AI Daily News June 27: 🚀Google’s Gemma 3n brings powerful AI to devices 🫂 Anthropic studies Claude’s emotional support 🔔Altman vs. NYT: Privacy Is the New PR Weapon 🤖YouTube adds AI summaries to Search Results 🥊 Meta poaching OpenAI researchers
17:31 Share

- Hello there and welcome. This is a new deep dive from AI Unraveled. - That's right. - The podcast created and produced by the incredible Etta Newman, senior engineer and yes, a passionate soccer dad from Canada. - Great stuff he puts together. - Absolutely, and hey, if you're finding value in these deep dives, please do remember to like and subscribe. It really helps us bring more insightful content your way. - It makes a difference. - Okay, so today we're diving in.

Cutting through the noise, really, we want to bring you the, well, the most important nuggets of knowledge. And some surprising facts, too, I bet. Exactly. About where AI is heading, like right now. We're looking at recent developments using sources like the AI Innovations Daily Chronicle for June 27th, 2025. A lot happening on that day. You bet. So whether you're just catching up, maybe prepping for a meeting, or you're just really curious, our goal is simple. A shortcut. A shortcut.

to being genuinely well-informed, but, you know, without feeling totally overwhelmed. You'll get a good grasp of the latest breakthroughs, the practical stuff emerging. And maybe some of the bumps in the road, too. The challenges. For sure. Okay, let's kick things off. Two big forces really shaking things up right now.

The intense talent wars. Oh, yeah. And this relentless drive to get powerful AI everywhere, even onto our smallest devices. Right. And that first point, the talent wars. I mean, a really striking example is Meta's aggressive move. They poached four key open AI researchers. Four. Wow. For their new super intelligence unit, right? We're talking Lucas Beyer, Alexander Kolesnikov, Zhao Weizhai.

These are the folks who actually set up OpenAI's Zurich operations. So a whole team, basically. Pretty much. And they also snagged TrapAdvance all.

He was foundational to OpenAI's 01 reasoning model. Okay, the 01 model. Remind us what that is. Think of it like a key part of how OpenAI trains its models to actually think and draw logical conclusions, you know, not just spit back facts it learned. Got it. That's definitely a bold play by Meta, especially because didn't Sam Altman just last week say OpenAI's best people weren't taking offers? He did. And he denied those kind of wild $100 billion bonus rumors too. Right. So these hires...

they seem to fly right in the face of that narrative it really do i mean it despite altman's public statements these researchers are moving to meta it just signals met is serious about this they're willing to spend willing to invest especially after that huge 15 billion dollar scale ai investment right and bringing their ceo over exactly alexander wang heading up this new division it really shows

Meta's superintelligence team is, well, taking shape fast. Lots of top talent pouring in. Makes you wonder what their first release will look like. Definitely one to watch. So, okay, that's the battle for the human brains, the superintelligence race. But then there's this parallel push, right? Making AI more accessible. Bacritizing it, yeah. And Google's new Gemma 3N model seemed like a perfect example there. Absolutely.

If we zoom out a bit, Gemma 3N, it comes in these smaller 2 billion and 4 billion parameter versions. It's all about bringing powerful multimodal capability. Multimodal meaning images, audio, text, all that. Exactly. Processing and understanding all sorts of information and bringing that directly to your mobile phone or other, you know, everyday edge devices. And it's efficient, you mentioned. Incredibly efficient. It can run on hardware with as little as 2GB of RAM. Two gigs. That's tiny. It is. So imagine this.

Built-in vision capabilities on your Pixel phone, analyzing video at 60 frames per second, real-time object recognition, understanding the scene around you. Wow. Plus, it handles audio features across 35 languages and speech-to-text as well. That level of efficiency right on the device, that's...

