🤖 OpenAI joins the race for humanoid robots

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 ‏‏‎ ‎

• Apple cancelling their autonomous car project

• Microsoft’s partnership with Mistral AI

• DeepMind’s work to create video games using GenAI

• … and much more

Let's jump in!

1. Microsoft invests in popular startup Mistral AI

We start with Mistral, who are an AI startup based in Paris and have been seen as Europe’s answer to OpenAI.

Microsoft has invested €15 million as part of a Series A extension that valued the company at $2 billion, but will convert into equity at Mistral’s next funding round.

Microsoft has been intentional in using such a small sum, as the real value is in the partnership between the two companies. Mistral will use their computing power to train future models and Microsoft will get early access.

Another point that stands out is how they will work to create custom models for the “European public sector”. Of course, these details and Microsoft’s small investment sum were designed to minimise the risk of upsetting the EU.

That hasn’t worked, as some in the European Commission are furious over this announcement.

This is because parts of the EU’s AI Act were recently changed to help companies like Mistral, after the French government pushed for concessions.

It’s likely that this deal will face scrutiny from EU regulators, who are already looking into Microsoft’s partnership with OpenAI.

2. Apple cancels their autonomous car project

It was known as Project Titan internally, but executives have decided to pull the plug and move some of their employees to other AI projects.

In many ways, this isn’t surprising news. While it’s been rumoured for over a decade, I’ve struggled to make sense of it.

Autonomous vehicle (AV) research doesn’t gel nicely with Apple’s branding for premium, reliable products, and their secrecy around future products.

First of all, AVs need to work 100% of the time to be seen as a success. Apple does not want news headlines about their cars driving into cement, blocking ambulances, or injuring someone. They’ve worked hard to build and maintain that brand over decades.

But to develop a self-driving vehicle that works in lots of different scenarios, it needs to be out in the real-world. And to get the data that you need, there has to be huge numbers of them on the roads.

This is tricky, as it raises the risk that an accident could happen and cause injury to an innocent pedestrian - generating those negative headlines you badly want to avoid.

Apple is also very secretive about their products and doesn’t like seeing leaks in the media. This is also difficult to do, since the goal is to collect more data and improve your autonomous vehicle. That can only happen in the open, where everyone can see it driving around.

Given that other self-driving companies have faced trouble in the last year, it looks like Apple have made the right decision here - especially as they shift their focus to create more AI products.

3. Alibaba's new AI can make any photo talk or sing

Soon we won’t be able to trust our own eyes. Researchers at Alibaba have demonstrated a new method that allows us to upload a photo of someone and then create a video of them singing.

It’s quite surprising how well it works, with the person’s face moving around - alongside their lips, eyes, and forehead. It’s not perfect and looks unnatural when you study it, but that doesn’t really matter for us right now.

Instead, we should be more concerned about how deepfakes will become more believable as time goes on.

A lot of media attention will inevitably be dedicated to the political ramifications, which makes sense given the potential impact on democracy - but we should be more worried about how this technology will impact children and teenagers.

Children can be cruel to each other. What happens when a teen decides to use this technology to publicly humiliate another and make it look as though they said something embarrassing?

We’ve recently seen similar cases happen to young girls in Spain, some as young as 11 years old.

Alibaba has been responsible in not releasing this model to the public, but it’s a glimpse of what’s to come. While laws won’t always deter people from doing things like this, they’re desperately needed to at least hold them accountable.

Intimate deepfakes have just been criminalised in the UK, but no such laws exist in the US at a federal level. Following the public letter that was signed by hundreds of AI leaders a few weeks ago, regulators will come under increasing pressure to change that.

4. Scottish events company uses AI to save costs, goes terribly wrong

It’s a textbook example of what not to do. You’re tasked with organising a Willy Wonka event for children, with tickets at a not-so-cheap £35 ($45) each.

You need to promote the event and have read the hype that AI can now do anything you want, so use ChatGPT to generate posters and write the actor’s scripts.

The only problem is that the scripts are “complete gibberish” and now your actors are furious that you gave them 15 pages of nonsense to learn.

Parents are even more angry that the event doesn’t look anything like the magical AI posters and have called the police.

TV reporters have arrived outside and are now asking if you’ll apologise to the families.

Of course, this is completely the event company’s fault, but some of the irresponsible hype around Generative AI is to blame too.

There needs to be a much more grounded conversation on what AI can do really well and what it struggles with.

In a world where algorithms and hyperbolic reactions reign supreme, that’s unlikely to happen on social media anytime soon.

