Even Tor Browser (/mullvad-browser) gave up trying to obscure the operating system though arguably they shouldn't have. There appear to be too many fingerprinting vectors.
I think they made the right call on that. It's unclear to me whether hiding the OS is even possible. There's just too much OS-specific behavior that happens inside (and outside) a web browser. It's hard to account for all of it.
OS rendering differences can likely betray you even when canvas extraction is blocked/noised. At least one tor-browser dev has publicly confirmed that you can't even hide the difference between X11 and Wayland[1], nevermind two entirely different OSes.
It's not surprising that JS would out you, what I am wondering if whether or not volunteering OS information is foolish when it may not be possible to determine without JS.
Why give up information when you don't have to? Some people disable JS at least some of the time.
Yeah, I generally agree with that viewpoint, but you are giving up that information, even without JS.
Your TCP stack can be fingerprinted through long enough observation. Windows, Mac, and Linux all look different on a network level. It's not as simple `if XYZ then OS = Windows`, it's more holistic/probabilistic, but it's possible nonetheless.
e.g. One thing that bugs me about arch linux is that they recently changed the kernel default TCP keepalive time to be shorter (much shorter[1]), making arch users stand out a lot. So, not only can a fingerprinter identify your OS, they might be able to pin down your exact distro based on TCP behavior alone.
I guess my point is that hiding the OS would be a massive amount of effort to plug a hole that cannot be plugged without effectively controlling the entire OS. TB/MVB is limited in what it can do by itself.
I believe they used to make the user-agent appear to belong to Windows but then they stopped doing it with the excuse that there are other ways to tell anyway.
The OS doesn't really matter, the amount of entropy it contains is very low. As long as the anonymity set of browser-users is large it's all good. And I believe Tor Browser accomplishes this objective.
When not doing privacy oriented cryptocurrency (cough money laundering cough) with ZKP's, if you really want private verification you are in a position where a single actor can authenticate the entire world and no one will know it happened. And to prevent it you assemble the pieces necessary to deanonymize anyone.
Make no mistake. ZKP age verification, as proposed, will just require multiple parties to collude to figure out your identity.
They can't even implement ZKP for remote attestation due to the auth-the-world problem.
If you read the link, that’s not necessarily how things have to work. You can make limited-use ZK schemes, plus you fall into the usual trap of binary thinking, that if something isn’t 100% effective it is necessarily 0%.
With ZKP age verification, services will not be able to track you without help from the CA. The CA will not be able to track you without help from the services. Both will contain the necessary information in their databases that when combined deanonymize you. The CA is the central authority/certificate authority.
So you should assume the government can track you, because you should assume both will be streaming those identifiers to it.
That's not really possible to explain in this space, unfortunately, but the overall idea is that there are mathematical techniques that allow you to prove that you have a valid certificate without revealing which one it is.
That would be ideal. However, this is tech proposal which takes so much of the slop out of the entire thing. With this implementation, there is no profit in it, unless your government is directly cooperating, aka a scandal in many countries.
The simplest worthwhile DIY sandbox you can have is to layer two tools: bwrap and gvisor.
bwrap args -- gvisor args do args -- /path/sandboxee args
bwrap will set up the environment and then gvisor elevates it into a true sandbox.
Standalone gvisor (not the 'do' subcommand) used to be a mess with the OCI json requirement, but recently they began work on presenting their own bwrap interface (likely to pursue AI agent uses) though I wouldn't use it myself yet.
People often look down on gvisor because they think it's some kind of syscall filter, it is not. It can use one of ptrace, seccomp or even KVM to intercept ALL syscalls and service them with it's own logic (which is in Go). Basically it's a VMM and kernel in one.
Any reason why you wouldn't use gVisor's bwrap interface yet? We're working on it precisely to make DIY sandboxing on Linux as easy as possible in order to get Linux-sandboxing-at-home to mature beyond the current syscall-filter-and-namespaces duct tape stage, so I'm curious to know what you'd like to see.
It just didn't seem fully baked yet, the 'do' subcommand works fine while the 'bwrap' alias has this problem: `bash: cannot set terminal process group (1): Not a tty`. When executing 'bash -li'. Also the EROFS feature of 'do' should probably be included in 'bwrap', it can be useful. Include overlay options.
Also some things you can do to make gvisor better are Wayland passthrough, vulkan support (or virtio native context). Being able to get gvisor to populate a network interface inside itself through a 'passt' (or 'containers/gvisor-tap-vsock') socket on the host would also be ergonomic. All of those are available on 'muvm' (based on libkrun) which if you have the time to set up is the next step in DIY sandboxing of graphical apps as well. See: <https://git.clan.lol/clan/munix>
Thanks. We're working on rootless network setup to make `runsc do --rootless` work with networking enabled when `passt` is installed right now. See issue #13337 (yes that's a cool issue number) which should unblock this.
