Latest / Deep Values: Value Investing Deep Dives / Naval Ravikant: The Golden Age of AI and Human
Transcript
- 0:00Welcome back to the Deep Dive. Today we are we're doing
- 0:04something a bit structurally different than usual.
- 0:07Yeah, quite. Different.
- 0:07Normally we're sitting here with, you know, a giant stack of
- 0:10papers, white papers, some highlighted books, maybe a
- 0:12quarterly report or two. We parse the text for you.
- 0:15But today, the source material isn't text at all.
- 0:18It's audio. And it's definitely not studio
- 0:21audio either, right? It's actually it's quite jarring
- 0:24when you first put the headphones on, you listen to the
- 0:27source material and you just hear wind, You hear cars passing
- 0:30by, you hear the literal scuff of shoes on the pavement.
- 0:33It feels incredibly raw. We are analyzing a recorded
- 0:37conversation between Naval, Ravikan and NIV, and they're
- 0:39literally just walking around town.
- 0:41And Naval kicks off with this line that I think, really it
- 0:45sets the entire tone for what follows.
- 0:47He says. Brains run better when they're
- 0:48being locomoted. Which is biologically true, by
- 0:51the way. Is it?
- 0:52Yeah. Increased blood flow, the rhythm
- 0:54of walking, It actually creates a completely different kind of
- 0:57synchronization in your brain. But the reason we are analyzing
- 1:01this specific walk today isn't because of the exercise tips.
- 1:04No, definitely not. It's because of the sheer
- 1:06density of the forecast they lay out while they're walking.
- 1:09They're essentially mapping out the next five years of the
- 1:12software economy. And they introduced a term early
- 1:14on that I really think is going to stick.
- 1:15Vibe Coding. Vibe coding.
- 1:18It just sounds so casual, almost dismissive.
- 1:20Right. Like you're just hanging out.
- 1:21Exactly, like you're just chilling, but the implications
- 1:24for you, for everyone listening, are massive.
- 1:26So the mission for this deep dive is to unpack that exact
- 1:29concept. We need to understand why the
- 1:32average software engineer might actually be in a lot of trouble,
- 1:35why agencies becoming the single most valuable currency in the
- 1:39world, and why Naval thinks we are all about to become product
- 1:43managers of a very strange new workforce.
- 1:46It is a complete restructuring of the labor market, and it's
- 1:49disguised as a tech update. Really, it's about the shift
- 1:51from syntax to semantics. But before we get into the
- 1:54actual mechanics of vibe coding, we have to talk about the market
- 1:57context, because none of this stuff happens in a vacuum.
- 2:00No, it never does if you. Were trying to understand where
- 2:02we sit in the grand economic cycle, like are we in a bubble?
- 2:06Is it value time? You need data.
- 2:09Right. And one of the most reliable
- 2:10long term metrics for that is the relationship between the
- 2:13total market capitalization and GDP.
- 2:16Gives you a really good sense of whether the market is
- 2:18overheating relative to actual economic output.
- 2:21Exactly and to track total market temperature.
- 2:24Welcome to search and download iOS app Buffett indicator from
- 2:27Apple App Store. It is a super clean, no nonsense
- 2:31way to keep an eye on that high level valuation.
- 2:33And just a quick disclaimer before we really dive into the
- 2:35meat of this. Full deep value research reports
- 2:37can be found in the description. All analysis is based on
- 2:40publicly available sources and may be subject to revisions or
- 2:43differing interpretations. All contents from this channel
- 2:46is for educational and research purposes only and does not
- 2:48constitute investment advice. Conduct your own due diligence
- 2:52before making investment decisions.
- 2:53Perfect, and for our contrarian friends who want the really
- 2:57advanced stuff. If you want to access more
- 2:59advanced contents, welcome to subscribe on Deep Values dot AI.
- 3:04We will share our latest researches and the most advanced
- 3:07AI agent tools for our contrarian friends.
- 3:10Don't forget to like this video. Much appreciated.
- 3:12OK, housekeeping is done. Let's get back to the sidewalk.
- 3:15Let's get back to Naval. Yes, Vive coding.
- 3:18When I hear that term, I immediately picture someone
- 3:20sitting in a beanbag chair, just just feeling the software into
- 3:24existence. But that's not quite what it is,
- 3:27is it? It sounds like absolute magic,
- 3:28but he's actually describing a very specific, very tangible
- 3:31workflow. To understand vibe coding, you
- 3:34really have to look at the shift in the stack for the last 40
- 3:36years. If you wanted to build an
- 3:38application, let's say a tool to track your household budget, you
- 3:42needed to understand the syntax. You needed the code.
- 3:44Right. You needed to know Python or
- 3:45JavaScript. You needed to know exactly how
- 3:47to set up a relational database. You needed to know how to handle
- 3:50API requests securely. Basically you needed to know the
- 3:52how. Exactly.
- 3:53You had the what the idea for the budget app itself, but the
- 3:56how was this massive unscalable wall.
- 4:00Either climbed it yourself by learning to code, which takes
- 4:02years of your life, or you paid someone 150 grand a year to
- 4:06climb it for you. Navel's main point here is that
- 4:09the how is completely collapsing.
- 4:11He specifically mentions the new models like clawed code as the
- 4:15main catalyst here. Yes, and the workflow he
- 4:18describes is fascinating because it entirely reverses the
- 4:21traditional dynamic we're used to.
