Imagine building the brain for a powerful AI model. Not over months, but maybe in just a couple of hours. Right. And doing it with something like, what, 200 ,000 graphics cards? That sounds like sci -fi, but it's actually happening now. Welcome to the Deep Dive. Today, we're going to unpack a really fascinating set of sources. They focus on this rapidly accelerating world of AI, specifically the intense battle that's going on, sometimes behind the scenes, for computing
power. Yeah, and our mission today is to explore what some people are calling the GPU arms race. We want to understand what it means for big tech, for everyone else. And then we'll dive into some of the latest kind of surprising AI developments that are popping up and shaping things. We'll try to highlight what's really important here, maybe how these shifts could impact you as you, you know. navigate this incredibly fast -moving landscape. So let's dig in. Okay, let's unpack
this first part. The core idea seems pretty straightforward, actually. Training the biggest, most advanced AI models. It now depends almost entirely on who controls the most computing power. Simple as that. It really is a new kind of arms race, isn't it? But instead of weapons, it's silicon and, well, massive amounts of energy. Exactly. And the biggest player kind of stepping out into the light right now, that seems to be XAI's Colossus
supercomputer. It's being called the Undisputed Heavyweight, packing, get this, 200 ,000 H100 equivalent GPUs, graphics processing units. 100 ,000. It's hard to even picture that. It's truly massive. And what's really fascinating, maybe even mind -bending, is this machine can apparently train a model like GPT -3 in under two hours. Under two hours, not weeks. Nope, not weeks or days, hours. Beat, whoa. Yeah. I mean, just imagine
scaling that kind of training capacity. Think about applying it to like a billion search queries or way more complex tasks. It's kind of staggering. That speed changes the whole development cycle. Completely. It allows for super rapid iteration experimentation, just fundamentally changes the pace of AI progress. And it's not just XAI, right? Other big tech players are scaling up incredibly fast, too. We're seeing Meta, Microsoft, and OpenAI, Oracle. They're all running these huge
clusters, 65 ,000 to maybe 100 ,000 GPUs. And Tesla's definitely in the game, too, 50 ,000 GPUs for their Cortex Phase 1 project. And it's interesting. Most of these top 10 players, the ones we know about anyway, are based in the U .S. Right. But, you know, what's really noticeable is who's not on that public list. Google and Amazon. Yeah, that's a good point. They're obviously dominant forces in AI model development. Totally.
But they're suspiciously quiet about their own supercomputer specs, given their history, their investments. It feels highly likely they're building huge clusters, too. Just maybe. In stealth mode. Or guarding trade secrets very closely. It suggests this foundational AI infrastructure is seen as deeply strategic, which brings up a really important question, especially when you look at the money involved. The sources say the cost of building and running these massive GPU clusters, it's
doubling every 13 months. Doubling every 13 months. Wow. That's not just a big expense. That's an accelerating financial strain, even for the biggest companies out there. Even for them. And the sources really emphasize this point. Whoever controls these chips, this compute, they essentially control the next wave of AI. Yeah, it's like the new strategic resource, like oil refineries or maybe fiber optic cables. These GPU clusters are that
critical infrastructure now. And that's probably why you see national labs like Lawrence Livermore also scaling up their own compute power pretty rapidly. They see the strategic importance, too. So thinking about these crazy costs and this intense competition, what does this whole escalating GPU arms race really mean for, say, smaller AI innovators or startups just trying to get going? Yeah, that's the big worry, isn't it? Is it creating this kind of AI oligarchy? That's exactly the
concern. I mean, this rising barrier to entry. It kind of implies that only the players with the deepest pockets can really afford to train and deploy these cutting edge frontier models. It definitely pushes innovation towards them, centralizing power. OK, so while that hardware battle is clearly intense, AI's impact goes way beyond just the raw compute, right? Let's maybe
pivot a bit now. Look at some broader ways AI is reshaping things like policy or new creative tools, even how we interact with tech itself. Yeah, sounds good. So policy wise, there was actually a pretty big move recently. The Trump administration released this really extensive AI action plan. Action plan. Yeah. And it's not just some quick memo. It's apparently a huge checklist over 90 different points, all aimed at making sure the U .S. stays, you know, an
AI superpower. 90 points. That's comprehensive. It is. And what's interesting is they apparently shaped it using feedback from over 10 ,000 public comments. So lots of different voices went into it. That's quite a bit of public input. Yeah. And alongside policy, there's also this push to maybe democratize the building of AI. Exactly. Like, look at GitHub Spark. It now lets developers build and deploy AI apps just using prompts, like plain English instructions. Just prompts,
not complex code. Right. Which is obviously super appealing to the, what, 150 million programmers already on GitHub. It basically turns coding into more of a conversation. Lowers the barrier significantly. Totally. And then you see these incredible creative uses popping up. There's an AI filmmaker, Lu Huang, who showed 10 really top tier examples of using JSON. JSON. OK, that's a data format, right? Makes things easy for chatbots
to read. Yeah, exactly. Using JSON to generate these stunningly realistic, really polished commercials. quality is wow it shows ai can handle seriously complex creative work now that's impressive and if we connect that to how we interact with technology meta seems to be taking a pretty bold step Kind of like Neuralink, but different. They've got
this AI wristband. Yeah, I saw that. It's designed to decode signals in your wrist nerve signals, basically to let you control devices with just tiny little hand gestures, almost invisible ones. So like minority report, but subtle. Kind of. It's definitely a big step towards making human computer interaction much more seamless, more intuitive, like tech becoming an extension of your thoughts. Imagine the possibilities there for accessibility, for just efficiency. For sure.
