🧬 DeepMind’s AI breakthrough can model “all of life’s molecules”

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 & OpenAI’s fund to tackle election deepfakes

• Why Stack Overflow is now giving OpenAI their data

• Amazon’s new tool to simplify GenAI development

• … and much more

Let's jump in!

1. Microsoft and OpenAI launch an education fund for deepfakes

Given that we’ll see 75% of democracies go to the polls this year, there are a lot of concerns about how voters could be influenced by misinformation campaigns.

With the rise of generative AI and deepfakes, those concerns have grown significantly. To counter this, all of the big tech companies have outlined their plans to prevent misinformation campaigns.

Microsoft’s new fund will be used to boost “AI education and literacy” among voters and elderly people. The money will be allocated to several organisations, such as Older Adults Technology Services (OATS).

It’s interesting how they’re putting more of an emphasis on educating the older generation, as previous initiatives have been focused on everyone.

However, $2 million is a tiny amount of money for these types of campaigns and is unlikely to make a big difference. There’s also the risk that this is only seen as something that older people can be tricked by.

That’s not the case at all. All generations are vulnerable to being deceived by deepfakes.

2. Amazon’s new tool to simplify GenAI app development

Bedrock Studio is a new offering from Amazon, which will allow them to compete with similar initiatives from Microsoft Azure and Google Cloud.

It will allow you to quickly prototype ideas, experiment with different AI models, collaborate with others in your team, and then build the app. Bedrock Studio is able to automatically deploy all the AWS resources you need.

What’s particularly good is that you can use Studio to directly set guardrails for your app. This means that you can restrict what the user can submit and prevent the model from responding with inappropriate content.

It’s great to see these new tools being announced, but I’m more interested in what they build for multi-agent apps. Having one LLM is good, but having multiple talk to each other is the real game-changer (if done correctly).

3. TikTok sues the US government over their potential ban

The social media platform has filed a lawsuit against the United States government, as it tries to stop a law that would ban the app.

The ban will come into force if TikTok isn’t sold to another company within 12 months. Their lawsuit argues that the new legislation is in violation of free speech.

Both Democrats and Republicans are worried about TikTok’s ability to collect data about millions of Americans, with this seen as a huge national security risk.

This is a legitimate concern, as TikTok’s parent company is based in China and could be ordered to provide that data to their security services.

That data is a gold mine, as it gives a clear insight into what each American’s interests, dislikes, and current political views.

It’s easy to imagine how they could be targeted in the future, in the hopes of swaying the US election, or even blackmailed.

4. Stack Overflow agrees to give OpenAI their data

This is a strange decision. Every software developer uses Stack Overflow to get suggestions from the wider community on how to tackle a problem.

As a result, it has a lot of data about how to solve obscure and difficult tasks. Most responses also detail how their solution works and why it’s better than others.

That’s incredibly useful for Large Language Models, which can write code that does a decent job - but fails to do anything more complex. That’s a problem, as it limits how they can be used by developers.

As a result, OpenAI have struck a deal with Stack Overflow to use their data and train future models. The weird thing about this deal is that it seems very one-sided.

Once OpenAI can create a model that can solve 80% of code problems, developers will rarely need to use Stack Overflow. Of course, the website will get a lot of money in the short term - but they’re helping to build their future replacement.

It seems like they’ve read the writing on the wall and are trying to cash in, while they still can.

5. Altera raises $9 million for AI agents that can play games

The startup has developed an agent that can play Minecraft like a human, with it even receiving funding from Eric Schmidt (former CEO of Google).

Altera's agents are designed to be companions, rather than assistants, which are capable of making their own decisions and can interact with players more naturally.

The company is currently testing the model with 750 Minecraft players and plans to officially launch later in the summer.

The intersection of AI agents and gaming is something that I’ve raised previously, as it allows companies like Microsoft’s Xbox to build out more subscription offerings for gamers - which is an accelerating trend in the industry.

OpenAI will let content creators ‘opt out’ of AI training

It has taken a long time, but OpenAI will finally create a new tool to let creators opt-out and stop their data being used to train new AI models. It’s hoped that the launch will happen by 2025, with other companies involved in creating an industry standard.

Ever since AI image generators became popular, artists have been deeply worried that their livelihoods are being destroyed. They also feel powerless, as AI companies use their copyrighted images to make new models - without permission and without payment.

When I was at SXSW this year, which features a huge number of people in the arts, there was a tense atmosphere. Videos that featured the Head of ChatGPT were booed by the audience and he was asked if artists should be paid for their work. Unfortunately, there was no answer.

I even heard one person ask an artist if they could take a picture of their display, with the promise that it “wouldn’t be used to train an AI model”.

The artist laughed a little, but it’s clear that attitudes are shifting and technology is no longer being seen as a net positive for everyone - and for valid reasons.

While this is a good and necessary first step, small artists shouldn’t have had to fight for this. The downside is that their images have already trained the current generation of models - so that can’t be undone.

DeepMind’s new AI can model “all of life’s molecules”

AlphaFold 3 is an advanced AI model that can predict the structure of proteins, DNA, RNA, ligands, and their behaviour.

It’s hoped that the new model will benefit researchers in various industries - such as medicine, agriculture, materials science, and drug development - by allowing them to test potential discoveries more effectively.

It works by using a diffusion method, which is similar to those used by AI image generators, to then generate 3D models of the new structures - based on a list of input molecules.

Interestingly, AlphaFold 3 can generate predictions that are 50% more accurate than its predecessors. Isomorphic Labs, which is another company owned by Google, is currently using the technology to discover new drugs.

DeepMind has also launched AlphaFold Server, which is a research platform powered by AlphaFold 3. It will be available for free to some researchers, which will let them generate biomedical predictions and without the need of huge computing resources.

This is a really good decision, as universities have struggled to keep up with the big tech companies and simply can’t match their spending power. With this free web platform, it helps to even the playing field for university researchers.

💬 OpenAI plans to announce updates to ChatGPT today

💲 FTX’s crypto fraud victims will get their money back — plus interest

🎨 OpenAI is working on a new AI image detector

📉 Rivian loses $1.45 billion as it continues to cut costs

💰 Lucid Bots secures $9 million for drones that can clean your windows

📢 Meta’s AI tools for advertisers can now create full new images

🤖 AI deepfakes dominate the Met Gala

🩻 Rad AI raises $50 million to help radiologists generate reports

🪖 Xtend raises $40 million to develop controversial drones for military use

Wayve

Wayve is a startup that’s focused on building software to enable autonomous driving. They’re based in Cambridge, UK and have just secured an impressive $1 billion in Series C funding.

The company was founded in 2017, after they were able to successfully drive a car around the city - without the need to move the steering wheel.

Their technology is already being used on delivery vans for both Asda and Ocado. With these new funds, the startup is aiming to sell their technology to car manufacturers and robotics companies.

Wayve is currently working with several of the top car companies, but aren’t in a position to publicly name them just yet.

They’ve received praise from the UK Prime Minister for their funding success and are working closely with the UK Government on regulations for self-driving vehicles.

Given that the UK is keen to get self-driving cars on their roads, and are funding several huge research projects in this area, Wayve is one to watch closely.

This Week’s Art

We’ve covered a lot this week, including:

• Microsoft and OpenAI’s education fund for deepfakes

• Amazon’s new tool to simplify GenAI development

• Why TikTok is suing the US government

• Stack Overflow’s strange deal with OpenAI

• Altera’s progress on AI agents that can play games

• OpenAI’s plans for an opt-out tool

• DeepMind’s new AlphaFold model

• And Wayve’s plans for autonomous vehicles in the UK

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