- 4:23It's not just you sitting there typing right me code code.
- 4:26It is a dialogue. You tell the AII want a budget
- 4:29app and the AI doesn't just spit out code blindly, it interviews
- 4:33you. This is the part that honestly
- 4:34blew my mind. The machine is the one asking
- 4:36the questions. Right, It asks you.
- 4:38Do you want to track expenses by category?
- 4:40Do you want to securely link this to your bank accounts?
- 4:42What should the user dashboard look like?
- 4:44Do you prefer a dark mode or a light mode aesthetic?
- 4:47It actively forces you to clarify your own intent.
- 4:50It's. Acting like a consultant.
- 4:51It acts exactly like a senior engineer or a really, really
- 4:54good product manager would. And once you answer all those
- 4:56questions, once you give it the vibe or the spec, it goes to
- 4:59work. But listen to what it actually
- 5:01does. Yeah, it builds the scaffolding,
- 5:04it downloads the correct libraries.
- 5:07It connects all the back end hooks.
- 5:09It writes the test harnesses to make sure the thing actually
- 5:11works. It does all the heavy lifting of
- 5:13construction. So the human is no longer the
- 5:16bricklayer in this scenario, the human is just the architect.
- 5:19Or as Naval puts it, the human becomes the product manager.
- 5:23And this is a really crucial distinction for you to grasp.
- 5:25In a traditional tech company, the product manager, the PM, is
- 5:29the person who owns the why and the what.
- 5:31They tell the engineers, hey, build this feature because the
- 5:33users need it. The engineers then go figure out
- 5:35the logic gates and the memory management.
- 5:37And in this new world we're entering the engineers.
- 5:40A piece of software. A piece of software and this AI
- 5:43has a very specific profile. Naval describes it as tireless
- 5:47and egoless that. Egoless part really struck me,
- 5:50because anyone who has ever managed a team of humans knows
- 5:53that friction is a huge, huge part of the cost of doing
- 5:57business. You have to manage morale.
- 5:59You have to manage burnout. You have to navigate creative
- 6:02disagreements. Right.
- 6:03Think about the social dynamics of a code review.
- 6:05If you tell a human engineer, actually I hate this interface,
- 6:09I want you to scrap it and start completely over, you're going to
- 6:12have a very difficult conversation.
- 6:13Oh, absolutely. You might lose that employee
- 6:15entirely. They might feel their work is
- 6:17undervalued or that you're just being capricious.
- 6:20But if you tell an AI, scrap it and start over.
- 6:23It just says OK, it enters a new prompt.
- 6:26It has 0 ego to bruise. It has absolutely no fatigue.
- 6:29It works 24/7, so the fundamental cost of iteration
- 6:33drops to 0, and when the cost of iteration drops to 0, the speed
- 6:36of product development just goes completely vertical.
- 6:38It changes the core constraint in the entire system.
- 6:41The constraint used to be engineering hours.
- 6:42A company would say we can only build 3 features this quarter
- 6:45because we only have 5 engineers.
- 6:47No, the constraint is just your ability to describe what you
- 6:50want clearly and your taste in judging if their output is
- 6:53actually any good. This brings up the question of
- 6:56taste, which Naval emphasizes incredibly heavily.
- 7:00If the coding part is essentially free, and the
- 7:03engineering part is fully automated, what is the human
- 7:06actually contributing to the process?
- 7:08Is it literally just the prompt? The human contributes judgement.
- 7:13Taste is the ability to look at the output and say this is
- 7:15clunky, or this user flow is confusing, or this color palette
- 7:19doesn't convey trust to the user.
- 7:21In the past you could get away with having really bad taste.
- 7:23If you are a great coder, you could build a clunky, ugly app,
- 7:26but if it was the only one that worked, people use it anyway.
- 7:29Oh man, think of early enterprise software.
- 7:31Horrible to use, just a nightmare of Gray boxes, but
- 7:34functional. We all suffer to it because we
- 7:36literally had no choice. Exactly.
- 7:37But now, if everyone can build functional software instantly,
- 7:40the differentiator becomes the user experience vibe, the nuance
- 7:44of the design. The human simply has to judge,
- 7:46is this good? Is this delightful to use?
- 7:48Does this actually solve the problem elegantly?
- 7:50So we are moving from a world of builders to a world of editors.
- 7:54In a sense, yes. But editors with extreme agency,
- 7:58editors who can manifest the product instantly, the gap
- 8:01between having an idea and holding a finished product has
- 8:05essentially collapsed. OK, so let's play this out a
- 8:07bit, because if I can vibe code an app on my walk to work and
- 8:11you can vibe code an app on your lunch break, aren't we about to
- 8:14just drown in apps? Naval literally calls it a
- 8:17tsunami. A tsunami of applications.
- 8:19That sounds incredibly messy. It's going to be incredibly
- 8:21messy. Imagine the App Store today.
- 8:23It's already insanely crowded. Now imagine it with 10,000 times
- 8:26the volume. Naval argues this fundamentally
- 8:29change in the economic structure of software.
- 8:31He compares it to what already happened to media.
- 8:33Like the YouTube model? Or the Amazon model.
- 8:35Yes, this is aggregator theory, and it's really worth unpacking
- 8:38this because it explains exactly who wins and who loses in this
- 8:42new economy. Before the Internet, you had
- 8:45independent local bookstores. They were the middle.
