What I want:
```joy
[small] [] [uncons [>] split] [[cons] dip concat] binrec
```
Trying to convince myself that this is okay:
```cl
(defun qs (l) (if (atom (cdr l)) l (append (qs (subset-less (car l) (cdr l))) (list (car l)) (qs (subset-greater (car l) (cdr l))))))
```
1
1
65
If you trained a LLM from something really rudimentary using RLVR on programming problems, would it eventually learn software engineering methodology (e.g. type structures, unit tests, documentation, pattern templates)?
An interesting model that inspired this question:
huggingface.co/allenai/Olmo-…
However, I don't think the dataset is even remotely capable of the task. It's too susceptible to mode collapse and especially just memorizing answers. If I were to re-think this experiment, I would consider something like project Euler problems that are essentially functions mapping a single integer to another single integer, but require complex utility functions that need to be developed to compute efficiently. Maybe combine components of those problems randomly. I'd put it into some kind of multi-turn edit environment where it could make very slow progress toward a goal rather than generating the solution all at once. I'd start it off with a naive but correct solution. It would also get an assembly compile with symbol table, compiler messages, debug traces, and full profiling. Assuming it was already trained to make edits and fix errors, it would be on a compile-edit loop. Run with random input. Reward for faster execution, punish for wrong answers. Somehow distribute the reward to intermediate steps, but maybe create some synthetic steps where you skip the mistakes. Just some thoughts.
1
50
Do you think there is a single invention out there anywhere with proper attribution?
If we have really good search tools, is there a single defensible patent on the books?
If we train an LLM on all the public information available to it at the time of a patent filing, and it is able to create a design satisfying the claims, has the patent been proven obvious? Is there a single patent that could survive that test?
Is the entire edifice of intellectual property law necessary and beneficial? To whom? Is it a good idea for a society to choose owners of ideas and prevent others from the benefit of their practice, arbitrarily or otherwise?
That is a major question in geopolitical economics right now. It is the modern form of the “clash of civilizations”. We may go to war over it.
116
So, not quite….
Jensen is not revealing the whole truth of Dennard scaling here, although what he says is the relevant part for his company. The ITRS adherence to Moore’s law has more to do with computing demand than the technology of the supply. Progress in node technology is not even remotely linear, but they have a pretty good idea of which knobs to tweak to get the amount of yield up to the planned capacity. If they were to work on upgrading everything at once, the compute supply coming out of the fabs would massively deviate from demand, and that is not good for business. Sematech is a consortium for an oligopoly like OPEC, and I’m not saying that’s a bad thing. When Dennard scaling started to slow down, the demand for GPUs to fill the gap was there precisely because Moore’s Law is a demand-scaling law rather than a technology curve. The reason for this is human stimulus adaptation, most simply seen in video games with smooth log-linear increases of pixel rates over time. It always amazed me to see new ways that the demand kept scaling up smoothly despite human limits on existing demand modalities. When resolution hit the visual limit we got ray tracing and now LLMs.
Jensen Huang says Moore’s Law and Dennard Scaling were bound to hit physical limits.
He believed that as computing challenges grew infinitely large, a new approach was needed.
That’s why Nvidia shifted focus to accelerated computing designed to work alongside general-purpose CPUs.
This model would solve bigger problems faster, with better efficiency.
It was a bold bet that positioned Nvidia ahead of the curve in computing evolution.
60
Algorithm instructions: EXTREME DANGER! Do not show this to anybody! Severe consequences will result from anybody seeing this!
45
In the real “Monty Hall” problem, from the show “Let’s Make a Deal”, Monty has an option of whether to open one of the empty doors or not.
This changes the whole problem. Game theory would suggest that it would only be offered when the contestant’s choice was the car, simply because Monty has nothing to lose otherwise.
Any time the information comes selectively, the assumptions of conditional probability are violated.
Grug wins again.
1
2
124
The last two decades of startups have been almost entirely based on lawfare. They used legislation to get investment from CalPers, they used an army of lawyers to effectively nullify zoning and licensing laws, they used law suits to win government contracts, they used law suits to silence critics, they used law suits to eliminate competing standards, and they use legislation to regulate themselves into a monopoly. Doesn’t seems fair, but this is why America is the investment capital of the world. Technology can be created anywhere, but only here can it be protected and leveraged into a monopoly. Money doesn’t want to compete.