And speaking of big moves. OpenAI looks like the next big thing is GPT -5. Expected maybe early August. GPT -5, okay. What's the word on that? Is it just a bigger GPT -4? Doesn't sound like it. It's described more as combining the powerful tech stuff from the GPT series that we know with their O -series models, like O3, which seem more focused on real -time, maybe multimodal interaction, seeing and hearing. Perhaps. So not just text, more integrated, more perceptive.
That seems to be the idea. So you won't just see it as a text model like GPT -4 was initially. It's expected to be a more. evolved system. Could change how we interact with AI quite a bit. Okay. Interesting. And one more on the business side. Yeah. Legalon. It's an AI legal tech company. They just raised another $50 million for their tools. $50 million. What do their tools do? They help with legal work, specifically contract review. Apparently they're already used by 7 ,000 organizations.
And get this, they speed up contract review by 85%. 85%. That's... Massive efficiency gain, right? They already serve like a quarter of Japan's public companies and they're expanding fast. So clear, tangible impact in a field that's usually kind of slow to change. Okay, so with all these things happening, the policy, the building tools, the interfaces, the business applications, how quickly do you really think these new AI tools could fundamentally change our day -to -day work?
our professional lives? I think very fast, honestly. Many of these seem designed for immediate practical use. They slot right into existing workflows, so the impact could be felt quite quickly. Okay, let's dive into maybe a few quick hits now. Things from the AI world that just kind of caught our eye, sparked some curiosity. And then let's end with something truly remarkable, connecting AI to human history. Sounds great. Yeah, a few things
popped out. There was a funny piece. Someone asked an AI if their job writing humor was at risk. Kind of meta, right? Chuckles softly. Yeah. And I'll admit it, I still wrestle with prompt drift myself sometimes, you know, where the AI kind of slowly forgets what you originally asked it to do. So, yeah, I get the anxiety. It's a real thing. We also saw this alert from scientists sounding a pretty light alarm, actually, saying AI might soon get smart enough to outsmart our
current safety checks. That's sobering. It definitely raises some immediate concerns. Yeah. On a lighter, more practical note, Google's got that new AI feature, right? Let's you... Virtually try on clothes you see in search results. Oh, yeah. The virtual try on. That could be a game changer for shopping online. Make it way more interactive. And Google's also rethinking search itself, aren't they? With this new AI curated web guide. Yeah. Trying to get more synthesized answers, not just
links. Right. Moving beyond just keywords. We also saw this really interesting comparison. Someone gave the exact same complex financial task to an AI and a human financial planner.
Oh, how'd that turn out? out well it just really highlighted their different strengths you know the ai was amazing at crunching the data finding patterns the human brought in like nuance emotional understanding strategic thinking about life goals different approaches fascinating comparison okay but here's the one that really got me connecting ai to our past google deep mind quietly released something called Aeneas. It's fully open source.
And it's pretty groundbreaking. What it does is it reads, analyzes, and actually reconstructs ancient Roman texts using images of fragments, bits of text. Wow. Wait, so you feed it like a picture of a broken stone tablet with faded letters. Exactly. Maybe just a blurry photo. Incomplete inscription. And Aeneas pulls from this huge database like 176 ,000 ancient writings. And then it intelligently guesses where it might be from, when it was likely written, and crucially,
what it used to say. That's incredible. How accurate is it? The numbers are pretty remarkable. It gets 72 % accuracy assigning inscriptions to the correct Roman province. 72%, okay. And it restores the broken or missing text with 73 % accuracy. Wow. And maybe just as important, 90 % of the historians who tried it said it significantly boosted their confidence in their own research by like 44%. That's a huge endorsement from the experts themselves. And here's the kicker. It's
totally free and open source. Open source. So anyone can use it. Adapt it. Yep. Researchers can download the model, fine tune it for their specific projects, maybe even adapt it to other ancient languages. That's amazing. It really raises this important question, doesn't it? How AI? through tools like Aeneas could just unlock vast amounts of human history, stuff that was basically inaccessible or would take lifetimes
to piece together before. Yeah, definitely. So thinking about that open source aspect, could Aeneas or tools built like it potentially unlock other forgotten historical texts, maybe even decipher completely lost ancient languages? Oh, absolutely. I mean, it's open source nature is practically an invitation for that, applying it to different scripts, different cultures. Yeah, it could potentially reveal countless historical insights we just don't have access to right now.
Sponsor. OK, so let's try to pull this all together. What does this all mean for you listening right now? We've seen today that this AI boom It's not just about clever software. It's deeply tied into this really high state's global arms race for raw computing power. And this battle, it's driving innovation like crazy, but it's also creating these immense financial pressures, these strategic pressures. Right. And if we connect that to the bigger picture, AI isn't just about
the hardware, the compute power. It's rapidly transforming almost everything from how governments make policy to how commercials get made, even down to our fundamental ability to like. preserve and understand ancient history. It really feels like a foundational shift happening across so many different areas. It's redefining what's even possible. So as you think about all this, there's maybe a question to mull over, a provocative
thought. As these AI models keep getting bigger and bigger, demanding more and more resources, where do you think the physical infrastructure is actually going to end up? The chips, the massive amounts of energy needed, these enormous data centers. where will they physically be built in the future? Yeah, that's a big question. And how might that geographical concentration, wherever it ends up being, how might that shift global
power dynamics in the years ahead? That's definitely something worth pondering as you, you know, engage with this AI world unfolding around us. Hopefully this deep dive gave you some valuable nuggets to think about. Thank you for joining us on this deep dive into the latest in AI OTRO music.