This article was completed before SpaceX released its Q3 2026 earnings report.
01 Where Grok Started
My day-to-day work doesn’t have much to do with images or video, so I’ve barely used Midjourney or Stable Diffusion. I’ve only tried models focused on video generation a few times, either. I wouldn’t say I know much about them.
I mostly use AI to get things done more efficiently, so most of the models I’ve used are general-purpose large language models. Among those, whenever image or video generation comes up, the first one that comes to mind is Grok.
One day, a question bumped into my head. My impression was that OpenAI’s and Anthropic’s models both started with chat. So why did Grok, a later arrival, seem to put its weight behind images and video right from the start?
I took a few guesses myself first. Maybe everyone was just at a different stage? Maybe, with everyone working on text, starting with images and video offered a slightly different way in? Maybe images and videos were more likely to win the algorithm’s favor on X: generate something, post it straight to X, and more people would see it, and perhaps learn that Grok had made it… I also thought of that far-off dream of Musk’s: putting computing power in space, and sending people up there to live.
From images to video, and then to embodied intelligence in the physical world, I felt there had to be some connection between these things—but I couldn’t quite explain what it was.
I had no idea which of my guesses were right or wrong, either. Still, with that vague feeling that “there must be something behind all this,” off I went to ask Grok.
When I looked things up later, I realized my impression needed adjusting. Grok also launched as a chat model in November 2023; xAI’s model card said it could help write code. Image generation wasn’t added until 2024, and its own Aurora model only came out at the end of that year. OpenAI, meanwhile, had already brought image generation into ChatGPT in October 2023.
So Grok had started with chat, just like the others. It’s just that, over the past couple of years, it has made image and video generation especially visible on X. Visible enough that even someone who barely uses these tools thinks of it first.
02 Grok’s First Answer
The first thing in Grok’s answer that struck me was something Musk wrote on X this March. He said the future of AI was mostly about video understanding and generation, because photons were the highest-bandwidth form of communication, and these were necessary tools on the way to AGI. In the same post, he also said Imagine already had a positive gross margin and wasn’t a money-losing product. That was his own claim; outsiders still couldn’t see a separate profit-and-loss statement for the product. The documents made public when SpaceX went public this year only disclosed figures for the AI division as a whole, which was still losing a great deal of money in 2025. One product making money and the whole division losing money aren’t contradictory.
The words “highest bandwidth” sent me looking further: how much information can a person actually take in, and how much can we express?
A paper published in Neuron in 2025 happened to address both questions. The authors estimated that our sensory systems collect at least a billion bits of information per second, while the information we express through behavior amounts to only about ten bits per second. The gap is so enormous that, as the authors openly acknowledged, no one has been able to explain it. But it made something more tangible for me: the world we perceive is far larger than the part we can put into words.
At this point, that vague intuition of mine finally began to take shape.
Language is like the world after compression: one sentence can sum up an entire afternoon. Video is a little closer to the world before that compression, preserving how objects move and collide, and the order in which things happen… Then a more vivid way of putting it came to me: video and language are like two ends of the same pipeline. One end is closer to what we can see; the other, to what we can say. This is just a metaphor I’m using to understand the question, though. The research confirmed the gap between seeing and expressing, but still couldn’t explain why it exists. Some AI researchers have also proposed that, as visual and language models grow larger, their internal representations of the world become more alike, as if approaching the same world from opposite ends. That, too, is still a hypothesis.
Grok also mentioned Ethan He, who once led the development of Grok Imagine. In a later interview, he said a few engineers on the team had taken about three months to release their first video model. Discussing the technical approach, he explained why image models can usually lay the foundation for video models, and said he believed much of visual intelligence rests on language understanding. As for the specific methods xAI used internally, he said he couldn’t disclose them. Put back into my metaphor, it becomes quite easy to follow: learning only from video isn’t enough, and neither is learning only from language. Both ends need to be there, and both need to be learned from.
In a 2024 interview, Turing Award winner Yann LeCun did a similar calculation: over the sixteen thousand hours a four-year-old child has spent awake, roughly 10¹⁵ bytes of visual information have passed through their optic nerves. The text used to train a large language model amounts to around 2×10¹³ bytes, which would take a person a hundred and seventy thousand years to read. From this, he argued that text alone cannot bring AI to human-level intelligence. That is his own judgment, and some researchers disagree.