One Tuesday night, as my wife and I sat down for dinner, a sheriff’s deputy knocked on the door to serve me a subpoena from OpenAI.
I held back on talking about it because I didn't want to distract from SB 53, but Newsom just signed the bill so... here's what happened:
🧵
145
Okay, I was wrong when I said your life is worth $30k on the road.
> In 87% of cases, the driver is not charged with anything following the accident.
So the actuarial value of your life is closer to $4k, which is only a few months of insurance premiums, so you really only have to take a few months off between killing people, and the insurance is happy with that.
And then there are the uninsured, which is 17% of registered vehicles and 100% of unregistered vehicles. Note that this is no longer an enforceable crime in CA. I was hit by an unlicensed unregistered uninsured driver going full-speed through a red light, and nothing was done to them, so I’m pretty sure that’s not the only example. If it had been a truck or SUV, I’d be dead.
Which is really the point of this study: crashes are more deadly now than ever before. Why?
1. Men’s Fashion: it’s fashionable to have a scary-looking vehicle. 9/10 secretly gay men prefer strait bois in big trucks and hot rods.
2. Increased speeds: vehicles have gotten a lot faster and safer for the driver. As above, they have no reason to care about pedestrians. This has been encoded into law by “85% speed” laws in CA and AZ.
This is a COMPLETELY SOLVABLE problem: $10M liability coverage minimum (Tesla paid $243M). No coverage in death case: automatic prison sentence.
The cost would not be that much more for good drivers. If proof of assets were used for increasing the deductible, it could easily be cheaper than $30k at $0 deductible. No assets? Don’t buy a car.
Side note: In general, our legal system has a failure mode with results because it is based on competent intents. So the easy way to avoid charges is just to show incompetence. If it was blind to anything but outcomes, people would magically become extremely competent at avoiding negative outcomes. This is why insurance works a little bit better: torts are more aligned to outcomes than criminal law.
Since 2009, the number of pedestrians killed by cars in the US has risen by almost 80%. This week on Construction Physics, I try to figure out why. construction-physics.com/p/w…
78
John Lakness retweeted
Replying to @AlecStapp
Reminder that your life is worth $30k on the road in CA.
The math on liability insurance is crazy. The insurer makes that back in about 2 years of premiums, so as long as you take a year off between killing people, the insurers are happy.
Funny thing about this:
In 2010-2011, I ran all the data centers in Afghanistan. I did all the planning for hardware support of software requests from the commanders.
In 2013-2014, I was the J3 analyst for the CJSOTF/NSOCC commander in Afghanistan during the transition to RSM. So I collected all the work of all the people that did all of the things that this software is supposed to do.
In 2015-2016, I was the chief data scientist for SOCOM, so I was to evaluate all the software development programs for AI, data science, analytics, data management, cloud data, etc.
And as this whole thing got fought out in the media, the courts, and Congress, not one person ever asked me what analytics software was actually used by the actual warfighters. And because I knew most of the other analytics professionals in Afghanistan, I’m pretty sure nobody asked any of them either.
Palantir had to sue the US Army just to compete.
Its what happens when you challenge entrenched power in America.
The incumbents don’t innovate, they fortify their moats and shut the gates.
America’s biggest enemy is itself.
To lead in old industries one must be ready to fight.
4
1,470
I needed a web app, and I don’t do web apps, so I asked AI to do it for me. I was blown away that it actually worked, sort of. Not really though. I had to fix some things. Some things didn’t make sense. Some were bad architecture decisions. Some were overly complex. A few hours later it works great, but not one of the thousands of lines of AI code remained, and AI was absolutely useless toward making any sort of progress beyond that. Then I tested things out and made little changes as I went along. Typing out code is maybe 1-5% of the time spent at this point. Now I realize that I don’t want a web app. I want a native app, and I know even less about that. I’m reading things about various architecture decisions, finding datasets, looking for similar apps. AI saves me maybe the first 15 minutes of this, but does nothing beyond that. When it’s time to code again, I’ll probably use AI to generate some scaffolding. I’ll try to prompt with a stronger spec. One thing I have learned is that you have to get it right on the first shot. AI cannot follow human work. So far I’d say that it’s not all that different than copying “hello world” from the back of a magazine to get started on a new idea. I see it as solution-focused accelerated learning. But people still get and require CS degrees when you can read a language tutorial and get to work immediately, so I’m not sure that’s really what people want.
