Who Will Control Artificial Intelligence?

Welcome to the Sentient Code, where intelligence is engineered, autonomy is emerging, and a line between human and machine grows thinner. Each episode, we decode the algorithms, explore the robotics, and examine the ideas shaping the future of artificial minds.

Hello, and welcome back. We are looking at something today that on the surface feels a little bit like a classic David and Goliath's story. But the more I look at the research, the more I start to think maybe David is actually being secretly funded by another Goliath.

That is a surprisingly, surprisingly accurate way to put it. It's definitely not the fairy tale version people might imagine, right.

It's way more complicated. Yeah, we're talking about what is and I don't think this is an exaggeration arguably the defining technological rivalry of our era. It is absolutely talking about the battle between open source AI and big tech.

It's huge, and honestly, the landscape is shifting so fast that by the time you're listening to this, the battle lines might have already moved again.

That's how fast this whole thing is going. And I think for a lot of people, you know, the narrative feels pretty simple It's the scrappy underdogs, right, the hackers, the independent researchers, the global open source community.

Right, the rebels, exactly, the rebels versus the big bad corporate giants. But looking at the material we have today, the reality is, well, it's a lot messier.

It's so much messier and frankly, so much more interesting because this isn't just about who sells the most software or you know, who has the highest stock price. This is really about the fundamental structure of power. It's about who holds the knowledge, who owns the infrastructure, and ultimately who gets to shape the twenty first century.

That is a heavy way to open the structure of power.

I think it needs to be, because we have to move past this simple binary idea of open versus closed. It's not that simple. Both sides are winning in different arenas, and you know, both sides are facing these huge, almost existential threats they might not even fully see yet.

Okay, so let's set the board. Then. If we're looking at this as a kind of conflict or maybe a game, who are the main players. Let's start with the heavy hitters.

The big tech camp For sure, these are the names everyone knows, you have open Ai, which is of course heavily heavily backed by Microsoft the tens of billions of dollars camp that's a good name for.

Them, exactly. Then you have Google deep Mind, which is the force behind the Gemini models. You've got Anthropic, which is a really interesting one because it was founded by ex OpenAI people, but now it's backed by both Amazon and.

Google at a bit of a tangled web.

There, a very tangled web. And then of course you have the quiet giant Apple doing its own thing, mostly focused on on device intelligence.

Okay, so that's the fortress, that's the establishment. And on the other side, who are the rebels.

The open source ecosystem? And this is where it gets much. It's more diffuse. It's not one company. It's a global network. You have incredible startups like Mistral and Friends. You have academic researchers and universities all over the world. You have hobbyists, and then you have this massive community coalescing on platforms like hugging face.

Hugging Face I always love that name. It just sounds so friendly and innocent for something that is essentially an arms depot for AI models.

It is a charming name. Isn't it. But the scale is anything but charming. It's deadly serious. They host hundreds of thousands of models. We're talking billions of downloads. It is, for all intents and purposes, the GitHub of the AI revolution.

So we have the giants and we have the swarm, and the mission for us today is to figure out, well, who's actually winning, because looking through all the source material, the answer seems to depend entirely on how you define winning.

Precisely, Winning the performance benchmark is one thing. Winning the enterprise market is another. Winning the hearts minds of developers that's a third. They're all different battles.

So let's start with the giants. Let's talk about why big tech is so dominant. What is this fortress they've built? What are the walls made of?

The first, and by far the thickest wall is what's called the compute mote.

The compute mote. Okay, I want to pause on this because I think when people hear expensive, they think, okay, like a really nice car expensive, or maybe a house expensive. But looking at the numbers here, we are talking about nation state level spending, aren't we.

We really really are to even begin to visualize what it takes to train a frontier model, the absolute bleeding edge systems like GPT four, Gemini Ultra. You have to start with the hardware, right, we aren't talking about the graphics card in your gaming PC. We're talking about the top of the line in Vidia H one hundreds, the AI chips.

The chips that there's a global shortage of the ones everyone is fighting.

Over the very same These chips are, you know, almost a strategic national resource at this point. A single one one of these costs as much as a luxury car.

A single chip.

Yes, now imagine you need to buy it, not one, but maybe twenty five thousand of them, twenty five thousand, and you have to tie them all together with specialized high speed networking cabling that costs more than most people's houses. You stick all that in a massive, custom built data center and you run it at one hundred percent capacity for ninety.

