Elon Musk：OpenAI 的背叛、他在 Tesla 的未来，以及下一个大机会——Grokipedia

Let’s get started. We wanted to try something new this week. Every week, I get a little upset. Things perturb me, Sacks, and when they do, I just yell and scream, “Disgrace!” So I bought the domain domain name.com for no reason other than my own amusement. But you know what? I’m not alone in my absolute disgust at what’s going on in the world. This week, we’re going to bring out a new feature here on the All-In podcast: Disgrace Corner.

This is fantastic. This is our new feature. Chamath, you look like you’re ready to go. Why don’t you tell everybody who gets your disgrace this week?

Wait, we all had to come with a disgrace?

You really missed the memo. All right, fine. Enough.

I got one. I got one.

Okay. All right. Just calm down.

My Disgrace Corner goes to Jason Calacanis and Pete Buttigieg, who, in the first 30 seconds of the interview, compared virtue-signaling points about how each one had worked at various moments at Amnesty International.

Literally affecting zero change, making no progress in the world, but collecting a badge that they used to hold over other people.

Absolutely.

We wrote a lot of letters. We wrote a lot of letters, which is good. That means it’s a good one, because behind the scenes, Jason and Pete Buttigieg: disgrace.

Great. I’m glad I get the first one, and you can imagine what’s coming next week for you.

I saw the Sydney Sweeney dress trending on social today. It’s too much.

What is it? I didn’t even know what this was.

You didn’t see it? Nice picture. Okay, bring it up.

It’s a little floppy.

How is this disgusting? What are you talking about?

It’s disgraceful. Look at this. Oh, my God. Too much. Too elegant.

Too much. In my day, Sacks, a little cleavage, maybe, perhaps in the 1990s or 2000s. Some side view. This is too much.

Very highbrow subject, madam. We were discussing our own politics and the Sydney Sweeney dress. I don’t know who’s trending.

We open-sourced it to the fans, and they’ve just gone crazy with what’s going on with the algorithm.

I’m getting Sydney Sweeney’s dress all day.

Well, maybe you should stop favoring it.

Fifteen times. And then Sacks—poor Sacks got invited to SlutCon for 2 weeks straight by the algorithm. What was your interaction? Did you bookmark SlutCon?

No, I say the algorithm has become—

Sacks will even tell you whether that’s a joke or a real thing.

It’s a real thing in San Francisco.

It’s all too real. It’s actually real. Yeah, for real.

But I’ve noticed that if you demonstrate interest in anything on X—if you click on it, God forbid, you like something—the algorithm is on it. It will give you more of that, a lot more.

Yes. We did have an issue—we still have somewhat of an issue—where there was an important bug that was identified and solved over the weekend, which caused in-network posts not to be shown. So, basically, if you followed someone, you wouldn’t see their posts.

Obviously, it was a big bug, a major bug. Then the algorithm was not properly taking into account if you just dwelt on something. But if you interacted with it, it would go hog-wild. So, as David said, if you were to favorite, reply, or engage with it in some way, it is going to get you a torrent of that same thing.

It was just going too far. Obviously, it was overcorrecting. It had too much gain on it; it turned up the gain way too high. Any interaction would then get you a torrent of that. It’s like, “Oh, you had a taste of it. We’re going to give you 3 helpings. We’re going to force-feed you. We’re going to give you the food funnel.”

And that’s all being done, I assume, with Grok now? So, it’s not like the old hard-coded algorithm, or is it using Grok?

What’s happening is that we’re gradually deleting the legacy Twitter heuristics. The problem is that, as you delete these heuristics, it turns out one heuristic, one bug, was covering for the other bug.

It’s like that meme with the internet, where there’s this very complicated machine and a tiny little wooden stick keeping it going, which was, I guess, AWS East or whatever had something like that. When somebody pulled out the little stick, everything stopped working.

I think it would be good if it showed 1 person you follow and then blended the old style, which was just reverse chronological order of your friends—the original version—with this new version. So you get a little bit of both.

Well, you can still—you still have the Following tab. Everyone still has the Following tab.

Now, something we’re going to be adding is the ability to have a curated Following tab. The problem is that if you follow some people and they’re a little more prolific than you, some people say a lot more than others. That makes the Following tab hard to use.

We’re going to add an option where you can have the Following tab be curated. Grok will say, “What are the most interesting things posted by your friends?” and then we’ll show you that in the Following tab. It will also show you everything, but I think having that option will make the Following tab much more useful.

It’ll be a curated list of people you follow—ideally, the most interesting things that they’ve said—which is what you would want to look at. We’ve mostly fixed the bug that would give you way too much of something if you interacted with a particular subject matter.

The really big change is that Grok literally reads everything that’s posted to the platform. There are about 100 million posts per day, so it’s 100 million pieces of content per day. I think that’s maybe just in English; it goes beyond that if it’s outside of English.

We’re going to start off by reading what Grok thinks are the top 10 million of the 100 million. We’ll actually read them, understand them, categorize them, and match them to users. This is not a job humans could ever do.

Once that’s scaling reasonably well, we’ll add the entire 100 million a day. It’s literally going to read through 100 million things and show you the things that it thinks are the most interesting posts to you.

How much of Colossus will that take? Is it tens of thousands of servers to do that every day?

My guess is it’s probably on the order of 50,000 H100s, something like that.

Wow. And that will replace search? You’ll be able to actually search on Twitter and find things with plain language?

We’ll have semantic search, where you can just ask a question, and it will show you all content—whether that’s text, pictures, or video—that matches your search query semantically.

How’s it been 3 years in? This was a 3-year anniversary like a couple years. Is this 3 years?

Yeah. Remember, it was Halloween.

Yeah. Halloween’s back.

Halloween’s back. We had a good time, yeah.

Wow. Yeah, 3 years. Where will things be 3 years from now? What’s the takeaway? 3 years later, you obviously don’t regret buying it. It saved free speech. That was good. You seemed to have turned that whole thing around, and that was, I think, a big part of your mission.

But then you added it to xAI, which makes it incredibly valuable as a data source. When you look back on it, the reason you bought it—to stop the crazy woke mind virus and make truth exist in the world again—great, mission accomplished. Now it has this great future.

Yeah, we’ve got Community Notes. You can also ask Grok about anything you see on the platform.

Just press the Grok icon on any X post, and we’ll analyze it for you and research it as much as you want. You can basically, just by tapping the Grok icon, assess whether that post is the truth, the whole truth, and nothing but the truth, or whether there’s something supplemental that needs to be explained. I think we’ve actually made a lot of progress toward freedom of speech and toward people being able to tell whether something is false or is false propaganda. The recent update to Grok is actually very good at piercing through propaganda.

We used that latest version of Grok to create Grokipedia, which I think is much more neutral and more accurate than Wikipedia, but actually has a lot more information than a Wikipedia page.

Did you seed it with Wikipedia? Actually, take a step back. How did you guys do this?

We used AI.

But meaning, like, totally unsupervised? Just a complete training run on its own? Totally synthetic data? No seeded set, nothing?

