Imagine an AI that can absolutely ace medical exams. I mean, outscore real doctors. But then when you or I try to use it for something simple like a stomachache, it actually makes things worse. That's not science fiction. That's the intriguing paradox we're unpacking today. Welcome to the Deep Dive. Yeah, we're really going to explore some fascinating corners of AI in this deep dive. We're pulling insights from a really rich source, a recent newsletter packed with
the latest developments. Today, we're looking at everything from why these AI tools, even the brilliant ones, can be kind of double -edged sword in the wrong hands. Yeah. All the way to the quiet revolution of AI running locally right there on your device. Right on your own machine. Exactly. And then there's this flurry of new tools changing, well, everything from how we see ads to how robots learn. Okay. Oh, and just so we're on the same page, when we talk about
LLMs, we mean large language models. Just think of them as like advanced chatbots that learn from truly massive amounts of text and data. Yeah. They understand, summarize, generate stuff. Got it. Okay, so let's really unpack this first one. The AI doctor dilemma. If AI is so incredibly smart, I mean, demonstrating near perfect knowledge, why does it seem to fall apart when an everyday person, you know, just needs help with a common sickness? It feels counterintuitive. It absolutely
does. And the research highlighted in our source material, it really drills into this. They ran this study involved about 1300 people using advanced LLMs like GPT -4 -0 trying to navigate. medical scenarios. Now, when these LLMs were just doing their thing, responding to clinical prompts, their accuracy was astonishing. We're talking 90 % to 99%. Wow. Incredible, right? But here's the twist. When humans got involved trying to self -diagnose using these same tools, the accuracy
just plummeted. Plummeted how much? Only 34 .5 % of the scenarios were correctly assessed. And here's the truly wild part. A control group just using Google, they actually did better. Better than the advanced AI. Yep. 47 % accuracy. The data suggests adding this powerful AI actually made people worse at self -diagnosing. That's, yeah, that's pretty sobering. What's going on there? Is the AI giving bad advice or is it how
people are using it? It's rarely about the AI giving flat -out bad advice, not with these professional -grade models anyway. The reasons the research uncovered were quite clear and, honestly, very human. Like what? Well, first, users often gave incomplete information. They didn't have the diagnostic discipline, you know. Second, frequent misinterpretation of what the AI spat out. The AI might list possibilities, but people kind of latched onto one or just didn't get the nuances.
And maybe most frustratingly, people just ignored good AI advice. Ignored it, even when it was right. Even when the LLM correctly flagged relevant conditions, only about one in three people actually use that information. I have to admit, I still wrestle with prompt drift myself sometimes, you know, where the AI's answers kind of shift subtly over time. So the idea of misinterpreting or overlooking AI suggestions, well, it isn't totally
foreign to me either. So it's less about the AI's raw intelligence and more about the human interface. And maybe the training needed to use that intelligence safely. Because like you said, in professional settings, AI seems to be making a huge difference. Exactly. It highlights that critical gap between raw, super accurate AI data and what you might call actionable, trusted guidance. A doctor brings years of training, right? deep context about a patient, the ability to ask precise
follow -ups. Right. An average user often lacks that. They might not even know what to ask next, or they might see a probability as a definite diagnosis. Which it isn't. No. Professionals, though, they have the framework to interpret that data correctly. Take open evidence, for example. It's a diagnostic engine trained only on peer -reviewed medical literature. Not for consumers. It's for pros. Okay. And the numbers. One in four U .S. doctors use it. On average,
10 times a day. Especially valuable in complex areas like oncology. It just shows how powerful these tools are when they're in the right hand, used the right way. So what does this all mean for us then? This tension between consumer use and professional success. It seems to hinge on who's using it and why. It's not about needing a robot doctor maybe, but something different. Right. It means AI can save lives? Absolutely.
Only if we design it to genuinely help people in a way that respects the complexity of human interaction and expertise, not just design it to ace a test. So if AI gives correct advice, but people don't listen, what's the real barrier then? It's about bridging that raw data with actionable, trusted guidance. That's the core challenge. Okay, shifting gears a bit, we often hear about AI in these giant data centers, you know, humming away in the cloud, needing constant
internet. But what if the next big shift is bringing that power much closer? Right. Right onto your device. That's exactly what's happening. There's a really strong push for AI that lives directly on your machine, your laptop, your phone, maybe even your robot vacuum. Huh. Why the shift? Well, a lot of users and developers, too, are getting kind of tired of unpredictable cloud AI updates, right? And the occasional outages. Local AI offers much better stability. Your workflows stay consistent.
Okay. Stability makes sense. And you get far greater control over the AI models themselves. And, crucially, enhance privacy for your data. Think of it like having your own personal AI assistant. Always there, always ready, and your info doesn't need to travel across the internet. That's a big deal for privacy. It is. It's a fundamental shift in how we might interact with AI. Moving the processing from some distant server farm right into your pocket or on your desk,
it gives you a level of consistency. Cloud services just can't always guarantee. Huge benefit. So what's the core advantage then for choosing local AI over the cloud? Greater control, consistent performance, and stronger data privacy, bringing the power to you. Okay. So beyond these fundamental shifts in where AI runs, the whole landscape is just exploding. New tools, new applications everywhere. It's not just helping doctors anymore. Oh, absolutely not. It's reshaping creative industries.
It's influencing public opinion, changing how we use tech every single day. The pace is incredible. Give us some examples. What's catching your eye? Okay, we'll look at voice AI. Eleven Labs, they're known for voice stuff, right? They just launched an AI assistant you interact with just using your voice. And it connects easily to other apps like Slack or Perplexity. So just talking to your apps. Yeah, imagine just talking naturally like they're another person. No typing, no clicking.