Days straight, ninety days, NonStop.

NonStop, twenty four to seven. And this is where the moat gets very, very physical. It's not just about buying the chips. It's about the physics, the heat.

I saw a note here about thermal densities.

That what you mean, that's exactly.

It.

If you put that much compute in one room, the air doesn't just get hot, it turns into a blast furnace. The racks would literally melt. So you need these incredible industrial scale liquid cooling systems, pipes of chilled water running to every single server rack.

So it's not just a server room. It's a power plant and a plumbing project it is, and you.

Need access to a power grid that can handle the load of a small city. We're talking hundreds of megawatts of sustained power. The electricity bill alone for a single training run, just the electricity can be tens of millions of dollars.

Get to wrap my head around that. So when we say the kid in the garage is locked out of this game, it's not because the GID isn't smart enough. It's because the garage would literally melt and the entire neighborhood grid would blow a transformer.

The garage would melt, the transformer would blow, and they'd get a bill for ten million dollars. So, yes, this is why the list of companies that can actually train a Frontier model from scratch is so incredibly short. It's basically Microsoft, Google, Meta, and Amazon. The barrier to entry is a capital expenditure of billions of dollars before you've written a single line of useful code.

You use the analogy of a really talented carpenter in their backyard workshop versus a massive industrial factory that spans ten city blocks. It feels like that.

That's a perfect analogy, and it doesn't even stop at the hardware. It's also the talent the people. Yes, the huge expertise required to orchestrate one of these massive training runs is incredibly rare. It's a kind of dark art.

What do you mean by that.

Well, when you have twenty thousand GPUs all trying to work in perfect parallel, things break all the time. Mysterious failures just happen.

Like a cable comes loose or a network switch fails.

It can be that simple, or it can be much weirder. A few GPUs might just start overheating and produce infinitesimally small errors in their calculations bad math. Now, in a normal computer program, you might get an error message it crashes. In a training run like this, those tiny errors can act like a poison. They can slowly destabilize the entire

So the engineers running these things, yeah, they aren't just coders. They're more like nuclear reactor operators staring at dials trying to prevent a meltdown.

They are absolutely high stakes problem solvers. They are staring at hundreds of graphs looking for tiny, almost imperceptible anomalies that suggest that AI's brain is getting sick. That level of operational expertise is a huge part of the moat. You can't just buy the chips and plug them in. You need the priesthood of engineers who know how to keep the beasts alive.

And I'm guessing those people are very, very expensive.

Extremely and they almost all work for the giants.

Okay, so that's the hardware in the talent. The first wall of the fortress is thick. But there's another piece, right, the data.

The data advantage. And this is where it gets a lot more subtle, but maybe even more.

Powerful, because I think most people's first thought is, well, the Internet is open, right, can anyway, just scrape the web and get the same data you can.

And the open source models do rely heavily on public data sets like the common crawl, but big tech has access to vast oceans of proprietary data that is completely invisible to the public web, like what for exams. Okay, just think about Google. They have over two decades of indexed search queries. They don't just have the text of the Internet. They know what billions of people are looking for, how they ferze their questions, and critically, what results they actually click on.

Right, they don't just see the words, they see the intent behind the words.

Precisely the intent. Think about Microsoft. They have deep integration into enterprise workflows. They'll see how businesses write emails and outlook, how they structure documents and word how they build presentations in PowerPoint. That's a huge unique data set about professional communication and meta the social graph decades of behavioral data, how people interact, what they like, what they share, the

It seems like the difference between having a library of every book ever written, which is the public web, and having that same library plus a secret video recording of every single person who ever read a book, yeah, showing what pages they lingered on, what they highlighted, what made them laugh.

That is a profound and perfect distinction. It's the quality versus quantity aspect. But there's an even bigger data advantage that builds on top of that, the feedback loop.

Explain that, what do you mean by feedback loop?

Big tech sees how users interact with their finished models in real time. Every time you use chat GPT and you don't like an answer, so you hit regenerate response.

Open ai sees that they log that.

They log it as a failure case every time you give a thumbs down to a response from Gemini. Google's reinforcement learning systems take note. They have this incredibly tight feedback loop with hundreds of millions of users constantly telling them what works and what doesn't. They are perpetually refining the model based on massive real world.

Usage, and the open source community doesn't have that same direct line, not.