As I read Grok’s answer, I also reminded myself: this is xAI’s own AI explaining xAI’s choices. It said the progression from images to video was “almost inevitable.” That phrase… I still had some doubts. I wasn’t ready to put too much faith in it. Researchers have tested some video models and found that they fail in physical situations outside their training data. Coherent footage doesn’t necessarily mean a model understands physics. And while video generation can help models learn patterns in scenes and actions—a path toward “world models” in embodied intelligence—getting a machine to act reliably in the real world still requires perception, control, safety validation…
Afterward, I copied the same question into Claude and ChatGPT. They added perspectives on distribution, competition, and business models, but Grok’s first answer was still the one that gave me the most to think with. It helped me first connect things I had only vaguely felt might fit together.
All this material was public, scattered across posts, interviews, and news stories. Yet it wasn’t until I asked Grok and followed its answer further that I strung it together. The little feelers of my curiosity had finally reached (the edge of) a whole chain behind it all.
For a moment, I thought I was “so slow”—how had I only thought of this today! But thinking about it more carefully, reporting in different fields often stays focused on its own questions: business coverage watches subscriptions, robotics coverage watches fingers, regulatory coverage watches controversies… These scattered pieces are what I usually come across. Or I dive headfirst into something else that catches more of my attention.
03 Listen to What They Say, and Watch What They Do
Looking at their answers together reminded me of something I often keep in mind: don’t just listen to what someone says; watch what they do, too. The same goes for understanding a company or an investment firm. Look at where it puts its money, whom it acquires, and what it has in hand.
So let’s start with what he said. In the early hours of October 18, U.S. Central Time, Musk wrote four sentences in a reply to a user on X. Two of them were:
The only things that will scale to incredible size (K2 and beyond) economically are super intelligence, robots, solar power and mass to space.
And, of course, if you have the first two, the rest will follow.
K2 means a Kardashev Type II civilization, one capable of using the energy of an entire star. The second sentence almost answered that intuition I’d started with. I had vaguely sensed a connection between images, video, embodied intelligence, and computing in space. His own sequence was: first superintelligence and robots, then solar power and sending things into space would follow. Of course, that is what he said. We still need to look at what he did.
Next, what does he have in hand? When a latecomer competes with those who got there first, the best place to compete is often where the resources they already have come in handy. Looking back, Grok was tied to X from the beginning. The images and videos it generates can circulate directly on X, which Musk has also called the world’s best source of real-time data. From 2025 onward, X and xAI came under the same holding structure, and in 2026 they both became part of SpaceX. The money and resources available to them probably grew along with that. In text chat, the early movers already had users and established habits. With images and video, X’s distribution could be useful right away. This is my guess as I look back now; I haven’t yet traced the real reasons behind that decision inside xAI.
Finally, what did he do? Actions carry different weights. Describing a vision at a launch event, posting something, or making a verbal promise is usually much easier than actually completing an acquisition or investment, and easier to walk back later. Acquisitions, investments of real money, and organizational restructuring are costly and hard to undo. They reveal more of what someone really wants.
I tried separating his actions over the past few years from the remedies that came afterward and the ideas he had put forward publicly, then comparing which were costly and which were hard to reverse.
Already done, at a high cost.
• In October 2022, he acquired Twitter, later renamed X.
• In March 2025, X and xAI were brought under the same holding company, putting the social platform and the AI company under one roof.
• In January 2026, Tesla signed an agreement to invest around $2 billion in xAI. After SpaceX acquired xAI, that investment became a stake in SpaceX, with the transaction completed in March.
• In February 2026, SpaceX acquired the holding company that housed X and xAI. That same month, xAI reorganized into four areas: chat and voice, coding, Imagine, and Macrohard.
Done, then needing restrictions.
• The nonprofit research organization CCDH estimated that, over eleven days spanning late December 2025 and early January 2026, Grok on X produced around three million photorealistic sexualized images of people, including about twenty-three thousand depicting figures who appeared to be underage. There were also cases of real people’s photos being turned into sexualized images without their consent. Regulators in multiple jurisdictions opened investigations. The platform first restricted Grok’s image generation and editing on X to paying users, then added safeguards a few days later to prevent real people’s photos from being edited to show their bodies exposed.
Publicly announced, still on the way.
