Naval pushes back on how the term world model is being used. Generating something that looks like a world you can wander around in is not a world model. A world model is an agent carrying an internal model of the world that lets it act, predict the consequences of its actions and adjust its behaviour from what happened, in a reinforcement-learning loop. He also grades current AI as jagged and weak at multimodal reasoning.

Common confusion: visual navigable output mistaken for an internal model.
Real test is predict, act, observe, adjust.
Naval calls today's AI jagged in intelligence and poor at multimodal reasoning.
New kinds of models, agents and intelligence are emerging, including Yann LeCun's recent work.

“The AI that I'm seeing at has jagged intelligence. It's also pretty bad at multimodal reasoning.”
Naval Ravikant · 08:30

“A world model is when you have an agent that has a model of the world inside its head, which allows it to take actions…”
Naval Ravikant · 09:00

#ai#world models#agents#reinforcement learning

Hot Take· 5

Hot Take01:00

Why Naval refuses to do organizational management at all

Asked whether he is restructuring his company around AI the way Jack Dorsey did at Square or Tobi is doing at Shopify, Naval says he has never been good at organizational management and actively hates it. His objection is to organizations themselves: at size you stop working with the best and the brightest and politics is always present. His answer is not better management but smaller groups.

Some founders are genuinely good at org experiments; Naval says he is not one of them.
His complaint about large groups is politics plus a diluted talent bar.
Small groups let you count on people operating independently.
The company runs on direct texts and one-on-one conversation, with GitHub as the only shared tool.

“I've never been good at organizational management. I actually hate organizational management cuz I hate organizations. I hate large groups.”
Naval Ravikant · 01:00

“So, I just prefer keeping groups small.”
Naval Ravikant · 01:00

#management#org design#startups#founders

Hot Take09:30

The nothing ever happens meme is over

Naval calls time on the X meme that nothing ever happens. He cannot fully articulate why, but says anyone paying attention can see the world moving faster post-COVID, whether because COVID caused a dislocation or because it broke an already unstable equilibrium and triggered a phase shift. The acceleration is geopolitical, economic and technological at once.

Post-COVID the pace of change is visibly higher across domains.
Possible cause: an unstable equilibrium that COVID broke, producing a phase shift.
VCs are being pushed into hardware, rockets, drones and AI.
Sci-fi technologies are in high demand while sci-fi scientists, authors and engineers are in low supply.
He frames the moment through the curse about living in interesting times.

“The famous meme I think at X was like nothing ever happens, right? I think that's over. I haven't quite been able to put my…”
Naval Ravikant · 09:30

“So I think sci-fi technologies are in high demand. Sci-fi scientists and sci-fi authors are in low supply. Sci-fi engineers are in low supply.”
Naval Ravikant · 10:00

#macro#technology#venture capital#covid

Hot Take12:30

AI democratizes bioweapons the same way it democratized coding

Naval names biological weapons as one of the real fears about AI. In the past, making one required both expertise and access, and very few people had both, though he argues even that number was too high. AI expands the pool by the same factor vibe coding expanded the pool of coders, which is hundreds of thousands of times. The counterweight is that the same AIs can research vaccines and countermeasures.

The old constraint on bioweapons was the overlap of expertise and access.
Vibe coding is the scale analogy: the pool of people who can code grew by orders of magnitude.
The same multiplier applied to biology is what makes it frightening.
The optimistic counterweight is AI-assisted vaccine and countermeasure research.
That defensive research is gated behind regulation while the offensive capability is not.

“Also, I think one of the fears with AI is biological weapons.”
Naval Ravikant · 12:30

“So, now that power is going to be democratized just like vibe coding is democratized.”
Naval Ravikant · 13:00

#ai risk#biosecurity#vibe coding#regulation

Hot Take13:30

Medical regulation is the bottleneck on AI solving biology

Naval calls AI applied to medicine, biology and therapies one of the real opportunities, but says it needs data that sits behind silos and rules. His proposal is anonymized, cleaned datasets plus a right to try. He argues almost no regulations are as bad as medical ones, points at how long COVID vaccines took because volunteer challenge trials were not allowed, and fears the unlock only comes in an emergency.

The opportunity requires looking at everyone's data and all the outcomes, not a sample.
Proposed unlock: anonymize, clean, release, then allow a right to try therapies.
COVID showed even emergency conditions did not produce speed.
Volunteer challenge trials were blocked by bioethics rules.
His framing of the core problem is too many people who can say no versus few trying to get things done.

“And there are almost no regulations out there as bad as medical regulations.”
Naval Ravikant · 13:30

“But, if you could anonymize, clean up, and allow that data set to get out there, and then you could let people test therapies with…”
Naval Ravikant · 14:00

#healthcare#regulation#ai#biotech

Hot Take17:30

Doom is easy to imagine, so optimism has to be nurtured

Naval's answer to why he does not get worked up about the future is that doom scenarios are simply easier to imagine than positive ones, because optimism requires creativity. Catastrophes are more legible to our minds so we hold them closer, and every decade of his life has produced a new environmental or war-driven end-of-the-world story. His conclusion is that optimism must be nurtured, rewarded and held irrationally.