In the same way. Now, when you download a model from hugging face and run it on your own laptop, the developers of that model have no idea what you're doing. They don't get that data back, because that's the whole.

Point, right privacy. You run it locally so they can't see your data exactly.

It's a feature, not a bug. But the cost of that privacy is that the model creators lose that invaluable telemetry. They're flying blind compared to Google or open ai, who have a god's eye view of how their creations are performing in the wild, so.

Hearing all of that, the billions in compute, the rare talent, the proprietary data, the real time feedback loops, it honestly feels like game over. How can open source possibly compete? It feels like you're bringing a well crafted knife to a thermonuclear war.

It absolutely looks that way on paper. The structural disadvantages are immense. But this is where the rebellion gets really really interesting, because despite all of that, open source is not only surviving in some very important areas, it's actually thriving.

Oh how is that even possible? Is it just the sheer number of people working on it.

That's a big part of it. But the main technical driver, the core reason they can compete, is a relentless focus on efficiency efficiency. The open source community, precisely because they don't have unlimited resources, has become world class at doing more with less necessity.

Is the mother of invention, as they say.

It's the perfect embodiment of that saying, when you don't have a billion dollars for your next training run, you have to get incredibly clever. And they've developed these amazing techniques like quantization.

Okay, I keep seeing this word in the research quantization. It sounds like something out of a sci fi movie, but it seems absolutely critical to the open source survival guide. What is actually happening there.

It is absolutely critical. The best way to think of it as like image compression. You know how a raw photo file from a professional camera can be huge, maybe fifty megabytes.

Sure, it has all the raw sensor data, every bit of.

Information exactly, But if you convert that photo to a JPIG it shrinks down to maybe two or three megabytes. You lose a tiny, tiny bit of detail. Maybe if you zoom in one thousand percent, the shadows aren't quite as perfect, But to the human eye the picture looks basically identicle and it.

Loads way faster. In a website and takes up a fraction of the space on your hard drive.

Precisely, quantization is basically doing that same trick to the AI's brain. These huge models are usually stored with incredibly high precision numbers. Think of every connection in the neural network having a weight that's like sixteen decimal places of accuracy, which takes.

Up a ton of computer memory and processing power massive amounts.

Quantization is a process that basically says, hey, do we really need sixteen decimal places of precision? What if we just round it down to four or even two?

And the model doesn't just get stupid when you do that.

That's the magic. It turns out for many of these models, you can round down those numbers very aggressively and the model only loses a tiny fraction of its overall intelligence. But suddenly a model that required a forty thousand dollars server with a ton of RAM can now run on a two thousand dollars gaming laptop.

So they are literally shrinking the giant's brain so it fits inside a regular person's head.

That is exactly what they're doing, and that's not the only trick. Then there's something called distillation, which is think of it like a teacher student relationship. You take a huge, smart, expensive model like GPT four and you use it to teach much smaller, cheaper model.

How does that actually work in practice?

You can, for instance, ask the big teacher model to generate thousands of perfect answers to questions on a specific topics a customer service. Then you take those thousands of perfect question answer pairs and you use them as the training data for the small student model.

Ah, so the student learns from the master's work exactly.

The student model might not know everything the professor knows about physics and poetry, but for that one specific task of customer service, it can get ninety percent of the way there for one percent of the cost to run.

That's a huge leverage point. It's like you're borrowing the intelligence of the giant to train your own litle army.

It is and because of techniques like this, the performance gap between the massive closed models and the nimble open models has narrowed again and again. The community is just relentless about optimization.

And this all leads to what one of our sources calls the deployment victory. I found this concept fascinating. The idea that winning isn't just about having the single smartest brain in a lab somewhere, but about being the one that actually gets used out in the real world.

This is such a critical distinction. There's a massive difference between the frontier, the absolute smartest, most capable model possible, and deployment what a company actually feels safe and comfortable putting into their production systems. Let's say you're a healthcare company you're handling sensitive patient records, or you're a bank

Probably not, if you can avoid it. Your compliance department would have a heart attack. You want to keep that stuff locked down on your own servers. You definitely don't want opening ISIS and IS potentially reading your patient files exactly.

You want to own the an entire stack. You need privacy, You demand security and control.

I would imagine total control.

Massive control. If you build your entire product on top of a proprietary model from a single company, you are taking on an enormous strategic risk. What happens if they decide to double the pricing next year?