• SpaceX filed an application with regulators proposing to launch up to a million satellites as orbital data centers, using solar power to supply computing power for AI. This is an application and a proposal; approval and construction are still some way off.
• In March 2026, Musk described Macrohard as a joint project between Tesla and xAI, aiming to have digital Agents operate computers and run software companies.
• He has also talked about having Grok coordinate Optimus robots. What we can see so far is still an idea.
My own speculation.
• I tried arranging these connections into a conceptual path: understanding images, generating images, writing code, operating computers, and finally acting in the physical world. Of course, this is only my own understanding. xAI has not publicly laid out such a path.
Putting his words and actions together, they line up at least on the intelligence side: bringing X and xAI together, Tesla’s investment in xAI, and SpaceX’s acquisition of xAI. Almost all of those most costly actions were bets on the first thing he named. On the robotics side, what we can see is still mostly ideas.
But the January incident also draws our attention to something: looking at what someone has done means looking not only at the bets they chose to make, but also at what they had to pull back afterward. Having fewer restrictions and being willing to generate things others wouldn’t allow was probably one reason Grok’s images spread so quickly on X. But it ended up hurting real people.
Some things aren’t thought through beforehand. Once done, they veer off the imagined course—and quite a long way off it.
Still, restrictions imposed afterward are remedies under pressure, a different kind of action from a deliberate acquisition. But they, too, tell us “something.”
Among the things already done, X’s place in all this also got me thinking. First, it entered the same holding structure as xAI; later, both became part of SpaceX. To get others to believe in a vision, a person needs a space that can hold it. X is such a space: videos generated by Grok circulate there, and the future-city footage Musk reposts is seen there, too.
On October 16, U.S. Central Time, Musk also announced that all ads on X would have to meet “reasonable aesthetic standards.” Advertisers with ugly designs, pushy sales pitches, or exaggerated product claims would be barred from advertising on X. I really liked that post. Not every ad deserves to be recommended, and I have a little more appreciation for someone willing to publicly set aesthetic standards for a platform they run. If he really enforces them strictly, he will have to be willing to turn away some paying advertisers. There is another story around this decision, too: in 2023, major brands pulled out of X in large numbers. In the first ten months of that year, U.S. ad spending on X fell by more than 60 percent from the previous year. This March, Musk ran a poll asking whether people had ever bought anything because of an ad on X. After seeing the results, he replied with a sigh.
Taste and circumstances may both be driving this decision. I also realized that, once one person’s taste becomes a rule determining what hundreds of millions of people can see, it becomes a form of power, too. Good taste also needs ways to be seen and questioned. For now, this standard is still just a public promise. Whether it can actually improve the ads people see on X will only be clear once it is put into practice.
04 Stuck in Traffic? Let’s Fly
Before asking Grok, I had actually already watched a video on X. On September 20, U.S. Central Time, Musk reposted creator Aze Alter’s short film AGE OF BEYOND: If Humans & AI United, with the words “This is the future we shall bring into being.” The film had come out in 2025. What it depicted was an imagined future of humans and AI expanding together to the moon, Mars, and beyond. Watching it left me stirred, feeling that the future was especially worth looking forward to. I didn’t think much further at the time. Only when I looked it up again now did I realize that Aze Alter had written, directed, and edited the film, using a range of AI tools to make it. In other words, the images that made me feel so hopeful about the future were made by a creator using video generation.
The cost of depicting a future keeps falling. The cost of actually bringing it into being hasn’t fallen along with it.
On October 15, U.S. Central Time, Tesla unveiled its next-generation Roadster at a SpaceX test site in Texas. The prototype used cold-gas thrusters developed with SpaceX to briefly lift off, flip, and jump over a row of parked cars. No driver was in the car during the demonstration, and Tesla explicitly stated that it could not fly on public roads. Reserving the version that uses thrusters for acceleration requires an upfront payment of $250,000. Tesla described another, invitation-only version as having a “full flight system”; its price has not been announced.
That evening, Musk wrote on X, “Turns out the future does have flying cars.” Later, he also said, “Future fun Roadster demos will include flying over the Snake River or maybe Grand Canyon and landing on its wheels and exceeding half the speed of sound.”