It was only recently possible for us to do this. We finished training a maximally truth-seeking version of Grok that is good at cogent analysis: breaking down any given argument into its axiomatic elements, assessing whether those axioms are likely to be true, whether they’re contradictory, and whether the conclusion naturally follows from those axioms.

We trained Grok on a lot of critical thinking, so it got really good at critical thinking, which was quite hard. Then we took that version of Grok and said, “Okay, cycle through the 1 million most popular articles in Wikipedia and add, modify, and delete.” That means researching the rest of the internet—whatever is publicly available—correcting the Wikipedia articles, fixing mistakes, and adding a lot more context.

Sometimes, the nature of propaganda is that facts are stated that are technically true but do not properly represent a picture of the individual.

This is critical because when you have a bio, as you do—and actually, we all do on Wikipedia—over time, it’s just the people you fired, beat in business, or who have an axe to grind. It slowly becomes the place where everybody who hates you puts their information.

I looked at mine, and it was so much more representative. It was 5 times longer, 6 times longer, and what it gave weight to was much more accurate. Much more accurate.

This opportunity was sitting here for a long time, and it’s great that you got to it because—

They don’t update my page much—maybe twice a month. And who is the secret cabal? There are 50 people who are anonymous who decide what gets put on it.

It was a much better, much more updated page in version 1.

Yes, this is version 0.1, as we put it—as we show at the top. By the time we get to version 1.0, I think it’ll be 10 times better. But even at this early stage, as you just mentioned, it’s not just correcting errors; it’s creating a more accurate, realistic, and fleshed-out description of people and events.

You think about subject matter, too? You can look at articles on physics in Grokipedia, and they’re much better than Wikipedia by far. This is what I was going to ask you: Do you think you can take this corpus of pages and get Google to deboost Wikipedia or boost Grokipedia in traditional search? A lot of people still find Wikipedia, and they believe it’s authoritative because it comes up number 1.

It really can. If Grokipedia is used elsewhere—if people cite it on their websites, post about it on social media, or, when they do a search and Grokipedia shows up, click on Grokipedia—it will naturally rise in Google’s rankings.

I did text Sundar because even a day after launch, if you typed in Grokipedia, Google would just say, “Did you mean Wikipedia?”

Wikipedia. Yeah.

And it wouldn’t bring it up at all.

Yeah, it’s true. So now, how’s the usage been? Have you seen good growth since it launched?

Yeah. It went super viral. We’re seeing it spread all over the place. I think we’ll see it used more and more as people refer to it.

People will judge for themselves when they read a Grokipedia article about a subject or a person they know a lot about and see, “Wow, this is way better than Wikipedia. It’s more comprehensive. It’s way more accurate. It’s neutral instead of biased.” Then they’re going to send those links around and say, “This is actually the better source.”

Grokipedia will succeed very well because it is fundamentally a superior product to Wikipedia. It is a better source of information.

And we haven’t even added images and video yet.

Yeah, we’re going to add a lot of video using Grok Imagine to create videos. If you’re trying to explain something, Grok Imagine can take the text from Grokipedia and generate an explanatory video.

If you’re trying to understand anything from how to tie a bow tie to how certain chemical reactions work—or really anything, including dietary or medical topics—you can just go and see the video of how it works, created by—

When you have this version that’s maximally truth-seeking as a model, do you think there needs to be a better eval or benchmark that people can point to, showing how far from the truth things are? If you’re going to start a training run with Common Crawl, or if you’re going to use Reddit, is it important to be able to say, “Hey, hold on a second. This eval just sucks. You guys suck on this eval. This is crappy data”?

Yeah, I guess I’m not—I think there are a lot of evals out there. I have complete confidence that Grokipedia is going to succeed because Wikipedia is actually not a very good product.

Yeah.

The information is sparse, wrong, and out of date. If you have something that is accurate and comprehensive, has videos, and where, moreover, you can ask about any part of it that you’re curious about—you can just highlight it and ask Grok right there—it’s great for learning something. It’s not going to be a little bit better than Wikipedia. It’s going to be 100 times better than Wikipedia.

Elon, do you think you’ll see good, uniform usage? If you look back on the last 3 years since you bought Twitter, there were a lot of people after you bought Twitter who said, “I’m leaving Twitter. Elon bought it. I’m going to go to this other—wherever the hell they went—and there are all these new creatures.”

Yeah, but Bluesky failing is my favorite.

I guess my question is: As you destroy the woke mind virus’ control of the system and bring truth to the system, whether through Grokipedia or through X, do people just look for confirmation bias and not actually accept the truth? What do you think? Or do you think people are actually going to see the truth and change?

Yeah, but I mean, is that like—

You thought Sydney Sweeney’s boobs were great? Let me see mine.

Looking good. Solid week there. Put a little something sheer, you know. I think we just got flagged on YouTube again.

Yeah, we did. That was definitely going to give us a censorship moment. Grade-A moves.

No, but do people change their minds? Can you actually change people’s minds by putting the truth in front of them, or do people just ignore the truth because they feel like they’re in some sort of camp? They want confirmation bias. They want to stay in a camp and be tribal about everything.

David, you were trying to ask a serious question. Go ahead.

I just want to know if people change their minds. Can you actually change people’s minds by putting the truth in front of them, or do people just ignore the truth because they feel like they’re in some sort of camp? They want confirmation bias, and they want to stay in a camp and be tribal about everything.

It is remarkable how much people believe things simply because they’re the beliefs of their in-group—whatever their political or ideological tribe is.

There are some pretty hilarious videos of some guy going around calling someone a racist Nazi or whatever, and then trying to show them videos of the thing they’re talking about, where he is in fact condemning the Nazis in the strongest possible terms and condemning racism in the strongest possible terms. They literally don’t even want to watch them.

So, yeah, at least some people would prefer to stick to whatever their ideological views are, whatever their political tribal views are.

No matter what. The evidence could be staring them in the face, and they're just going to be a flat-earther. There is no evidence that you can show a flat-earther to convince them the world's round, because everything is just a lie: the world is flat.

I think the ability to hit Grok in a reply and ask it a question in the thread has really become a truth-seeking missile on the platform. When I put up metrics or something like that, I reply to myself and say, “Grok, is the information I just shared correct? Can you find any better information? Please tell me if my argument is correct or if I’m wrong.” Then it goes through, DMs Sacks, and Sacks gets in my replies and tries to correct me.

It does a really good job, and that combined with Community Notes means you’ve got 2 swings at bat: the community’s consensus view and then Grok coming in. I think it would be really interesting if Grok, on really powerful threads, did its own version of Community Notes and had it sitting there ahead of time. You could look at a thread and see something next to it, or maybe click on a specific statistic, and it would show you, “Here’s where that statistic is from.”

Pretty much every post on X, unless it’s advertising or something, has the Grok symbol on it.

Yeah.

You just tap that symbol, and you’re 1 tap away from a Grok analysis. Literally, you’re 1 tap away. We don’t want to clutter the interface by providing an explanation, but if you go on X right now, it’s 1 tap to get Grok’s analysis. Grok will research the X post and give you an accurate answer.

