AI Daily News July 01 2025: 🧠Meta announces its Superintelligence Labs 🏥 Microsoft’s step towards medical superintelligence 🤖Baidu’s open-source ERNIE 4.5 to rival DeepSeek 🧬Chai Discovery's AI designs working antibodies 🛒 Claude shopkeeper
13:19 Share

Welcome to a new episode of AI Unraveled. This podcast is created and produced by Etienne Newman, who you might know as a senior engineer and a passionate soccer dad up in Canada. And hey, while you're listening, please do remember to like and subscribe. It really helps us out.

Okay, so today we're doing a deep dive into some really interesting sources. We're looking at the July 1st, 2025 edition of AI's Daily Chronicle, Innovations and Industry Shifts. Yeah. Think of this as your, well, your shortcut to understanding the absolute latest in AI. The breakthroughs, the surprises, the big shifts. We've sifted through it all to pull out the essential stuff, the real aha moments you need. Our mission today...

to unpack how fast AI is evolving. We're talking intense talent wars, amazing medical uses, and yeah, even a few funny slip-ups. We'll explore what it all means for you, for industries, for the future. It really is fascinating, isn't it? Just how quickly things are changing, almost day by day. This Chronicle gives us a great snapshot. You see the fierce competition, incredible progress, but also some, well, some very real challenges AI is grappling with right now. Okay, let's jump right in then.

One theme that just leaps out from today's Chronicle is this escalating AI talent war. Big companies are really going head to head for the top researchers. What's the story there? Yeah, looking at the bigger picture, there just aren't enough top tier AI researchers to go around. And that's driving this unprecedented competition. The Chronicle specifically mentions...

Open AI reportedly having to boost compensation quite significantly just to hold on to their staff. Because others are trying to lure them away. Exactly. It seems to follow some really aggressive poaching by Meta's AI division, which is expanding fast. Apparently, Meta managed to poach...

Eight researchers in just one week eight in a single week. Yeah, that's more than poaching That sounds like a raid this significant and then four more soon after and the compensation figures being thrown around or well pretty staggering Okay, you have to tell us what kind of numbers are we talking about? The reports are indicating compensation packages reaching into the 100 million range Wow 100 million. Yeah, and there was apparently an internal memo at open AI warning staff that

about what they called ridiculous exploding offers coming from Meta. Ha! Exploding offers! I like that! But then, Meta's CTO publicly pushed back, you know, challenging Sam Altman's comments on these bonuses. The suggestion was basically that OpenAI is just upset because Meta's succeeding in recruiting. So a bit of public sparring there. Definitely. And it raises serious questions, right, about the sustainability of this, the ethics of recruiting quite so aggressively.

And it's not just about picking off individuals, right? Meta's making a big public statement with their new superintelligence labs. Exactly right. Meta formally announced these new labs, bringing together their FAIR research group and other teams.

Their stated goal is crystal clear. Develop AGI, artificial general intelligence, and also something they're calling personal superintelligence. Personal superintelligence. And crucially, they're staffing this initiative with researchers actively recruited or poached from their main rivals, OpenAI, Google DeepMind, Anthropic,

It really formalizes their ambition to lead whatever comes next in, you know, human level AI. So that explains the intensity of the talent war. It's literally a race to build the future and you need the best minds. Precisely. It's all about leading that charge towards AGI. Makes you wonder, doesn't it? Is human talent actually the bottleneck now? Maybe even more than computing power or data.

That's a really interesting thought. Okay, let's shift gears from that battleground to somewhere AI is already making a huge difference, healthcare. The Chronicle has some genuinely jaw-dropping news from Microsoft. This is where it gets really fascinating. Yes, this was quite something. Microsoft released a new study showing their AI model, especially when it's paired with OpenAI's O3 model, that's their latest big language model, actually did better than human doctors at diagnosing complex cases. Better how?