Imagining the methods of doom is easier than imagining the methods of rising up.
Catastrophic outcomes are more legible, so people fixate on them.
Every decade has produced a fresh environmental or war-based doom story; some were genuinely close.
Optimism is the only way out, so it needs active reward rather than mere tolerance.
The crabs-in-a-bucket doomers may be right and are still not who you want beside you.

“Yeah, I don't get worked up about it because I think it's just so much easier to imagine doom scenarios than it is to imagine…”
Naval Ravikant · 17:30

“And so I think we have to nurture optimism. We have to reward optimism. We have to be irrationally optimistic because that's the only way…”
Naval Ravikant · 19:00

#mindset#optimism#risk#psychology

Explainer· 2

Explainer07:00

The questions Naval is actually trying to figure out about AI

Two to four companies dominate AI, five if you count Nvidia on hardware, and Naval says he does not know whether that is stable. He lays out the live questions: commodity, monopoly or oligopoly; do models top out on data or run to AGI; does the world consolidate past the Mag 7 or fragment. He calls the emerging consensus on centralized training probably right, and names the contrarian bet he cannot yet justify.

Conventional wisdom: centralized training, two to four dominant firms, data centers and power as the limits.
The contrarian bet would be distributed training, but Naval says he does not yet see the evidence.
Open questions include whether models stop improving as data runs out.
People inside the frontier labs believe all value disappears into the labs; he declines to be in the futurist business.
Users may trade privacy and open source away simply to get the smartest model.

“Is this going to be a commodity business, or is this going to be a monopoly business, or is it going to be an oligopoly…”
Naval Ravikant · 07:00

“I think now the conventional wisdom is going centralized training, two to four companies dominating, data centers and power are the limits, and everyone is…”
Naval Ravikant · 08:00

#ai#market structure#open source#agi

Explainer10:30

Why drone defense is structurally harder than drone attack

Naval thinks drones remain under-leveraged even after their battlefield prominence, and that nothing close to the end game has been seen. Defense is the hard part: an attacking drone has kinetic energy on its side coming down, plus surprise, and the attacker can mass drones at one point while the defender is spread thin. The defender's single structural advantage is short range.

Drones have come to prominence in recent fighting but are still under-leveraged.
Attacker advantages: kinetic energy from above, surprise, and the ability to concentrate force.
Defender disadvantage: always spread thin across the whole perimeter.
Defender advantage: much shorter distance to traverse than the incoming drone had to cover.

“I think drones are still under leveraged even though they've come to prominence in the battlefield recently.”
Naval Ravikant · 10:30

“The defender has one advantage which is short range.”
Naval Ravikant · 11:00

#drones#defense#warfare#hardware

Takeaway· 2

Takeaway15:30

Apple does two things well, and that is unusual

Naval uses Apple and Google to make a point about corporate competence. Most companies do one or two things well. Apple builds great hardware and great software, which is why its devices work, but it is not that good at cloud and AI. Google is strong at cloud and AI and at cloud software, weak at hardware and at consumer software.

Competence is narrow: one or two things well is the norm.
Apple's edge is the hardware plus software combination, not either alone.
Apple is explicitly weaker at cloud and AI.
Google's software strength is cloud software, not consumer software.

“You know, most companies do one or two things well. Apple does two things really well. They build great hardware, they build great software. They're…”
Naval Ravikant · 15:30

#big tech#apple#google#strategy

Takeaway17:30

You can see the jobs that die and not the ones that get born

Naval treats AI job loss as the clearest case of an imagination asymmetry. Listing which existing jobs disappear is easy; predicting the next job is very hard, yet the next job has always arrived. Nobody 200 years ago, when nearly everyone farmed, could have imagined 10% of today's jobs or the level of technological and economic advancement that followed.

Existing jobs are visible and enumerable; future jobs are not.
The historical record is that the next job inevitably appears.
Two hundred years ago almost everyone worked on a farm and could not have imagined today's work.
The asymmetry is about imagination, not about evidence.

“It's very easy to look at existing jobs and see how they will go away, but it's very hard to predict what the next job…”
Naval Ravikant · 17:30

“They couldn't have imagined 10% of the jobs that exist today cuz back then everybody was working on a farm.”
Naval Ravikant · 18:00

#ai#jobs#economics#forecasting

Frameworks in this episode

Commoditize Your Complement

Reports on Demand: AI as the Implicit Intranet

The Fully Interconnected Graph Organization

The Logic of Violence

Insights & moments

⚡Myth Buster· 1

A navigable generated world is not a world model

◆Hot Take· 5