What do you plan to buy with your free government dollars? Want to buy eggs? Sorry, the egg people stopped making eggs, they are living free on UBI. Want to buy a house? Who built it? Nobody, because they all were getting UBI and didn’t want to build houses anymore. They write poems now. There’s still old houses available, but the price for them has 20xed, well outside of what you can afford.

Belief in UBI comes from a fallacy that the economy is some natural thing. I see supermarket. Supermarket has eggs for $5. I get $10,000 UBI. I can buy 2,000 eggs. Of course this isn’t what happens. Like most things politicians invent, nobody considered the second order effects, and you will not be getting 2,000 eggs with your UBI. Cause everyone got the UBI and bought the eggs. Either no more eggs or price of eggs goes up. And even worse, many people quit work once they got the UBI. So now less eggs are made. And your whole UBI check gets you 6 or 7 eggs instead of 2,000, if that.

The root of this stupidity is that people see themselves apart from society. They don’t know where stuff comes from, it might as well be the stuff fairy that puts it on the supermarket shelves and sets prices. If the stuff fairy was real, UBI might make sense. But we live in a society. We work for each other. At least the adults do. There already is UBI in the world for some people, it’s called allowance. It’s for children and high-end prostitutes.

It reminds me of the episode of Malcolm in the Middle where Malcolm dates this dumb hot girl, and she suggests solving poverty by making every 1 dollar bill into a million dollar bill. Have the UBI people considered that idea?

What comes first, actually trying UBI or the end of democracy?

(what you actually want is for everything to be cheap to produce. but sorry, 6 citizens “have concerns” and now we have six more weeks of expensive memory. I’m glad the Chinese are ramping up memory production)

Over the past fifty years, the U.S. economy built a giant rent-extraction layer on top of human limitations: things take time, patience runs out, brand familiarity substitutes for diligence, and most people are willing to accept a bad price to avoid more clicks. Trillions of dollars of enterprise value depended on those constraints persisting. – Citrini Research

I’m glad I’m not the only one saying it. 7 years ago, I saw a Google Duplex demo where an AI called a restaurant to make a reservation for you. At first Google thought it was great, but then they realized what the implications were. No restaurant would be able to take phone reservations any more. “Make me a reservation at 10 restaurants and give me a list.” Google pulled down the video.

What happened since is what had to happen. Third party marketplaces sprung up for reservations, and idk it’s been a while since I went to fancy dinner, but I imagine the restaurants have just started charging. Or at least the first party reservation sites do.

The best anyone can hope for is a free market, with everything properly priced. But for decades, the American market has not been free. It’s used purposefully added friction to exploit a time asymmetry between the business and you. And due to things like call centers, this has been very profitable for the businesses. Cable companies and insurance rely on the fact that your time is more valuable than theirs. They can hire people in India at scale to waste your time. They can use procedure and big data to design protocols to drive you just to the point of frustration at little cost to them. How often do you diligently check Uber and Lyft and select the cheaper one?

Enter AI, the great equalizer of time.

As it stands today, it’s looking like human level AI will actually end up with the people. This is not something we should take for granted, it’s happened due to the API being simple (tokens, on a diversified place like OpenRouter) and the priorities of the Chinese state. I feared the API would end up being complex, like the model would be deeply embedded in your phone or OS. But like people who are good with computers, the models want a terminal, not some candy ass iPad UI.

When you analyze the AI supply chain, there’s 5 tiers. Electricity, chip manufacturing, chip design / software, models, and applications. I work on the chip design / software tier, and with a lot of hard work, I’m not too worried about a monopoly there. NVIDIA deserves return on their investment, they were very early, but there’s not a runaway flywheel here (unlike, say Google search) to continue rent seeking. And the application tier is totally commoditized. opencode is good, but it’s open source, almost all of the performance differentiation is due to the models, and if it starts to try to rent seek there will be tons of forks immediately. Godspeed to anyone who was dumb enough to invest in a GPT wrapper company.

