This makes the Moon a very high altitude aircraft.
Not according to the rectal boost regime of one of Dyson's infamous astrochickens.
The K'rman line lies at an altitude of 100 kilometres above the Earth's sea level, and commonly represents the boundary between the Earth's atmosphere and outer space.
The line is named after Theodore von Karm'n (1881–1963), a Hungarian-American engineer and physicist.
He was the first to calculate that around this altitude, the atmosphere becomes too thin to support aeronautical flight, because a vehicle at this altitude would have to travel faster than orbital velocity to derive sufficient aerodynamic lift to support itself (neglecting centrifugal force).
[*] I decided to split the difference on the two instances of the twin a-acutes.
Today some other Slashdot story had "zebra cosplay" in the story summary. Sheesh. Our collective nerd mojo is a quart down.
If you're determined enough as a 3-letter agency to get in, then you can also disappear the person.
Yes, of course. The good, old TLA infinite-budget porn.
Your position in the security food chain determines how much they are willing to spend. Even well-healed Q-class spooks answer to an ROI at scale.
Of course, part of the signal about your rung on the security food chain is determined by how effectively you armour yourself with effective prophylaxis.
This is why security culture can only work as a public good, wherein everyone on principle uses the highest caliber of security practical. When security is practiced exclusively on an as-needed basis, it only helps to paint a more accurate bull's eye on your backside.
All the TLAs must surely love the useful idiots who distract from the economic model that prevails here, by ranting at high pitch about naked capabilities, as there are no endemic constraints on their side of the fence.
I guess I do more cut and paste than your average keyboard monkey, but here's one more tip.
When you need to select more text than fits into your window, don't do that horrible slither-drag where you take your mouse outside of the text window, causing your window to continuously smooth scroll in the desired direction.
At least on my system, the smooth scroll is never smooth enough, and if you're the least bit impatient, you never see the target coming until you overshoot.
For long selections, I almost always start by positioning my target at the bottom of my 23" portrait screen, then begin the selection by double-clicking the center of the last word of the last sentence, thus to begin an upward word-mode selection drag. Go up an inch or two to establish a visible region, then pause.
Maintaining your drag operation (with the mouse button held down), press Page Up until your target scrolls onto your screen, then complete the mouse drag with the page at rest. I begin at the bottom of my monitor so that ending the selection is almost always an upward motion into my primary field of view (the very bottom edge of my display is barely an inch above desk height, which is not my favourite sight line). Of course, Page Down will correct an overshoot.
[*] In my Firefox, there's a bug where scrolling under your mouse with the keyboard does not update your selection region if the mouse remains motionless relative to your display (rather than your document). If you experience this problem too, jostle the mouse a bit while doing this to the selection region visible.
[**] There's a second bug in my Firefox: I can't do Pg Up select in a text input box without into disjoint, multi-select mode, and for incomprehensible regions, too.
One more thing. My security mode in Firefox means that many scripts don't run, and my layout is often fubar.
It's not unusual for one end of my desired selection to wind up underneath some other text object, where the other text region grabs the initial double-click.
Again, start at the other end, if possible, and extend your selection into the overlap (your selection is now securely anchored to the right text layer). For some reason, it's often hard to get the mouse to end such a selection precisely where you mean to.
Once again, the shift arrows to extend your selection in the text-flow domain instead of layout domain are your eternal friends. But here you'll find many good reasons to use ctrl-shift arrow to advance your selection in full word units.
By starting your drag at an internal letter boundary within a word, you can get half a word at the starting end, and still advance your selection region in speedy word units. And then you can release the ctrl key at the end, and fine tune the end selection in character unit, too.
New discovery
Actually, I already knew that you could resume mouse drag by pressing the shift key before clicking down, and that if your original selection was in word mode (or line/para mode) it would resume your original mode as soon as the mouse began to move.
What I hadn't noticed is that right at the point of click, it selects exactly where you click (middle of a word, no problem).
