.

It’s been a good few weeks for the startup. They have raised $150 million in Series B funding, with backers including Airbus, and they were selected by NASA for a contract worth up to $300 million.

5. Microsoft gives a big re-design Copilot, now supports voice and vision

Microsoft is introducing a huge overhaul to its Copilot AI assistant, with voice and vision features added to make the experience more personalised.

The redesign has been heavily influenced by Inflection’s own chatbot, with the startup’s former CEO now working at Microsoft.

Microsoft is making a significant bet on voice, which will allow users to chat with Copilot in a more natural way - similar to OpenAI's advanced voice mode.

Copilot Vision is another key feature. It enables the AI assistant to see and “understand” what the user is looking at - such as webpages or screenshots.

You can also use “Copilot Daily” to get audio summaries of the news and weather forecasts.

For some time now, Copilot has been seen as too restrictive and not very useful for most tasks, especially when you compare it to competitors like ChatGPT and Claude.

Microsoft is hoping that this re-design will change that reputation and see it become a “personal” AI that can be used for almost any topic.

Students used Meta’s smart glasses to dox people in real-time

Earlier in the week, two Harvard students have raised privacy concerns about Meta’s new AR glasses.

The students were able use to the glasses’ cameras and find out information about people on the street.

In several examples, ordinary people will walk past and are unwittingly “scanned” by the glasses.

It then uses facial recognition to identify their name, address, phone number, and job role.

To prove that it worked, the students walked up to several people and correctly guessed the person’s name and job.

It’s a pretty frightening demo. Anyone could be recording you in the street, scanning your face, and using it to find out who you are.

Online databases, such as Clearview and Pimeyes, have made this possible by scraping millions of photos from the internet.

If you’ve ever posted a public photo on Facebook, it’s likely that your image and name are searchable in the same way.

Unfortunately, there’s very little we can do to prevent these types of privacy breaches.

All you can do is apply for your image to be removed from online databases. To find out more about how you can do this, check out this document.

Meta's video generator blows Sora out of the water

The new AI video generator is simply incredible. The videos are high quality, very consistent as the camera moves around, and you can even edit the video with text.

For example, you could upload a video of a man running and ask Movie Gen to “change the background to a sports stadium”.

Surprisingly, it can do that perfectly. There aren’t any glitches or strange behaviour in the background. The results are pretty flawless.

It’s likely that Meta’s recent research, including their progress on video segmentation, are enabling these fantastic edits to happen.

Earlier in the year, they announced SAM 2 and this tool is capable of identifying key objects within a video.

It also allows you to select specific items and change their appearance. This is what’s powering Movie Gen in the background.

When the user asks to “add the northern lights” to a video, SAM is identifying the section of video that needs to change and then assigns another AI model to show the northern lights.

It’s very, very impressive. It will undoubtedly have some impact on the wider film industry.

I don’t think it’s realistic for Hollywood to use this technology, because the quality is not good enough, but student filmmakers could really benefit from this.

Tools like Movie Gen will help them to quickly shoot a video, add multiple effects, and test what the end result would look like.

While Movie Gen is expensive to run and isn’t being released to the public, it offers a glimpse into how the film industry will change.

🌲 Black Forest Labs releases an updated FLUX1.1 image generator and API

📸 Meta will train AI models on any image you upload to Ray-Ban Meta AI

🎨 OpenAI launches new Canvas interface for ChatGPT

📏 Apple an AI model that generates incredible 3D depth maps

🕶️ Microsoft HoloLens 2 has been discontinued, with no successor planned

💰 OpenAI raises $6.6 billion, is now valued at $157 billion

🚪 WordPress cofounder is paying employees to leave if they disagree with him

🚗 Waymo robotaxis got stuck on the Vice President’s motorcade route

👉 FacePoke allows you to bring portrait photos to life

🎥 Helm.ai reveals a new GenAI model for developing autonomous cars

Serve Robotics

Chances are that if you’ve seen one of those robots crossing the street, it’s probably owned by Serve.

The company is based in Redwood City and now has over 2,000 delivery robots that travel on the sidewalk.

It was started back in 2017 and spun out of Uber a few years later. Recently, they have signed a partnership with Wing to test robot-to-drone deliveries.

The system will involve Serve's robots collecting orders from merchants and transporting them to Wing's AutoLoader for drone pickup.

It’s a really interesting idea and could lead to more people using online deliveries, as greater distances that can be reached.

If a drone can land near your house, the sidewalk robot could then be used to reach your door.

The two companies will be trialling the idea in Dallas, Texas, which could be expanded into other cities if it is a success.

I’ve added a link to Serve Robotics below, if you want to learn more.

This Week’s Art

A lot has been covered this week, including:

• Microsoft’s research that can create medical X-rays with Gen AI

• Toyota’s huge investment into an air taxi startup

• How Ukraine is using a “dragon drone” to burn tanks and trenches

• What the US Space Force is doing to identify new threats in space

• Microsoft’s big re-design for Copilot, which now supports voice and vision

• How Harvard students used Meta’s smart glasses to dox people in real-time

• Why Meta's video generator is so impressive and the impact it could have

• And how Serve Robotics are using drones to transport their sidewalk robots

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.