- 8:48They curated books for their specific local town and.
- 8:50They made a decent living doing it.
- 8:52They were the gatekeepers of their local geography.
- 8:54They did. Then Amazon comes along.
- 8:56Amazon is the aggregator. It effectively has infinite
- 8:59shelf space. So what happens?
- 9:01The middle gets completely hollowed out.
- 9:03You have the head, the massive best sellers, the Harry
- 9:06Potter's, the James Patterson's selling millions and millions of
- 9:08copies. And then you have the Long Tail,
- 9:10the highly obscure treatises on 18th century knitting that maybe
- 9:15sell 5 copies a year. But the local bookstore that
- 9:17sells a mediocre mix of both? They're gone because they can't
- 9:21stock the long tail and they can't discount the head enough
- 9:24to ever compete with Amazon. Completely wiped out and Naval
- 9:27sees this exact same pattern coming for software.
- 9:30We are very used to the head of software.
- 9:32Google, Facebook, Salesforce, Microsoft Excel, these are the
- 9:35Harry Potter's of software. They are massive, highly
- 9:38polished, multi billion dollar products.
- 9:41And those aren't going anywhere. No.
- 9:42In fact, they get even stronger because they can use these
- 9:44internal AI tools to make their own products even better.
- 9:47The engineers working on Google search are now 100 times more
- 9:50productive because of AI, so the head just gets bigger and more
- 9:53dominant. But what about that long tail of
- 9:55software? What does that actually look
- 9:56like for you and me? This is where vibe coding really
- 9:59shines. Navel gives a great kind of
- 10:01funny example of a lunar phase tracker.
- 10:04A lunar phase tracker. But not just any generic
- 10:06tracker. A lunar phase tracker
- 10:08specifically designed for, say, an INTJ personality type who
- 10:12also happens to really like gardening.
- 10:13OK, that is niche. That is painfully niche.
- 10:16It's hyper niche in the old world, let's say even just back
- 10:18in 2023, you could never ever build that.
- 10:21To hire an engineer to build a gardening Lumer tracker for
- 10:24INTJS will cost you 100 grand minimum.
- 10:28The total addressable market for that app is maybe 5000 people
- 10:31globally. The math simply doesn't.
- 10:33Work right, You'd never make your money back.
- 10:35The customer acquisition cost would be way higher than the
- 10:37lifetime value of the user. Exactly.
- 10:39But in the world of vibe coding, you can build it in a single
- 10:42afternoon for the cost of your electricity bill and a basic AI
- 10:45subscription fee. Suddenly the economics of the
- 10:48long tail become entirely viable.
- 10:51So we really are going to see a zillion tiny apps.
- 10:53A zillion of them. You'll see an app built just for
- 10:56your specific homeowners association, an app for your
- 10:59family's personal recipe collection that directly with
- 11:02the API of your specific local grocery store software becomes
- 11:06disposable and hyper customized to the individual.
- 11:09But here. Is the really dark side of that
- 11:11curve? If the head gets stronger and
- 11:14the long tail absolutely explodes, what happens to the
- 11:17middle? What happens to the average
- 11:19everyday app naval? Says there is no demand for
- 11:22average. That's the direct quote.
- 11:23It's a pretty. Brutal line it is.
- 11:25Brutal, but it's the strict reality of digital distribution.
- 11:29Think about it in the physical world, Physical geography
- 11:31protects the average. If I live in a small town, the
- 11:34local pizza place doesn't actually have to be the best
- 11:36pizza in the world, it just have to be the best pizza within a 10
- 11:39mile radius. It has a geographic.
- 11:40Mode protecting its business exactly.
- 11:42But on the App Store there is 0 geography.
- 11:45I can download the absolute best calorie tracker in the world
- 11:48just as easily with the exact same tap of a finger as I can
- 11:52download the 10th best. So why on earth would I ever
- 11:55download the 10th best? You wouldn't.
- 11:57I wouldn't. I download the one everyone else
- 11:58is already using or the one with the objectively best features.
- 12:01This. Creates a pure winner take all
- 12:03dynamic. Naval actually quotes that
- 12:06famous scene from Glen Ross to explain it first place.
- 12:09Gets a Cadillac second place. Gets a set of steak knives. 3rd
- 12:12place you're fired in the. Software world, the second best
- 12:16search engine doesn't even get steak knives.
- 12:18It just gets 0. It gets.
- 12:19Completely ignored. So the pressure on quality is
- 12:23immense now if you are trying to compete in the head, if you are
- 12:26sitting there trying to build the next Instagram or the next
- 12:28Spotify, you have to be perfect. You have to be better than the
- 12:32massive incumbents who already have billions of dollars in
- 12:35armies of custom AI models. That sounds.
- 12:37Literally impossible for a normal startup.
- 12:39It is nearly. Impossible.
- 12:41That's exactly why the opportunity for you, for the
- 12:43listeners, for the new Vibe Coders is entirely in the
- 12:47niches, the long. Tail the long.
- 12:49Tail Don't try to be the second best Google.
- 12:52Be the absolute best application in the world for tracking
- 12:55inventory in a vintage comic book store.
- 12:57Be the complete monopoly of your specific micro niche naval.
- 13:01Has that piece of advice, he says keep redefining what you do
- 13:04until you are the best in the world at it.