You can actually start your selection with the shift key depressed (one to three clicks), release the mouse button, travel to your destination as a letter boundary, then single-click to extend your selection to exactly that point.
In fact, you can single-click to drop an invisible selection anchor, then shift click at the other end, and the whole selection appears all at once (but you don't get any feedback on the accuracy of your initial click if you do this, not until the full selection is complete).
I suppose I could monkey around with visible text carets in my Firefox settings. But that doesn't seem necessary for now.
It would actually be nice in some ways to double click on the first word alone (release the mouse), then double click on the final word (this
I've been using a Logitech G5 Laser as my primary desktop mouse for close to a decade now, on an indigo Func Industries Surface 1030 mouse pad of the same vintage (smooth side).
Both remain optimal for desktop work.
Apparently there were some G5 mice that lacked the thumb buttons. Mine has two thumb buttons, and a tilt wheel, and the DPI controls.
I'm right handed, but I switched my mouse to my left hand a long, long time ago to reduce back pain. By doing so I position my mouse hand closer to my midline: my ancient Compaq keyboard has a full numeric keypad on the right hand side, which adds an extra seven full inches to my hand travel distance (6" home-to-mouse on the left side, 13" from home-to-mouse on the right side).
Irony: this makes the "thumb" buttons relatively useless.
While the weight cartridge can be important to some, the heart of the G5 is its 2000 dpi laser engine via a 6.4 megapixels/second image processor.
The ability to customize the dpi and USB polling rates is another huge plus with the G5.
You can configure up to five different dpi sensitivity settings between 400 and 2000 with the ability to set the horizontal and vertical sensitivities separately.
The USB polling rates can be set from 125 Hz to 1000 Hz based on six preset numbers.
The G5 defaults to 500 Hz which is a setting that we found offered the best combination of performance and compatibility across several different chipsets.
Reading more closely, I discover that the my G5 is the second edition of the G5, which is basically an MX518 upgrade product, with the same internal engine.
———
My experience
I'm using maximal ballast weights which for me greatly improves proprioceptive feedback. Because this is a super sensitive mouse, I have my mouse response cranked up almost to ludicrous speed.
I have 45" of horizontal display travel (three 23" monitors, one in landscape, two in portrait). My full-bore mouse flick (45" bezel-to-bezel) measures just under 2" on my mouse pad. My 23" vertical throw on my portrait monitor measures just under an inch. (These are consistent numbers.) When I creep across, my horizontal throw measures closer to 3" instead of 2". (Turns out, I have far more speed and far less acceleration than I believed, prior to making this measurement just now.)
Interesting algebra: 100 dpi screen resolution * 20:1 fast-movement mouse response ratio = 2000 dpi mouse resolution requirement to address single pixels. Three resolutions available on mouse: 400/800/2000. For the 13:1 slow-movement response ratio (usual speed at the landing site), I'd need 1300 dpi for single pixel address. (Meaning that I do have to switch down to slow speed to access individual pixels, but once I do, I have a decent margin.)
I'm using roughly 5 square inches of mouse mobility to manage three 23" screens in text-selection mode, and ranging over all of 2 square inches for window and focus management.
Super important tip: use double-click drag to select text in full-word mode as often as possible. I always aim for the middle of my target word, double-click to select the full word (do not release the double-click) and then drag to select a word region, aiming for the middle of the final word, where I finally release the mouse button.
Sometimes one end of your text selection contains a weird punctuation mark, which makes for a narrow target. There are three solutions for this. The first solution is painfully precise initial aim (down to a single letter). This is bad. Try reversing the selection by starting at the other end (it's fairly rare that both ends are problematic). Painfully-precise final aim is still better than painfully-precise initial aim, because you're usually coming in much slower after starting the selection, and
For some reason, I dumped a clause in a weird place. It should have read:
If some current implementation of AVX-512 has two issue pipelines, that works out around eight double precision terms per clock cycle, if you really maxed this out.