The tty issue is known, should be fixed soon too, though contributions welcome as it sounds like it should be simple fix and we love more contributions :)
FWIW, X11 apps work well, I have a personal hacky project in which I've been running Librewolf in gVisor, with the window being reflected as a native Wayland window. It uses `Xvfb -fbdir` aimed at a bound tmpfs mount to get a shared memory region containing the window's pixel data which can be read directly from out of the sandbox, has Pulseaudio audio passthrough, and a socket server passing through mouse/keyboard events to make the window interactive. Works smoothly even for YouTube playback, and I successfully played a game of Unreal Tournament 2004 at 24fps in it, with no noticeable mouse/keyboard latency :)
We're basically making baby steps to get there less hackily.
That's good to hear! Hopefully the passt approach you are pursuing will include the ability to use an existing passt socket and not just launch one for you.
Wayland is tricky because there are memory buffers being shared between the compositor and the client. crosvm (also by google) adopted 2 custom solutions to it of which one got merged into mainline.
Achieving audio passthrough is trivial as it's just a unix socket. `-host-uds=all`
That's the approach I initially took, but experienced some combination of noticeable stuttering and latency regardless of which buffering strategy I tried... Had to switch to a shared memory ring buffer, along with some adaptive playback speed shenanigans (sometimes imperceptibly speeding up playback when falling behind production of audio samples, sometimes imperceptibly slowing down when there's less than a few milliseconds' worth of samples left in the ringbuffer), in order to achieve actually-gapless playback.
Ah, I see the distinction. In my case, I was using the PulseAudio client+server with a network sink running in-sandbox, producing a set of audio samples that are then replayed by a PulseAudio client in the out-of-sandbox Rust application receiving these samples. Exposing the socket directly to the sandbox is dangerous, e.g. it would let the sandbox record the host's microphone directly.
My goal was for the sandbox to only have access to a single dumb playback sink and no other access by default, and to gate any further access (e.g. microphone access) in something the out-of-sandbox software can catch and ask the user about before it is granted to the sandbox.
Properly configured (including strict seccomp) bwrap on its own will be sufficient 99% of the time. But ultimately you are at the mercy of the enormous kernel attack surface and the 0days that result from it.
If you do anything valuable and are compromised it may get brought to the attention of whoever organized the automated attack (ex. AI agent doing interesting proprietary work that installed something it shouldn't have, chat logs got uploaded and analyzed) and they will then sell you to someone with the 0days to extract more value from you. Assuming you didn't screw up and leave a back door open somewhere of course.
Identity theft is a thing. And if you gate a desirable commodity behind an identity it will become even more of a thing. There are 100's of millions of identities to steal.
Alpine is a good system for the boot/main rootfs because it's rock solid and has the most recent kernel. When glibc is required or when you just want to access the repositories of other distros you can unshare and pivot_root into the respective rootfs (using bubblewrap). There is also flatpak (uses bubblewrap internally).
You can also go the extra mile and use bubblewrap to initialize the desired mount environment and then enter a sandbox from it using virtualization (/dev/kvm). Some VMM's such as muvm[1] even allow for hardware acceleration and performant Wayland pass-through.
I believe it's a great deal worse than that. All the metacognitive insight we do have may just be confabulation and we are fooled into believing that we have it because the process for conjuring it is good at finding a plausible answer.
When you read about and observe the split-brain patient experiments the appropriate response is abject horror at the implications.
The Opus model as usual impresses. Gave it a paper link with bullet point instructions and constraints (while baiting it to perform some mind reading of my intentions) and it implemented production ready code + the requested attack simulations: <https://gist.github.com/coppsilgold/00d3cd490cb7f8ffc3fe5c1c...>
I'm sure the frontier labs figured out very clever ways to leverage user input and actions as data for training and signals for RL. DeepSeek wants in on the game.
Agreed. The amount of effectively annotated (possibly very sensitive) data that users are voluntarily shoving across the line seems worth losing some money over. I imagine that data is also not exactly safe from the Chinese government.
> Almost all human traits are partly genetic and partly due to the environment and/or random. If you could change the world and reduce the amount of randomness, then of course heritability would go up.
There has been a lot of effort to determine systematic environmental factors that would influence things like intelligence and while it's easy to do harm (lead exposure) it's all but impossible to do any good.