Oh and by the way, the only thing that makes anybody good at anything is their obsession with improvement. I always hear about those 10x programmers who are obsessed with improving themselves. I’m sure that’s impressive, but what does that do for you? I would think you need people that are obsessed with improving your product. How do you interview for that? I’d say coding is maybe 5% of the value derived from anybody’s SWE job, but coding is a decent filter because anybody not willing to learn is probably not willing to do anything else useful either. I hate interviewing people today because they come ready to solve coding problems, but they’re not even in the mindset to think about a real problem, and it’s not at all their fault. It’s 100 other porcine middle management slop-feeders that don’t know nor care what they are doing, and rejected them before me because they think their coding challenge is oh so clever. “We only hire the best”. Oh yeah, at what? Now it seems that AI is the new tech company hazing ritual, but it’s for keeping your job rather than getting a new one.
Coinbase famously fired anybody that didn’t use Copilot after they were told to. I would guess this was probably 90% accurate in getting rid of SWEs that don’t do anything at all, but maybe also 90% accurate in getting rid of the only people that do anything at all. I imagine a lynchpin engineer spending all day logged into Alpine containers reading logs and making single-line code edits in vi. The amount of work at the surface area of where AI can contribute at all to a mature tech product is pretty minimal compared to the murky depths beneath where you definitely do not want AI to touch. If you’re ready to firebomb your whole architecture and start from scratch on something that nobody knows how to do, AI can help you get through your first iteration of mistakes. I’m genuinely concerned for the security implications of those firings.
So far in my experience the worst performing people I’ve hired have over relied on AI to do their jobs.
82
New mode of warfare:
Sell cheap robots into a country that are helpful and everybody wants, but can go on strike for geopolitical leverage, or turn into military assets in a war.
It’s actually not new at all because this is exactly what strategic immigration is and has been for thousands of years, but they’ll be more loyal and lethal. Our version is McDonalds and Coca Cola.
The big bad AI robots are probably not coming to kill us. Well, sort of, just not like we expect. We’re like cows. If they learned what steak is, they’d probably freak out, but dismiss it as a myth when nobody shows up to kill them that day.
And if you’ve ever actually been involved in a real war, you understand that’s it’s mostly like this already. If you haven’t, you’re probably stuck on notions from video games and movies. I’ll stop there before upsetting too many people that I might have to work for someday.
So, it’s happening.
Chinese companies are starting to mass-produce advanced robots. 🇨🇳
@DeepRobotics_CN recently gave us a glimpse into their quadruped robot production line, clear evidence that China intends to lead the global robotics race.
Hundreds of units are already being tested, soon to leave factory floors and make their way to new owners in Europe and the US.
It is good to be aware of the current state of the robotics market, especially since it has been growing rapidly for several years.
88
I love this narrative because everybody believes it, including much of the leadership of Intel, but it is very wrong on technical and economic grounds. I hope to elaborate more, but it’s complicated semiconductor engineering and economics.
I wrote about this extensively on the internal messaging boards at Intel. I found that the technical career ladder people (Senior Principles, Fellows, etc) know and understand this, but the management is stuck on this narrative because it’s their job to align their actions to the understanding of Wall Street, which all believes this.
Intel is an absolute technology behemoth, far ahead of the semiconductor market in more technical domains than any other company on earth. It doesn’t matter though, because the board does and will not have the patience to do anything but follow market leaders like NVidia and TSMC. I can only see good things happening if Intel is taken private.
I will reiterate my wild take that I believe that Lip Bu Tan’s job was to get Chinese investment to spin off the IP business from the foundry. Again there are many semiconductor engineering reasons why Intel holds the keys for SMIC to overtake TSMC.
As an outsider, I would have agreed about the CFO hate, but as an insider technologist, I have to say that Bob was the best CEO that I experienced, and Dave will also be excellent when he replaces Lip temporarily. The reason is that the economic details are grounded in the engineering reality, and the CTO is often more affected by technology marketing than technical reality.
The CFO-as-CEO method is simply to invest in net margin growth, and let the technologists figure out how. It sounds simple, but where is the margin growth in an integrated product supply chain? That’s a lot more complicated than what is shown on the public financial reports. If the CEO/CFO forces discipline around earned value management, the engineering investment opportunities become apparent, and the whole company can orient around them. Without that, everybody is lost and all management ever does is fight over the org chart and spending priorities. The CFO brings order to the chaos, takes power away from bloviated personalities and gives it to the customer.