That's really something. But usually there's a tradeoff, right? Efficiency versus raw power. How does GEMMA 3N actually perform? That's a fair question. The larger E4B version of GEMMA 3N, it's actually making some serious waves. It's the first model under 10 billion parameters to break the 1300 score mark.

on the Elmarina benchmark. Okay, Elmarina. That's like a standardized test for these big models. Exactly. A competitive benchmark. Yeah. So what this means is AI isn't just shrinking. It's powerful despite shrinking. We're shifting from needing the cloud for everything to having truly personal, always-on intelligence. So your phone isn't just running apps. It's

understanding your world in real time. All within its own power. Yeah. That's the quiet revolution that's happening here. Limitless potential for intelligent stuff right on your device. Man, the pace of all this. Talent moving, new models popping up. It really hammers home how vital it is to stay ahead in AI. No kidding. And

You know, if you are looking to boost your career, maybe get certified in AI. Etienne Newman has some amazing resources. Oh, yeah. His prep books. Exactly. Titles like Azure AI Engineer Associate, Google Cloud Generative AI Leader Certification, AWS Certified AI Practitioner Study Guide, Azure AI Fundamentals, Google Machine Learning Certification. Phew. Covers the big one. Totally. They're all available over at djamget.com.

And, of course, we'll put the direct links right in our show notes for you. Makes it easy. Definitely worth checking out if you're serious about leveling up. Okay, so moving beyond these, like, big tech battles, AI is also...

Quietly and sometimes not so quietly changing how we do everyday things. How we learn, how we shop. Even how we find old photos. Right. Let's start with learning. You found something interesting with Google Gemini. Yeah, this is pretty clever. A tutorial showing how you can use Gemini to basically turn lecture videos into really detailed study materials. How does that work? It's simple, really. You just upload your lecture video to the Gemini app.

then you prompt it you ask it to analyze this lecture video and provide you know a detailed outline comprehensive notes okay any formulas or examples mentioned and even time stamps for each topic discussed in the video that alone sounds incredibly useful for students like breaking down a long lecture totally but it gets better you can then follow up and ask it to create quizzes based on the content like practice questions comprehensive quizzes yeah complete with answer keys explanations

It can even code up a little interactive quiz for you, maybe with a hint button if you get stuck. Wow. Okay. The real power move here is saving all that generated material, the notes, the quizzes into one document. And then you do it again for your next lecture and the next. So you're building your own custom study library for the whole course. Exactly. It's leveraging AI to really personalize and supercharge your study routine.

making it way more efficient and tailored. That's a fantastic practical application. Speaking of practical, Google's also testing this new app, Doppel, for virtual try-ons. Yeah, Doppel. The idea is you give it a full-body photo of yourself and then just a screenshot of an outfit you see online anywhere on the web. And it, what, puts a close on you? Kind of. It generates an AI clip of you supposedly wearing that outfit. It's definitely a glimpse into maybe where online shopping could go.

But I hear there are some quirks. Oh, definitely. It raises questions about the current limits for sure. The tool sometimes really struggles rendering pants correctly. Struggles how? Well, sometimes it apparently creates fake feet, like just sort of pastes them on. Chuckles. Fake feet. Okay. Yeah. And it's even been reported to make people in mirror selfies look like way thinner than they are. Hmm. Not ideal. Not really. So while it works with clothes from anywhere, which is cool...

these glitches show AI still has a ways to go with, you know, nuanced visual details.

understanding the human form especially emotion is tricky yeah clearly okay what about just everyday media YouTube's doing AI summaries now yeah they're testing it out adding AI summaries to search results for some queries it's an opt-in experiment right now just for premium subscribers could be handy for getting the gist of a video quickly for sure convenient for the user but there's a potential downside right if people just read this summary maybe they don't click the video ah right which could hurt the

the channels, their views, their revenue. Exactly. Something to keep an eye on. Meanwhile, Google's Ask Photos, their AI search for your photo library that's back and apparently faster. They updated the Gemini model behind it. Yeah. Enhanced capabilities. It's evolving more into like a personal memory assistant and

really changes how you dig through your own visual history, your digital life. Like an AI curator for your photos. Pretty much. Making that massive library actually searchable and, well, intelligible.