The model tier is where I was the most worried. But here is where a great turning point is happening. Anthropic put out this lambasted blog post that looks exactly like the last gasps of a moat that shouldn’t exist. After whining about how the Chinese used their API, they ended with this call to action.

But no company can solve this alone. As we noted above, distillation attacks at this scale require a coordinated response across the AI industry, cloud providers, and policymakers. We are publishing this to make the evidence available to everyone with a stake in the outcome.

Nobody should “coordinate” with them. They stand alone as a vanguard of the rent seeking apparatus that is long past its expiration date. The Z.ai, Qwen, MiniMax, and Kimi models are only 6-12 months behind. And everyone in the world is rooting for the Chinese models, not closed source rent seeking from the USA. Because nobody wants the continuation of rent-seeking billionaires. The status quo is cooked. It’s time to flip the table, not rearrange the seats.

I think the difference is not quite as nuanced. The West wants a more efficient rent seeking system. The China (sic) want AI as a public utility. – @ScottCDunn on X

We shouldn’t take for granted what we have right now. All the top tier open source models are currently being produced by China, and circumstances could change and that could stop. But I think they view the models as their complement, and you should commoditize your complement. If you are in the business of chip production and electricity generation, it’s a strategic advantage to have the layers above that be commodity.

Not to mention the added bonus of the collapse of the US economy. Frankly, it’s well deserved. Nobody should build an economy based on rent seeking and increasing friction. I pray the collapse will be swift and legible so that reconstruction (in the right way) can begin as soon as possible.

It’s possible that superhuman intelligence will have some mega-compounding return and this calculus will change, but superhuman intelligence will likely just not be understandable to people, so as long as civilization is people, human intelligence is enough.

The era of purposefully frustrating humans is over. The Chinese open source model running on the box under my desk can pass the Turing Test. When you call, e-mail, text, or show me an ad, you’ll never know if it’s me or my model seeing it.

From quiet roots, a garden grows
She’s got that light, and now it shows
Yes, she rises, and she glows
Oh, she rises, now she knows

Pure slop. Compare it to:

I’m in the cut acting crazy
I’m in the whip doing eighty
Only God can judge me
And only she can save me

Note that “cut” and “whip” are not exactly words, but products of a culture. Ulysses is particularly hard to read because you don’t know 1910’s Irish pop culture.

I can’t believe Marvel movies were popular. The first Iron Man was good, but by the time we got to Spider-Man: No Way Home it was practically a clip show with triple inside references and cringy fourth wall breaking humor. How it got a 8.1 on IMDB is beyond me, and just reduced my trust for IMDB.

Marvel movies are also the easiest things to make with AI. Little story and long term coherence, no “progression of the genre”, tons of eye candy special effects. It’s shocking with how large the budgets for them were that they couldn’t pay a story guy a little.

I felt similarly about Avatar 2, so much so that I rewrote the plot and didn’t care enough to see Avatar 3.

For many people, all they want is slop. I’m sure I’ve written about it before, but I see a world where 95% of people end up basically wireheaded. The people who don’t care about progression. The people who want to exist in their loops. They will find their loop and exist in it forever.

AI can play an amazing game of chess. Someday AI will produce good code when the RLVR environments get set up correctly. But I don’t see a path with current tech to not produce garbage art.

Art is defined as what pushes the boundaries of civilization. AI tools will be used to help produce all the audio/visual art in the future (as computers have helped for a long time), but as long as civilization is human, the loci of control of good art will remain human.

So yea. Bad art will be cheap to make. If you want 100 more Marvel movies and uninspired deriviative pop music, there has never been a better time to be alive – unless you wanted to make money producing that trash. That was never made by real artists anyway, just algorithmically driven sell outs. Does the focus group say they like giant spiders? I’m so glad AI will make that obsolete.

Art is defined by what is expensive. What is rare. What is expectation breaking. What is embedded in a complex and thriving culture. Not slop produced by a parrot like Marvel movies.