You can even ask it to do further research and further due diligence. You can go as far down the rabbit hole as you want. I do think this is consistent with wanting X to be the best source of truth on the planet by far, where you hear any and all points of view, but where those points of view are corrected by human editors with Community Notes.

The essence of Community Notes is that people who historically disagree agree that the Community Note was correct. All of the Community Notes code is open source, and the data is open source, so you can recreate any Community Note from scratch independently.

By and large, it’s worked very well.

Yeah.

I think we originally had the idea to have you back on the pod because it was the 3-year anniversary of the Twitter acquisition. I just wanted to reminisce a little bit.

I remember.

Where’s that sink? Well, yeah. Elon was staying at my house. We had talked the week before, and he told me the deal was going to close, so I was like, “Hey, do you need a place to stay?” He took me up on it.

The day before he went to the Twitter office, there was a request made to my staff: “Do you happen to have an extra sink?” They did not.

Who has an extra sink, really?

But they were able to locate one at a nearby hardware store. I think they paid extra to get it out of the window or something.

I think the store was confused because my security team was asking for any kind of sink. Normally people wouldn’t ask for any kind of sink because you need a sink that fits in your bathroom or connects to a certain kind of plumbing. They were asking what kind of faucets we wanted. “No, no, I just wanted a sink.”

And didn’t care what the sink connected to.

That was it. They were almost not letting us buy the sink because they thought maybe we’d buy the wrong sink. It’s just rare that somebody wants a sink for meme purposes.

One of my favorite memories was when Elon said, “Hey, swing by. Check it out.” I said, “Okay, I’ll come by.” I drove up there, and I was looking for where to park the car. I realized there were just parking spaces around the entire business building, and I was like, “Okay, this can’t be legal parking.” But I parked, and it was legal parking.

Yeah.

You’re in downtown San Francisco, so you might get your window broken, but—

Yeah.

I might not be there when I get back. We got in there, and the place was empty.

Yeah. It was seriously empty except for the cafeteria.

The Twitter headquarters was 2 buildings. One of the buildings was completely and utterly empty, and the other building had about 5% occupancy. At 5% occupancy, we all went to get something to eat, and we realized there were more people working in the cafeteria than at Twitter.

There were more people making the food than eating the food in this giant, really nice cafeteria. This is where we discovered that the actual price of lunch was $400.

The original price was $20, but at 5% occupancy, it went 20 times higher. They still kept making pretty much the same amount and charging the same amount, so effectively lunch was $400.

That was a great meeting.

Yes. And then there was that—where we had the initial meetings, trying to figure out what the heck was going on, in these buildings. There were 2 Twitter buildings, and 1 of them had literally no one in it. That’s where we had the initial meetings.

We tried drawing on the whiteboard, but the markers had gone dry. Nobody had used the whiteboard markers in 2 years.

So sad.

None of the markers worked. This was totally bizarre, but it was totally clean because the cleaning crew had come in and done their job and cleaned an already clean place for 2 or 3 years straight. Honestly, this is more crazy than any Mike Judge movie or Silicon Valley or anything like that.

I remember going into the men’s bathroom, and there was a table with—

Hygiene—

Menstrual hygiene products.

Yep.

Yeah, refreshed every week.

Tampons. A fresh box of tampons.

We were like, “But there’s literally no one in this building.”

No, they hadn’t turned off sending fresh tampons to the men’s bathroom in the empty building.

No.

So every week they would put a fresh box of tampons in an empty building for years. It must have been very confusing to the people who were being asked to do this because they were thinking, “Okay, I’ll throw them away.”

I guess they’re paying us, so we’ll just put tampons.

You seriously have to consider the string of possibilities necessary for anyone to possibly use that tampon in the men’s bathroom at the unoccupied second building of Twitter headquarters. You’d have to be a burglar, a trans man burglar, who was unwilling to use the women’s bathroom, which also had tampons.

Statistically, there’s no one in the building. So you’ve broken into the building, and at that moment you have a period.

Yes. You’re on your period. You’re more likely to be struck by a meteor than need that tampon.

I think it was shortly after that you discovered an entire room at the office filled with Stay Woke T-shirts. Do you remember this?

An entire pile of merch.

Yeah.

Yes.

#StayWoke.

Stay Woke. There were also a big sort of buttons, like magnetic buttons that you put on your shirt, that said, “I am an engineer.”

I’m like, “Look, if you’re an engineer, you don’t need a button.” Who’s the button for? Who are you telling that to? You can just ship code. We would know. We can check your GitHub.

There were scarves, hoodies, all kinds of merch that said #StayWoke.

Yeah, a couple mugs.

When you found that, I was like, “My God, man. The barbarians are fully within the gates now.”

The barbarians have smashed through the gates and are looting the merch.

Yes, you are rummaging through their holy relics and defiling them.

When you think about it, David, the amount of waste that we saw there during those first 30 days—

Just to be serious about it for a second, this was a publicly traded company. If you think about the financial duty of those individuals, there was a list of SaaS software we went through, and none of it was being used. Some of it had never been installed, and they had been paying for it for 2 years.

They had been paying for a SaaS product for 2 years. The one that blew my mind the most, which we canceled, was that they were paying a certain amount of money per desk to have desking software in an office where nobody came to work.

So they were paying—

Nobody.

There were millions of dollars being paid for.

For analysis of pedestrian traffic—software that uses cameras to analyze pedestrian traffic to figure out where you can alleviate pedestrian traffic jams—in an empty building.

Right.

That's like 11 out of 10 on a Dilbert scale.

Yeah. It was pretty—shout-out to Scott Adams.

You've gone off scale on your Dilbert level at that point. Let's talk about the free speech aspect for a second, because I think that is the most important legacy of the Twitter acquisition. I think people have short memories, and they forget how bad things were 3 years ago.

First of all, you had figures as diverse as President Trump, Jordan Peterson, Jay Bhattacharya, and Andrew Tate. They were all banned from Twitter. I remember when you opened up the Twitter jails and reinstated their accounts, freeing all the bad boys of free speech.

The best deal.

Yes. So you basically gave all the bad boys of free speech their accounts back. But beyond just the bannings, there were the shadowbannings. Twitter had claimed for years that they were not shadowbanning, that this was a paranoid conservative conspiracy theory.

Yeah. There was very aggressive shadowbanning by what was called the Trust and Safety group, which, of course, naturally would be the one that is doing the nefarious shadowbanning. I just think we shouldn't have a group called Trust and Safety. This is an Orwellian name if there ever was one.

Hi, I'm from the Trust Department.

Oh, really? We want to talk to you about your tweets. Can we see your DMs?

Say that you're from the Trust Department. It's literally the Ministry of Truth right there.

Yeah.

Executives had maintained for years that they were not engaged in this practice, including under oath. On the heels of you opening that up and exposing that—

Because, by the way, it wasn't just the fact that they were doing it. They created an elaborate set of tools to do this. They had checkbox sets of tools to deboost accounts.

Yes. Subsequently, we found out that other social networking properties have done this as well, but you really exposed it. This is still being done at the other social media companies, including Google, by the way. I don't want to pick on Google, because they're all doing it, but for search results, if you simply push a result pretty far down the page—or to the second page of results—the joke used to be, or still is, I think: Where do you hide a dead body? The second page of Google search results, because nobody ever goes to the second page of Google search results.