It implies that the only environment that matters is either purely random (truly random accidents, circumstances) or non-systematic (results from non-linear interaction of environment and genes).
When stated that way it almost feels like a tautology because this is what genes exist to do in the first place. To control the interactions of their vessel and environment to the maximum degree. And from the perspective of an individual gene, all the other genes are part of the environment too.
> There is no such thing as “true” heritability, independent of the contingent facts of our world.
It's uncomputable (need to run Monte Carlo simulations on a human life). All efforts are to approximate it.
What you're saying is completely accurate, but I'd add that it's all relative. Are you falling towards the ground, or is the ground falling towards you? For instance malnutrition lowers IQ, in both directions. There is an inverse correlation between IQ and BMI, but what's most interesting is that that correlation has maintained just as strong even as obesity rates skyrocketed, which is suggestive that there's probably something causal, in some direction, somewhere in there.
And so in modern times if it turns out that eating less than most people apparently want to contributes to IQ, are you doing something good by eating less, or are they doing something bad by eating more? I think it's basically the same thing, just looked at in different ways.
Or are smarter people better able to regulate their food intake? (Either innately, or because society gives them other privileges which makes them less likely to overeat)
I would say, that on the whole, this has to do with habituated impulse control and self-restraint.
Classical writers speak of this as well, things like how inordinate and undisciplined appetites (not just for food, mind you; sex, too, and undue acquisitiveness of all sorts, for instance) darken the mind. What is inordinate and undisciplined is not proportioned or directed by reason. So, such character traits are rooted in fidelity to reason which means that not only do they avoid the aforementioned darkening of the mind by moderation of appetite, but the very character strength of being able to do so enables rational existence in other things.
Innate intelligence doesn't secure discipline. Indeed, it gives the person a bigger footgun and allows for more elaborate rationalizations of vice.
Which then begs the question, what is IQ actually measuring - something more like innate intelligence, or a fairly big slice of learned, habituated test-taking ability?
Regardless of what underlying trait it's actually measuring, the habituation factory is a big component of its supposed bias - that is, has your background taught you the kind of problem-solving habits that will help you to post the best possible score?
> Which then begs the question, what is IQ actually measuring - something more like innate intelligence, or a fairly big slice of learned, habituated test-taking ability?
> "It is a variable that summarizes positive correlations among different cognitive tasks, reflecting the assertion that an individual's performance on one type of cognitive task tends to be comparable to that person's performance on other kinds of cognitive tasks."
In other words people who are good at these tests are also good at real world tasks. Meaning IQ measures much more than one's ability to pass an IQ test. There are of course many examples of poorly structured IQ tests (including people re-taking the same IQ test and doing better at it the second time around). However a well-structured IQ test presenting novel questions (absent popular culture references and trivia) provides a very good approximation of the g factor in almost all cases. This means high IQ is highly correlated with things like income, unemployment, crime, homelessness, addiction, divorce, and many other objectively measurable life outcomes.
There is room for a philosophical debate about what g factor is, but it is beyond contest at this stage that g factor is real, and IQ almost always does a very good job of measuring it.
It's beyond contest that the g factor is real in part because it's a statistical inevitability in any series of related tests, be they for intelligence or product/market fit in automobiles. It's an exploratory statistic, a hint at underlying causality; it is not a dispositive revelation about the structure of human thought.
Sure: there's a battery of general cognitive tests, and if you smush data sets together a dominant factor will emerge. And?
This is exactly my point. It's tautological to argue that IQ isn't correlated with intelligence. It is, definitionally. You seem to acknowledge this so perhaps I don't understand your original comment.
(I misread your original comment --- no, it's not tautological. It's tautological to say that IQ isn't correlated with IQ test results. "g" has nothing to do with "intelligence" as a construct.)
If IQ tests measure one's ability to do most other cognitively demanding tasks then it's measuring more than the test itself. Whatever you want to call that thing is irrelevant. In sociology and psychology we call that the g factor. The vast majority of people call that intelligence.
That's the tautological argument! Note that I'm not even really engaging with the validity of IQ tests, just the surface logic that you're using. You literally just argued, in effect, that IQ tests measure intelligence because people call them intelligence tests!
That's a semantic point you're making, and I'm not contesting it. To repeat myself, you can call the correlation anything you like. I'm merely explaining that the correlation exists and is incredibly material.
We agree the correlation exists. The incredible materiality of it is what's in dispute. The obvious problem with these arguments is that people point to the correlation and say "see, I'm right about the materiality!".