But it’s a reminder for us to double check what these tools are generating, as we can’t put complete trust in them.

5. Data centres and power grids are struggling with the AI boom

With the rising demand for compute to train new AI models, data centres are seeing their power consumption increase significantly.

Just a decade ago, these facilities would typically consume 10 megawatts of power - but the average today is 100 megawatts, with some of the biggest even reaching over 600 megawatts.

This rapid growth in demand is placing strain on the US power grid and those in other countries.

Ireland is particularly exposed, since data centres are using almost 20% of the country’s electricity - with demand rising by 400% since 2015.

To counter this, the industry has made significant investments in renewable energy but the demand still outpaces supply.

But as we shift focus to develop ever-more powerful AI systems, this will become a much bigger problem going forward.

OpenAI partners with Figure to develop humanoid robots

Earlier this week, the robotics company announced a huge $675 million funding round to support their research. It’s attracted top investors from Microsoft, OpenAI, Nvidia, and others.

Figure’s main goal is to bring humanoid robots into warehouses and factories, rather than in your own home. That makes sense, given the huge costs involved in setting them up and also from a safety aspect.

These robots are primarily being developed to pick up and move heavy items around the warehouse, along with other generalised tasks.

However, one thing that hasn’t been talked about much is Figure’s partnership with OpenAI. Both companies are working to “develop next generation AI models for humanoid robots”.

The significance of this can’t be overstated. Over the last 2 years, there has been a lot of research into how generative AI can be used to train and control robots. Microsoft, Google, and Meta are all looking into this and made good progress.

It’s something that I flagged back in January as a theme to watch out for in 2024, which you can read here.

It’s a new sector that OpenAI are keen to get into, as they’ve also invested in Figure’s rival 1X and are hedging their bets between the two companies.

DeepMind shows that GenAI can make video games

We’ve seen text-to-image models a lot in the last year, but it’s possible we could also get text-to-video games in the next few years.

DeepMind has trained a new model that can create 2D games from text descriptions, sketches, or even photos.

Most of those games look really similar to Super Mario Bros. - with the character jumping up on ledges and avoiding traps.

To do this, they gave the model thousands of hours of footage and asked it to work out what button was pressed to make the character move.

This differs to previous techniques, which instead focused on providing the model with labelled data that specified the exact button pressed. As you can imagine, that’s a lot of work and limited the amount of training data that could be used.

While Genie can only make games at 1 frame per second, it shows clear promise for the future. In a few years, we could have the ability to create our own interactive GenAI games.

🇮🇳 India approves $15 billion in semiconductor plant investments

🍎 Tim Cook says Apple will “break new ground” in GenAI this year

🎧 Adobe reveals a GenAI tool for music

📈 Fintech giant Stripe’s valuation rises to $65 billion

🔓 Database exposes 2FA codes used by the world’s tech giants

🔍 US SEC is investigating OpenAI over last year’s power-struggle

📸 Photoroom, an AI image editor, raises $43 million

Glean

This startup is focusing on enterprise customers and helping employees to efficiently search through several data sources.

It can be quite frustrating when you’re looking for a specific file or topic, as you can easily spend over 10 minutes looking in different apps and filesystems.

This is what Glean are primarily working on. They will connect to your company’s data sources and then use these to create a custom model.

It will then analyse the data and suggest relevant info based on the employee’s team, job role, and project work - leading to more personalised results.

Although, plenty of companies will inevitably be quite worried about having a 3rd party analysing their confidential data.

Glean claims the data is secure and can give the company some control over this process. They also provide the option to add restrictions on what employees can search for.

They’ve already got companies like Duolingo, Grammarly, and Sony using their platform - and Glean has just announced that over $200 million was raised from investors.

They might be worth exploring if you’re looking to move away from other tools, like Microsoft SharePoint.

This Week’s Art

We’ve covered a huge amount this week, including:

• Microsoft’s new deal with Mistral and how it’s upset the EU

• Apple’s move to cancel their autonomous car project

• Alibaba’s interesting research on deepfakes

• How you shouldn’t use AI to promote a children’s event

• The growing pressure on data centres and our power grids

• DeepMind’s work to create video games using GenAI

• Figure’s partnership with OpenAI on humanoid robots

• And how Glean are aiming to take on enterprise search

Have a good week!

Liam

Feedback

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 Software Engineer and works within Kainos' Innovation team. He identifies business value in emerging technologies, implements them, and then shares these insights with others.