nostrademons

Are you confusing Tor with a VPN? With a VPN the IP is a commercial server run by the VPN provider, and there's usually just one hop (you run an encrypted connection between yourself and the VPN provider that all your traffic is routed over, then the VPN provider makes outgoing connections on your behalf). You're relying on the VPN provider's promise to not log customer connections, but if they are honest, all connections seem like they're coming from the VPN provider and aren't traceable beyond that.

With Tor, you connect to a random Tor node (which may be another residential computer), which connects to a node in the middle, which connects to an exit node, which forwards your requests on to their final destination. All connections look like they come from the exit node, which might be someone's residential IP.

Everybody is saying that the Great Lakes region is a climate haven, but it's intensely vulnerable to both polar vortex events coming down from Canada and wet-bulb heat waves coming up from the Gulf of Mexico. The middle of America is flat; there is no natural barrier to stop a frigid or sweltering airmass from moving in. This has already happened multiple times within just the last few years.

In terms of death toll, a power outage during a severe deep freeze (with subzero temperatures and -50 wind chills) or a wet bulb event has a far greater lethality potential than wildfires. The most deadly wildfire in California history was the Camp fire that leveled the town of Paradise; it caused 85 deaths. The Oakland Hills firestorm of 1989, immortalized in SimCity and several made-for-TV movies, only caused 25 deaths. By contrast, the 1995 Chicago heat wave caused 740 fatalities, and even the 2024 polar vortex event that barely made headlines killed more people (90) than the Camp Fire.

(Also note the low absolute numbers of even very severe, headline-grabbing natural disasters in developed countries. And that good construction can mitigate almost all of them. You are probably better off, in strict actuarial terms, by settling in a place where you can make a lot of money and then using it to prepare your home for disaster than by settling in a place that is cheaper but where you can't afford the latest building technology.)

It's beneficial for the largest firms, the ones that can forge personal relationships with the administration, because it's detrimental to all other businesses. Competitive advantage and all that.

Note that Google stopped using it in IIRC 2006.

It wasn't because of the adversarial link farms, though, which are usually handled by trying to identify fake links and take them out of the computation in a preprocessing step. (There are many others parts of Google's '00s ranking algorithm that relied upon backlinks as well). It was because the web scaled to the point where PageRank couldn't process it, because the exact matrix solution to it is O(N^3). They replaced it with an iterative graph traversal algorithm from a set of ~1000 seeds which were themselves chosen through the original PageRank algorithm. I suspect this is published somewhere, because Gemini alludes to it when I ask what's the algorithmic complexity of PageRank.

Interestingly this is a common pattern that Google uses. Come up with a heuristic algorithm that works well enough to get your first million users. Then, train a machine-learned algorithm on the actual behavior of your first million users to scale to your next billion. Assistant's NLP was similar, where the first version had all these linguists hand-inputting grammars for all the different ways you might say a command, and then they just trained a much simpler neural net mapping utterance -> command once they had enough users to get dense training data.

The first time it happened (2023) it was a non-issue; he did exactly as you said, appointment happened as planned, and then he was back on his scheduled plane trip a week later.

Everything post-2025 has been a shit-show. There's no predictability or clear guidance, either from our lawyers or from the state department.

Predictability matters a lot in business. If you know what the procedure is and how long it will take, you can plan around it. If the rules change in the middle of the game you're left scrambling to keep up.

I've had a couple of reports that this applied to. It's incredibly disruptive. "I've got to fly back to India to renew my visa." "Okay, when will you be back?" "I have no idea, it depends when I can get an appointment. It could be next week, it might be in a month." "[State department] We're pausing all visa appointments." "Well fuck."

Meanwhile I've got a tech lead who is 12 hours away from the rest of the team. He was at least willing to work US-East hours while in India (basically staying up all night and sleeping during the day), but there was a huge amount of management overhead and inconvenience for the employee involved.

There are two purposes to jobs:

One is the obvious one, to produce things. AI can somewhat help with that, although less than you might think because it's largely producing knowledge output rather than physical things. If AI lets one knowledge worker do the work of 10, the effect of that is to free up 9 knowledge workers to take up manual labor, it's not to make manual laborers 10x more productive.