You're stuck.

What if they change their roadmap and deprecate the version of the model you rely on. What if they go out of business or get acquired.

You're building your entire house on land that you're just renting. They could evict you at any time.

Precisely, open source offers stability and sovereignty. You download the model, weights there are yours. You run it on your own infrastructure, behind your own firewall. No one can ever take it away from you.

So for the enterprise, you know, the big serious companies that are actually paying to implement this stuff in the real world, open source is often the better strategic bet.

For many regulated industries. It's often the only acceptable option.

There's another factor here too, which is just the sheer speed of innovation, the swarm intelligence. I want to dig a little deeper into that. Can you give me a concrete example of a time the open source community just completely outran the giants? Oh?

Absolutely. The perfect example is the rise of mixture of experts architecture or MOE.

Sounds fancy, but what is it? In simple terms?

Okay, So, a traditional model is what we call dense. It's one single giant brain. Every time you ask it a question, the entire brain has to light up and think.

About it, which is computationally expensive.

Very and MOE model is different. Instead of one giant brain, the model is made up of many smaller expert submodels. Maybe you have eight different experts, and when you ask a question, there's a tiny router network at the front that decides which one or two experts are best suited to answer it.

Like a receptionist at a big hospital directing calls.

Exactly like that. Oh you have a coding question, I'll send you to the python expert. You have a question about ancient Rome, let's talk to the history expert.

And why is that better?

It's massively more efficient for any given thought. You're only activating a small fraction of the total brain, so it's much faster and much much cheaper to run, but you still get the benefit of the combined knowledge of all the experts.

Okay, I get it. So how did the swarm outrun the giants on this?

Well? Big tech has been using this technique internally for a while, but when the open source company Mistral released their first MOEE model, the community just exploded. They dissected it, they understood it, and they started building their own improved variants instantly. Within weeks, we had people figuring out clever hacks to run these incredibly complex architectures on consumer grade hardware, on.

MacBooks, a laptop, a model with multiple expert brains. How is that even possible.

It's possible because the community became obsessed with optimization. They found these brilliant ways to load the different experts in and out of your computer's memory so fast that you could run a model that on paper should never be able to fit on your machine.

As that classic hacker spirit I love.

Yeah.

Big text answer is buy a bigger, more expensive server. The open source answer is no, let's rewrite the code to make it fit on the server we already have.

That is the fundamental cultural difference right there.

And look at the timeline in a big tech company. If a researcher has a brilliant new idea like.

That, it has to be written up in a proposal, It has to go through a committee review, it has to get prioritized and put into the next quarters product roadmap. It can take months or even years to see the light of day.

And the open source world.

Someone reads a new academic paper on a Tuesday morning. By Tuesday night, some brilliant person has a working implementation of it on GitHub. By Wednesday, someone else has forked it and made it ten percent faster. By Friday, it's been integrated into the main community tools and is available for everyone in the world to download and use.

It's like evolution on fast forward, a hyper revolution it is.

It's a decentralized that iterates and improves immediately in parallel. Big tech has deep pockets and amazing researchers, but they simply cannot match that sheer, chaotic velocity of global experimentation.

And then there's the geopolitical angle, which was something I hadn't fully considered before digging into this.

It's huge. It's easy to forget that not every researcher or developer is sitting in Siliton Valley or London. If you're a brilliant AI researcher in a country that doesn't have easy access to US proprietary tech, maybe because of sanctions or just because of economics, you can't use the closed APIs. OpenAI might be blocked in your country or just be too expensive.

So for a huge portion of the world, open source is the only game in town.

It's a force for democratizing access. It means a student in India, a researcher in Brazil, a startup in Nigeria. They all get access to the same cutting edge tools that were until recently locked away inside a handful of American corporations.

So we have this picture of the big tech fortress with all the money and data, and we have the open source rebellion with all the speed, the privacy in the global reach. But then then we have to talk about the meta paradox.

Ah, Yes, the agent of chaos, the spanner in the works.

This brings us to the weirdest, most confusing part of the whole map. You have meta Mark Zuckerbert's empire, which sits squarely, undeniably in the big tech camp. But then you look at what they're doing with their Lama models, and it looks for all the world like they're the primary arms dealer for the rebels.