I first saw someone share that video on WeChat Moments, and the first thought that popped into my head was so childishly simple: if traffic ever gets completely stuck, just fly up! Fly somewhere that isn’t jammed, land back on the ground, and keep driving.
But actually, the last sentence of Musk’s reply about K2 was:
But Roadster is super cool, so we're doing it anyway.
So in his sequence toward K2, this flying car wasn’t on the main path. They were choosing to make it simply because it was super cool. Funny, though: the very thing he treated as something to do “on the side” was what sent my imagination wandering~
What we can see, of course, is only a brief lift off the ground. A future where cars can “take off and land whenever traffic gets stuck” is still unknown. But I remembered a scene from that short film: a little cabin flying above the ground, between green trees and a row of transparent domes, with no tracks and no cables holding it. Add that thought of flying up whenever traffic jams, and a future city like that doesn’t seem so hard to imagine. Roads on the ground occupy a flat plane. The bigger the city, the more crowded that plane gets. If traffic could extend into the sky, the city would have more dimensions of space to use.
Let the thought run a little further, though, and a question appears: if everyone wants to fly up and head somewhere without traffic, wouldn’t that place soon get jammed, too? If everyone takes to the sky, how can they each find their way without crashing into anyone else?
In a transportation economics study, the economists Duranton and Turner found that highway mileage and vehicle miles traveled in U.S. cities grew in almost the same proportion. New traffic soon filled the extra roads. But their research was about highways on the ground. How much demand for air travel would grow depends on prices, noise, weather, takeoff and landing sites, regulations… That is another question. Still, I couldn’t help wondering again: once we open up the possibility of using the sky, might new demand appear, too?
So the hardest part of a city where people can fly probably lies beyond the vehicles themselves. If large numbers of flying cars enter the skies above cities one day, a coordination system will need to continuously track each car’s position, movement, and surroundings, predict possible conflicts, and give each one a way through without collisions. This shares some challenges with the question I’d been pursuing through Grok: “How do we get AI to understand a dynamic world?” They don’t necessarily have to be solved in the same way, though.
And then there was that huge cloud of white mist when the car took off, looking like steam from a rocket launch. Reports said these thrusters expelled high-pressure gas, without combustion. Exactly how the white mist formed still awaits an official explanation. Looking at that cloud, I imagined a version just a little further into the future: one day, a car takes off without a cloud of white mist or a deafening roar, as if an enormous invisible hand were lifting it steadily from the ground. It reaches a certain height and then, just like driving normally, off it goes with a little “whoosh.” That steadiness might feel like what I imagine a maglev train feels like, or like the little cabin flying between the trees and domes in that film. What technology would make that happen? I don’t know yet. I even feel that I probably won’t live to see that day.
05 Wan Hu’s Cloud of Smoke
That cloud of white mist from the Roadster reminded me of a very old story. Legend has it that a Ming dynasty official named Wan Hu tied forty-seven rockets and two kites to a chair, sat down, and had his servants light the rockets. There was a huge bang. When the smoke cleared, both the man and the chair were gone.
The story’s origins are quite winding. An early English version that can be traced appeared in Scientific American in 1909. Its character was named Wang Tu, the setting was before the Common Era, and the ending was different, too. The familiar version about Wan Hu in the Ming dynasty spread later through other books, then made its way back into the Chinese-speaking world. Whether Wan Hu had a historical counterpart remains unclear. Yet in 1970, the International Astronomical Union named a crater on the far side of the moon after him.
Someone who may or may not have existed, whose name really did reach the moon—all because of a thought about flying into the sky.
Looking further back, Chinese mythology already held two impulses. Pangu splitting open the primordial chaos tells of a world that can be opened up and created. Chang’e flying to the moon tells of the wish to reach places the body cannot yet go. Thousands of years later, China’s lunar exploration program is called Chang’e, its Mars probe Tianwen, and its Mars rover Zhurong. Ancient myths and literature left these names behind; many years later, people gave them to probes that actually flew into space.
From Jules Verne sending people to the moon with a cannon in a novel, to the Wright brothers’ first twelve seconds off the ground, to Apollo 11 landing on the moon, only a little over a hundred years passed. Between the latter two, just sixty-six. It is knowing and learning about things like these that gives me such awe for human creativity. It seems people first solve survival problems like food and shelter, then reshape nature, build cities… then connect the internet, and now build AI… It seems someone always imagines something “impossible” first. Then, after who knows how long—it actually becomes real.