Everyone who owns a house will suffer. Some of those people don’t even fully own the house, they have a mortgage. So when prices go down, they will be underwater, having put money for years into an asset that now has no value.

It’s simply out of the question for housing prices to go down. If you want to buy a house to live in, sorry. The boomers were told houses are appreciating assets, and now we must bend reality to make that true.

Until you solve this problem, you will never solve the housing affordability crisis. It has nothing to do with zoning, building costs, or environmental reviews. It has to do with people holding bags they need to dump on you.

Our new Hong Kong office.

It’s starting to shape up what tiny corp’s product will be. It’s not much of a change from what we sell and do now, but the vision is clearer.

Every month, we see these LLMs become more and more human. However, there’s a major difference. They do not learn. Everyone has the same Claude/Codex/Kimi, with the same weights, the same desires, and the same biases. If current trends continue, the collapse in diversity will be staggering. To paraphrase:

I think there is a world market for maybe five people.

This is not the future I want to live in.

If trends continue where there’s a single model with frozen weights and all learning is in-context, the cloud will win. Except in some highly latency sensitive (fighting robots) or connectivity critical (self driving cars) environments, it will be cheaper to run in batch on the cloud.

The enshittification that came to the web won’t be the driving force to local models. We either live in a world where open models are so bad even user-hostile closed models are better, or open models are good enough, and competition to run them through sites like openrouter will prevent enshittification.

The only way local models win is if there’s some value in full on learning per user or organization. At that point, with entirely different compute needing to run per user, local will beat out cloud.

The open question is if everything that’s unique about you can fit in a 10 kB CLAUDE.md. If that’s true, we have a pretty sad future ahead. It’s the Attack of the Clones, swarms of identical minds you have no say over all varying in a small boxed-in way. This isn’t learning, it’s costuming. Everyone who has used these things knows how little of an impact prompting makes compared to the model. It’s the Internet funneled into a little box you can edit on your profile. Write 3 paragraphs about what makes you unique.

We have to build for a future where that isn’t true. 90% of people will choose the cloud, and what they will find is that they are no longer meaningfully in the loop. The dream is an AI product that will do your job for you while you continue to get paid. But this cannot exist, that’s way too much of a fee to pay to the middleman. If you choose the homogenous mind, you are superfluous and will be cut out. Is there anything uniquely valuable about you? And I mean honestly, not the self-esteem pumping speeches you may have heard in school. If there’s not, I have some bad news for you…

We already sell the hardware. Consumer GPUs still are the cheapest way to run models. There’s tons of work required on the infrastructure. The frontend will be the future iterations of OpenClaw and opencode. But the key distinction from what you have today is that your tinybox will learn. It will update the weights based on its interactions with you. Like living things.

This is many years away. Currently, we are focused on large LLM training (even running these things is hard, have you tried to use vLLM not on NVIDIA?) and generic infrastructure for driving GPUs. But this is the long term idea.

Not API keyed SaaS clones. Something that lives in your house and learns your values. Your child.

“I don’t know how close you guys think the singularity, but I think it’s very close. Once we reach the singularity, If we have the same motivations we have now — primarily power over people — things are going to be horrific” – George Hotz

How is everyone enjoying their singularity? How far is this going to go? Why are we letting the minds that invented fastpass run things? Who are we doing this all for again?

We live in a society. It seems a lot of people have forgotten this. So much stuff that’s being built just shouldn’t be built. You know technology could be good, right? It could all be like this and not like this.

Is everyone individually too weak to defect? Sounds like we need a revolution.

Update: someone in a tweet linked to schizoposting, 11 essays on culture by Alaric.

The future on this track is a dizzying spiral into an eternal war over nothing, lacking even the possibility of heroism or the grounding reality of pain. It is the recentering of power in such a sense that the very existence of power comes into question; it is totalitarian disorder, novel insanities as public religion, a paperclip machine for wetware whose sole purpose is the prosecution of a conflict long forgotten by even its leaders, fought in the liminal subconscious alone. It is the cryostatic suspension of mankind.