- 13:06That is the. Entire strategy right there.
- 13:08You keep narrowing the scope of what you're offering.
- 13:10I'm the best at software. No, that's too broad.
- 13:12I'm the best at financial software.
- 13:14Still way too broad. I'm the best at financial
- 13:16software for dentists. Getting warmer.
- 13:19I'm the best at financial software for dentists who
- 13:20specialize in orthodontics in the UK.
- 13:23Bingo. Now you have a monopoly.
- 13:25And because. You can build it basically for
- 13:27free with AI. That tiny monopoly is actually
- 13:31highly profitable. You can make a great living just
- 13:33owning a micro niche. Precisely.
- 13:35The barrier to entry is super low, but the barrier to
- 13:38excellence in a specific niche is still high because you
- 13:41actually need that deep domain knowledge.
- 13:43You need to know what UK orthodontists actually struggle
- 13:45with every day. I want to pivot.
- 13:47Here just a bit because we have been talking a lot about Vibe
- 13:50coders, the nontechnical people entering the space, but I know
- 13:54we have actual highly trained software engineers listening to
- 13:58this right now. People who sent four years in
- 14:00university learning memory pointers, Big O notation and
- 14:04compiler design, and they are probably sweating a little bit
- 14:07listening to this conversation. Yeah, it's.
- 14:09A completely valid anxiety. If English is the new
- 14:11programming language, what happens to the people who spent
- 14:14a decade mastering C++ house? Did they just.
- 14:17Waste their time? Is their degree entirely
- 14:19worthless now? Naval.
- 14:20Says no. Emphatically no.
- 14:22In fact, he argues that traditional software engineers
- 14:24are about to become the most leveraged people on the planet.
- 14:27Leverage is a very big word in the naval lexicon.
- 14:29It is. Leverage really just means
- 14:31getting drastically more output for every single unit of input
- 14:34you provide. He argues that engineers aren't
- 14:37being replaced at all, they are being given a superpower.
- 14:40But why? Do we still need them?
- 14:41If the Vibe coder can just talk to Claude and get a fully
- 14:44functioning app, why do I need the guy who deeply understands
- 14:47the Linux kernel Because of? Leaky abstraction.
- 14:50OK this. Is a fairly technical concept,
- 14:52but it's really crucial for everyone to grasp.
- 14:54We need to break this down. What exactly is a leaky
- 14:57abstraction? Think of a.
- 14:59Car. A car is an abstraction of a
- 15:01highly complex combustion engine.
- 15:03You have a gas pedal and a steering wheel.
- 15:05That's your interface. You don't need to know how the
- 15:08fuel injection works or how the spark plugs fire in sequence
- 15:11just to drive to the grocery store.
- 15:13All that mechanical complexity is abstracted away from you
- 15:16until the. Car breaks down until the car.
- 15:18Breaks, or until you try to drive that standard sedan up a
- 15:2190° rocky Cliff when you push the tool to its absolute limit.
- 15:26When the tool simply malfunctions, the abstraction
- 15:28leaks. The messy reality of the machine
- 15:30pokes right through the interface.
- 15:32And if you don't? Actually know how the machine
- 15:33works under the hood. You are just stranded on the
- 15:36side of the road waiting for a tow truck.
- 15:38Exactly. AI tools like Claude or ChatGPT
- 15:41are massive, massive abstractions.
- 15:44They completely hide the code, they hide the underlying logic.
- 15:46And most of the time, for basic tasks, that's totally fine.
- 15:50But AI makes mistakes. It hallucinates entire concepts.
- 15:54It uses highly inefficient older libraries.
- 15:56It accidentally introduces major security vulnerabilities into
- 16:00your database. So the vibe.
- 16:02Coder builds an app and it works perfectly for 100 users.
- 16:05But then they get popular, they scale to 10,000 users and the
- 16:08whole thing just crashes entirely because the database
- 16:10architecture was sloppy and the vibe.
- 16:12Coder just stares at the screen and types Claude fix it, and
- 16:15Claude either says I don't see the problem, or it provides a
- 16:18fix that completely breaks something else in the app.
- 16:20Now what? The vibe coder is entirely
- 16:22stuck. They've hit the hard limit of
- 16:24the abstraction. Enter the.
- 16:25Traditional engineer Enter the. Wizard Naval uses this fantasy
- 16:29analogy that I absolutely love. AI is essentially a magic wand.
- 16:33And now everyone gets a wand. You can wave it around and make
- 16:37some pretty sparks fly. But the traditional engineer,
- 16:40they are the ones who actually know the spells.
- 16:42They understand the arcane underlying laws of the magic.
- 16:45They know why the sparks are flying.
- 16:46So when the vibe. Coders app crashes and burns.
- 16:49The engineer can actually pop the hood, look at the raw code
- 16:52and say ah here is the memory leak, here is the race
- 16:55condition, here is exactly where the AI hallucinated a back end
- 16:59function that doesn't actually exist exactly.
- 17:01The engineer is the only one who can fix the wand itself.
- 17:04There's also. The concept of edge cases that
- 17:06Navel brings up in the walk, right?
- 17:08AI is trained on what we call in distribution data.
- 17:11That means it has seen billions and billions of examples of
- 17:13standard corporate websites, standard user login forms,
- 17:16standard e-commerce shopping carts.