Earlier, in “What Are People Who Can See the Future Actually Looking At?” (part of my “见外手记” writing experiment), I wrote about a line by Kuriyagawa Hakuson: “Apart from the present, the future does not exist.” Perhaps those who first imagine the “impossible” are also looking at the present: what is still missing now, and which wishes have yet to be spoken.
06 What Does a Thought Have to Pass Through to Become Real?
“Work hard enough, give it enough time, and things will work out.” I really want to believe that. But looking at Grok, the Roadster, and Wan Hu together, I realized time is only one of the conditions.
For a thought to become real, it has at least four layers to pass through.
Capability.
A video model can generate coherent footage. Thrusters can briefly lift a car off the ground. These things that carry capabilities are the easiest to see, and what launch events mostly talk about.
Environment.
The reach of Grok’s videos is inseparable from its growing inside X. The Roadster can already briefly lift off and land again, but regulations and cities aren’t ready to deal with it.
Coordination.
One car flying is an engineering problem. Ten thousand cars flying at once becomes a social problem. Who goes first? Who yields? Who is responsible if something goes wrong? All of this needs new rules.
Feedback.
Every demonstration, every reposted glimpse of the future, could change the direction of the next attempt: influencing the next group of people’s attention, willingness to buy, and investors’ assessments, which may in turn change where teams put their resources. Musk’s four-sentence reply was soon recast by a fan account as “Elon Musk just laid out the next thousand years of the economy in four sentences.” What began as a judgment in a reply can become a more certain, grander narrative as it is retold again and again. Feedback can also push things out of control. January’s image-generation incident is one example.
And as a thought passes through these four layers, there are two common illusions along the way.
One is about time. People often overestimate a technology’s short-term impact and underestimate its long-term impact. This observation is called “Amara’s Law.” It reminds me not to mistake short-term promises for long-term results. The Roadster first appeared in 2017. Demonstrations were delayed again and again, and it only lifted off this year. Mass production is said to be coming in 2028. For now, this example only shows the distance between a promise, a demonstration, and a product actually delivered into people’s hands. Whether it will have a long-term impact is still unknown.
The other is about distribution. Science fiction writer William Gibson once said that the future is already here, just very unevenly distributed around the world. Even reserving the Roadster that only uses thrusters for acceleration requires $250,000 upfront. The first to experience futures like these are often those who can afford the high price. The costs fall elsewhere, through different mechanisms: people whose real photos are turned into sexualized images without consent suffer actual harm; large data centers draw on limited local electricity and water supplies, potentially adding to the pressure on surrounding communities’ resources; the more satellites there are in orbit, the more crowded space becomes…
There is one last thing I need to be honest with myself about. We remember the Wright brothers; we rarely remember the machines that couldn’t fly. In the 1950s, the U.S. and Canadian militaries funded a flying-saucer-shaped aircraft called the Avrocar. It could hover, but once it rose more than about ninety centimeters above the ground, it became unstable. I said earlier that I am in awe of human creativity. Part of why I feel that awe is that we tend to hear the success stories. The failed experiments rarely make it into the telling. But they, too, are part of what made this path possible. It is also this context that helps those who come later see more clearly which ways don’t work.
07 A Future I (Maybe) Won’t Live to See
Before wrapping up, I want to return to the most personal question: why do I still believe in the arrival of a future I probably won’t get to see?
This July, Musk wrote on X, “Any civilization that loses faith in the future will die.” Reading that, I thought: people really do need to believe in the future to dive headfirst into things whose results they may never live to see, to create, to make things happen. But I also want to keep a few questions in mind: who is drawing this future, who will bring it about, and who is paying the price…
I think I am willing to believe because so many things were never going to be completed within one person’s lifetime. I am just running one little stretch of a relay whose length none of us can accurately predict. The astronomers who named a lunar crater “Wan Hu” had never met Wan Hu. The people who named the lunar exploration program “Chang’e” knew Chang’e had never existed. But what they were doing was much alike: bringing an old, old thought over, carrying it a little further, then handing it to the next group of people. Those who plant trees probably won’t live to sit in their shade. Still, one by one, the trees keep being planted.