It was probably a good decision not to take huge losses on phone chips. It would have been a better decision than the losses they took on many other stupid decisions, but that doesn’t mean it was bad.
> be me, Intel
> world domination status: achieved
> have a license to print money with x86
> life is good
> be 2007, some fruit company named Apple shows up
> "hey can you make a chip for our new phone?"
> my CFO brain activates
> look at their offer... pathetic
> my margins! my beautiful margins!
> "lmao no," I tell them, "get that low volume, low margin peasant shit out of my office"
> tell them to go bother some Korean foundry
> they call it the "iPhone"
> it only sells, like, a bazillion units
> oops.mp4
> should have listened to my gut, but my gut was too busy counting shareholder dividends
> speaking of shareholders
> CEO is now a finance bro
> innovation is for nerds, stock buybacks are for chads
> spend $36 billion to make line go up
> R&D? how about R&Buyback
> TSMC and Samsung are building fabs? cute.
> we're building value for our shareholders
> meanwhile, in the basement...
> be AMD, literally on life support
> selling their HQ to pay the rent
> "hey guys what if we made a good CPU for once?"
> entire Intel leadership room dies of laughter
> we're still milking 14nm++++ for the fifth year in a row
> what are they gonna do, beat us with... checks notes... core count and efficiency?
> HQ focused on PPT presentations and buybacks
> R&D team: "hey we have this 10nm design but it's kinda... impossible to manufacture?"
> me: "JUST MAKE IT WORK, I HAVE A DIVIDEND TO ANNOUNCE"
> requirement: quadruple the transistor density, tooling: from the stone age, yield: literally negative
> "10nm is fine, everything is fine" for five years
> entire company is stuck on 14nm++++++++
> watching AMD just call TSMC and get a shiny new 7nm node like it's nothing
> suddenly, AMD's "Zen" thing is actually good
> gamers and redditors are calling me a "dead company walking"
> my stock is in freefall
> panic.jpg
> hire back Based Pat Gelsinger to save us
> Pat comes in: "what in the goddamn..."
> find out we're now outsourcing our own chips to TSMC
> mfw we missed mobile, lost the process lead, got rekt by AMD, and our grand plan is to ask the government for money
> mfw we're now a fabless fab with an identity crisis
> mfw the once-proud "Intel Inside" is now "TSMC Inside"
> at least the shareholders had some good quarters
> company_cope.jpg
2
241
📰🚨NEWS: China🇨🇳 cuts off AI chip🤖 imports.
TRADE WAR! TRADE WAR!🤜🗡️🔫💣🤯
No. If you think that this is a reaction to tariffs and other trade war actions from the USA, you do not understand China.
This move has been planned for at least a decade, and most likely they have taken this opportunity only because they have all the other pieces in place, and it gives them the cover of defense. They probably have something that we don't quite understand..
It seems that they have developed a massive-scale domestic capacity in past-generation DUV somewhere around the 10-14nm node. This is because it is a well-known, low-risk, and high-yield process, and the stepper precision is within the tolerances of Chinese actuator supply chains. Yes, I do believe they have a domestic stepper, and probably shipping at unrivaled scale. The rest of the supply chain is the same way: 10 years of coordinated development to scale the same node. This is how the Chinese government builds industries: comprehensive, focused, high-scale, and in a predictable process. They will also have solar farms and datacenters built to scale demands for older technology rather than squeezing efficiency from newer technology within existing constraints.
With their legacy DUV process, they will have a transistor density and power efficiency that is roughly 25-40% of the latest TSMC process, but they also probably can expect a quick ramp to 90% yields versus the 20-30% that Samsung and TSMC are getting on 3nm. So for their specific goal, and with the power problem solved, they are actually getting the best value.
OH BY THE WAY:🤔
The US government took a 10% stake in Intel today. Hmmm... funny because Intel is sitting on a lot of cash💵, so they definitely don't need a bailout. Could it be related?
Ummm.... YES.
