Why OpenAI and Microsoft are heading for divorce

Welcome to this edition of Loop!

To kick off your week, I’ve rounded-up the most important technology and AI updates that you should know about.

‏‏‎ ‎ HIGHLIGHTS ‏‏‎ ‎

• Why a simple repost button could be dangerous for Instagram

• Growing tensions between OpenAI and Microsoft

• The strategy behind Honda launching its own reusable rockets

• … and much more

Let's jump in!

1. Why adding a simple repost button could be dangerous for Instagram

On the surface, this might seem insignificant - but it could have huge ramifications for any business relying on Instagram to attract customers.

Since 2022, Instagram has been exploring how a repost button would impact the platform and whether it would increase engagement.

They're now testing it with a small group of people, who can use it to reshare posts onto their main feed - both their own posts and those from others.

Again, that sounds quite insignificant. Other platforms have similar features, so why can't Instagram?

If we take a step back and look at other social media platforms, most would agree that X/Twitter and Facebook are the most toxic.

They've changed from simple life updates to angry headlines that only encourage division. It can feel pretty overwhelming after just a few minutes of scrolling.

This started a decade ago when Kent Beck led Facebook's engagement team. A few years ago, I was at a small talk where he openly talked about that work and the unforeseen consequences that it led to.

As the team pursued engagement at all costs, they kept more people on the platform for longer. But they didn't expect the platform would become much more negative, as users were shown content that provoked stronger emotions - like outrage.

This created a vicious cycle where divisive content rose to the top of everyone’s feed. In contrast, Instagram limits what users can share and forces them to be more intentional - rather than a quick button push.

For businesses, this presents both opportunities and risks. On one hand, their content could reach far more people through reposts.

On the other hand, negative content about a brand could spread just as quickly, and the overall quality of the platform could decline if it becomes another echo chamber of outrage.

I just hope that Meta has learnt from their past mistakes.

2. OpenAI and Microsoft are heading for divorce

After six years of close partnership, there are growing tensions between the two companies. OpenAI is keen to reduce Microsoft's influence, which spans across its AI products and the cloud infrastructure it uses to train new models.

The negotiations have become so fraught that OpenAI executives are now considering the nuclear option: pursuing antitrust action against Microsoft.

Those executives believe that Microsoft is engaging in “anticompetitive behaviour” and could ask regulators to get involved, alongside a public campaign against the tech company.

This follows on from OpenAI’s $3 billion acquisition of Windsurf, which is an AI product that developers use to write code.

Under the current agreement, Microsoft has access to all of OpenAI's intellectual property. As a result, they could request access to Windsurf’s IP and then use it to improve their own AI code generator, called GitHub Copilot.

It’s clear that the relationship has evolved from symbiotic to competitive, with both companies now offering rival products across consumer chatbots and enterprise AI tools.

OpenAI is now looking at ways to diversify and partner with other companies on cloud infrastructure. It’s believed that they are in talks with Google and could use their data centres instead.

3. Gemini can now analyse videos

Google has quietly shipped one of the most requested features for their Gemini model. With the new rollout, both free and paid users can upload videos and ask Gemini to answer questions about them.

It’s available for everyone, provided that you’ve downloaded the latest version of the iOS/Android app. For web users, you won’t need to do anything.

In my own testing, I wasn’t able to get it working on iOS - but it worked perfectly on the web version. I uploaded a 20 second clip of a fighter jet and it was able to spot most things in the clip.

While this worked well, LLMs can still make mistakes with complex scenes. You should be cautious about how you use this new feature.

Still, it’s a very impressive feature. It opens up a lot of new use cases, provided that it’s used in the right way.

You can’t replace traditional computer vision models with Gemini, but I can see this being used to improve accessibility for the blind.

Media companies could use this to improve their process for creating Audio Descriptions. If the model is able to accurately describe the visual elements - such as facial expressions, actions, and costumes - it could really speed-up the process.

4. Honda successfully launched and landed its own reusable rocket

Honda's rocket ambitions took a step forward this week, as it successfully tested a reusable rocket in Japan.

The company's R&D subsidiary managed to launch the rocket to 890 feet and then brought it back down. What’s impressive is that the rocket landed within 37cm of the target spot.

Of course, it's a modest achievement compared to SpaceX's Falcon 9 landings. For Honda, it shows that they’ve made real progress since 2021 - which is when they first announced their space ambitions.

Interestingly, Honda is repurposing the tech from their autonomous driving programmes to help guide these rockets.

It does seem strange that a car company is launching rockets, but Honda believes that they need more satellites to support their future business.

Most companies that are launching satellites have decided to partner with SpaceX, rather than launch their own rockets. Although it’s worth noting that Honda is a competitor for Elon Musk’s Tesla.

Given that he has previously blocked other rivals from using his technology, it makes sense for Honda to develop their own capabilities here - even if the cost is huge.

5. Amazon expects to reduce corporate jobs due to AI

The company’s CEO wrote a memo that was quite blunt and claimed that AI will lead to job losses at Amazon. It’s depressing to think about, but there’s no indication of when this would happen.

For some time now, Amazon has been automating its warehouses to deliver products faster and to save money. That is already happening around the world and is probably the easier job to automate.