You could hide a dead body there and nobody would find it. You haven't made the result go away; you've just put it on page 2.

Shadowbanning, I think, was number 2. First was banning; second was shadowbanning. I think third, to me, was government collusion and government interference. You released the Twitter Files. Nothing like that had ever been done before, where you actually let investigative reporters go through Twitter's emails unfettered.

I wasn't looking over their shoulder at all. They just had direct access to everything.

They found that there was extensive collusion between the FBI and Twitter's Trust and Safety group. It turns out the FBI had 80 agents submitting takedown requests, and they were very involved in the banning, the shadowbanning, and the censorship. I don't think we ever had definitive evidence of that before. That was pretty extraordinary.

Yeah. The U.S. House of Representatives had hearings on the matter, and a lot of this was unearthed. It's public record. Some people on the left still think this is made up. I'm like, this is literally the Twitter Files. They're literally the files at Twitter. We're talking about emails that were sent internally that confirm this. This is what's on the Slack channels, and this is what is shown in the Twitter database as where people have made either suspensions or shadowbans.

Has the government come and asked you to take stuff down since, or do they just have to—the policy is, “Hey, listen, you've got to file a warrant. You've got to come correct,” as opposed to just putting pressure on executives?

Our policy at this point is to follow the law. The laws are obviously different in different countries. Sometimes I get criticized for, “Why don't you push free speech in XYZ country that doesn't have free speech laws?” I'm like, because that's not the law there. If we don't obey the law, we'll simply be blocked in that country.

The policy is really just to adhere to the laws in any given country. It is up to us to agree or disagree with those laws. If the people of that country want laws to be different, then they should ask their leaders to change the laws.

But as soon as you start going beyond the law, you're putting your thumb on the scale. I think the right policy is just to adhere to the laws within any given country.

Sometimes we get into a bit of a bind, like we got into with Brazil, where this judge in Brazil was asking us—or telling us—to break the law in Brazil and ban accounts contrary to the law of Brazil. Now we're somewhat stuck. We're like, wait a second, we're reading the law, and it says this is not allowed to happen. The judge was also giving us a gag order, so we weren't allowed to say it was happening.

We had to break the law, and the judge was telling us to break the law. The law is breaking the law. That's where things get very difficult. We were actually banned in Brazil for a while because of that.

I want to make one final point on the free speech issue, and then we can move on. I think people forget that the censorship wasn't just about COVID. There was a growing number of categories of thought and opinion that were being outlawed. The so-called content moderation, which is another Orwellian euphemism for censorship, was being applied to categories like gender and even climate change. The definition of hate speech was constantly growing.

Yes.

More and more people were being banned or shadowbanned, and there were more and more things that you couldn't say. This trend of censorship was growing. It was galloping, and it would have continued if it wasn't, I think, for the fact that you decided to buy Twitter and open it up. It was only on the heels of that that the other social networks were willing to be a little bit chastened in their policies and start to push back more.

Yeah, that's right. Once Twitter broke ranks, the others had to. It became very obvious what the others were doing, and so they had to mitigate their censorship substantially because of what Twitter did. Perhaps to give them some credit, they also felt that they had the air cover to be more inclined toward free speech.

They still do a lot of shadowbanning and whatnot at the other social media companies, but it's much less than it used to be.

Yeah. Elon, what have you seen in terms of governments creating new laws? We've seen a lot of this crackdown in the UK on what's being called hateful speech on social media, and folks getting arrested and actually going to prison over it.

It seems like when there's more freedom, the side that is threatened by that comes out and creates its own counter, right? There's a reaction to that. Are you seeing more of these laws around the world in response to your opening up free speech through Twitter and those changes, and what they're enabling? The governments and the parties that control those governments aren't aligned, and they're stepping in and saying, “Let's create new ways of maintaining our control through law.”

Yeah. There is an overall global movement to suppress free speech under the guise of suppressing hate speech. The problem with that is that your freedom of speech only matters if people are allowed to say things that you don't like, or even things that you hate.

If you're allowed to suppress speech that you don't like, then you don't have freedom of speech. It's only a matter of time before things switch around, the shoe is on the other foot, and they will suppress you. Suppress not, lest you be suppressed.

There is a very strong movement to codify speech suppression into law throughout the world, including the Western world—Europe and Australia.

The UK and Germany are very aggressive in this regard.

Yes. My understanding is that in the UK there are something like 2,000 or 3,000 people in prison for social media posts. In fact, there are so many people who were in prison for social media posts, and many of these things are like—you can't believe that someone would actually be put in prison for this.

They have, in a lot of cases, released people who have committed violent crimes in order to imprison people who have simply made posts on social media, which is deeply wrong.

Mhm. And it underscores why the founders of this country made the First Amendment the First Amendment: freedom of speech. Why did they do that? Because in the places they came from, there wasn't freedom of speech, and you could be imprisoned or killed for saying things.

Can I ask you a question just to move to a different topic? If you came and did this next week, we'd be past the Tesla board vote. We talked about it last week, and we talked about how crazy ISS and Glass Lewis are. We used this one insane example where Ira Ehrenpreis didn't get the recommendation from ISS and Glass Lewis because he didn't meet the gender requirements, but then Kathleen Wilson-Thompson—

It also doesn't make sense.

Can you—so the board vote is on the African-American woman.

Yeah. True. They recommended against her, but then also recommended against Ira Ehrenpreis on the ground that he was insufficiently diverse.

Yeah. These things don't make any sense.

I do think we've got a fundamental issue with corporate governance in publicly traded companies, where about half of the stock market is controlled by passive index funds. Most of them outsource their decisions to advisory firms, particularly Glass Lewis and ISS. I call them corporate ISIS. All they do is basically—they're just terrorists—and they own no stock in any of these companies.

I think there's a fundamental breakdown of fiduciary responsibility here. Any company that's managing, even though it's passively managing index funds or whatever, does have a fiduciary duty at the end of the day to vote along the lines of what would maximize shareholder returns. People are counting on them. People have all their savings in a 401(k) or something like that, and they're counting on the index funds to vote in company votes in the direction that would ensure their retirement savings do as well as possible.

But the problem is that this is outsourced to ISS and Glass Lewis, which have been infiltrated by far-left activists. Where far-left political activists go, they go where the power is. Effectively, Glass Lewis and ISS control the vote of half the stock market. If you're a political activist, what a great place to go work would be Glass Lewis. And they do.

My concern for the future is that this Tesla thing is called compensation, but really, it's not about compensation. It's not like I'm going to go out and buy a yacht with it or something. It's that, in order to build up Optimus and have all these robots out there, I need to make sure we do not have a Terminator scenario and that I can maximize the safety of the robots.

I feel like I need to have something like a 25% vote, which is enough of a vote to have a strong influence, but not so much of a vote that I can't be fired if I go insane. My concern would be creating this army of robots and then being fired for political reasons because ISS and Glass Lewis—or the activists at those firms—effectively fire me, even though I've done everything right. That's my concern. Then I cannot ensure the safety of the robots.