It's the single biggest plot twist of the last decade in tech. They are spending those billions and billions of dollars we just talked about building the data centers, burning the megawatts of electricity to build these massive state of the art brains and then they just put the weights on the internet for free.

It feels like charity, which I'm going to go out on a limb here and assume Meta is not a charity.

You would be correct. This is not altruism. This is one of the most ruthless and brilliant strategic moves in the history of the tech industry. It's a concept that economists call commoditizing the compliment.

Commoditizing the compliment. Okay, break that down for us in plain English. What does that mean?

Okay, think about it this way. If your core business is selling hot dogs, what do you want to be true about hot dog buns?

I want them to be as cheap and as widely available as humanly possible free if I can manage it. Why Because the cheaper and easier it is to get buns, the more of my hot dogs people will buy. The bun is a necessary part of the equation. But it's not where I make my money.

Right, The buns are the compliment to your main product. Now, look at Meta. What is their core business? What do they actually sell?

They sell us, They sell our attention. They sell advertisements on Instagram and Facebook and WhatsApp.

Correct, They do not sell cloud computing services. They don't sell subscriptions to AI models. For Google and Microsoft, the powerful AI model is the hot dog, it's the product. They need that model to be expensive and exclusive so they can charge you twenty dollars a month for hy access to it.

But for Meta, the AI is just the bun.

The AI is the bun. It's the complementary good, it's the infrastructure. By giving away their state of the art LAMA models for free, Mark Zuckerberg is trying to drive the market price of raw intelligence down to zero.

Which completely scorches the earth. For his primary competitors, Google and Microsoft.

It annihilates their business model. If any startup in the world can grab Lama three for free and build a fantastic business on top of it, why would they ever pay open Ai or Google huge licensing fees. Meta is strategically trying to make sure that no one else can build a gatekeeper position in the AI layer of the Internet that could one day threaten their advertising empire.

That is absolutely rusiless. Okay. Another analogy, imagine if you were a company that made all its money selling shaving cream, and your main competitor made all their money selling expensive proprietary razors. You might start giving away really good razor handle for free. Yes, perfect, because if everyone has a free razor handle that works great, nobody needs to buy the competitors' expensive ones. But everybody still needs to buy your shaving crank.

That is the perfect analogy. Meta is giving away the razor handles to protect it shaving cream business. They are scorching the earth so that their rivals can't build castles that might one day charge Meta a toll.

And I'm guessing there are other benefits to Meta too, right, beyond just kneecapping their competitors.

Oh, massive benefits. Remember that swarm intelligence we were just.

Talking about, Yeah, the hyper evolution.

The second Meta releases a new Lama model, thousands of the world's smartest developers immediately start hacking on it. They find bugs and they fix them. They figure out new ways to make it faster, they discover how to run it on cheaper hardware.

And Meta gets all of that research and development for.

Free, completely free R and D. They are effectively outsourcing a huge chunk of their innovation pipeline to the entire world. Plus they build up this incredible reservoir of goodwill with the developer community. Suddenly Meta isn't the bad guy from the privacy scandals. They're the good guy who is enabling open science and empowering the little guy. It's a masterful repositioning of their brand. They become the neutral Switzerland like party in the AI wars.

That is some serious forty chess right there.

It really is, and it completely complicates the whole narrative. It's no longer just corporations versus the people. It's corporation A versus corporation B, with corporation A using the people as a strategic labor against corporation B.

So with all these complex pieces on the board, the fortress, the swarm, the chaotic agent in the middle, what is the current score If we were to freeze the game right now today, who is actually winning?

Well, we have to be realistic about it. If we're judging purely on the frontier, the absolute cutting edge of raw capability, big tech still has the lead.

The big proprietary models are still smarter.

On the whole. Yes, if you look at the leaderboard for complex reasoning, for advanced coding challenges, for multimodal tasks that involve understanding both images and text. The flagship models from open Ai, Google and mpropic still consistently beat the best open models.

So the gap is narrowing, but it's still there.

It hasn't closed yet. And for users who need the absolute best in class performance, think of a doctor using AI to help diagnose a complex disease or a lawyer using it for intricate legal analysis, that small performance gap can really matter.

And what about integration just getting it into the hands of normal people.

Big tech wins there hands down. It's what we call the distribution.

Advantage, which means what exactly it means.

The AI is showing up almost invisibly inside the tools you already use every single day. It's Copilot appearing in Microsoft word, it's Gemini helping you write an email in Google Docs. It's AI photo editing build directly into your phone.