- 17:18It can build those things in its sleep.
- 17:20But what if you are trying to build something that has
- 17:22literally never been built before?
- 17:23Like a completely. New high frequency trading
- 17:25algorithm or a novel zero knowledge crypto protocol?
- 17:29Or a totally new physics engine for a virtual reality game, Or
- 17:33operating software for a quantum computer.
- 17:36The AI hasn't seen that data because that data doesn't exist
- 17:39yet. It can't accurately predict the
- 17:41next token because the tokens for that specific problem
- 17:44haven't been written by humans yet.
- 17:45So it just. Starts guessing it starts.
- 17:47Hallucinating wildly in those specific moments, the extreme
- 17:50edge cases, the truly novel groundbreaking problems, The
- 17:53human engineer is still the absolute only one who can solve
- 17:57it. You still need to sit down and
- 17:59hand code that critical infrastructure from scratch.
- 18:02So the message. To the the engineer's listening
- 18:03is clear. You are not obsolete.
- 18:05You are the safety net. You are the architect of the
- 18:08truly complex systems. And you?
- 18:10Should be using the AI to handle all the boring stuff.
- 18:12Let the AI write the boilerplate code.
- 18:13Let the AI write the tedious documentation.
- 18:15Let the AI write the unit tests. You save your actual human brain
- 18:18power for the hard novel problems.
- 18:20That is what 100 X leverage looks like.
- 18:23You are effectively managing a massive fleet of junior
- 18:25developers who work at the speed of light.
- 18:27This leads. Us to a much deeper
- 18:29philosophical point that Naval makes we're talking about jobs
- 18:33here, the job of a coder, the job of APM.
- 18:36But Naval draws a very sharp distinction between having a job
- 18:40and having entrepreneurial agency.
- 18:41This was. Honestly my favorite part of the
- 18:43entire conversation. He casually drops this line.
- 18:46No entrepreneur is worried about an AI taking their job.
- 18:49Which when? You first hear it sounds a
- 18:51little arrogant, like, oh, I'm an entrepreneur.
- 18:53I'm too special and unique to be replaced by a machine.
- 18:56But let's unpack why he says that it really comes.
- 18:58Down to the fundamental definition of a job.
- 19:00A job, generally speaking, is a predefined slot in an
- 19:03organization. It's a strict set of tasks.
- 19:06Process these specific invoices, write the specific software
- 19:09module, manage the specific team of five people.
- 19:12If your daily work can be defined by a rigid set of
- 19:15instructions and AI can eventually and probably will do
- 19:18it. But an?
- 19:19Entrepreneur is different an entrepreneur.
- 19:21Operates entirely in the unknown.
- 19:22They're actively trying to create something that does not
- 19:25currently exist in the world. They are exercising what Naval
- 19:27calls extreme Agency Agency. That is the keyword for this
- 19:32entire era agency. Is the ability to act upon the
- 19:35world to achieve a specific desire.
- 19:38And this is exactly where a naval identifies the hard,
- 19:41immovable limit of artificial intelligence.
- 19:44AI as it exists today and for the foreseeable future, has
- 19:48absolutely zero agency. Because it has.
- 19:50No desire. It literally doesn't want
- 19:52anything it wants. Nothing.
- 19:53It doesn't have mortal fear. That's a really.
- 19:55Interesting connection he makes the link between fear and desire
- 19:58naval. Argues they're deeply
- 20:00intrinsically linked, You and I, we have evolutionary biological
- 20:03drives. We want to eat.
- 20:05We want social status. Want to reproduce?
- 20:07We want to not die tomorrow. These underlying drives are what
- 20:10create our goals. I want to start a company so it
- 20:12can be financially free. I want to build this cool app so
- 20:14I can impress my peers. I want to solve this specific
- 20:17problem because it personally annoys me everyday.
- 20:20The AI. Does not care if you turn it
- 20:22off. It doesn't.
- 20:23Care at all. It has no survival instinct,
- 20:25therefore it has no intrinsic self generated goals.
- 20:29It only has the goals that we explicitly give it.
- 20:32Write a poem about a cat. Calculate this massive sum.
- 20:35It is merely a tool. It is a highly advanced missile.
- 20:39But a missile is useless without a guidance system and the human.
- 20:42Is the guidance system the human?
- 20:44Is the guidance system. That is exactly why the true
- 20:46entrepreneur isn't worried. If you are trying to go to Mars
- 20:49like Elon Musk, you aren't worried that the rocket ship is
- 20:52going to somehow take your job. You are the one who actually
- 20:55wants to go to Mars. The rocket is just the vehicle
- 20:57getting you there. But wait.
- 20:59What about the whole idea that AI is intelligent?
- 21:02If it's eventually much smarter than us, won't it just naturally
- 21:04develop its own agency, Naval? Pushes back really hard on the
- 21:07standard definition of intelligence here, he says.
- 21:09The only true, valid test of intelligence is if you actually
- 21:13get what you want out of life. That's a very.
- 21:15Pragmatic real world definition. It's not about scoring high in
- 21:19an IQ test or winning at chess, right?
- 21:21It is highly pragmatic. If you are a certified genius
- 21:24but you are miserable and broke, are you really acting
- 21:26intelligently? But applying that metric to AIAI
- 21:30completely fails the test because it wants nothing.
- 21:32It physically cannot get what it wants because there is no what
- 21:35it wants. But what about the?