If you don't get that vote, if it doesn't go your way—it looks like it's going to—would you leave? Is that even in the cards? I heard the board was very concerned about that.

Let's just say I'm not going to build a robot army if I can be easily kicked out by activist investors.

Yeah. No way. No way. It makes sense. Who is capable of running the 4 or 5 major product lines at Tesla? This is the madness of it. It's a very complex business. People don't understand what's under the hood there. It's not just a car company: you've got batteries, you've got trucks, you've got the self-driving group. This is a very complex business that you've built over decades now. It's not a very simple thing to run, and I don't think there's an Elon equivalent out there who can just jump into the cockpit.

If we take a full turn around the corporate governance corner, also this week, what was interesting about the OpenAI restructuring was that I read the letter, and your lawsuit was excluded from the analysis by the California attorney general, basically saying this thing can go through. That means your lawsuit is still out there, right? I think it's going to go to a jury trial.

Yes.

So that corporate governance issue is still very much in question. Do you have any thoughts on that?

I believe that will go to a jury trial in February or March, and then we'll see what the results are there. But there's a mountain of evidence that shows that OpenAI was created as an open-source nonprofit. That's literally the exact description in the incorporation documents. In fact, the incorporation documents explicitly say that no officer or founding member will benefit financially from OpenAI.

They've completely violated that. Moreover, you can just use the Wayback Machine and look at the website of OpenAI: “Open-source nonprofit,” “open-source nonprofit,” the whole way through. Then it looked like, wow, there's a lot of money to be gained here, and suddenly it starts changing. They try to change the definition of OpenAI to mean “open to everyone” instead of “open source,” even though it always meant open source.

I came up with the name.

Yeah, that's how I know. If they open-sourced it, or if they gave you the percentage ownership in it that you would rightfully be owed—which, for $50 million for a startup, would be half at least—they must have made an overture toward you and said, “Can we just give you 10% of this thing and get your blessing?” You obviously have a different goal here.

Essentially, since I came up with the idea for the company, named it, provided the A, B, and C rounds of funding, recruited the critical personnel, and told them everything I know, if that had been a commercial corporation, I'd probably own half the company.

I could have chosen to do that. It was totally at my discretion. But I created it as a nonprofit for the world—an open-source nonprofit for the world.

Do you think the right thing to do is to take those models and just open-source them today? If you could affect that change, is that the right thing to do?

Yeah, I think that is what it was created to do, so it should. The best open-source models right now are generally from China, which is bizarre. Fate seems to be an irony maximizer.

I think the second best are—one is maybe better than second best—the Grok 2.5 open-source model is actually very good. We'll continue to open-source our models. But try using any of the recent so-called OpenAI open-source models. They're out; they don't work. They basically open-sourced a broken, nonworking version of their models as a fig leaf. Do you know anyone who's running OpenAI's open-source models?

Exactly. Yeah. Nobody. We've had a big debate about jobs here. Obviously, there's going to be job displacement. You and I have talked about it for decades. What's your take on the pace of it? You're building self-driving software, and you're building Optimus.

We're seeing Amazon take some steps here where they're saying, “We're probably not going to hire for these positions in the future.” Maybe they're getting rid of people now because they were bloated, but maybe some of it is AI. It's all debatable. What do you think the timeline is, and what do you think we're going to need to do as a society to mitigate it if it goes too fast?

I call AI the supersonic tsunami—not the most comforting description in the world.

Fast and big.

There is a tsunami, a giant wall of water, moving faster than the speed of sound. That's AI.

When does it land?

Yeah, exactly. This is happening whether I want it to or not. I actually tried to slow down AI.

The reason I wanted to create OpenAI was to serve as a counterweight to Google, because at the time Google essentially had unilateral power in AI. They had all the AI, essentially. Larry Page was not taking safety seriously.

Jason, were you there when he called me a speciesist?

Yes, I was there. You were more concerned about the human race than you were about the machines.

And, yeah, you had a clear bias for humanity.

Yes. Yes, exactly. I was like, “Larry, we need to make sure that the AI doesn’t destroy all the humans.” And then he called me a speciesist—like a racist or something—for being pro-human intelligence instead of machine intelligence. I’m like, “Well, Larry, what side are you on?” I mean, that’s kind of a concern. And at the time, Google had essentially a monopoly on AI.

Yeah. They bought DeepMind, which you were on the board of and had an investment in. Larry and Sergey had invested in it as well, and it’s really interesting.

I found out about it because I told him about it. I showed him some stuff from DeepMind, and I think that’s how he found out about it and acquired them, actually. I have to be careful what I say.

But the point is that Larry’s not taking AI safety seriously, and Google had essentially all the AI, all the computers, and all the money. I’m like, “This is a unipolar world where the guy in charge is not taking things seriously.” And he called me a speciesist for being pro-human. What do you do in those circumstances?

Yeah.

Build a competitor.

Yes.

So OpenAI was created essentially as the opposite: an open-source nonprofit, the opposite of Google. Now, unfortunately, it needs to change its name to Closed for Maximum Profit AI.

Yeah.

Or Maximum Profit, to be clear—the most amount of the company, the most amount of profit you possibly get.

I mean, it’s— it’s comical when you hear—

Fate is an irony maximizer.

You have to say, what is the most ironic outcome for a company that was created to do open-source, nonprofit AI? It’s super closed-source. It’s tighter than Fort Knox. The OpenAI source code is locked up tight in Fort Knox. And they’re going for maximum profit. Get the bourbon and the steak knife—they’re ready.

Yeah. I mean, they’re going for the buffet and just diving headfirst into the profit buffet. Or, at least, the revenue buffet. Profit, we’ll see.

I mean, it’s like ravenous wolves for revenue.

Ravenous revenue buffet.

No, no, it’s literally like—

Supervillain.

It’s like Bond-villain-level flip. It went from being the United Nations to being SPECTRE in James Bond land.

When you hear Sam say, “I’m going to raise $1.4 trillion to build up data centers”—what do you think?

Yeah. No, but I think he means it.

Yeah. I mean, it’s audacious, but I wouldn’t want to insult the word. Actually, I have a question about this. How is that possible? In the earnings call, you said something that was insane, and then I think the math actually nets out. You said we could connect all the Teslas and, during downtime, allow them to offer up inference. You can string them all together. I think the math is that it could actually be 100 gigawatts. Is that right?

If ultimately there’s a Tesla fleet that is 100 million vehicles—which I think we probably will get to at some point—and they have mostly state-of-the-art inference computers in them, each at, say, 1 kilowatt of inference compute, and they have built-in power and cooling and connect to Wi-Fi—

That’s the key. Yeah, exactly.

Yeah, exactly. And that’d give you 100 gigawatts of inference compute.

Elon, do you think that architecture—there was an attention-free model that came out last week. There have been all of these papers, all of these new models that have been shown to reduce power per token of output by many, many, many orders of magnitude. Not just an order of magnitude, but maybe 3 or 4. What’s your view, given all the work you’ve been doing, on where we’re headed in terms of power per unit of compute or per token of output?