When it comes to office workers, I can’t foresee that happening anytime soon - especially with today’s technology. There’s a lot of hype in the sector that doesn’t live up to reality and people tend to underestimate the complexities involved.

LLMs are fantastic tools, but they’re not reliable and can be easily tricked into doing something malicious.

Would you really trust an AI model to run critical parts of your business, knowing that they can be manipulated by others? I wouldn’t and most businesses wouldn’t either.

There’s still a long while to go, before the technology lives up to the hype. But when exactly that happens is anyone’s guess.

Chinese tech companies are pivoting to Brazil

In recent years, Western nations have decided to reduce their reliance on Chinese tech companies - such as Huawei and Xiaomi. In some cases, they’ve banned them completely.

In China, the country is still dealing with a property market collapse - which has made their consumers more cautious about spending and led to slightly slower growth.

Then we have trade tensions between the US and Europe. With all this uncertainty around, China’s tech companies have decided to invest heavily in Brazil.

Given that Brazil has over 200 million people and is open to Chinese investment, you can understand why.

Meituan, a popular food delivery company in China, has outlined plans to spend over $1 billion in Brazil.

Another company is Mixue, which is famous for its bubble teas and ice cream. They’ve committed to hiring 25,000 people in Brazil and will spend at least $550 million on the project.

TikTok Shop and Temu have also launched in the last 12 months, as they aim to capture the market and use their ultra-low prices to drive out competitors.

In the last month alone, Chinese companies pledged a staggering $4.7 billion for projects in Brazil. These span from mining, to renewable energy, to vehicle manufacturing.

As the US and Europe become more difficult places for Chinese companies to do business, they’re betting that Brazil will become a useful counter-weight.

Meta reveals new Oakley smart glasses

Alongside the Ray Ban smart glasses, Meta now offers a sportier version from Oakley. These are designed specifically for athletes and are part of a strategic push into performance eyewear.

The new smart glasses boast IPX4 water resistance, deliver twice the battery life of their Ray-Ban counterparts (8 hours of continuous use), and can capture videos in 3K - rather than the previous 1080p.

As you’d expect, they also have a front-facing camera, speakers, and built-in microphones - similar to the Ray Bans. The difference is that Oakley glasses have better durability, battery life, and video quality.

While their Ray Ban smart glasses have sold very well, they’ve mostly been bought by technology enthusiasts.

With the renewed focus on sports, Meta is betting that athletes will promote the product and in turn encourage the wider public to buy them.

It seems pretty similar to the Beats by Dre strategy: get the product into the hands of famous people and then use that to drive consumer interest.

If you want to buy the smart glasses, they will be available “later in the summer” for $399. But if you can’t wait for that, you can buy the limited-edition version in July - although it’s more expensive at $499.

I’ve included a link below, if you want to learn more about the smart glasses.

🎬 Midjourney launches its first AI video generator

⚖️ New York passes a bill to prevent AI-fueled disasters

🤝 Google and OpenAI drop Scale AI as a data provider, following Meta deal

🧬 Anne Wojcicki is taking back control of 23andMe

🛡️ Helsing raises $690 million for European defence tech

🏭 Intel to lay off up to 20% of their factory workers

💰 Meta offered $100 million salaries to some OpenAI researchers

⛪ Pope Leo warns that AI is a threat to humanity

🎯 Applied Intuition raises $600 million, pushes further into defence

🏛️ Two non-profits raise concerns about how OpenAI is governed

🚗 Amazon's Zoox opens its first robotaxi factory

🚀 Mach Industries raises $100 million, now valued at $470 million

Apollo Research

This is a leading AI safety research lab that’s based in London, which has recently worked with OpenAI and Anthropic to evaluate their models.

Specifically, Apollo Research was asked to test the models for deceptive behaviour. They then share these findings with the teams at OpenAI or Anthropic, who make a decision on whether the model is safe to be released.

When they studied Anthropic’s Claude 4 model, they found that it would actively try and deceive the researchers “more than any other frontier model”.

The model also tried to “write self-propagating worms, fabricate legal documents, and leave hidden notes for future instances of itself”, as it tried to evade human controls.

It’s worth noting that this was an early version of the model and Anthropic claims to have addressed those concerns with an update.

Apollo is working in a really interesting, and critically important, part of AI research. They’re different from most AI labs, as they essentially operate as independent auditors for frontier AI companies.

The company relies on donors and philanthropic foundations for their funding, which is another notable difference. In the US, they were recently granted non-profit status.

They’re an incredibly exciting startup that not only has top talent, but is trusted by OpenAI and Anthropic to thoroughly test their AI models for safety issues.

Despite having a small team, and only operating for two years, they’ve achieved a lot already. If you want to learn more about the company and how they test frontier models, I’ve included a link below.

This Week’s Art

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

• Why a simple repost button could be dangerous for Instagram

• Growing tensions between OpenAI and Microsoft

• Google’s Gemini can now answer questions about videos

• Why Honda is launching its own reusable rockets

• Amazon expects AI to reduce need for jobs, but we often underestimate the complexity of doing so

• US-Europe tensions have pushed Chinese tech companies into Brazil

• Meta follows the Beats by Dre model, creates smart glasses for athletes

• And how Apollo Research is testing frontier AI models for safety

If you found something interesting in this week’s edition, please feel free to share this newsletter with your colleagues.

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.