The second purpose is to distract people from attempting to work on their own interests and instead corral them into working for others'. You see examples of this in the form of acquihires (where they obviously don't need you to do the work, they already have a competing product, but they do need to take your product off the market), and in bullshit jobs, and in think tanks and consulting firms and legal machinery. It's a form of social control and a means to maintain social order, not a requirement for actual productivity. When you don't give people a job to do, they start working for their own interests, and their own interests frequently conflict with others. And we've seen that over and over again: the Great Depression nearly led to a fascist coup in the U.S. [1] and did result in fascism in Germany, Italy, and Spain, and large communist movements in France and China. The COVID pandemic and subsequent lockdowns led to widespread protests and civil unrest in the U.S. and other countries.

Popular discourse concentrates almost exclusively on the first purpose, because the second one is icky to talk about. In reality most people are individually better off being independently wealthy and not working, but society can only absorb so many free agents with their own ideas and wills to power. (This is also why retirement and stay-at-home parenting are broadly institutionalized: it's okay for people to not work when they no longer have the energy to work, it's only dangerous when people are not working and yet have the energy to do things other than working). But the second reason actually has more explanatory power. If you assume that the true reason for employment is to take people out of the competitive will-to-power market, lots and lots of irrational moves suddenly make sense, like high-paid executives, bullshit jobs, bureaucratic incompetence, projects where billions are invested and then the whole thing is canceled, competing internal divisions, people not getting fired for doing nothing, etc. The whole point is for them to do nothing.

There's a Chesterton's Fence [2] issue at work here. Many people, particularly young people, grow up with the first explanation for work and take it at face value, and then a new technology comes around that saves all this work and they're like "Of course this is a good idea! I want to be part of it!" But if the actual purpose of work is to distract people from pursuing their own interests, this is incredibly dangerous. The actual consequence is going to be widespread unrest and quite possibly war.

[1] https://en.wikipedia.org/wiki/Business_Plot

[2] https://fs.blog/chestertons-fence/

Yann LeCun's response is interesting. I've heard that the transformer architecture underlying LLMs is actually really effective for robotics, but progress is a lot slower here because instead of just scraping the web and Anna's Archive, you have to outfit workers with motion-capture suits that record their every move, and then have them physically do enough jobs that you can gather good training data on how the job is done. This is a little challenging when no blue-collar worker wants to be replaced by a robot.

Long-term there's probably a lot more potential, and much bigger markets, in robotics than in LLMs, but it will really take the long term to get there.

They can't, really. Putin's domestic power base will not let him suffer a defeat. His power is predicated on seeming all-powerful; once he doesn't, then other internal threats within Russia which have been kept at bay by fear will be emboldened, and the Russian state collapses. This is a worse security outcome for the U.S. than having a hostile dictator in charge of Russia; with the status quo, at least the enemy is visible, we know who has the nuclear launch codes, and his future actions can be reasoned about, while if a nuclear-armed state descends into anarchy we have no idea who the threat is and they have no incentive not to launch.

The mistake here is in seeing this as a clash of two nation-states that are each unified actors. Humans have agency; states are social constructs. It is quite possible for a state to disintegrate if the ideas holding it together are proven false. The Soviet Union did just that 35 years ago, and it could easily happen again this generation.

His strategy was to give Ukraine just enough weapons to prevent annihilation without giving them so many weapons that would allow them to win. Hence, quagmire. The weapon sales also came with conditions like not allowing them to strike deep within Russia itself, all of which prolongs the war without leading to a clear victor.

Biden gets Russia and Ukraine stuck in a quagmire in Ukraine, so Putin retaliates by getting the U.S. and Iran stuck in a quagmire in the Middle East. Wars are won by the countries that stay out of them. The Russia-Afghanistan war played a very significant role in the downfall of the Soviet Union.

The irony is that China is the big winner from all of this, simply by staying above the fray and making trade deals with all the countries that more bellicose superpowers start fighting. When your adversary is making a mistake, let him.

Not honest-to-god Tesla tiles, but a friend of mine put up Chinese knock-offs that he's pretty happy with. I think the problem here is "Tesla" rather than "solar roofs"; friend has a non-Tesla solar roof and it looks great, produces great, and has required little maintenance since he put it in 4 years ago.

