Anthropic's Claude can now create games and play them

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 ‏‏‎ ‎

• OpenAI’s co-founder starts a new superintelligence company

• Why Anthropic’s artifacts feature is a huge deal

• Runway’s video generator makes a big leap forward

• … and much more

Let's jump in!

1. US bans the sale of Kaspersky software, citing security risk from Russia

The US has raised new sanctions against 12 executives at Kaspersky, which is a Russian cybersecurity company that provides antivirus software.

This follows a growing trend as Western nations try to de-risk their industries and move away from Russian and Chinese technologies.

Foreign tech companies are seen as much higher risk, since it’s possible that their national government will order them to share sensitive data.

The US government has been wary of Kaspersky for some time and in 2017 they banned all their federal agencies from using the tool.

Kaspersky software will be banned in the US from July 20th and users are being urged to quickly move to other providers.

For those companies who run critical infrastructure, the US CISA will provide them with support and help to find an alternative security tool.

2. Apple is in talks to partner with Meta

This follows on from their partnership with OpenAI, which will see ChatGPT integrate with a revamped version of Siri.

Apple is also in talks with Meta for a similar deal, but nothing has been finalised yet. This is being done to reduce their reliance on OpenAI, which almost collapsed last November after their CEO was briefly ousted.

It’s an interesting move, as Apple seems to be positioning itself as a “distribution partner” for AI companies.

This means that they don’t have to pay for access to the technology, but instead allow the AI companies access to hundreds of millions of users - with the possibility that some will sign up for premium subscriptions.

Given that billions of dollars are now needed to train these models, this could be a very good revenue stream for AI companies.

It also works well for Apple, as it gets an inside look at the next-generation of AI models from different providers.

3. OpenAI’s former chief scientist has launched a new AI company

Ilya Sutskever was one of the co-founders of OpenAI and his new company is called Safe Superintelligence Inc. (SSI).

Their sole focus is to achieve superintelligence in a way that’s safe and can be controlled.

During his time at OpenAI, Sutskever played a crucial role in the company's efforts to improve AI safety and worked closely with their internal Superalignment team.

However, Sutskever and most of that team left OpenAI in May. This was due to the company not investing enough on AI safety, which impacted the work they were doing.

Unlike OpenAI, which initially launched as a non-profit organisation before restructuring, SSI has been designed from the outset as a for-profit entity.

It will have offices in Palo Alto and Tel Aviv, with a recruitment drive now starting to lure technical talent from other AI companies.

4. DeepMind’s new model can generate soundtracks for videos

This is a really interesting area, as today’s video generators can’t synchronise sound effects with the video that they created.

DeepMind’s research shows that they can create music, sound effects, and dialogue that match the characters and tone of your video.

While the generated audio isn’t convincing right now, we’ve seen rapid progress in the last few years and I’m sure the same will be true for this technology.

To guard against misuse, the generated audio is watermarked by DeepMind to clearly indicate that it was created using AI.

Google’s watermarking tool is far better than their competitors, but there’s still the possibility that bad actors can get around it.

The model will be made available to the public in the future. In the meantime, DeepMind is working with the film industry to ensure it’s safe.

5. OpenAI buys Rockset to bolster its enterprise offering

Rockset’s technology could help OpenAI to reduce hallucinations in their models, which is preventing a lot of enterprises from fully adopting the technology.

We don’t know the financial terms of the deal, but Rockset had previously raised over $117 million from investors.

OpenAI is aiming to sign more enterprise customers, as it aims to rapidly grow revenue to over $3.4 billion this year.

Rockset is OpenAI's second public acquisition, after it purchased Global Illumination last year, which was a startup that developed AI tools for the creative industry.

Anthropic’s new model beats GPT-4o

Claude 3.5 Sonnet is now available for free on the web and iOS. It’s being positioned as the middle option in Anthropic's lineup between Haiku 3 (smallest model) and Opus 3 (biggest model).