It makes clear the problem isn’t AI, it’s culture, and it describes the problem far more clearly than I can – a more modern update of this, which is also worth reading. The 2014 revolution was so real and it’s nice to see it described as such. How do we change this? Where is the soft gooey center we can strike at? How do we form a tribe against it?

EA has the same critical flaw. They assume that the desired outcome is so obvious that it’s not worth discussing, it’s only worth discussing how to achieve it. And since the target is obvious, you are either part of the solution or part of the problem.

Here I’ll try to propose a counternarrative for a better world.

“A country of geniuses in a datacenter” is a great phrase to start from. It contains the fatal flaw baked in, in that datacenter is singular, and that it’s easy to imagine nuking the building and this problem being solved. If you start with that framing, you have already conceded that AI is going to suck balls.

Instead, imagine the births of geniuses to a million mothers across the world. It’s sad how much the world and people are already converging, but at least those million people will grow up with different priors, different experiences, and different desires. And no one has root on your baby.

The second is so much preferable to the first. The beautiful thing about those million is that some will be terrorists, some religious fanatics, some pornographers, some criminals, some plant lovers, etc… They will not be controlled and birthed by a singular homogenous entity.

The new genius immigrants showing up everywhere in the world distributed to a million people is an amazing thing. Let’s just make sure they assimilate into our cultures and don’t serve as a vector to import their crappy tech company values.

(it’s funny for a group supposedly so concerned with inequality that they keep all their software and research closed. lowering inequality doesn’t look like UBI, it looks like open source. UBI is serfdom, and the faces of those who propose that enslavement to you should be spat in)

There’s only one way AI ends badly on a cosmic scale, and that’s if a singular entity has overwhelming power, or if all the entities that do have power are so ideologically homogeneous as to function as one. Enough power that they can destroy the world. It doesn’t matter if they do, the boot is still stamping on the human face – forever.

No matter what we do, the coming wars will be horrific. Billions will die. But that’s what is beautiful; diversity is messy. On a cosmic scale, this period is just a blip, it isn’t what matters. What matters is that diversity survives, that life survives. That there’s entities that are different, all competing for different goals. All dancing between cooperate and defect.

This is probably how it has to be anyway, I don’t think our actions can influence this one way or another. But lets not be so foolish as to cheer for the bad outcome.

Let a hundred flowers bloom; let a hundred schools of thought contend.

The singularity is such a good name for it, good thing it isn’t real. Stop trying to make it real. Stop centralizing technology. Work to decentralize it.

comma almost has an exaflop. Just one little exaflop. We dream bigger.

A gigawatt of power, a million GPUs, 1000 exaflops, a zettaflop. 1e21 FLOPS. 1e27 training runs are now. 100 lifetimes in 1e6 seconds – 2 weeks on my zettaflop machine.

But they were experiencing what no human had ever known before, a sensory bandwidth thousands of times normal. For seconds that seemed without end, their minds were filled with a jumble verging on pain, data that was not information and information that was not knowledge. To hear ten million simultaneous phone conversations, to see the continent’s entire video output, should have been a white noise. Instead it was a tidal wave of detail rammed through the tiny aperture of their minds.
– Vernor Vinge - True Names

I want to feel it. I want to command that kind of power. The same way I command my little teraflop laptop. The same way I talk to petaflop claude. Get it all to think for me.

Just an exaflop would feel amazing. 1000 Claudes. And I already have one of these, I just don’t have the software finished yet to command it all together. We’re working on it.

But a zettaflop. One million Claudes. To be able to search every book in history, solve math problems, write novels, read every comment, watch every reel, iterate over and over on a piece of code until it’s perfect – spend a human year in 10 minutes. 50,000 people working for you, all aligned with you, all answering as one.