Well, we have a clear example of efficient, power-efficient compute, which is the human brain. Our brains use about 20 watts of power, but of that, only about 10 watts is higher-brain function. Most of it is housekeeping functions—keeping your heart going, breathing, and that kind of thing. So you’ve got maybe 10 watts of higher-brain function in a human.

We’ve managed to build civilization with 10 watts of a biological computer. And that biological computer has, like, a 20-year boot sequence, but it’s very power-efficient.

Given that humans are capable of inventing general relativity and quantum mechanics, inventing aircraft, lasers, and the internet, and discovering physics with a 10-watt meat computer, essentially, there’s clearly a massive opportunity for improving the efficiency of AI compute. It’s currently many orders of magnitude away from that.

It’s still the case that a 100-megawatt or even a gigawatt AI supercomputer at this point can’t do everything that a human can do. It will be able to, but it can’t yet.

We’ve got this obvious case of human brains being very power-efficient and achieving and building civilization with 10 watts of compute. Our bandwidth is very low. The speed at which we communicate information to each other is extremely low. We’re not communicating at a terabyte; we’re communicating at, like, 10 bits per second.

That should naturally lead you to the conclusion that there’s a massive opportunity to be more power-efficient with AI. At Tesla and at xAI, we continue to see massive improvements in inference-compute efficiency.

Do you think there’s a moment when you would justify stopping all the traditional cars and going completely all-in on Cybercab if you felt the learning was good enough and the system was safe enough? Is there ever a moment like that, or do you think you’ll always dual-track and always do both?

I mean, all of the cars we make right now are capable of being a robotaxi. There’s a little confusion in the terminology because our cars look normal. The Model 3 or Model Y is a good-looking car, but it looks normal.

It has an advanced AI computer, advanced AI software, and cameras. We didn’t want the cameras to stick out, so we put them in unobtrusive locations. The forward-looking cameras are in front of the rearview mirror. The side-view cameras are on the side repeaters. The rear camera is just above the license plate, typically where the rearview camera is in a car. The diagonal-forward ones are in the B-pillars.

If you look closely, you can see all the cameras, but you have to look closely. We just didn’t want them to stick out like warts or something.

All of the cars we make are hyperintelligent and have the cameras in the right places. They just look normal. All of the cars we make are capable of unsupervised, full autonomy.

We have a dedicated product, which is the Cybercab, that has no steering wheel or pedals, which are obviously vestigial in an autonomous world. We start production of the Cybercab in Q2 next year, and we’ll scale that up to quite high volume. I think ultimately we’ll make millions of Cybercabs per year.

But it’s important to emphasize that all of our cars are capable of being robotic taxis.

The Cybercab is gorgeous. I told you I’d buy 2 of those if you put a steering wheel in them. And there’s a big movement online—

Putting a steering wheel in.

People are begging for it. Why not? Why not let us buy a couple? Just the first ones off the line and drive them. They look great. It’s like the perfect model. You always had a vision for a Model 2, right? Isn’t it the perfect Model 2, in addition to being a Cybercab?

Look, the reality is people may think they want to drive their car, but the reality is that they don’t. How many times have you been, say, in an Uber or Lyft and said, “You know what? I wish I could take over from the driver. I wish I could get off my phone and take over from the Uber driver and drive to my destination”? How many times have you thought that to yourself?

No, it’s quite the opposite.

Zero times. Okay.

I have the Model Y, and I just got FSD 14.1. I have Juniper, and I got the 14.1, and I put it on Mad Max mode the last couple of days. That is—

Mad Max.

A unique experience. I was like, “Wait a second. This thing is driving in a very unique fashion.”

Yeah.

It assumes you want to get to your destination in a hurry.

Yeah. I used to give drivers an extra $20 to do that—

For a medical appointment or something. I don’t know.

Yeah, but it feels like it’s getting very close. You have to be very careful.

You know, Uber had a horrible accident with the safety driver. Cruise had a terrible accident. It wasn't exactly their fault, except that somebody got hit, and then they hit the person a second time and dragged them.

Yeah. Yeah.

This is pretty high stakes, so you're being extremely cautious. The car is actually extremely capable right now.

But we are being extremely cautious and paranoid about it because, to your point, even one accident would be headline news.

Well, probably worldwide headline news, especially if it's a Tesla. Waymo, I think, gets a bit of a pass. I think half the country—or a number of people—would probably go extra hard on you.

Yes.

Yeah, exactly.

Not everyone in the press is my friend.

Hadn't noticed.

Yeah. Some of them are a little antagonistic.

People are pressuring you to go fast, and I think everybody's got to take their time with this thing. It's obviously going to happen, but I get very nervous that the pressure to put these things on the road faster than they're ready is a little crazy. I applaud you for putting the safety monitor in and having the safety driver. No shame in the safety driver game.

It's obviously the right decision, but people are criticizing you for it. I think it's dumb. It's the right thing to do.

Yes. We do expect the cars to not have any sort of safety occupant—or, really, a driver who just sits in the car and doesn't do anything.

Monitor.

Safety monitor. He just sits in the car and doesn't do anything.

Safety dude.

Yeah. We do expect that the cars will be driving around without any safety monitor before the end of the year, sometime in December.

In Austin, yeah. I mean, you've got a number of reps under your belt in Austin, and it feels like you guys have done a pretty good job figuring out where the trouble spots are. Maybe you could talk a little bit about what you learned in the first three or four months of the Austin experiment.

Actually, it's gone pretty smoothly. A lot of things we're learning are just how to manage a fleet, because you've got to write all the fleet-management software, right?

Yeah.

You've got to write the ride-hailing software. You've got to write basically the software that Uber has. It's just summoning a robot car instead of a car with a driver.

A lot of what we're doing is scaling up the number of cars to see what happens if you have 1,000 cars. We think we'll probably have 1,000 cars or more in the Bay Area by the end of this year, and probably 500 or more in the greater Austin area.

You have to make sure the cars don't all, for example, go to the same Supercharger at the same time.

Right.

Or don't all go to the same intersection. There's the question of what these cars do. Sometimes there's high demand, and sometimes there's low demand. What do you do during those times? Do you have the car circle the block? Do you have it try to find a parking space?

Sometimes, say it's a disabled parking space, but the writing or the markings have faded. The car's like, "Oh, look, a parking space. We'll jump right in there."

Yeah, get a ticket.

You've got to look carefully and make sure it's not an illegal parking space. Or it sees a space to park and it's ridiculously tight, but it's like, "I can get in there."

But with 3 inches on either side.

Bad computer. But nobody else will be able to get in the car if you do that.

There are just all these oddball corner cases.

Regulators are all very—yeah, they have different levels of pickiness and regulations depending on the city and the airport. It's just very different everywhere. That's going to be a lot of blocking and tackling, and it just takes time. Elon, let me ask you another—

In order to take people to San Jose Airport, you actually have to connect to San Jose Airport's servers. You have to pay a fee every time you drop someone off.

So the car actually has to do a remote call. The robot car has to do a remote procedure call to San Jose Airport's servers—

To say, "I'm dropping someone off at the airport. Charge me whatever, 5 bucks." There are all these quirky things like that. Airports are somewhat of a racket.