Right. I don't have to go find a special website or installing it. It's just there. The friction is zero, zero friction.

The open source alternatives might be fantastic, but you still have to make a conscious choice to go find them, download them, maybe configure them. Most people, most of the time, will take the path of least resistance. And big tech owns those paths.

And this is just the simple brand factor.

Right heask your neighbor, ask your parents to name a AI. They will say chat GPT. They won't say mistroll or Lama or mixt roll eight by seven B.

That brand recognition drives usage, which drives.

Revenue, and that revenue funds the next billion dollar training run. It's a powerful self reinforcing cycle.

But there's some big X factors here, wild cards that could still tip the scales, and one of the biggest is the government regulation.

The regulatory dimension. This is where the battle moves from the server room to the halls of Congress and Brussels.

How does regulation play into this? My first assumption would be that everyone hates regulation, that it just gets in the way.

Actually, no, and this is one of the most counterintuitive parts of the whole story. Big tech often quietly welcomes certain kinds of regulation.

Really, why on earth would they want more rules?

Because compliance is expensive?

Wow.

Of course, if the government passes a law that says, to legally release a powerful AI model, you must first perform ten million dollars worth of safety testing, formal auditing, and red teaming.

Who can afford to do that, Microsoft, Google, the big guys.

Exactly, and who can afford.

To do that the two college students in a garage or a small, bootstrapped open source startup.

Precisely expensive. Regulation, while sounding good on the surface, often ends up creating a massive barrier to entry. It protects the incumbents from new competition. It's a way for them to pull up the ladder behind them.

So big tech might publicly say, oh, yes, please regulate us, this technology is so dangerous, while privately thinking this is fantastic. This will crush all the small fry competitors.

It's a very common dynamic in mature industries. It's a form of what's called regulatory capture.

And what's the argument from the open source side?

They argue that you simply cannot regulate software in that way. Their core argument is essentially math is speech. You can't put the genie back in the bottle. You can't make it illegal to share a file of numbers.

If I download it, I can run it.

And if you regulate the US and European developers too heavily, the top talent and the innovation will just move to a country with looser rules. You don't stop it, you just offshore it.

So there's a real fear that heavy handed regulation could actually hurt the good guys in the open community. Yeah, far more than it hurts the giants they're trying to rain in.

That is the central fear. The final shape of regulation could end up deciding the winner of this war more than any single technological breakthrough.

There's one more X factor I want to touch on, which is synthetic data. We talked earlier about how big tech has this huge data mode, but isn't there a movement now to just create your own data.

Yes, this is a major frontier for the open source world and a potential moat killer.

What is it exactly?

Well, the problem is that you can run out of high quality text on the public internet to train your next.

Model, on which I hear we are actually getting close to doing.

We are getting surprisingly close. The really high quality, well written stuff is finite. So the idea is this, what if you use a really smart existing model like GBT four or claude to write new training data. You ask it to generate textbooks or to create thousands of high quality examples of Socratic dialogues, and then you use that perfectly clean AI generated data to train your next new model.

So the AI is teaching the next generation of AI.

Yes, exactly, isn't that.

A bit incestuous? Don't you risk degrading the quality? It feels like making a photocopy of a photocopy of a photocopy. Eventually it just turns into a blurry mess.

That is the number one risk. It's a very real phenomenon called model collapse or habsburg AI. If you aren't incredibly careful, the models do start learning their own mistakes, They start hallucinating more, their knowledge becomes weird and distorted.

So how do you prevent that from happening?

Meticulous curation and this is another area where the open source community swarm intelligence shines. You have organizations like a Luther AI and literally thousands of volunteers who treat building these data sets like a massive science project. They are manually cleaning data, filtering out the bad stuff, and rating the quality of the synthetic examples.

So it's human verified synthetic data. A hybrid approach.

Exactly, It's a hybrid and this is a very powerful technique for helping to close that data mode. If you can generate your own high quality training data. You don't need access to Google's private search logs nearly as much as you thought you did.

That is fascinating. It really challenges that old idea that data is the new oil and only the giant oil barons have access to it. It turns out you can kind of synthesize your own oil if you're smart enough, if.

You are smart enough and extremely careful.

So let's try to bring this all together. Let's try to reach a verdict. We've looked at the advantages of stre the paradox as the wild cards. Who wins? Is there a clear winner in this battle for the future?