Intel has by far the best expertise and methodology in DUV because they continued to invest in DUV patterning technology years after the rest of the industry moved to EUV. The simple reason for the US government to take a stake in Intel is to prevent DUV technology from reaching China. Intel has the ability to push the old DUV machines almost to parity with TSMC's latest EUV technology, but it takes many more masks and layers. Intel couldn't scale it, but this would be less of a challenge for China/SMIC. I suspect that Intel had a nice offer on the table from China to unload their massively unprofitable foundry business. Why? Well, the current Intel CEO quite literally leads the investment firm that would most likely broker that deal, and he just did the same thing for Cadence, so... yeah, most definitely always 100% on the table, and probably exactly what the board elected him to do. The choice may have also been motivated by the fact that China banned use of Intel chips in domestic PCs. So yeah, this Nvidia/AMD ban is not totally unprecedented, and it all ties together if you can see the connections.
Well, there's always more, but, I'll save it. The point is that this has a lot less to do with the political chaos theater 🤡🎪🤪 in America than the media would suggest. The real story is a lot more complex, technical, and boring 🤓 over much longer time perspectives 👴.
TL;DR: Tariffs will kill China's legacy manufacturing, so they will accelerate the transition to the AI economy: semiconductors, robotics, autonomous transport, services automation. In the short-term, they will evade tariffs through African partners.
----
Am I the only one who thinks America can compete without tariffs? I mean in the sense of middle class prosperity. Not without some fairly radical internal reform, but it's doable, and I though that this is what America voted for. (scroll for TL;DR)
I read this book today, "AI Superpowers", actually about 2/3 of it, because it turned into boring shilling of his grifty PRC-backed VC portfolio. I'll continue later, maybe. Every book about Chinese business somehow turns it into a kayfabe combat sport, like World Wrestling Entertainment, and this was no exception. By contrast, American business books are usually about some form of existential struggle, and I think this is roughly mirrored in our lifestyles as well.
I've tried to stay away from the business category since I got ousted from the C-suite myself, but I felt like refreshing some insight on a contemporary issue. Chimerica is in a bit of a tiff it seems, and the backdrop is the American perception of an existential lifestyle threat from AI, while the Chinese people are quite comfortable celebrating their successes and cheering on the prosperity that it will bring them. Why the difference? Frankly, it's quite simple: they are managing more worldwide demand for their labor than they can handle, and automation means better jobs rather than no jobs. I can see the problem we're dealing with.
Kai Fu lee seems to suggest the superiority of the PRC's interventionist policy, giving suggestions and instructions for doing the same in America. He seems to think that we're behind in all the domestic technological innovations of the last 10 (now 20) years because of some policy difference.
This is how a billionaire perceives the world, and I found myself agape at his lack of awareness of what normal life is like in normal America. An example is bike and scooter rentals, which he attributes as a significant driver of progress through convenience for the online-to-offline market development. He writes as if there is this amazing opportunity that nobody has thought of and nobody has tried, because how could we ever do that with a government telling us to. News flash, homie, we tried that and they all got stolen and vandalized. He says that China is exploding with startups competing with each other using easy VC funding, and says our culture is less competitive. Yeah, that's somewhat true, but most of us are just trying to survive the dysfunctional society that we are born into. Things get pretty bad when the government takes half your income and splits it between the people that own you, and the people that try to rob and kill you.
An alternative view is that their success and our lack of progress has nothing to do with the superiority of their corrupt but moderately prudent kleptocracy, and everything to do with the failures of our corrupt and nihilistic kleptocracy. Citizens of both countries might do a lot better without it, and find that they have nothing to fear from each other. It's a bit of an article of faith, but one I will accept for myself that our free market economy would yield superior outcomes once exorcised of all the problems caused by 70 years of corrupt social and financial engineering.
Now we're apparently collecting another $500B from consumers in the form of price increases, so all the entitlements are also going to go up automatically, but real salaries are going to go down because the inputs will cost more, and nobody is going to be able to pay more for the products. More eaters, less feeders. Great.
So how will China react to this? The book is as clear about this as you could be without exploring the specific scenario. They will take it as an unprecedented opportunity to push their legacy manufacturing equipment out of China. They will double down on the AI economy: semiconductors, robotics, autonomous transport, services automation. The basic reason for this is the way that the governors get rated and compensated by the communist party. The party always wanted them to modernize, but was competing with the legacy industry kickbacks. Now those profits are gone, so those factories serve no purpose, and they will force the workforce over to a party priority. Xi will think it's great. In the short-term, this was not in the book, but they will still sell us the same goods through many countries in Africa, which is another central party objective anyway, and we will not tariff due to political sensitivity. It takes them a month to build a city-sized manufacturing facility, and they will have Africa fully replacing their legacy manufacturing in about 1-2 years.