Also, I kind of ignored integer floor thinking there might be a trick as simple as a walking bit mask (perhaps not).
However, the resultant store should be fully parallel, although we might be exceeding L1 store bandwidth hammering these wide registers out on nearly every cycle).
Also, there's horrific precision loss here on my assumed double precision, and it actually turns out everything much past this row is integer per force: [76] 3.416455e+15 5.527940e+15 8.944394e+15 1.447233e+16 2.341673e+16
Beyond this point you don't even need the add 0.5 part (you might as well add 0).
Anyways, forget the algorithm, and hire the guy or gal who posts the "actual correct rounding is god awful" link as part of his or her solution.
Before getting to your comment, I wrote three lines of R code, and determined that the pow(-0.618034,n)/sqrt(5) term is below rounding error for all n.
If you scroll down in the link you gave, above, to the "rounding" section, turns out my "amazing" discovery was previously known:-)
If some current implementation of AVX-512 has two issue pipelines, you basically need one vector multiply by constant register (rep(phi^k)) and one vector add with constant register (rep(0.5)) to compute eight terms (k=8) double precision.
A minor amount of cleverness to achieve roughly a 4-way interleave accounts for pipeline latencies (one way to slice the cat, you can think of this as k=32 over four working registers concatenated).
Close to eight double precision terms per clock cycle, if you really maxed this out. Which is sad, because prior to term 1500, your values are in well in excess of e+307. [1471] 1.178511e+307 1.906872e+307 3.085383e+307 4.992255e+307 Inf [1476] Inf Inf Inf Inf Inf
Nice, in one way, because now you don't have to analyze for numeric instability.
But damn. The fully optimized AVX-512 implementation is a thirty hard minutes of brow sweat to achieve those 200 gloriously productive clock cycles (storing the resultant vector in memory, definitely not printing the values out).
Assuming AVX-512 at 2 GHz (in practical silicon, this mode is likely to be deturboed), the 30 minute programming time to 100 ns execution time ratio comes in at a hefty 18 billion to one (just using one core).
By 1975, its researchers had invented a personal computer with a graphical user interface that was almost a decade ahead of its time.
With a price to match: ten of thousands of dollars per workstation at that point in time (likely in excess of $100,000 each in 2019 USD). Moreover, you couldn't run this without a dedicated support staff, because it was extremely raw technology.
It wasn't just a decade ahead of its time: it was a full decade ahead of any viable market. Commercializing this beast was a creative act of the first magnitude, all by itself. Xerox had very little expertise to offer in pinching pennies to hit consumer price points.
If they form a joint venture with some Steve Jobs figure down the road, Xerox probably turns into Daddy Bigbucks as the project goes over budget time and time again. You can't license this to a young, upstart, thrifty company like Apple Inc., because Apple certainly did not have the cash on hand at that time to pay hefty licensing fees.
What actually took the market by storm was the IBM PC shitbox, where tiny amounts of memory were suited to an appalling limited operating system. (We're looking at you, MS-DOS.)
How do you win the installed-base software war of the early 1980s, bootstrapping the world with Smalltalk? You can find a price and performance point for a Smalltalk system that will move hundreds of thousands of boxes per month, as the IBM PC later did? Before something else an order of magnitude less sophisticated (at a quarter to one tenth of the price) gains complete market control?
Normally, in technology, polishing something up for market is the other 90%.
But in this case, Xerox was multiple 90% efforts away from a viable sales model, if there was any such model at all.
Probably their best inroad to the future was to build a line of Xerox LaserWriters spanning desktop to enterprise, while pricing the desktop model so attractively that they rarely ever sold the enterprise model (except to displace a fleet of expensive Xerox copiers).