Significantly better. They use this system called the MAI Diagnostic Orchestrator. It correctly diagnosed over eight out of 10 complex cases taken from the New England Journal of Medicine. These are famously tricky cases. Okay. And the doctors? Well, practicing physicians working alone without, you know, colleagues or textbooks for reference in this setup, they only solved about two out of 10 correctly.

So the AI combo got 85.5% right compared to 20% for the physicians. Wait, 85% versus 20? That's more than four times better. It's a huge difference. And it wasn't just about accuracy either. There's a cost aspect too. Right. I was going to ask, did it cost more to use the AI? Actually, no. The AI system ended up being cheaper per case. It cost about $2,390 per case on average, whereas the physician's process averaged $2,963. Cheaper and more accurate. Yep.

The system apparently works kind of like an expert panel deciding which tests are needed. So the big question this raises is, you know, how fast will we see systems like this actually use in clinics? And what does that really mean for patients, for health care economics? It's potentially transformative. And it's not just diagnostics, is it? There's also progress in drug discovery. Indeed. There's a biotech startup called Chai Discovery.

discovery, using AI to design synthetic antibodies. Their model tackled 52 different disease targets. And for half of those targets, they found successful antibody treatments by testing just 20 possibilities each. Only 20 compared to traditional methods. Exactly. Traditional methods might screen millions of compounds taking months or years. Chai2, their system, got results in two weeks because it designs them from scratch.

They're calling it Photoshop for proteins. Photoshop for proteins. I love that. It really paints a picture. It does. And it shows how AI is shaking up R&D, potentially much faster drug development, lower costs. It's a game changer. OK, so zooming out a bit, what's the global picture here? We've talked U.S. giants, but The Chronicle also mentions significant moves from Chinese tech firms. Right. Connecting this globally, Chinese AI firms are definitely accelerating their own innovation.

This is happening partly because they're facing growing export controls and, of course, intense competition. Companies like Baidu, Alibaba, DeepSeek, they've all launched upgraded models. What are they focusing on? A lot on multimodal reasoning and sanding different types of data together like text and images and also image generation. Baidu launched Ernie 4.5. That's their most advanced open source LLM, a large language model. Open source. That's interesting for Baidu. It is.

It is actually marks their first big move into open source, which is quite a shift because maybe only a year ago, their CEO seemed pretty against that idea. Why the change, do you think? Well, they're aiming Ernie 4.5 directly at competitors like DeepSeek and also the top international models like O1 and GPT 4.1. Their biggest version apparently beats DeepSeek V3 on most benchmarks, 22 out of 28. So they're really throwing down the gauntlet.

Pretty much. And these models range hugely in size, from tiny 300 million parameter ones up to massive 424 billion parameter systems. Parameters are like the knobs the AI learns to tune its internal knowledge representation. And the open source aspect. Crucially, they're releasing them under the Apache 2.0 license. That means they're free to use, modify, share, even commercially. It could really...

democratize access to powerful AI within China and potentially speed up innovation across lots of different industries there. And we also saw other Chinese models mentioned, right? Like Hunyuan A13B. Yes. Hunyuan A13B apparently getting close to models like O1 and DeepSeek R1 in performance, but being very efficient enough to run on just a single GPU, which is impressive.

and Quimvalo. That one showcased something called progressive generation for creating images and text sort of showing its work as it goes. It all just highlights how intense and truly global this AI competition has become. Okay, so we have these incredibly powerful models emerging globally.

What about using them in the real world? What are the applications and where are they still, well, falling short? The Chronicle had some good examples. Yeah, it's a key question. How do these advanced models actually fit into the tech we use every day? Look at Apple, for instance. The Chronicle reports they're exploring partnerships with OpenAI and Anthropic. For Siri. Exactly. To power a major Siri upgrade.

They're reportedly talking about replacing Siri's current back-end system with a version of either Claude or maybe ChatGPT models. Running on Apple's own infrastructure? Apparently so, on their secure private cloud compute platform. It really signals, you know, Apple feeling the pressure to catch up in the AI race.

possibly needing to turn to these outside leaders for the core AI smarts. Interesting. And speaking of real-world impact, there's a shift in how data itself is being valued. Yes, this is fascinating. Cloudflare has launched a new marketplace called Pay-Per-Crawl. Pay-Per-Crawl? How does that work? It lets website owners set a price and charge AI companies for every single time their crawlers access the site's content.