Yeah.

We had to solve that thing. It's kind of funny: the robot car is calling the airport server to charge its credit card or whatever.

To send a fax. "Yeah, we're going to be dropping off at this time." But it will soon become extremely normal to see cars going around with no one in them.

Yeah.

Yeah. Elon, just before we lose you, I want to ask if you saw the Bill Gates memo that he put out. A lot of people are talking about this memo. I guess Billy G is not my lover.

Oh man. Did climate change become woke? Did it become woke, and is it over being woke? What happened with Billy G?

That's a lot. Great question. Great question.

Yeah. You'd think that someone like Bill Gates, who clearly started a technology company that's one of the biggest companies in the world, Microsoft, would be really quite strong in the sciences. But actually, at least in my direct conversations with him, he's not strong in the sciences.

This is really surprising. He came to visit me at the Tesla Gigafactory in Austin and was telling me that it's impossible to have a long-range semi. I was like, "But we literally have them. Pepsi is literally using them right now."

You can drive it yourself or send someone. Obviously, Bill Gates can drive it himself, but you can send a trusted person to drive the truck and verify that it can do the things that we say it's doing. He's like, "No, no, it doesn't work. It doesn't work." And I'm kind of stuck there.

I was like, "Well, it must be that you disagree with the watt-hours per kilogram of the battery pack. You must think that perhaps we can't achieve the energy density of the battery pack, or that the watt-hours per mile of the truck chassis is too high, because when you combine those 2 numbers, the range is low. Which one of those numbers do you think we have wrong, and what numbers do you think are correct?"

He didn't know any of the numbers. I'm like, "Well, then doesn't it seem premature to conclude that a long-range semi cannot work if you do not know the energy density of the battery pack or the energy efficiency of the truck chassis?"

But yeah, he's now taken a 180 on climate. He's saying maybe this shouldn't be the top priority.

Climate is gay. It's just the climate is gay. That's wrong.

It's totally—

Bill Gates said the climate is gay? Come on.

Maybe he's got some data he's got to put up. Does he have to stand up a data center for Sam Altman or something? I don't know. What is Azure?

I don't know.

He changed his position. I can't figure out why.

The reality of the whole climate-change thing is that you've just had people who say it doesn't exist at all, and then people who say it's super alarmist, saying Los Angeles is going to be underwater in 5 years. Obviously, neither of those 2 positions is true.

You can measure the carbon concentration in the atmosphere. You could literally buy a CO2 monitor from Amazon. It's like $50, and you can measure it yourself. You can say, "Okay, the parts per million of CO2 in the atmosphere has been increasing steadily at 2 to 3 per year."

At some point, if you continue to take billions, eventually trillions, of tons of carbon from deep underground and transfer it to the atmosphere and oceans, you're transferring it from deep underground into the surface cycle. You will change the chemical constituency of the atmosphere and oceans. You literally will.

Then you can argue to what degree and over what time scale. The reality, in my opinion, is that we've got at least 50 years before it's a serious issue. I don't think we've got 500 years, but we've probably got 50. It's not 5 years.

If you're trying to get to the right order of magnitude of accuracy, I'd say the concern level for climate change is on the order of 50 years. It's definitely not 5, and I think it probably isn't 500.

The right course of action is actually just the reasonable course of action, which is to lean in the direction of sustainable energy, lean in the direction of solar and a solar-battery future, and generally have the rules of the system lean in that direction.

I don't think we need massive subsidies, but we also shouldn't have massive subsidies for the oil and gas industry.

Okay. The oil and gas industry has massive tax write-offs that they don't even think of as subsidies because these things have been in place for, in some cases, 80 years. But they're not there for other industries. When you've got special tax conditions in one industry and not another industry, I call that a subsidy. Obviously, it is. But they've taken it for granted for so long in oil and gas that they don't think of it as a subsidy.

The right course of action, in my opinion, is to remove subsidies from all industries. But the political reality is that the oil and gas industry is very strong in the Republican Party but not in the Democratic Party. So, you will not see even the tiniest subsidy being removed from the oil, gas, and coal industry. In fact, there were some that were added to the oil, gas, and coal industry in the big bill, and a massive number of sustainable-energy incentives were removed, some of which I agreed with, by the way. Some of the incentives have gone too far.

The correct scientific conclusion, in my opinion—and I think one can back this up with solid reasoning; ask Grok, for example—is that we should lean in the direction of moving toward a sustainable-energy future. We will eventually run out of oil, gas, and coal to burn anyway because it's finite. There's a finite amount of that stuff, and we will eventually have to go to something that lasts a long time, that is sustainable.

But to your point about the irony of things, it seems to be the case that making energy with solar is cheaper than making energy with some of these carbon-based sources today. The irony is that it's already working. The market is moving in that direction. This notion that we need to force everyone into a model of behavior is just naturally going to change because we've got better systems.

You and others have engineered better systems that make these alternatives cheaper, and therefore they're winning. They're actually winning in the market, which is great. But they can't win if there are subsidies to support the old systems, obviously.

Yeah. By the way, there are actually massive disincentives for solar because China is a massive producer of solar panels. China does an incredible job of solar-panel manufacturing—really incredible. They have roughly 1.5 terawatts of solar production right now, and they're only using 1 terawatt per year. By the way, that's a gigantic number. The average U.S. power consumption is only 0.5 terawatts.

Just think about that for a second: China's solar-panel output, maximum capacity, is 1.5 terawatts per year. U.S. steady-state power usage is 0.5 terawatts. Now, you do have to reduce that. If you produce 1.5 terawatts a year of solar, you need to add batteries to account for the differences between night and day, the fact that the solar panel is not always pointed directly at the sun, that kind of thing. So you can divide by 5-ish, but that still means that China has the ability to produce solar panels with a steady-state output roughly two-thirds that of the entire U.S. economy from all sources.

That means that, with solar alone, China can, in 18 months, produce enough solar panels to power the entire United States—all the electricity of the United States.

What do you think about near-field solar, a.k.a. nuclear?

I'm in favor of—look, make energy any way you want that isn't obviously harmful to the environment. Generally, people don't welcome a nuclear reactor in their backyard. They're not championing, “Put it here. Put it under my bed. Put it on my roof.” If your next-door neighbor said, “Hey, I'm selling my house, and they're putting a reactor there,” the typical homeowner response would be negative. Very few people will embrace a nuclear reactor adjacent to their house.

Nonetheless, I do think nuclear is actually very safe. There's a lot of scaremongering and propaganda around fission, if we're talking about fission. Fission is actually very safe. The U.S. Navy obviously has this on submarines and aircraft carriers, with people really working right next to it. A submarine is a pretty crowded place, and they have nuclear-powered submarines. So I think fission is fine as an option. The regulatory environment makes it very difficult to get that done.

It's also important to appreciate the sheer magnitude of the power of the sun. Here are some important basic facts. Even Wikipedia has these facts right.

Yeah. Even Wikipedia got it right.