And then somehow you try to cram your LaserWriter authoring software onto any cheap-ass PC client that comes along. Not that IBM wouldn't change the API underneath your hands if you got too big and powerful as a result. So it's better if you own the cheap PC client hardware, too. But this is not a business Xerox could feasibly have entered. $$$ ran in their blood. Good grief, what other kind of company would have a research center with a $100,000 toy stuffed under every desk, ten years ahead of any viable market strategy?
Sure, Xerox built PARC because they were secretly Walmart at heart.
And I've got an Ethernet bridge to sell you, with 16 glorious switched ports of 10BASE5 coax.
Deep neural networks barely made it through a decade-long siege of Leningrad where it became so unfashionable it was almost left to die in the snow. Is that your definition of "instant gratification"?
Humans are equally terrible at articulating many of our fundamental skills. Even grand master chess players only manage to articulate a pedagogical narrative, and not the real thing.
It does bug me sometimes that people forget that 90% of the reason we like our machines is they provide complementary abilities: massive databases with total recall, blinding fast arithmetic, rarely ever making an error, sub-microseconds reaction times rather than tens of milliseconds. Where we're at now is substituting mechanical systems that overlap key human competences, where the mechanical system is nowhere near as good on many dimensions, but nowhere near as erratic as human performance, either.
Finally, wherever did this idea originate that big messy systems were going to have clean analytic decompositions?
Back in the 1950s the excuse for this view was that when you only have a hammer, everything looks like a nail. When you're limited to a few kilobytes of memory, the computer is applicable to a few classes of extremely analytic systems, where no part is giant and messy. But actually, DNN systems for machine translation require hundreds of megabytes. Because human language is extremely messy. In NLP, the GOFAI agenda was only ever aimed at some kind of highly constrained conlang, which encapsulated a dense, proposition nucleus (completely bereft of metaphor) entirely unlike any human language ever spoken.
At no point in the last forty years have I not regarded GOFAI as some kind of adolescent SF fantasy reified.
Do you look at Winograd's work from 1970 and see a glass half full or a glass half empty? It was cool for its day, but as a software engineer, I always thought to myself "this dog doesn't scale". And I was right. There was no era of SHRDLU 2.0 or SHRDLU 3.0. The analytic complexity in this domain scaled far faster than the analytic ingenuity of Terry Winograd's graduate students.
So much for Lisp. Then along came Prolog: another scaling disaster.
Perhaps once we refine the DNN and invent the first DNN rectifier (mapping a messy world onto a clean, orthogonal conceptual world) maybe we'll finally find a good home for the kind of cleverness we once thought of as the whole AI cheese plate.
I'm not going to read that article again just now, but as I recall it, the detection algorithm is not detecting objects at the two main LIGO installations independently, so that the detections corroborate each other, but combining both signals into what amounts to a single instrument (basically into a single sigma budget, rather than separate sigma budgets).
Secondly, the search is template-driven, scanning for exactly the kinds of ring-downs they expect (hope) to find.
Between these two things, it's certainly possible into deceiving yourself into thinking you've detected something you haven't detected.
(I haven't followed up on this data analysis challenge recently.)
Finally, the cosmic directionality of the two LIGO machines is terrible. I forget the exact number, but between the two machines, you get something like a giant banana whose length is 20% of the sky.
That provides an introduction, but does not quantify the banana in square degrees that I can see on a quick revisit.
Ideally we would have four machines, and the machines would be partitioned for independent detection. Once the detection is confirmed to the same sigma twice, then all four machines can be combined into a single directional assessment, and then we can get hot onto neutron-star mergers in visible light.
Double funny, because as science writing goes, this is good. Every sentence has a job, and does a job. No mealy mouthed phrasing. Round pegs in round holes, square pegs in square holes. Easily in the top quartile of Slashdot science summaries.
However, one nit: compared to the abstract someone else posted, they err in talking about "the" drug, when the experiment seems to have covered three related compounds, only two of which exhibited the vaunted memory-improvement profile.
Use common sense when you choose things you put into your body.