I think it makes sense as soon as students have the fundamentals down. You have to be able to read, listen, make connections, think critically, devote sustained attention to a task, and manage your time effectively. Some students develop these skills in elementary school, and then starting in middle school, the flipped classroom approach can work. Some adults never develop these skills. Although arguably, people that never develop these skills will likely be sitting there zoning out in a traditional classroom lecture and not actually learn anything.

Also - 90s tech generally didn't try to charge you rent for existing. Either you paid them a price for some fairly tangible benefit to your life, or they gave you the service for free fueled by VC dollars (which we now recognize as nurturing dependence so they could charge you rent for existing, but it felt very different in the moment).

I think a lot of the complaints about tech these days come from the feeling that it's trying to own your life rather than improve your life. Things like tractors that won't go if you don't pay their subscription fees; cars that won't go if you don't pay their subscription fees; music collections that you don't actually possess and that go away if you don't pay their subscription fees; photo & backup services that make all your childhood memories go away if you don't pay their subscription fees; proprietary defaults that auto-opt-you-in to these services so you don't realize you can't access the photos without paying their subscription fees; SaaS vendors that hold your business hostage until you pay their subscription fees; search advertising that holds the path to your customers hostage unless all your profits go to them; search/AI engines that are ~70% accurate but don't provide any good way of seeking out alternative information; E-commerce platforms that take your money but give you junk instead; building contractors that take your money but give you junk instead; insurance companies that take your money but don't actually pay out claims; rent that keeps increasing by the maximum amount each year but never bothers to service habitability concerns with the place; and so on.

I think a lot of the frustration today comes from the feeling that people are treading water not to improve their lives, but to make somebody else rich, oftentimes people who aren't working at all.

AI is now more powerful than the people doing the prosecution. After all, those folks are using AI to make their legal briefs, and also for burning peoples' houses down with drones for that matter.

Welcome to our 21st century dystopia. Hope you survive.

This is typical at the college/university level - when I attended 20-25 years ago, we were expected to learn the material through readings outside of class, and then class time was spent on discussion, application, questions, and working through problems.

IMHO there's a lot to recommend this style. A lot of students learn better when they can wrestle with the material at their own pace, on their own time, in their own setting. It quells a lot of anxiety about "I'm not following what the teacher is saying, am I stuped, will I look bad in front of all my peers?" And then class time can be spent identifying holes in your knowledge and getting instant feedback from an expert, which is where they are most useful.

Not GP, but one of the challenges with debating whether LLMs are "conscious" is that we don't even really know what it means for a human to be "conscious", or even if consciousness is experienced by other humans the same way it is for ourselves.

What we do know: neurons carry electrical impulses across their synapses to trigger other neurons to fire, and more frequently used synapses are strengthened while infrequently used ones are pruned. This is not all that dissimilar to how a multi-layer perceptron is trained: it's floating point numbers in a big matrix rather than biological structures and electrical impulses, but there is still that element of frequently used connections being strengthened and infrequently used ones being pruned.

What we hypothesize but do not know: there is a thin brain structure of grey matter called the claustrum that has tendrils that reach into nearly every other brain structure. In many ways, this is similar to the attention mechanism of the transformer architecture. It is hypothesized that this may be the seat of consciousness, owing to experiments where electrical stimulation of the claustrum caused patients to immediately lose consciousness. However, there is no way to prove this, owing to the difficulty of otherwise removing or disabling the most connected structure in the brain and observing its effect on consciousness without permanently killing the patient.

Beyond that, we don't know much. I've got a family friend that's been a practicing therapist for 50 years, and I asked him what was the most interesting observation he made in his career. It was that "Everybody experiences the world in a different way, and yet everybody assumes that everyone else experiences the world the same way they do."

The optimal fit doesn't remain static.

The reason for all of [job hopping, fluid labor markets, bankruptcy, startup formation, inflation] is because the world doesn't stay the same. Desires change. New technologies are invented. Resources get depleted, and substitutes need to be found. Bottlenecks emerge. Old people die, and young people are born.