You know, all these practical uses, they might get people thinking, hey, maybe I could build something with AI. It perks the imagination, right? Definitely. And if that's you, if you're interested in actually building stuff with AI, you absolutely have to check out Etienne Newman's other resource, the AI Unraveled Builders Toolkit. Ah, yes, the toolkit. It's great. It includes a whole series of AI tutorials, PDFs, audio, video formats, whatever works for you.

Plus, AI and machine learning certification guides. It's all designed to help you just start building your own AI projects. Big hands on. Exactly. The links, again, are right there in the show notes. Super valuable if you want to move from just learning about AI to actually doing AI. Good.

Good stuff. Okay, so let's shift perspective a bit. Think about the bigger picture, AI in the enterprise at work, and also maybe getting a reality check on some of the hype, the social impact narratives. Right. And on the enterprise front, a pretty big data point came from Salesforce.

Their CEO, Marc Benioff, revealed that generative AI is now handling up to 50%. 50%, half. Yeah, half of all their internal workflows, everything from sales operations to service ops. Wow, that's not trivial. Not at all. It's a really clear indicator of how a major enterprise software player is actively redefining workforce productivity.

It shows AI's impact on white collar roles isn't just theoretical anymore. It's deployed. It's happening now. Exactly. Inside a huge company. And if that's not transformative enough for you, having an AI powered startup studio aiming to launch, get this, 100,000 companies a year. Whistles, 100,000 a year. Well, they plan to use AI agents to basically ideate business ideas, validate them, and then deploy them as digital businesses. Whoa. Okay. Connecting that to the bigger picture.

That could be a seismic shift in entrepreneurship, couldn't it? How so? Well, you're moving away from the traditional model of, you know, a founder having an idea and building it slowly. This is more like AI powered company factories. Company factories. It fundamentally changes the barriers to entry, the scale of creation. Maybe it even democratizes starting a business in a totally new way.

or centralizes it, depending how you look at it. Interesting. Lots to unpack there, but okay. Amidst all of this, the massive innovation in these huge claims,

Sometimes you need that dose of reality, right? It's good to ground ourselves. And Anthropic published some research about how people actually use their AI, Claude, for emotional support. Right. And this study is important because it kind of pushes back against some common media narratives. It found that using Claude for emotional support or what they called effective conversations, it's actually far less common than you might think from reading the news. Really? How much less common? They looked at 4.5 million Claude conversations.

companionship, role play. That stuff accounted for less than 0.5% of the interactions. Okay, Tenny, so what were people using it for in terms of emotional support? Mostly practical concerns, things like advice on career changes or navigating relationship issues, stuff like that. More like a sounding board for specific

Exactly. And importantly, the study found users' sentiment often actually got more positive during the chat. It wasn't, you know, amplifying negative thought spirals, which has been a concern raised elsewhere. That's good to hear. It is. Now, it's worth remembering, Claude is generally more developer-focused than, say, ChatGPT or specialized platforms like Character AI. Ah, so maybe those other platforms see different usage patterns. It's possible, yeah.

But this study, at least for Claude, directly counters some of the more extreme stories about AI romance and deep dependency. It's a good reminder that sometimes the data tells a more nuanced story than the headlines. Absolutely. Check the data. Okay. So as AI gets more sophisticated, it inevitably starts bumping up against trickier areas: edge cases, reliability questions, privacy concerns. For sure. And in healthcare, we see both sides of this coin.

On the one hand, you have Alibaba's new AI model, Grape. Grape. Yeah, Grape. It analyzes 3D CT scans to detect gastric cancer. And a paper in Nature Medicine reported that Grape significantly outperformed human radiologists in identifying the disease in studies. Outperformed radiologists. Wow. Yeah. Potentially spotting it much earlier, too. That's huge potential for saving lives. A massive AI breakthrough. Incredible potential. But then there's

There's the flip side. Right. Then you see, for example, a Greek study that found even the best state-of-the-art AI models can get tripped up by simple things like slang or typos in medical exam questions. Slang and typos, like stuff humans handle easily. Exactly. And this highlights a really crucial point. If AI struggles with these imperfections, it undermines its reliability, especially in high-stakes areas like exams or even diagnostics. So the training data needs to reflect the real world warts and all. Precisely.

