Meta’s new tool will change how we analyse videos

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 AI impersonations are a growing threat

• Neura’s exciting demo of their humanoid robot

• The fusion power startup that’s backed by Bill Gates

• … and much more

Let's jump in!

1. US Copyright Office tells Congress to outlaw AI impersonations

The agency is concerned that AI will be used to impersonate people, with the aim of damaging their reputation and destroying their livelihood.

This is the first part of its report, which explores how rapid advances in AI will impact copyright laws - especially when it comes to digital replicas of people, such as celebrities, politicians, and ordinary citizens.

Given that digital replicas are becoming incredibly realistic and growing in scale, their report recommends that new laws are introduced to combat this emerging threat.

Specifically, they want people to be held accountable for both creating and distributing these deepfakes.

Online platforms would also fall under the scope of the law, as they need to quickly remove content after they’re alerted that it’s a fake.

2. Perplexity outlines plans to share ad revenue with publishers

Perplexity is an AI search engine that provides users with summarised answers, rather than dozens of links.

Several news organisations have been angry that their content is being used in Perplexity’s summarised answers.

This means that less people will visit the news articles, leading to falling ad revenues for businesses that are already struggling to survive.

Whenever a news organisation is cited in an answer, Perplexity says it will share a "double-digit" percentage of the ad revenue.

It’s a pretty rotten deal and given that trust is so low in the company, I doubt that many will agree to it.

This is simply a tactic to reduce the backlash against them and to prevent any further lawsuits.

3. Fusion power startup raises another $53 million

Type One Energy is backed by Bill Gates and its funding has now reached a total of $82.5 million.

The company aims to finalise its reactor design by 2030 and then licence it to a third party for construction.

That design is based on a concept used by Germany's Wendelstein 7-X, but they have used supercomputers to further optimise it.

In the short term, they plan to prototype and validate the reactor design - before proceeding with a final version.

Given some of the recent comments from Type One’s CEO, it seems like their biggest focus is on decarbonising Southeast Asia.

4. Canva acquires Leonardo.ai to boost its generative AI tools

Leonardo produced their own image generators, with a focus on game development and designers.

That specialism works perfectly with Canva, as it’s one of the most popular platforms for creatives and has 180 million users.

While Canva has been offering some image generation tools, they haven’t been able to keep up with recent advances from Midjourney and Adobe.

This acquisition will help Canva to differentiate themselves versus their competitors, like Adobe and Microsoft, who are building very similar platforms to Canva’s.

It will also ensure that they’re not reliant on models from their rivals, which would have been a serious risk to their business’ future.

Leonardo is a startup that I covered several months ago, which you can read here in my spotlight section.

5. GitHub allows developers to easily test models with just a few clicks

GitHub has launched GitHub Models, enabling its more than 100 million developers to become AI engineers and build with industry-leading AI models.

The new model playground allows users to explore popular private and open models from Meta, Mistral, Azure OpenAI Service, Microsoft, and others.

This move is solidifying GitHub’s position as the go-to-platform for development.

They now offer products for hosting, running, and testing code - alongside the AI models themselves.

This will be embraced by developers, as it reduces the barrier to entry and makes it easier to explore new models.

But it will also make it much harder for companies to move away from GitHub, as almost everything is now hosted on the platform.

For Microsoft, they will get incredibly valuable data about what the most popular models are - which can then be used to inform their investments in new AI companies.

Meta’s SAM 2 is a big deal for the tech industry

Just last year, Meta announced their Segment Anything Model (SAM). This was a huge breakthrough, as it allowed us to identify objects in an image and draw lines around them (segment them).

That doesn’t sound very exciting, but it’s incredibly important as there are a lot of downstream impacts.

For example, SAM 1 led to massive advances in how we analyse satellite imagery. Because SAM is incredibly accurate and fast, it meant that we could search for “plane” and every aeroplane in the image would be highlighted.

It’s not an ability we had before and it’s like something you would see in CSI Miami, where the investigators simply type in what they want to find.

With SAM 2, it promises to be just as big as the first version. We can now segment both images and videos.

There’s lots of ways this could be used, such as creating new video effects or to help us train new computer vision models.

It could also lead to improvements in how we pixelate and redact videos. Since this model is better at detecting objects and their outlines, we will be able to more accurately redact specific objects from videos.

This is incredibly useful, as people often have to do this manually.

There’s no doubt that SAM 2 will have an even bigger impact than the previous version, which will be very exciting to see.

ChatGPT’s new voice mode allows it to sing and mimic accents

OpenAI has rolled out ChatGPT's new advanced voice mode to a small group of ChatGPT Plus subscribers.

In one example, ChatGPT tells a story as if it were an airline pilot speaking to passengers on a flight, with the audio changing to sound like it's coming from an intercom.

ChatGPT can even correct the pronunciation of French words and gives you advice on how to adjust your voice inflection.

The AI model does a decent job at mimicking regional US accents, including New York, Boston, and Wisconsin.

ChatGPT's advanced voice feature can also sing in different styles, producing a blues-style take on "Happy Birthday" and imitating what animals like frogs and cats would sound like singing the same tune.

The initial rollout will be slow, but they hope to make it available for all ChatGPT Plus subscribers sometime this fall.

💼 Intel will lay off over 15,000 employees

🎨 Midjourney V6.1 has been released, with big improvements to image quality

🚛 Aurora Innovation, a self-driving truck startup, is about to sell $420 million in shares

🎬 Runway announces an even faster, cheaper AI video model

💊 Pharma giant Cencora is alerting millions about a recent data breach

🇬🇧 UK’s nuclear submarine software was secretly developed by Russian and Belarusian engineers

🍎 Apple says it was responsible when training its AI models

🧪 NIST releases a tool that lets us test an AI model's risk level

Neura

This is a German robotics company that has got a lot of attention recently.

They’ve unveiled their humanoid robot, called 4NE-1, which is able to do different tasks - such as ironing or moving boxes.

Interestingly, Neura now have early access to Nvidia’s new tools for humanoid robots. They join other prominent companies who have early access, such as Boston Dynamics and ByteDance.

While Neura’s demo is pretty impressive, I would treat it with some level of scepticism until we see more evidence.

It has been very common for companies to release demo videos that are highly selective, without showing their product’s true limitations.

Regardless, they’ve got a clear vote of confidence from Nvidia - which is interesting given how competitive this sector is.

Neura are joining 1X, Figure, Tesla, Agility, and Boston Dynamics in the race to create humanoid robots that can perform almost any task - from the home to the warehouse.

This Week’s Art

We’ve covered a lot this week, including:

• Why the US Copyright Office wants to outlaw AI impersonations

• Perplexity’s new ad revenue programme for publishers

• The fusion power startup that’s backed by Bill Gates

• Why Canva bought Leonardo to boost it’s GenAI offering

• GitHub’s new Models product for developers

• How SAM 2 could have even more of an impact than the first version

• ChatGPT’s advanced voice mode and what it can do

• And how Neura are working on humanoid 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.