Changing wages and periodic layoffs are ways of adapting the jobs that people do to the new realities of which jobs need to be done.

This really depends on mindset.

If your mindset is "The economy is an incredibly dynamic, living thing whose purpose is to satisfy the consumer desires of the moment", then job-hopping can be all of very satisfying, very lucrative, and very purposeful. Your purpose is to do whatever is most needed. You don't get attached to any one task, but treat yourself as malleable and adaptable, and think of your past roles as a portfolio of skills and experiences that you can draw on to meet new challenges. You could describe your approach to work as "Work is something I do, not what I am."

If your mindset is "The economy is the society that I grew up in, and I'm seeking my place in it, and then I want a role where I can grow and build expertise", this is extremely unsettling. You view your job as an identity, a part of yourself. To leave that job is to leave a part of your identity behind, and to be fired or laid off is to have a part of your identity ripped away. And so you'll fight hard (and take many poor bargains) to avoid being put in that situation. It's not simply a matter of economics; it's a matter of being and belonging. Work is not just what you do, it is who you are.

Commerce vs. Guardian syndrome [1], or growth vs. fixed mindset [2]. There isn't really a right answer, but American culture, society, and business favors commerce syndrome over guardian syndrome, while many other cultures (really, most of the rest of the world) is the opposite.

[1] https://jebkinnison.com/2016/04/29/jane-jacobs-monstrous-hyb...

[2] https://online.hbs.edu/blog/post/growth-mindset-vs-fixed-min...

A product that lets airlines keep annoying people off them isn't worth enough to either the airlines or Google for it to be worth releasing.

My very first project at Google made them about $180M. It was canceled after about 3 years because it didn't move the needle. If it doesn't make at least a billion, it's not worth the distraction to the executives. I was like "Well can I quit and buy it back, I would love to make $180M", but no, it was too tied in to Google infrastructure to make a clean split, and would cost more in lawyers and exec attention to be worth it.

Most likely Google is using this so that they have actual training data on how a company runs internally. What kind of emails do people send? What actions are taken in response to them? What starts out as a chat and then becomes an email? What do communication architectures inside the company look like? Whose job is superfluous and doesn't actually add to the bottom line? Who does the same thing over and over again, which can be replaced by a computer and LLM?

They can then build this into the next release of Gemini and bundle it with their Cloud offerings, where it will generate much more than a billion dollars in sales to all sorts of organizations, without anyone much caring about the individual people who make up the data set.

It's a response to the GP, which makes the point that time in the field would still be spent with family. Time in the field with family is different from "spending time together" in the same sense that time on the computer adjacent to family is different from "spending time together". In both cases, your attention is on a task rather than on a person, you may be performing different tasks, and you're in close physical proximity but not necessarily doing the same thing.

Also I think many extroverts misunderstand introverts in this respect. My mom (introvert) and me (introvert) would often just read in the same room, and that was enough for us: I assure you, it felt very different than sitting with my butt against a complete stranger on the subway. My younger kids (introvert and ambivert) insist on digging their toes under my back while we sleep to make sure I haven't gone anywhere; they are not even conscious, but the proximity is important to them. This particular kid is an extrovert and has made about a dozen bids for my attention in the time between my previous comment and now (most of which have been reciprocated), but that is again the huge difference between sitting next to a family member and sitting next to a stranger on the subway. With family, that bond of trust is already there, and so you feel okay interrupting the silence to ask what your next move should be in Magic Sort, or why you can't download Township, or how do you turn the phone off silent mode, or showing me how he just beat his personal best score on 1 To 100, while you would never ask those of a stranger on the subway.

Time on computer is often spent "with family" too, eg. my son is home from school today and noodling around on my phone next to me while I handle a bunch of errands online.

I think that's going to be the central tension of the next few years. The article suggests that software becoming more secure is going to lead to renewed calls for a "legitimate" backdoor for law enforcement to access everything about the device. Well, that security means nothing if someone phishes Kash Patel, something which has already happened. You go through the weak link, which is almost always the human.

I could see a future where the government is replaced by the software system which has the fewest security flaws, simply because it is the last thing left standing as everybody hacks everybody else. We end up with a Matrix-like or Terminator-like future where humanity now lives in service of the machines.