- 21:360 sum trap Naval talks about how this agency dynamic applies to
- 21:41highly competitive things like dating or stock trading.
- 21:44This is a. Fascinating thought experiment.
- 21:46There are certain domains in life that are naturally
- 21:48positive. Some like building houses.
- 21:51If we have AI robots build houses, everyone gets a cheaper,
- 21:54better house. We all collectively win.
- 21:56But dating? Dating is.
- 21:57Strictly 0 sum. If I marry the person of my
- 22:00dreams, you cannot marry them. It is a direct competition.
- 22:03Stock trading is 0 sum. If I buy a stock at the absolute
- 22:06bottom, someone else had to sell it to me at the bottom.
- 22:07They lost out on the gain I made.
- 22:09So what? Happens when literally everyone
- 22:11brings a super intelligent AI to a 0 sum game.
- 22:15The. Alpha completely cancels out.
- 22:18Imagine a near future where everyone has an AI dating coach.
- 22:22You are sitting at dinner on a date.
- 22:23You have a tiny earpiece in. The AI is whispering the
- 22:27absolute perfect, most charming, witty thing to say based on
- 22:30millions of successful interactions.
- 22:32But my date. Also has an earpiece.
- 22:34Exactly. And her AI is actively analyzing
- 22:38your voice patterns, parsing your syntax, and whispering to
- 22:41her. He is currently using a standard
- 22:43GPT 4 flirtation script. Disregard his comments, it's
- 22:46just a. Ridiculous arms race to the
- 22:48bottom or a race? To the exact middle.
- 22:50If everyone has access to the exact same super intelligent
- 22:53tool, the baseline competence of the whole market rises, but the
- 22:56relative advantage completely disappears.
- 22:58If everyone is super smart, no one is super smart.
- 23:01So where does? The actual advantage come from
- 23:03then. If we all have the same AI
- 23:04whispering in our ears, it comes right back.
- 23:06To the human, it comes back to the weird, the genuine
- 23:08creativity, the unpredictability, the things the
- 23:10AI would never, ever suggest because they simply aren't in
- 23:13the statistical manual. This is what we call.
- 23:14The photography momentum for creativity.
- 23:16Exactly. This is the historical analogy
- 23:18that actually makes me feel incredibly optimistic about all
- 23:21of this. Think about the mid 19th
- 23:23century. Imagine you are a professional
- 23:26painter. Your entire job, your core value
- 23:29proposition to society is realism.
- 23:32You are paid handsomely to paint a portrait of the Duke that
- 23:35actually physically looks like the Duke, and that was a.
- 23:38Highly technical skill. It required incredible hand eye
- 23:40coordination. It took decades of practice to
- 23:43master. It did.
- 23:44And then click the camera is invented.
- 23:46Suddenly a relatively simple machine can produce a perfect,
- 23:49completely realistic image in a fraction of a second, better and
- 23:53more accurate than you ever could with a brush.
- 23:55So did the art. Of painting.
- 23:56Just die, no? But realism completely ceased to
- 23:59be a unique selling point. So art had to move.
- 24:01It was forced to go where the camera physically couldn't go.
- 24:03It went. Inward it went weird.
- 24:05We got. Impressionism.
- 24:05We got Cubism, we got surrealism, Abstract
- 24:08expressionism. Human artists completely stopped
- 24:11trying to capture reality and started trying to capture raw
- 24:13feeling or complex concepts or even just madness and.
- 24:17Naval thinks the exact same thing is happening right now
- 24:19with AIAI. Fully democratizes the basic It
- 24:22can write a perfectly standard corporate cover letter.
- 24:25It can write a standard predictable Hallmark movie
- 24:27script. It can code a standard
- 24:29e-commerce website. That is the realism of our
- 24:32current era. So human creativity has to move
- 24:35to the surrealism of our time. Exactly.
- 24:37We have to purposefully move out of distribution.
- 24:40Explain that. Term again for us out of
- 24:41distribution AI models are. Fundamentally trained on a
- 24:44specific distribution of data, which is basically everything
- 24:47humanity has already produced and uploaded up to roughly 2023
- 24:51or 2024, they are at their core statistical prediction engines.
- 24:56They simply predict what word or pixel comes next based entirely
- 24:59on what came before in the training data.
- 25:01They're. Conformist by design.
- 25:03They are literally the ultimate conformists, yes.
- 25:06They are mathematically mandated to conform.
- 25:08If you ask an AI to write a pop song, it will write something
- 25:12that sounds exactly like the mathematical average of every
- 25:14single pop song ever written. It will be catchy, sure, but it
- 25:18will not be revolutionary. It won't be The Beatles in 1964.
- 25:22It won't be Nirvana in 1991. Because those.
- 25:24Bands completely broke book the existing pattern their music
- 25:28sounded wrong to people at first they.
- 25:30Introduced a completely new signal into the noise true human
- 25:34creativity, Navel argues, is a leap it is finding a strange
- 25:38connection that simply was not present in the historical
- 25:40training data he actually. Contrast this with Steve Jobs.
- 25:43Famous definition of creativity, right?
- 25:46Yeah, Jobs. Famously said, creativity is
- 25:47just connecting things, Navel says.
- 25:49Well, actually, AI is really, really good at connecting
- 25:52things. It can seamlessly connect a
- 25:54Shakespearean's on it to a toaster repair manual if you
- 25:56prompt it to. So creativity.