When I finish this article, I’ll post it on X. That also means I am entering the loop I described in section six: my sharing it might bring that vision to more people. But even as I’m drawn to the vision, I want to keep a little room for my own reservations, a place between “I’m moved by this” and “what I choose to believe” where I can slow down and wander freely along the paths my mind takes.
Some ideas will be redesigned; others will be ruled out by new conditions. Only the ones that make it through all those constraints will finally land in reality. I can’t guarantee that the car lifting quietly off the ground will ever exist. What I can do, when that question happens to pay me a visit, is give it a little attention, ask it, go chat with a few different AIs, then set the answers alongside public facts and compare: what did someone say, and what did they actually do—and then, what did I think of?
So, to sum up, I originally just wanted to know why Grok cared so much about image and video generation. Following that question all the way through brought Musk’s space plans to mind, along with things imagined that might take generations to become real (or might remain imagined)… My brain can’t help declaring: this has been a most satisfying trip!
If something has made you curious lately, maybe you, reading this, could give it a little attention too, and see what interesting things you can dig up~
And finally! Wan Hu’s smoke cleared long ago, and the Roadster’s white mist will clear too. But there will always be someone to pick up those thoughts and pass them to the next generation. Even if I may never see that car quietly leave the ground, I am willing to believe someone will.
This article began with a conversation between Grok and me. Claude helped with the writing, and ChatGPT helped with the review. The questions, judgments, and choices all came from Helia (that’s me!).
Sources
01 Where Grok Started
1. xAI, Grok-1 Model Card, November 2023
https://x.ai/news/grok/model-card
2. OpenAI, DALL·E 3 is now available in ChatGPT Plus and Enterprise, October 2023
https://openai.com/index/dall-e-3-is-now-available-in-chatgpt-plus-and-enterprise/
3. xAI, Grok Image Generation Release (Aurora), December 2024
https://x.ai/news/grok-image-generation-release
02 Grok’s First Answer
4. Elon Musk’s post on video understanding and generation, March 30, 2026 (U.S. Central Time)
https://x.com/elonmusk/status/2038756516048916578
5. AI Weekly, xAI Lost $6.4B in 2025 as Capex Surges (relaying TechCrunch’s reporting on SpaceX’s IPO filings), May 2026
https://aiweekly.co/alerts/xai-lost-64b-in-2025-as-capex-surges
6. Zheng & Meister, The Unbearable Slowness of Being, Neuron, 2025
https://www.sciencedirect.com/science/article/pii/S0896627324008080
7. Latent Space podcast, Why Video Agent models are next — Ethan He, June 2026
https://www.latent.space/p/video-agents
8. AOL, Elon Musk says Grok's latest feature is the new Vine
https://www.aol.com/elon-musk-says-groks-latest-002432138.html
9. Kang et al., How Far Is Video Generation from World Model, ICML 2025
https://proceedings.mlr.press/v267/kang25g.html
10. Huh et al., The Platonic Representation Hypothesis, ICML 2024
https://arxiv.org/abs/2405.07987
11. Lex Fridman Podcast #416, Yann LeCun interview transcript, 2024
https://lexfridman.com/yann-lecun-3-transcript/
03 Listen to What They Say, and Watch What They Do
12. Elon Musk’s “K2” reply, October 18, 2026 (U.S. Central Time)
https://x.com/elonmusk/status/2111683964721037767
13. Decrypt, Musk Says Real-World Data From X and Tesla to Drive AGI Efforts, November 2025
https://decrypt.co/349920/musk-real-world-data-x-tesla-drive-agi-efforts
14. SiliconANGLE, xAI’s acquisition of X, March 2025
https://siliconangle.com/2025/03/28/elon-musks-xai-acquires-x-33b-valuation/
15. Tesla, Form 10-K/A (related-party transactions: investment in xAI converted into SpaceX shares)
https://ir.tesla.com/_flysystem/s3/sec/000110465926053166/tm2611837d1_10ka-gen.pdf
16. Shacknews, SpaceX’s acquisition of xAI and orbital data centers, February 2026