Then there's advertising. Did you see that Kelshi ad during the NBA finals? I might have missed that one. It was wild. 30 seconds. Made entirely with AI. Reportedly cost a tiny fraction of a normal ad shoot. It sparked this huge debate about the future of advertising. I bet. Disrupting traditional industries. Totally. Democratizing it, maybe, but also challenging creative workflows. Sounds powerful. That kind of power, it often brings you challenges, right? Especially around
information. Absolutely. And this is more concerning. We're seeing AI generate... misinformation on a really alarming scale. The newsletter points to two highly realistic AI generated pro -Iran propaganda pieces. Oh, wow. Flooding TikTok, Instagram, Facebook, YouTube. Over 30 million views on TikTok alone. Posted hundreds of times. It raises these really serious questions about manipulation, misinformation, figuring out what's even real online anymore. Yeah, that's a huge
ethical challenge. We're just starting to grapple with that. We really are. And then on the productivity side. AI is making a big push there, too. How so? Well, ChatGPT just rolled out new features. Real -time document collaboration. Speedy PDF export with citations. Ah, so competing with Google Docs, Microsoft Word. Exactly, a direct play. The productivity sweep battle is definitely heating up, and AI is right in the middle of it. And connecting back to our first point about
professional use, look at a bridge. The medical AI app. Yeah. It's now valued at a staggering $5 .3 billion, just raised $300 million, $800 million total. It just reinforces the immense value when AI is applied professionally. That's serious money. It is. And Quick Hits on some other tools. RunBear lets you use custom GPTs inside Slack or Teams. Dubbing 3 .0 translates videos into like 30 languages, one click. Sounds natural. Great for creators. Overflow AI turns
questions into instant charts. makes data easy. Lots of tools emerging. And one last thing, a big legal and ethical point. Anthropic, using copyrighted books to train its AI. It's currently being argued as fair use in court. That's huge. Yeah, the outcome could have massive implications for how all AI models get trained in the future, what data they can use. Okay, considering all that advertising, misinformation, productivity, ethics. What's the most surprising area AI is
impacting right now for you? Its ability to generate media from ads to propaganda at scale and low cost. That speed and accessibility is just transformative. Okay, let's pivot to the physical world now. Let's talk robots. Google, just kind of quietly, dropped something pretty big, something that could fundamentally shift how robots operate. Yeah, this is potentially a huge lead forward, a real game changer for robotics. It's called
Gemini Robotics on Device. On device. So running locally, like we were talking about earlier. Exactly. Robots running complex AI models locally. Think about that. No cloud connection needed constantly. No lag. No Wi -Fi dependency. That's different. Most robots need that connection, right? Until now, yeah. Most AI -powered robots had to send signals back and forth to huge servers just to, you know, move an arm or pick something up. The computation happened off -board. This
new Gemini model lives inside the robot. So what does that mean practically? Faster responses, better privacy because the data stays on the robot, and way more potential for real -world use, especially where internet might be spotty or non -existent. Google's saying this on -device model performs almost as well as its bigger cloud sibling and even beats some unnamed rivals in benchmarks. And this isn't just theory. They've shown it working. Oh, yeah. The live demos were
pretty impressive. These weren't just lab tricks. Robots using this new Gemini brain were doing surprisingly delicate adaptive things in real time. Like what kind of things? Things like unzipping bags, carefully folding clothes, even tackling totally new tasks they hadn't seen before, like assembling parts on a moving conveyor belt. It's not just about force. It's about nuanced interaction, adaptation. Wow. And what's really cool for developers, Google's releasing a full Gemini Robotics SDK.
It's a software development kit. It means devs
can fine tune robot tasks using just like. 50 to 100 examples that few yeah you basically just show the robot what to do a few times and it gets it's like stacking lego blocks of data for them teaching them fast okay connecting this to the bigger picture yeah it feels like everyone wants to build the gp2 of robotics right that foundational model for truly smart autonomous machines could this local ai be that leap whoa yeah imagine scaling this robots that truly understand
and adapt in our homes our workplaces operating reliably without needing that constant internet umbilical cord. That's a massive game changer for getting robots out into the real world, responding instantly to unpredictable stuff. So how does on -device AI fundamentally change how we'll interact with robots day to day? It enabled true autonomy and real -time responsiveness, freeing them from network limits. More natural collaboration
becomes possible. Sponsor. So reflecting on our discussion today, we've really explored this fascinating tension at the heart of AI right now. It's incredibly powerful, yes. Capable of passing the hardest tests, mastering complex data. Yet its actual value, its true worth, so often depends on how we interact with it, how we understand its limits, and maybe most importantly, how we design it to genuinely help, not just
impress. Yeah, and we're definitely seeing that clear shift towards AI living closer to us on our devices or embedded right into robots. That gives us more control, better privacy, and just incredible stability. Right. And the sheer range of new AI tools. I mean, from shaking up creative fields and boosting productivity to raising these really tough ethical questions around misinformation. Yeah. And this shows how fast this field is moving and touching, well, pretty much every corner
of our lives. Which brings up, I think, an important question for all of us. As AI gets smarter and as it moves physically closer running on our phones, in our homes, via robots, what new responsibilities do we take on? You know, the users, the developers. How do we ensure it truly serves humanity, helps us thrive responsibly? Two secs silence. Lots to think about there. We hope this deep dive gave you some new insights to mull over. Out to your own music.