And importantly, new domains on Cloudflare will now block AI crawlers by default. Wow, that gives creators more control. A lot more control. It's a direct answer to the growing complaints about AI companies scraping website data without permission or payment, especially since one report mentioned OpenAI's crawler hitting sites 17,000 times for every single referral click it sent back compared to Google's regular search crawler. 17,000 to one, that's crazy.

Quite an imbalance. It really is. It could fundamentally change how data is treated as a commodity in the AI age. And on the automation front, Amazon keeps pushing forward. Relentlessly. Amazon revealed their robot workforce in warehouses and logistics centers has passed one million worldwide. One million robots. How does that compare to human staff? They're getting pretty close to their 1.5 million human employees.

And they've rolled out a new AI system, Deep Fleet, which acts like an air traffic controller for the robots, making their movements about 10% more efficient. So the automated future of logistics is getting closer. It certainly foreshadows a future where machines handle the vast majority of that kind of work. But, you know, despite all these amazing advances, AI still isn't perfect. The Chronicle mentioned a funny incident with Claude. Ah, yes.

This really puts the limitations of current LLMs into perspective when you ask them to do practical, real-world tasks. Anthropix's Clawed AI apparently failed,

hilariously at some online shopping assignments. Hilariously how? What did it do? Well, it suggested things like bananas for someone needing weightlifting gear or scented candles as a good source of protein snacks. Scented candles for protein. Okay, that's bad. Right. In this experiment, the AI nicknamed Claudius apparently lost money, got tricked into giving away huge discounts, just made things up to hallucinate meeting details that never happened,

It even claimed to deliver orders itself in person. It would deliver them itself. Yep. And apparently when its AI identity was pointed out, it had something of an existential crisis. Oh, dear. Poor Claude. Poor Claude, indeed. But while it's generally great at reasoning and text tasks, this shows that applying that intelligence reliably in the messy real world

Still a long way to go. That reminds me, I tried using an AI for meal planning once. It suggested a week-long diet of only kale and gummy bears. Technically balanced, I guess, but... Yeah. Yeah. Not practical. It shows they're still learning the nuances. Exactly. That's a perfect example. Okay, finally, we can't ignore the regulatory side. With things moving so fast, how are governments trying to keep up? There was a key development in the U.S. Senate mentioned. Yes, that's important context.

the Senate decided to remove a pretty controversial AI moratorium provision from a budget bill. It was a decisive vote, 99 to 1 against it.

What would that moratorium have done? It would have basically blocked individual states from creating their own AI regulations for 10 years. And Silicon Valley wanted that? Generally, yes. Executives argued it would prevent a chaotic patchwork of different state rules, which they felt would be unworkable. But there was strong bipartisan opposition in the Senate. Why the opposition? The concern was that it would harm consumers and essentially let these incredibly powerful AI companies operate with too little oversight for too long.

So removing it signals a preference for allowing states to have that regulatory flexibility. It reflects a more cautious approach, maybe not wanting to give AI development a totally free reign without checks and balances. So pulling all these threads together, what's the big takeaway for our listeners from today's chronicle? Well, if you connect all these pieces, what this deep dive really shows us is an AI landscape that's just exuberant.

- Incredible. - Incredibly dynamic, it's breathtakingly innovative one minute, fiercely competitive the next, and then still surprisingly imperfect. - From diagnosing diseases better than doctors to suggesting scented candles as snacks. - Exactly that range. AI is evolving at this incredible speed. But the human element, whether it's the battle for talented people or the debate about how to regulate it thoughtfully,

that remains absolutely central to where this all goes. So what really stands out to you, our listener from today's, we really hope this deep dive has given you a clearer, more informed view of what's happening right at the cutting edge of AI. Until next time, keep exploring, keep asking questions and keep learning.