Time for a catch-up!

I think that in years to come, we will look back on this week as being one of the most significant in the history of AI development for potentially opposing reasons.

So let me set the scene. At Apple’s World-Wide Developers Conference (WWDC) there was a sense of disappointment that there was still no update of Siri’s AI update, a year after it was promised at WWDC 2024. Apple’s position is that their approach to AI development is incremental in nature, rather than “leap-frogging”. Given the fact that this approach had not failed them in the past, you would be a brave person to bet against them.

Contrast that to Meta’s stated intention to invest $65 billion in AI by the end of 2025 (and we’re half-way through already!). They underlined that this week with their purchase of 49% of Scale AI stock for $14.3 billion, and the announcement of an intention to create a Superintelligence Project, led by Scale AI’s CEO, Alexandr Wang. The aim of the project is to finally achieve the “Holy Grail” of Artificial General Intelligence (AGI).

Why is Scale AI so important? Well, the most crucial resource for building advanced AI is meticulously organised training data, because if AI models want to learn, they require vast datasets that have been carefully annotated by humans. Scale AI just happens to be the leading tool for doing that and is also used by the likes of Microsoft and OpenAI. Add to that the fact that only 10% of the internet remains freely trainable for models, with more and more platforms (Reddit, etc.) limiting access, and you can see the logic. In the “Gold Rush” for AI training data, Meta have bought the “Caterpillar digger’, and everyone else has picks and shovels.

But here’s the twist. This week, Apple and some researchers quietly released a paper that basically says that the latest form of artificial intelligence technology suffers “fundamental limitations” that result in a “complete accuracy collapse”. Without delving into the complex scientific background, the research basically said that the current ways of training AI have effectively hit a glass ceiling, something that Gary Marcus has been warning about for the last few years.

Does that mean that AI development will stop – definitely not. It does, however, perhaps explain why Apple have not pushed forward so quickly with the Siri upgrade. It also underlines why Meta’s massive investment and radically different approach to the issue might be the only thing to move the “needle” in terms of progressing towards AGI.

As always, I’d like to say that was all that’s happened this last week, but they’ve been loads more things, so let’s dive right in …

Big Sharks vs. small fish

Apple and Meta were also both in the news due to new releases this week. Apple’s machine learning researchers revealed a new AI image generation system, STARFlow, which could rival OpenAI’s DALL-E and MidJourney. Unlike diffusion models, STARFlow uses a combination of transformative algorithms to achieve competitive results with high efficiency.

Meta also released a new AI model, V-JEPA 2 What’s different about this is that it is designed to change how machines understand the physical world. By teaching it to understand concepts like gravity and object permanence, V-JEPA 2 can predict real-world outcomes and adapt like humans. This is likely to have a big impact on the development of robots but will also impact on issues like autonomous vehicles.

Staying with the theme of vision, after beta-testing its Video Generator, Amazon in the USA is now offering US sellers free tools to create realistic, eye-catching ad videos in less than five minutes. Interestingly, Amazon are also looking at expanding its drone-delivery service to more cities in the USA, meaning that it will continue to pose a threat not just to the creative advertising industry, but also the logistics industry.

We also had the news that Qualcomm unveiled plans for futuristic “smart glasses” integrated with generative AI. This underlines the importance of Qualcomm’s recent $2.4B deal with Alphawave, which indicates that AI devices are likely to become increasingly important in terms of both investments, but also revenue streams.

Talking of revenue streams, despite the fact that OpenAI continues to rake in hundreds of millions, it agreed this week to co-operate with Google on the development of data centres. This again underlines the other continued key area of AI investment, namely infrastructure.

Meanwhile, over at Google, the Gemini 2.5 Pro AI suite has quietly taken the crown for coding tasks, outperforming OpenAI’s GPT models and Elon Musk's xAI tool Grok Beta.

Sticking with software development ElevenLabs also unveiled Eleven v3 (Alpha), a powerful text-to-speech model offering precise emotional control, multilingual support, and nuanced non-verbal expressions.

It is worth remembering that with regards to things like vision and voice, the current (incredible) level of development is the worst it will be moving forward.

Legislation, policy and other news

Much to the disappointment of many in the creative industries, the UK’s boringly named Data (Use and Access) Bill was finally passed this week. Concerns remain over the level of intellectual property protection it offers for artists.

In contrast xAI announced that developers can't use its API to train AI models. It’s a huge reversal from X’s earlier stance where public data was fair game for training.

In terms of ethics, concerns are being raised about AI companies like OpenAI and Google aggressively pushing AI tools onto college campuses, offering services tailored for students (e.g., career advice, tutoring).

Critics argue these tools erode critical thinking and social skills, with some studies showing AI reliance can lead to misinformation and shallow learning.

Equally worrying was the news that 44% of IT professionals report having no formal governance policy for AI tools in their workplace.

Unfortunately, this week I have no medical breakthrough supported by AI to report on, but at least AI has been helping unlock the contents of scrolls buried by Mount Vesuvius in 79 AD, so it’s not all doom and gloom.

Stay informed. Stay critical. And wherever possible—stay ahead.

Regards

Tom Carter