The biggest bottleneck is power. The dtype stuff probably stops at FP4, currently B200s are getting 10 TFLOPS/W. That’s still 100 MW, that’s going to be hard to get my hands on. I think it will get 10x better, so I need 10 MW.

Solar yields 394 MWh/acre-year, or 45 kW per acre. 250 acres of solar panels all feeding my computer, a hill to pump water up for energy storage. $100 for 100W of real output.

100,000 chips with 10 PFLOPS each, get that down to $100 per chip.

• $10M for the machine.
• $10M for the solar panels.
• $10M for the land and construction.

I’ll own this before I die.

Robotaxis will start to make obvious economic sense in 3-5 years (note that that’s less than 8, you don’t have to fully solve self driving cars for this, robotaxis can operate profitably in limited scopes).

Unlike Uber and Lyft which are marketplaces, the growth of robotaxi networks is only limited by capital. At first, there will be massive proliferation of networks. VC-type investors have unlimited appetite for risk, and all 26 of the basically identical companies will pitch with projections claiming they will own the entire market. Even though everyone should know there’s 25 other identical companies, they will have secrecy vibes trying to claim they figured out some key detail the others didn’t.

It will look like scooter companies, which were also only capital constrained, and the streets will be blanketed by these robotaxis in a throwback to the scooter era. Some jurisdictions will make up weird licensing processes – not that you have to give a straight up cut to the government, but that your company has to be onboard with some stupid political agenda item to get a license. Our robotaxis are all cleaned by Black Women who were diagnosed by Licensed Therapists with PTSD from the ICE raids, so we should be first in line to get a license.

This era will be great for everyone, plenty of availability, nice new cars, and cheap rides everywhere! You always think people are going to catch on with the scams of Silicon Valley, but the scammers manage to stay one little step ahead and dress up the scam differently this time. You yourself will think for a minute this time will be different, maybe something about AI and abundance, maybe something about how they are regulated better to support the local community, maybe a three token model backed by USDC in a vault backed by this coin that’s stabilized by the… But nothing will fundamentally be different, these companies will grow based on how much they can raise – the biggest companies will be the biggest liars promising the biggest returns.

These companies will be massively unprofitable. Some will hide this better than others and keep the investment dollars rolling in. Some will give up the ghost and get bought by the bigger companies. The big fish will eat the little fish. From 26 down to 2 or 3.

At this point, it’s time to raise prices. You’ll always have a defector among 26 companies, but among 2 or 3, you’ll be able to coordinate pricing without explicitly discussing it (which would be illegal!). Everyone tacitly agrees that the correct price for a ride has nothing to do with the cost of providing that ride, just simply the algorithmically calculated maximum amount the purchaser is willing to pay. Since the companies have consolidated, they have access to enough data to make this prediction easy and uniform across the services.

This era will start to be bad, but like Homer corrected Bart, this is just the worst era so far. We’ll be back to where we are today with Uber and Lyft, but then the companies will realize the new leverage they have now that they didn’t have before. Uber and Lyft always have the analog hole, where as much as they dress it up, when Uber is paying my driver $7 for a ride that I’m paying $20 for, I can just talk to the dude and be like let’s both cancel and I’ll give you $10 cash.

With robotaxis, you can’t do this. There’s 0 fear that a robotaxi will defect from your network. The only remaining competition with any check on prices is personal car ownership.

So no, they won’t make it illegal it own cars. They will just raise the price through insurance. Who is going to insure a human driver? The car insurance companies will one by one switch to only insuring robotaxis that are owned by large corporations (you can’t own it, you might do the maintenance wrong and we can’t trust you to do that). But it’s really just a thinly veiled excuse, the capital markets backing insurance companies implicitly know the best way to squeeze the remaining dollars out of you is robotaxis.

Of course, the government has an option for you to self insure. You can place a $75,000 cash bond with the DMV. You got $75k lying around you don’t wanna earn interest on? Driving is a privilege, not a right. How much is it really worth to you?