https://www.shacknews.com/article/147689/spacex-acquires-xai-elon-musk
17. Reuters, SpaceX seeks FCC nod for solar-powered satellite data centers, January 2026
https://www.reuters.com/business/aerospace-defense/spacex-seeks-fcc-nod-solar-powered-satellite-data-centers-ai-2026-01-31/
18. xAI’s reorganization into four divisions (a compilation based on reporting by The New York Times)
https://texxr.com/entity/macrohard/launch
19. Reuters, Musk unveils joint Tesla-xAI project Macrohard, March 2026
https://www.reuters.com/business/autos-transportation/musk-unveils-joint-tesla-xai-project-macrohard-eyes-software-disruption-2026-03-11/
20. Tesla Q4 2025 earnings call transcript
https://storage.googleapis.com/magi-store/earnings/TSLA/2026-01-28_Q4_FY2025/03_outputs/transcript.html
21. CCDH, Grok floods X with sexualized images, January 2026
https://counterhate.com/research/grok-floods-x-with-sexualized-images/
22. Reuters, Grok limits image generation to paid users after backlash, January 2026
https://www.reuters.com/sustainability/boards-policy-regulation/musks-ai-bot-grok-limits-image-generation-x-paid-users-after-backlash-2026-01-09/
23. Bloomberg, xAI blocks Grok from creating sexualized images of real people, January 2026
https://www.bloomberg.com/news/articles/2026-01-15/xai-blocks-grok-from-creating-sexualized-images-of-real-people
24. Elon Musk’s post on aesthetic standards for ads, October 16, 2026 (U.S. Central Time)
https://x.com/elonmusk/status/2111131160348688395
25. Reuters, After Musk tirade, X faces prospect of more advertisers fleeing, November 2023
https://www.streetinsider.com/Reuters/After+Musk+tirade%2C+X+faces+prospect+of+more+advertisers+fleeing/22473903.html
26. Yahoo Finance, Musk asked if anyone actually bought something from an ad on X, March 2026
https://finance.yahoo.com/news/elon-musk-asked-anyone-actually-173256522.html
04 Stuck in Traffic? Let’s Fly
27. Elon Musk’s repost of Aze Alter’s short film, September 20, 2026 (U.S. Central Time)
https://x.com/elonmusk/status/2101601873115681037
28. Aze Alter, AGE OF BEYOND: If Humans & AI United (YouTube)
https://www.youtube.com/watch?v=vp7xoPeWzEw
29. Reuters, Tesla’s next-generation Roadster unveiling, October 15, 2026
https://www.reuters.com/business/autos-transportation/tesla-set-unveil-next-generation-roadster-sports-ev-after-years-delay-2026-10-15/
30. Elon Musk, Turns out the future does have flying cars, evening of October 15, 2026 (U.S. Central Time)
https://x.com/elonmusk/status/2110913107094606330
31. Elon Musk’s post on future Roadster demonstrations, October 17, 2026 (U.S. Central Time)
https://x.com/elonmusk/status/2111507158215057620
32. Benzinga, Tesla Preps SpaceX Rocket-Powered Flying Roadster Demo
https://www.webull.ca/news-detail/15402037474960384
33. Duranton & Turner, The Fundamental Law of Road Congestion, NBER
https://www.nber.org/papers/w15376
05 Wan Hu’s Cloud of Smoke
34. Wikipedia, Wan Hu
https://en.wikipedia.org/wiki/Wan_Hu
35. The Encyclopedia of Science Fiction, Rockets (the 1909 Wang Tu version)
https://sf-encyclopedia.com/entry/rockets
36. China National Space Administration, From the poem “Tianwen” to the mission “Tianwen”
https://www.cnsa.gov.cn/n6758824/n6759009/n6760412/n6760413/c6840458/content.html
37. China National Space Administration, China’s first Mars rover named Zhurong
https://www.cnsa.gov.cn/n6758824/n6759009/n6760412/n6760413/c6840433/content.html
06 What Does a Thought Have to Pass Through to Become Real?
38. Captain Eli’s quote-post of Musk’s “K2” reply, October 2026
https://x.com/thecaptaineli/status/2112073115949809743
39. National Museum of the US Air Force, Avro Canada VZ-9AV Avrocar
https://www.nationalmuseum.af.mil/Visit/Museum-Exhibits/Fact-Sheets/Display/Article/195801/AFmuseum/avro-canada-vz-9av-avrocar/
07 A Future I (Maybe) Won’t Live to See
40. Elon Musk, Any civilization that loses faith in the future will die, July 25, 2026 (U.S. Central Time)
https://x.com/elonmusk/status/2081154036338692413