Kinda, yes. Osama Bin Laden's express goal was to bankrupt the U.S. by provoking it into a war it could not win. Three wars in the Middle East later, and $33T in additional government debt, and hear we are.

This article (and the previous one) is a little weird, because nowhere in either of these is a discussion about why you would use an arena, and I'm not certain the author understands that very fundamental concept.

You use an arena when you have computation that might consume a bunch of scratch memory, and then you want to free all of that memory at once when the computation is done. It's basically dynamically-scoped memory, very similar to stack-allocating a big structure and letting the stack pointer free it, but without the requirement that you know the size of all the memory you're allocating at compile time, or that allocations follow a strict LIFO ordering. They can be very fast for the same reason that stack allocation and copying GC is very fast: it's just bumping an allocation pointer. But unlike copying GC, deallocation is fast too, because you just free the whole arena at once.

free() being a no-op with arenas follows naturally from that, and is basically the whole point.

But ArrayLists also not playing nicely also follows naturally from it. An ArrayList (or Vector) has its own dynamic memory management; it transparently reallocates when you run out of space. This is useful for convenience, but if you're aiming for performance - which is basically the main reason to use an Arena - you want to be a little bit more careful about copies and allocations.

Typically, the common pattern for arena-based allocation is that you have one pass to plan out & compute where everything goes (where you're allocating just metadata, and metadata is storing pointers, sizes, and lengths of the actual data), and then you do one pass to allocate and write out the output data structure. That way everything is copied no more than once: you copy and compute your results and write them directly to the output data structure, then free the arena with all the scratch work all at once. Object serialization is the canonical example, and indeed common serialization libraries like Protobufs or Apache Arrow are big users of arenas. But if you're just accumulating things in an ArrayList and letting it automatically resize, you're doing it wrong.

My wife wouldn't have considered this a business before LLMs. She's mostly just doing friends' parties, and oftentimes getting paid in-kind (i.e. instead of bringing a gift, she just facepaints all the neighborhood kids).

Remember that in the mid-00s, this is what YouTube and the social web was. Folks doing little, productive economic activity, usually unpaid, which can now reach an audience through technology. And now the top career aspiration in young kids is "Be a social media influencer." That's how disruptive innovations start: with little transactions that aren't considered business at all, which then grow until they subsume the economy that everybody knows of.

Makes me wonder how many of the complaints about capitalism are really about corruption, and if people would have a more favorable view of it if we didn't have this system of interlocking directorships, revolving public/private appointments, regulatory capture, and backroom handshakes between government and corporations that seems to shut many Americans out of the capitalist wealth-creation engine.

Actual example: my wife wanted a logo for the facepainting business she's starting with our 5yo, so she asked Gemini "I want a logo with these objects in it, drawn in these colors, with this business name." And it spits back a perfectly passable logo in 5 seconds. Done. No designer needed. No Figma needed. It cut out the entire value chain.

Is it as good as a professional could've done with Photoshop? No. But it's about 90-95% of the way there, which is good enough that neither her nor any customers would care.

A great deal of business is precisely this, tasks that need to be done but where you only care about "good enough" solutions. After all, a key principle of business is you don't outsource your competitive advantages. You pay people for the table stakes, the things that everybody needs but that they only need to be "good enough". LLMs can generate "good enough" facsimiles of a wide variety of fields.

If you want determinism, you ask Claude Code to write you a computer program to the desired specs, and then run that program over and over again, instead of renting the software from a SaaS vendor. I've done that a bunch - I'll show Claude the inputs, I'll tell it what output format I want, tell it the tech stack, and then have it write the program. Then I'll ask for a systemd service or crontab entry to run it repeatedly.

For a lot of tasks, classical deterministic computer programming is better than an LLM. It's just that LLMs are very good at classical deterministic computer programming now.

The leaked internal Google memo "We have no moat" [1] was really prescient. It foresaw all of these proprietary models losing out to open-weight models.

Entirely possible that none of the big incumbents ends up winning, economically.

[1] https://simonwillison.net/2023/May/4/no-moat/