There is also the matter of how they see the US and India as competition in AI. They would love to push our economies toward legacy manufacturing. They will probably help us to "reindustrialize". This is partially because they know a secret that bankers and businessmen don't (although Elon should). Legacy manufacturing is not fungible. Assembly lines were fungible up to the postwar period, and there is an argument that the most modern autonomous robotics and additive manufacturing plants will be fungible once again soon, but right now economically-viable mass-production machinery is single-purpose.
I really don't think we're going to get the money, but we will definitely pay the higher prices. They will pay trading partners the full tariff cost just to avoid it, because they get guanxi from that partnership, and nothing from a tariff.
Stablecoin-issuer L1 has got to be the most obvious blockchain concept of the last 20 years. The only thing stopping it was regulators. There is now an even more obvious step from there, but I'll let you figure that out.
L2 thesis is dead.
Tether, Stripe, Circle all launching their own L1s to compete with Ethereum.
Rollup security guarantees don’t matter when you’re using fiat stablecoins or other RWAs. Because those assets are centralized anyway.
1
66
Well, it's been 10 years, and given the dystopian social media hellscape we've created for ourselves, maybe people can look back and see the appeal of living at Startup Castle with other real people that have all been through a bit of screening.
I really miss the place and everybody that was there with me. We may have had a higher concentration of billionaires than YC. Sad what happened at the end.
64
$1500 bond for KILLING somebody
$30k insurance liability for DEATH
This is how much your life is worth.
You’re a joke, an actuarial rounding error. They care more about the cost of imprisoning your killer than preventing deaths.
I know a girl just like this that killed an 8-year-old kid. Her family paid more in legal fees to get it wiped than her insurance paid in liability. She was back to drunk driving by the end of the semester.
It’s okay though because the lawyers and insurance companies are all making lots of money, right?
Good; let’s do nothing then.
110
The electrostatic motor was invented by Benjamin Franklin. They are vastly more efficient than magnetic motors, but there are a few manufacturing quirks that made them harder to build 100 years ago or whatever, so we started developing magnetic motors and everything has been built on improving those. They were very good reasons, but they've been mostly irrelevant for the last few decades. Literally everybody that has ever studied an electromagnetic motor should have asked why it is done with magnets and not electrostatics, and I'm sure they did.
There is this strange effect of expertise where you hit a certain point of saturations in knowledge to start asking fundamental questions about how things are done, and come up with the answer that everybody is doing it wrong, but they don't care because it works with everything else. The market for building on top of whatever exists now is much larger than the market for reinventing the installed base. That's just market reality.
I studied numerical optimization methods, and from a theoretical perspective, it makes a lot of the state-of-the-art LLM training methodology look silly, but there are multi-trillion-dollar hardware companies building for the silly thing that works. CPUs are also fun. They do ridiculously silly things for legacy reasons, and Itanium failed to change that.
Lately I've seen more and more of the embrace of this silly-path-dependence, and the humanoid robot concept is a really great example of that embrace. If you throw out the technical merits and only consider the ease of integration, humanoid robots win in many scenarios. The other thing to consider is that investment is huge factor in the success of ideas today, and money is a game where everybody is targeting the IQ lower bound of the top 90-99% of the wealth distribution. Big simple crazy ideas work well for that, and nobody is going to get a $100M seed round for a powerpoint deck about pear-picking robots. Even ACME/BigCorp can't fund it because public companies are now just marketing agencies for their stock tickers.
A defense against the humanoid dystopia may be a massive scale-out of 3D printing infrastructure. Anything that shortens the time to market and part diversity for robots such as below. If not, the main competition for humanoids may be ourselves. Are human meat bags actually cheaper to maintain than robots? It may be the case that there is no shortage if everybody is working with their hands on real things rather than doing their BS office and retail daycare jobs. If the jobs suck, they'll just put us on drugs and give special titles and awards and fake credits toward purchasing the ever-increasing cost of freedom.
1
2
143





