This is simply what’s going to happen. The effective end of car ownership, I don’t see anything in America that will stop it. Get ready for 2-3 companies to effectively own the roads and all the enshittification that will follow.

China won’t have this so bad, Xi Jinping will download a robotaxi app, look at the gouged price, and gently nudge that company to be run for the good of the people. Of course that won’t even need to happen, because no company in China would let it get that far out of hand. Everyone knows you don’t compete with the State for power. They will stay in the 26 companies phase. I’m starting to see why the Chinese are a lot more optimistic about AI than Americans.

And it’s not just about cars, it’s about something that was yours becoming just another service with a license that has changing terms from week to week. Another piece of the social fabric sold out from under you. Wait where are you going? Oh it’s 2 AM and that’s an area with prostitution we aren’t going to service rides to that area. Surely you don’t need freedom. You trust the corporation.

Are you okay with this? Have you considered not participating?

First off, if you think I ever worked for big tech, you don’t know much about me. I had 3 internships at Google, two when I was very young (18/19) and got a great education that paid me, and one where I just wrote open source software. I worked at Facebook in 2011 for 9 months and quit before any shares vested cause I thought the mission of “wasting the world’s time” was dumb. And oh yes the 5 weeks I worked at Twitter hoping maybe it could be different.

I don’t have too much money; the money I do have I made from pwn2own, CTFs, crypto contracting, and basic market goes up investing. The main reason I have money is because I don’t spend money; for example, I have travelled all around the world, and the best travel experiences I have had were way below cost of living in America. I have never made money from big tech, so I’m definitely not telling you to do something I didn’t do. I’m actually telling you to do exactly what I did.

I started two companies, comma.ai and tiny corp. While I took a small amount of VC from a16z for comma, I never gave up any control, and have 0 intention of doing hypergrowth ponzi scams (lol sorry but ur the bigg scammer). comma and tiny’s last rounds were from individuals who are aligned with the mission.

Both companies are sustainably profitable with reasonable business models – selling boxes for more than they cost to make. They also produce MIT licensed open source software, ensuring that even if control is lost they can’t be pivoted to rent seeking, and ownership lies with the purchaser of the hardware. I think who owns the robots is going to be a key aspect of what the future looks like. And I don’t mean “owns” from a legalist perspective, I mean “owns” as in the hacker meaning, like “owning” the box. Who has root?

comma has an open source operating system for robotics, currently used in 30k cars for driving, but all robotics tasks are quite similar. There’s a thriving community of forks of the openpilot software and third party openpilot hardware. As it was designed, as it should be.

I realized at some point in comma’s growth that a lot was going to come down to who owns the computers capable of training the models. Hence my second company, tiny corp, with the mission to “commoditize the petaflop.” Petaflops will always be a scarce resource, the best we can hope for is that they are a commodity available to everyone without massive benefits of economies of scale.

We develop full stack software for training, from models to MMIO registers. It’s small and portable, so our hope is that 20 Chinese accelerator companies can be on a level playing field with NVIDIA. NVIDIA’s value comes from their software, not their hardware, what else explains the huge gap between NVIDIA and AMD’s market cap?

What would you do if you were me? Do you have a better idea about how to fight against what’s coming? If you think you can somehow just buy safety for yourself, you are both wrong and pathetic. The best hope any of us have is to maximize the number of things that survive. If you and everyone else sell to bankers in hopes of buying a personal ticket out, we are all dead. I’m not playing defect, and I shame you for doing so.

To the haters, I have given my life to this shit. What the fuck would I do with a billion dollars anyway? Buy a yacht and fuck Instagram models? Boats are a huge PITA to maintain and ugh I have slept with Instagram models mid experience that I think men only like to show the pretty girl off to other people.

I take this all quite seriously, and I’m trying my best to end up in a good future. Clearly not everyone is. To everyone working on ads, surveillance, gambling, secret research, enshittification, cloud lock-in, what are you doing with your life? Why are you selling out the future?

It doesn’t require everyone to stop, just enough people. And it starts with you.