- 25:58Isn't just basic connection anymore, it is.
- 26:00Unpredictable connection. It was a Black Swan event.
- 26:03It is the crazy idea that makes absolutely no sense to anyone
- 26:07until it actually exists and then it retroactively changes
- 26:10everything. So if.
- 26:11You want to stay relevant in your career.
- 26:13Don't try to be the AI at being consistent or fast.
- 26:17Be the AI at being chaotic. Be the glitch.
- 26:20In the system, be the human anomaly I really love.
- 26:22That be the glitch. OK, let's move to the final
- 26:25major theme of their walk, which is learning.
- 26:28Because if we are all going to be these new product managers
- 26:30and Wizards, we clearly have a lot to learn.
- 26:32And naval calls AI The ultimate tutor this is.
- 26:35Easily the most democratizing and hopeful part of the entire
- 26:38technology. Think about the biggest barrier
- 26:40to learning deeply complex subjects like quantum physics,
- 26:43advanced math, or coding. It's usual just that you get
- 26:45stuck on one specific concept. You hit a wall.
- 26:49The textbook explains it in a very specific academic way that
- 26:51you just don't understand, and you're reading it at 2:00 AM and
- 26:54there is no one around to ask or even.
- 26:56Worse, you feel stupid asking if you are 40 years old and you
- 26:59suddenly realize you don't actually understand how basic
- 27:02fractions work. You aren't going to raise your
- 27:03hand in a crowded night class, you are just going to suffer in
- 27:06silence and give up shame. Is a massive underdiscussed
- 27:10barrier to learning AI. Entirely removes the shame.
- 27:13It acts as the endlessly patient tutor.
- 27:16It meets you exactly where your current comprehension level is.
- 27:19Navel gives this brilliant example.
- 27:21You can tell it explain the theory of relativity to me using
- 27:24an eighth grade vocabulary, but assuming I know a whole lot
- 27:26about residential plumbing, and it can.
- 27:27Actually do that. It'll say.
- 27:28Imagine gravity is like a clogged pipe.
- 27:31It can do. That instantly it can perfectly
- 27:33modulate its output to your precise, weird context and it
- 27:36never, ever gets tired. You can ask at the exact same
- 27:39dumb question 10 times in a row in 10 slightly different ways
- 27:42and it will never roll its digital eyes at you, that is.
- 27:45Incredibly powerful. It is truly personalized
- 27:47education at a global scale and naval.
- 27:50Uses a really beautiful visual metaphor to explain why AI is
- 27:53such a profoundly good teacher. The pachinko machine I love.
- 27:57This image from the audio walk us through how he describes it.
- 27:59OK. Imagine a giant glass front
- 28:03pachinko machine. You know, the vertical Japanese
- 28:06arcade game where little metal balls drop through a dense maze
- 28:10of brass pins. You take millions and millions
- 28:12of these little metal balls and pour them into the top.
- 28:15These balls represent all the raw data in the world.
- 28:18Pixels of images, words from every book ever written,
- 28:20millions of lines of code snippets.
- 28:22That's the training. Data being fed into the model,
- 28:24right? And as the balls fall down
- 28:26through the pins, the machine uses back propagation to
- 28:29constantly adjust the angles of those pins.
- 28:31It is desperately trying to find a path, a specific set of
- 28:34mathematical weights and biases, that allows all the balls to
- 28:37flow through the bottom in a way that effectively compresses the
- 28:40data. Compression is.
- 28:41The absolute key concept here why is data compression so
- 28:44important to intelligence? Because.
- 28:45Compression literally is understanding to compress
- 28:48massive amounts of data. You are forced to find the
- 28:51underlying pattern. If I want to send you a computer
- 28:53file containing 1,000,000 blue circles, I could send you the
- 28:56exact X&Y coordinates of every single blue pixel, but that's a
- 29:00massive multi GB file that's just raw uncompressed data.
- 29:05Or you could. Just send me the basic geometric
- 29:06formula for a circle and the hex color code for blue exactly.
- 29:09Π * r ^2 color hex 0000 FF. That tiny one line formula
- 29:15accurately represents the million circles that is
- 29:18compression. But to find that tiny formula,
- 29:20the machine first had to truly understand what the abstract
- 29:23concept of a circle actually is. So by.
- 29:25Forcing the AI to mathematically compress the entire Internet.
- 29:28We accidentally forced it to understand the deep concepts
- 29:31underlying the Internet, yes. It didn't just rigidly memorize
- 29:34the text of a physics textbook, it had to actually learn the
- 29:37underlying concept of gravity in order to accurately predict the
- 29:41next word in the physics textbook.
- 29:43That deep conceptual mapping is exactly why it can explain it to
- 29:46you using a plumbing analogy. It has mapped the fundamental
- 29:50structure of human knowledge itself.
- 29:52So. Practically speaking, for our
- 29:53listeners, how should we be using this giant pachinko
- 29:57machine? Naval has some pretty strong
- 30:00opinions on the current trend of prompt engineering.
- 30:03Oh yeah, he. Calls them boomer queries
- 30:05Boomer. Queries.
- 30:06That's hilarious, he. Genuinely thinks the whole
- 30:08burgeoning industry of prompt engineering.