What’s fascinating is that their 3.5 version can outperform the bigger Opus 3 model, especially when it comes to tasks like analysing images or writing code.

If you want a model that’s good at writing text, you should still use Opus 3. It’s my go-to model for analysing documents and research papers.

However, their new Artifacts feature is even more exciting and could have a huge impact on how people complete their work.

An Artifact is a window that opens beside the chat interface. It will show images or run code for you, which you can easily interact with.

This opens the door to a lot of new use cases, as it makes it much easier to analyse documents, convert information into interactive graphs, and run basic computer programs - all within the same browser window.

Some have already used it to create 2D games, play Texas Hold ‘em, or even simulate musical instruments - such as the piano.

It’s clear that Anthropic is trying to position Claude as the single place to complete your work and immediately see the results, without the need to run other software.

What that means for other tools, and how they integrate with these LLMs, remains to be seen - but it doesn’t look like it will take long for us to find out.

Runway’s video generator catches up with OpenAI

Runway is a tech company that’s developing generative AI tools for film and content creators.

They’ve been popular in the industry for some time now, but OpenAI soon grabbed the headlines as their tool created videos that were much higher quality.

Runway have now released their latest model, which can now generate videos faster, at much better quality, and offers greater control over how the videos will look.

While their videos are limited to a maximum of 10 seconds, the new model seems to be a lot better at showing human expressions - with a wider range of actions and gestures possible, compared to the last version.

The startup hasn’t outlined what data it has collected to create this model, due to the highly-competitive landscape and the growing risk of lawsuits.

Runway and OpenAI aren’t the only big players here. Luma, Adobe, and Google have made similar leaps forward in recent times - with video generators now able to create content that’s high definition and incredibly realistic.

There’s no doubt that the film industry will be impacted by this technology, although it’s not quite at that stage.

The main issue is that the videos are short-form and can look noticeably “strange” when blown up onto a larger screen.

However, given the rapid progress we’ve seen in recent years, that might not be the case for long.

📈 Nvidia overtakes Microsoft as the world's most valuable company

⚛️ Congress votes to advance nuclear energy development in the US

🚨 Security bug allows anyone to spoof Microsoft employee emails

🤔 As media companies sign deals with AI firms, journalists question how it benefits them

👚 This chatbot can suggest outfit ideas based on your colour type

💻 Poolside is raising over $400 million to build a co-pilot for programmers

🇫🇷 France leads the pack for generative AI funding in Europe

🔍 Daydream raises $50 million to build an AI search engine for e-commerce

Waabi

Waabi is an interesting autonomous trucking startup that’s based in Toronto.

The company is currently being supported by Geoffrey Hinton, who is known as one of the “Godfathers of AI” for his contributions to the industry.

It was founded in 2021 and recently raised over $200 million in Series B funding - with also Uber joining as an investor.

Unlike Tesla's Full Self-Driving system, which uses "imitation learning," Waabi's AI has been trained using a closed-loop simulator called Waabi World.

Waabi's simulator will automatically build a digital twin of the world and can teach their AI system to learn from its mistakes - without the need for human intervention.

What’s particularly fascinating about their approach is that significantly less data and computing power is needed to create their self-driving models.

Their system is currently being tested in Texas, as the state offers more favourable weather conditions and regulators are supportive of their testing plan.

The company plans to launch fully autonomous trucks in 2025, which is an ambitious target. If you want to read more, you can check them out using the link below.

This Week’s Art

The week has been dominated by Anthropic’s release of Claude 3.5 Sonnet, which has seen huge improvements in performance for programming and data analysis.

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

• Why the US has banned the sale of Kaspersky’s antivirus software

• Apple’s talks with Meta on a potential partnership

• Ilya Sutskever’s new superintelligence startup

• DeepMind’s model that can generate soundtracks for videos

• Why OpenAI has bought Rockset

• Anthropic’s new Claude upgrade and Artifacts feature

• Runway’s new video generator that matches Sora in quality

• And how Waabi are planning for autonomous trucks in 2025

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.