Yes. I'm saying even Wikipedia's got these facts right. The sun is about 99.8% of the mass of the solar system. Jupiter is about 0.1%, and everything else is in the remaining 0.1%. We are much less than 0.1%.

If you burned all of the mass of the solar system, the total energy produced by the sun would still round up to 100%. If you just burned Earth—the whole planet—and burned Jupiter, which is very big and quite challenging to burn, and turned Jupiter into a thermonuclear reactor, it wouldn't matter. Compared to the sun, the sun is 99.8% of the mass of the solar system, and everything else is in the miscellaneous category.

Basically, no matter what you do, the total energy produced in our solar system rounds up to 100% from the sun. You could even throw another Jupiter in there. Somehow, if you could teleport 2 more Jupiters into our solar system and burn them, the sun would still round up to 100%. As long as you're at 99.6%, you're still rounding up to 100%.

Maybe that gives some perspective on why solar is really the thing that matters. As soon as you start thinking about things at a grander scale, like the scale of civilizations, it becomes very obvious. I'm not saying anything that's new, by the way. Anyone who studies physics has known this for a very long time.

In fact, I think it was a Russian physicist who came up with this idea in the '60s, just as a way to classify civilizations: Kardashev scale 1 would be that you've harnessed most of the energy of the planet. Kardashev scale 2 would be that you've harnessed most of the energy of your sun. Kardashev scale 3 would be that you've harnessed most of the energy of the galaxy.

We're only about 1%, or a few percent, of Kardashev scale 1 right now, optimistically. But as soon as you go to Kardashev scale 2, where you're talking about the power of the sun, you're really just saying everything is solar power and the rest is noise. The sun produces about a billion times—or call it well over a billion times—more energy than everything on Earth combined.

It's crazy. It's mind-blowing, right?

Yeah. Solar is the obvious solution to all this.

In the short term, we have to use some of these other sources. But hey, there it is. An hour and a half. Star-powered. Maybe we've got a branding issue here.

Yeah. Star-powered.

Instead of solar-powered, it's starlight.

Yeah. Starlight.

It's the power of a blazing sun. How much energy does an entire star have?

More than enough. All right.

That's for sure.

Also, you really need to keep the power local. Sometimes people—honestly, I've had these discussions so many times—say, “Would you beam the power back to Earth?” I'm like, “Do you want to melt Earth?” Because you would melt Earth if you did that. We'd be vaporized in an instant.

You really need to keep the power local—basically, distributed power. I guess most of it we use for intelligence. The future is a whole bunch of solar-powered AI satellites.

But the only thing that makes the star work is that it happens to have a lot of mass. So it has that gravity to ignite the fusion reaction, right? But we could ignite the fusion reaction on Earth now. I don't know if your view has changed. I think we talked about this a couple of years ago, where you were like, “We don't know if or when fusion becomes real here.”

I want to be clear about my opinion on fusion. I studied physics in college. At one point in high school, I was thinking about a career in physics. One of my sons is actually pursuing a career in physics, but my problem is that I came to the conclusion that I'd be waiting for a collider or a telescope. I do have an actual degree in physics, but I have a strong interest in the subject.

My opinion on creating a fusion reactor on Earth is that this is not a hard problem. Actually, it's a little hard. It's not totally trivial, but if you just scale up a tokamak, the bigger you make it, the easier the problem gets. You've got a surface-to-volume-ratio issue where you're trying to maintain a really hot core while having a wall that doesn't melt.

There's a similar problem with rocket engines. You've got a superhot core in the rocket engine, but you don't want the chamber walls to melt. So you have a temperature gradient where it's very hot in the middle, and it gradually gets cold enough as you get to the perimeter—the chamber walls—so it doesn't melt.

If you just scale up the donut reactor, the tokamak, and improve your surface-to-volume ratio, that becomes much easier. In my opinion, anyone who looks at the math can see that you can make a reactor that generates more energy than it consumes, and the bigger you make it, the easier it is.

In the limit, you just have a giant, gravitationally contained thermonuclear reactor like the Sun, which requires no maintenance and is free. So why would we bother making a little, itty-bitty sun that's so microscopic you'd barely notice it on Earth when we've got the giant, free one in the sky?

Yeah, but we only get a fraction of 1% of that energy on planet Earth. We have to—

Much less than 1%.

Yeah. Right, so we've got to figure out how to wrap the Sun if we're going to harness that energy. That's our longer—

If people want to have fun with reactors, that's fine. Have fun with reactors. But it's not a serious endeavor compared to the Sun. It's a fun science project to make a thermonuclear reactor, but it's just peanuts compared to the Sun.

Even the solar energy that does reach Earth is 1 gigawatt per square kilometer, or roughly 2.5 gigawatts per square mile. That's a lot. Commercially available panels are around 25%, almost 26%, efficiency. If you pack them densely, with an 80% packing density—which I think you could get in a lot of places—you effectively have about 200 megawatts per square kilometer.

You need to pair that with batteries so you have continuous power, although our power usage drops considerably at night, so you need fewer batteries than you think.

A rough number that's easy to remember is 1 gigawatt-hour per square kilometer per day. That's a roughly correct number.

But then doesn't your technical challenge become the scalability of manufacturing those systems? You have to access the raw materials and get them out of the ground on planet Earth to make enough of them to reach the scale and volume you're talking about. Do we have the ability to do that with what we have today? Can we pull that much material out of the ground?

Yes. Solar panels are made of silicon, which is essentially sand.

I guess I mean more on the battery side.

On the battery side, I'd like to throw out some interesting factoids. If most people don't know, what is Earth made of, as measured by mass? Actually, it's iron.

Iron, yeah. We're, I think, 32% iron, 30% oxygen, and then everything else is in the remaining percentage.

We're basically a rusty ball bearing. That's Earth, with a lot of silicon at the surface in the form of sand.

Iron-phosphate lithium-ion cells use iron, which is extremely common—the most common element on Earth, even in the crust. Phosphorus is also very common. The anode is carbon, which is also very common, and lithium is very common as well.

You can do the math. In fact, we did the math and published it, but nobody looked at it. It's on the Tesla website. It shows that you can completely power Earth with solar panels and batteries, and there's no shortage of anything.

All right, so on that note—

Go get to work, Elon, and just power the Earth while you're getting implants into people's brains and satellites and other good-fun stuff. Good to see you, buddy.

Yeah, good to see you guys.

Thanks for stopping by. Thanks for doing this. You have the Zoom link, so stop by anytime.

Thank you for coming today, and thank you for liberating free speech three years ago.

Yeah.

That was a very important milestone.

I see you guys are all in different places. I guess this is a very virtual situation.

It's always been that way. I'm at the ranch.

Are you ever in the same room?

We try not to be, except when we do that summit. Otherwise, we're each in our own places.

Yeah, your summit is pretty fun.

We had a great time recounting SNL sketches that didn't make it.

Oh, God, there are so many good ones. We didn't even get to the Jeopardy! ones.

Yeah, yeah. Those are so offensive.

I think we skipped a few that would have dramatically increased our probability of being killed.

You can take this one out, boys. I love you all. I'm going to poker later.

Take care.

Bye-bye.