- 30:10This idea that you have to learn special magical syntactical
- 30:13phrases to get the AI to work properly is just a temporary
- 30:16artifact of early software. It's like taking a class on how
- 30:19to use a punch card in the 1960s.
- 30:21It's a bridge. Technology that will be obsolete
- 30:23in the year? Exactly.
- 30:24He says don't learn syntax, just talk to it.
- 30:27He literally just rambles at the AI AI into his phone.
- 30:30He treats it exactly like a very smart, very patient colleague.
- 30:34If the AI doesn't understand what he meant, Navel says,
- 30:36that's the AI's fault, not his. And the models are improving so
- 30:40incredibly fast that the need for strict prompt engineering is
- 30:43just vanishing overnight. Plain English.
- 30:45Is the new interface English? Is the code.
- 30:48If you can clearly articulate your complex thought, you can
- 30:51program computers now. He also mentions.
- 30:53A very practical tip about paying for the best models
- 30:55available. Don't be cheap.
- 30:56With intelligence, that is his golden rule it makes.
- 30:59Total sense when you think about it.
- 31:01If you were hiring a human consultant to run your business,
- 31:04would you hire the guy who gets AC minus average just because he
- 31:07charges $10 or the guy who gets an A+ who charges $20?
- 31:11Especially when the actual cost difference is literally just
- 31:14$20.00 a month for a premium subscription.
- 31:17Naval uses the advanced thinking models, the ones that actually
- 31:20take a minute to reason through a problem, like the O1
- 31:23equivalents for absolutely everything, even trivial
- 31:25everyday stuff. But why?
- 31:27Use a multibillion dollar supercomputer to just check if
- 31:30your fridge temperature is right because.
- 31:32In the real world, reliability matters more than speed.
- 31:36A 95% correct answer is often completely useless.
- 31:39If I give you a complex code snippet that is 95% correct, the
- 31:42app is broken. It will not compile.
- 31:45You want the model that is 99.9% correct.
- 31:48The value of true intelligence is exponential, not linear, so
- 31:51his. Advice is just get the premium
- 31:53subscription, use the biggest brain available.
- 31:56It is. Arguably the highest ROI
- 31:58investment any individual can make right now we have covered.
- 32:00So much ground today. The incoming tsunami of hyper
- 32:03niche apps. The complete death of the
- 32:05average software product. The rapid rise of the
- 32:08non-technical vibe coder. The deep philosophical nature of
- 32:11human agency versus machine logic.
- 32:13As we wrap this deep dive up, I really want to address the
- 32:16elephant in the room for a lot of people.
- 32:18The underlying anxiety it is. Palpable out there, you hear all
- 32:21this rapid change and you naturally, I think the ground is
- 32:23shifting under my feet. My hard earned skills are
- 32:25expiring. What if I get completely left
- 32:27behind in this new economy naval?
- 32:29Addresses this directly. He says all anxiety simply comes
- 32:32from the unknown. We deeply.
- 32:34Fear what we don't understand. If AI remains a magical black
- 32:37box to you, it becomes a monster, it becomes Skynet, it
- 32:41becomes this unstoppable force that just happens to you so.
- 32:44How do you actually turn the monster back into a tool?
- 32:46You open the. Hood you don't have to go back
- 32:48to school to become a machine learning engineer.
- 32:50You don't need to do the complex calculus, but you do need to
- 32:53understand the basic principles we discussed today.
- 32:56Understand the pachinko machine analogy.
- 32:58Understand that it is a statistical predictor, not an
- 33:01omniscient God. Understand its limits with edge
- 33:04cases right once you. Understand mechanically how a
- 33:06chainsaw works. You stop being terrified that it
- 33:08will randomly jump out and bite you, and you start respecting it
- 33:11as a highly efficient tool for cutting wood precisely.
- 33:14Action is the only cure for fear and the final.
- 33:16Call to action he leaves us with is actually pretty inspiring.
- 33:19Naval basically calls us all spellcasters.
- 33:22Now we are. Entering a golden age of
- 33:24creation, we have all been handed the magic wand.
- 33:27The traditional barriers to building a business, to creating
- 33:30art, to solving hard technical problems, they have all
- 33:33completely collapsed. The only absolute limit left is
- 33:37your own personal curiosity, the means of.
- 33:39Learning are abundant the desire to learn is scarce that.
- 33:43Is the ultimate take away from you.
- 33:45If you aren't actively building something or learning something
- 33:48new right now, you really can't blame the lack of tools or
- 33:51access anymore. The bottleneck is just you, so
- 33:54to. Everyone listening.
- 33:56Stop worrying about the robot taking your current job.
- 33:58Start thinking very deeply about the mission you actually want to
- 34:01lead. Be the.
- 34:02Entrepreneur be the pilot, let the AI just be the engine that
- 34:06gets you there and go. Find your specific niche in the
- 34:08long tail. Go build that highly specific
- 34:11lunar tracker for gardening introverts.
- 34:13Seriously. Someone please build that and
- 34:14send it to us. Thank you for.
- 34:15Diving deep with us today, this is one of those conversations
- 34:18that I truly think we will look back on in five years and say
- 34:21that was the exact moment it all shifted I completely.
- 34:24Agree, it's a whole new world. Go explore.
- 34:26It go vibe code something weird. We'll see you on the next deep
- 34:29dive. Take care.
- 34:30Everyone.