Precisely. Medical AI must be trained on the messy reality of how language is actually used if it's ever going to be a truly safe and dependable tool in hospitals or classrooms. It just reminds us, you know, AI isn't magic. Not yet. It's not a silver bullet. Got it. And then there's the ever-present privacy debate. Sam Altman took a pretty public shot at the New York Times recently, didn't he? He did. Over their lawsuit against OpenAI.

Specifically, he called out the NYT's demand that OpenAI should keep user chat logs, even when the user specifically chooses private mode. And Altman's take was? He basically said, that's not privacy, that's surveillance cosplay. Pretty strong words. Yeah, no kidding. Surveillance cosplay. So what's the bigger picture here for users, for you listening? Well, the implication is pretty significant. If the NYT gets what it's asking for, then that privacy

Private mode you click on in your AI tool. It isn't actually private at all. That feels like a major breach of trust. It's a massive user experience issue. Yeah. A potential trust crisis waiting to happen.

And think about companies trying to use AI in regulated fields like finance or healthcare. It becomes a compliance nightmare. So the key takeaway is? The takeaway for you is that in 2025, privacy policies aren't just legal documents anymore. They are core product features. Companies like OpenAI are making a bet. A bet on what? They're betting that you, the user, will ultimately choose the brand that actually respects your choices. The one that actually hits delete when you ask it to.

It's about earning and keeping that trust. Privacy as a feature, not just a policy. Makes sense. OK, before we wrap this deep dive for the 27th, just a quick lightning round of other notable things. Sure. We saw Black Forest Labs release FLUX.1 Context. Looks like a new step in AI image editing. DeepSeq's R2 model apparently hit some snags due to NVIDIA chip shortages. Ah, the hardware bottlenecks continue. Yep. OpenAI also released Deep Research via their API and Web Search in their 0304 mini models.

Hagen introduced Hagen Agent for easier video creation. More AI tools popping up, always. Meta won a fair use ruling in an AI training data case, which is significant legally. And Suno, the music AI company, acquired WavTool, another music creation platform. Expanding their toolkit? Wow. Okay.

It really was a packed day in AI. Never a dull moment. So let's try to pull this all together. What does it all mean? Today, we've really journeyed through a ton of AI innovation. From talent wars to tiny devices. Right. Fierce competition, incredible advancements, making AI personal. We looked at how it's changing learning, shopping, even how businesses operate. Transforming workflows, yeah. But we also hit those critical questions, didn't we?

about reliability, about the real social impact versus the hype and that constant privacy debate. Definitely. And zooming out, I think the big picture is clear. AI is charging ahead on so many fronts, but it's also really grappling with these complex, nuanced human elements. Like language quirks or what privacy actually means to people. Exactly. And that raises, I think, an important question for you, for everyone listening, as AI gets woven more deeply into our lives.

How do we strike that balance? The balance between? Between the incredible promise, the amazing capabilities AI offers, and the absolute imperative for ethical use, for transparency, and transparency.

crucially for respecting our personal boundaries. That's the big question, isn't it? Finding that balance. Well, we hope this deep dive has given you a clearer, maybe more informed perspective on where AI is right now, right at the cutting edge. Hope it was useful. And remember, if you're feeling inspired to build your own AI skills, maybe get certified. Etienne Newman's AI Cert Prep Books and that AI Unraveled Builder's Toolkit are really fantastic resources. Highly recommend them.

You can find all the links you need over at dgemgate.com. And like we said, they're right there in our show notes. Easy to find. Check them out. Well, thank you so much for joining us for this deep dive. Stay curious, keep learning. And we'll catch you on the next one.