You know. The honest answer, and it might be a little unsatisfying, is that it's not binary. It's not going to be one side standing victorious on the ashes of the other.

So it's a split decision.

I think it's a permanent coexistence. They're going to win different layers.

Of the stack. Okay, break that down for me. Who wins what.

Big tech wins the frontier for the foreseeable future, the absolute smartest, most powerful, most capable systems are going to come from the groups with the billions of dollars to spend on compute and the unique data feedback loops.

Okay, so they own the top of the mountain.

They will likely also win the broad consumer layer, the apps on your phone, the voice assistant in your car, the AI built into your search engine. That will be a big.

Tech because of that convenience and distribution advantage.

We talked about it, right, But open source is going to win the deployment.

Layer, the infrastructure, the plumbing.

The plumbing of the world. Yes, enterprise back ends, specialized tools for science and medicine, the infrastructure for finance. Most of that will eventually run on open models because of the critical need for cost effectiveness, control, privacy, and stability.

So, if I'm hearing you right, big tech pushes the absolute ceiling of what's possible higher.

And higher, and open source raises the floor for everyone and spreads that technology everywhere.

That's a really interesting, almost symbiotic dynamic. They need each other in.

A weird way, they really do. And frankly, I think the most important victory for open source might not be market share at all.

What is it?

Then, It's the establishment of a principle.

What principles say, The principle.

That AI capability doesn't have to be concentrated in a handful of corporations, that we don't have to live in a world where three companies in California hold all the keys to the future.

It's an insurance policy for the world.

It's exactly that open source ensures that the knowledge, the tools, and the town are distributed globally. It means that no matter what happens with corporate roadmaps or government regulations, the fundamental capability to build and understand these powerful systems belongs to everyone, not just to a few boardrooms.

That's a really powerful thought. It shifts the focus from who makes the most money to who gets to participate in building the future.

Exactly, it's about preserving human agency in the age of AI.

You know, we've talked about winning in these broad strokes, but I want to ask you personally, as someone who analyzes this space every day, where do you put your chips if you had to bet on what the AI landscape looks like and say five years.

Five years is an absolute eternity in AI time.

Humor me, what's your gut feeling.

I think we're going to see what I'd call a barbell distribution of intelligence. A barbell What do you mean On one end of the barbell, you'll have these massive, centralized, super intelligent cloud models owned by maybe two or three companies. This is what you'll use when you needed to solve cancer or design a fusion reactor, or plan a global logistics network.

The really heavy lifting.

The planetary scale heavy lifting. On the other far end of the barbell, you will have billions of tiny, highly efficient, highly personalized open source models running locally on our phones, our glasses, our laptops, maybe even our home appliances.

The edge, the personal intelligence.

Layer, the edge, yes, and the middle. I think the middle gets squeezed out. The medium sized general purpose proprietary models might just disappear because they won't be as smart as the giants or as cheap in private as the local models.

So you're either a god in the cloud or you're a trusted personal assistant running in my pocket.

That is where the economics and the user needs seem to be pointing.

And open source in your view, owns that personal assistant layer.

I think it has to, because I do not want Google or any single corporation owning the AI that runs on my glasses and sees everything I see, And here's everything I say I want that to be mine, verifiably private and under my contry, and open source is the only technological path to guarantee that that.

Makes a ton of sense. It's the privacy and control argument again. It always seems to come back to that in the end.

It almost always does.

So as we wrap this up, what's the one thing our listeners should take away from all this when they see the next big headline about open AI or Google or nu Lama model, what lens should they view it through?

I would say, try to look past the simple versus narrative. Don't just ask who has the higher school on the latest benchmark today, ask a different set of questions. Where is this technology actually being deployed and who controls that deployment?

Follow the power and remember the meta paradox. Nothing is ever as simple as good guys versus bad guys.

Never The chaotic strategic middle is where the real story is almost always hiding, and.

I think that's a perfect place to leave it. We often think about technology as this linear race to a finish line, but maybe it's more about the kind of landscape we're building as we go. Who gets to build the roads, who owns the land and Are we allowed to build our own houses if we want to?

That is the fundamental question. What kind of world do we want this AI to help us build? And who gets a real say in the blueprints?

Something to mull over. Thank you for listening, Stay curious, stay skeptical, and we will see you next time.