We have strong data to show that people who score higher on IQ tests also earn more, have higher net wealth, have lower rates of unemployment, higher graduation rates, lower suicide rates, lower rates of addiction, lower rates of crime, lower rates of divorce, lower rates of fatherlessness, lower rates of household abuse, higher rates of home ownership, longer life spans, have better general health on almost every metric, and a host of other positive QoL indicators. I don't see how you could argue that is not material.
> which is suggestive that there's probably something causal, in some direction, somewhere in there.
Perhaps suggestive, but far from conclusive (I know you know this too). To me, it is suggestive that there is likely some other factor that may explain the relationship better, but then again, I am wrong more often than right, so what do I know? ;)
For example, compare that to growing wealth inequality, and I wouldn't be surprised if that is a potential factor. Less income = less access to care, less access to healthier food options, perhaps less time to for self-care, etc., and if wealth/career potential is gatekept by academic achievement, economic utility, or intelligence, then I can see the two, intelligence and BMI, being correlated, but not directly causal. Though, no study would give people large sums of money to improve their lives, so I doubt we will know for certain.
The point is not that it is something hard to compute that we can only approximate. The point is that there is no well-defined heritability independent of the environmental distribution.
I am not sure I understand the comparison. But let's say we found out that a rock is too heavy for a person to lift on Earth. Does it mean that it is too heavy to lift on the moon? No. And if we find out that something is or is not heritable, does it mean that in new circumstances such as a new policy or some sort of change of norms the heritability will stay the same? No. That's something that many people find counterintuitive.
I have no problem finding whether this rock is too heavy to lift for me in these specific circumstances. But if we try to define if it's too heavy in general, it becomes not well defined. Specific traits have causal development pathways that theoretically should allow to predict genes/environment interaction, but heritability uses population-level statistics.
This is demonstrably untrue. IQ has increased consistently for decades, far faster than genetic factors can explain. Environmental factors like education, nutrition, and medical care are the obvious explanation.
Maybe across the whole population because most people were struggling to eat enough and received almost no education.
If we compared average modern humans against average well fed and educated ones from 200 years ago would that still hold up?
I suspect the average college educated human from 1800 would obliterate the average college educated human from 2026.
> At the elite level, marathon performance is defined by energy availability as much as physiology.
> Maintaining a pace of 2:50 per kilometer requires a constant supply of fuel. Even small disruptions in energy delivery can result in significant time loss.
coppsilgold is the one who made a hard-line, clear-cut dichotomy when they said "it's easy to do harm [but] it's all but impossible to do any good". bglazer referenced several interventions that are known to increase IQ which challenge this dichotomy. Saying that it is difficult to separate "doing good" and "stop doing harm" is agreeing with the point that coppsilgold created a distinction without a difference.
This also assumes that IQ testing has remained static. It has not. IQ tests continue to evolve and there are >1 of them and they do not all agree. I.E. the tests themselves might be responsible for some of the variance.
it's hard to separate IQ decreasing and return to mean with IQ stabilizing
in 20th century most of the world moved past famine and toxins - did any factor of similar scale happen in 21st century as well to start looking for opposite processes?
> while it's easy to do harm (lead exposure) it's all but impossible to do any good.
That's just a meaningless statement no different from "while it's easy to subtract negative numbers, it's all but impossible to add positive numbers."
> or non-systematic (results from non-linear interaction of environment and genes).
Non-linear interaction does not mean non-systematic. Computer programs are fully deterministic (and therefore "systematic") while being non-linear (and therefore generally unpredictable). It is true to say that when things are non-linear it's hard to tell with certainty what effect some policy will have, but given that most human systems are non-linear, this is true for just about everything.
I was actually curious about this myself back when everyone was chiming in about how it was physically impossible.
This is first and foremost an engineering problem as you need to design a system that will both tolerate high heat and be able to pump even more heat to the radiators. The high temperature seems to be the primary objective to design for unless launch costs become absurdly low.
It's not "impossible" but so hard, complex, and expensive that any "gains" you get from being in space are nothing compared to the costs you pay for being in space.
I.e. it's not worth it.
The cost of launching 100K servers, each of which needs 20m^2 each of radiator (for a single H200 server), or 250 m^2 for a GB200 rack!
Ok but these numbers are for a single server or single rack, now what about a standard cluster size of like... 50k GPUs?
You would need (with optimal idealized efficiencies) roughly 64000 m^2 of space to cool down your space data-center. That's 9 American football fields of double sided radiator panels! For a single data-center, and realistically there would be inefficiencies and wastage so it could end up more like 20 American football fields of cooling needed.