Good lord, welcome to the middle ages.
Is it common sense that the fructose half of sucrose is metabolized in the liver by much the same pathway that processes ethanol (which if abused, in either case, contributes to fatty liver disease)?
No, it is not.
Is it common sense that the pancreas contains a melatonin receptor, so that your metabolic response to carbohydrates varies throughout the day?
Is it common sense that ulcers are mainly caused by Heliobacter pylori?
Is the effect of Toxoplasma gondii on motorcyclist and mouse behaviour common sense?
Is it common sense that fecal microbiota transplants would prove more effective in treating C. difficile than vancomycin?
Is it common sense that wholesome fresh fish potentially contains toxic levels of methyl mercury that bio-accumulate in adipose tissues?
Is it common sense that the high-productivity crops introduced during the agricultural revolution (not yet using GMO breeding techniques) remain as nutritious as the original heirloom crops?
(Besides, that was a trick question. There were three separate agricultural revolutions as human population exploded, so there are—logically—three entirely different tiers of heirloom throwbacks; the only reason this hasn't shown up at a Whole Foods near you is that Amazon's marketrons have yet to figure out how to make Silver Heirloom, Gold Heirloom, and Platinum Heirloom sound appetizing—though it does accurately reflect viable price points, given the associated yields.)
Diet is super important. We can't go around making naive assumptions. Neither can we trust failed epidemiology to untangle these incredibly complex signals. However, from the microbiome (and proteomics) much truth shall flow, even if it proves to be slow going.
By the way, Einstein was no Einstein in school, and not really a very good mathematician.
I know someone very smart, who is currently doing graduate work in logic, who actually bothered to go to a library (perhaps it was Princeton) to take a boo through some of Einstein's original manuscripts.
Somewhere in there, he found something like nine pages of notes over the course of which Einstein essentially taught himself four-dimensional differential geometry. He said it was an extremely efficient self-course, setting a pace he couldn't imagine himself.
Not long ago I audited about five hours of Susskind's introduction to GR. (About 200 hours of Susskind's lectures are available on YouTube.)
Adding the Lorentz transform was pretty straightforward, but then when you add accelerating frames of reference, you're left with a deep problem, which actually stumped Einstein for some while.
Eventually, he wrote down the Einstein metric:
G mu nu = R mu nu - 1/2 R g mu nu and the rest was history.
Susskind commented that this was quite a bit worse that QED, because gravitation self-interacts far more than EM (I think his analogy was to imagine photons that also carry charge).
Neither of these anecdotes in any way supports the idea of Einstein as a weak mathematician, though clearly his intuition in writing down the right problem greatly exceeded his formal abilities.
My friend concluded that what Einstein really when he commented something to the effect of "if you think you have problems, they're nothing compared to mine" was relative to the task at hand: inventing a whole new metric tensor.
Furthermore, Einstein probably was Einstein in school, it's just that no teacher ever set a test in writing down the right problems (rather than the right answers). Having such a gift at writing down the right problem, one can imagine why he didn't exert himself in the competition to write down answers to the tired problems of yesterday.
This can be viewed through the economic lens of comparative advantage at an individual scale. You might just be the best person alive on the planet at writing down the right problem (this is not easy). Should you invest your marginal effort in developing that capacity, or in polishing apple's for your teacher, using a skillset where you are definitively ordinary (formalism) as compared to Poincare or Riemann? Where being merely Poincare or Riemann would be a definite step down, as compared to your one true gift.
I don't really care what these companies charge, so long as every point of payment offers a discount to people electing to pay by a competitive method with a lower cost structure (one of these may involve cash).
Is trying to get an interview on an ageing and declining website the best you guys can think of for advertising?
I don't agree. There's an question in the air about whether the social media world needs to slap the hand of one giant bad apple (Facebook), or seriously retrench from a failed, life-destroying technology across the board.
If not Facebook, then what else? And so you interview the "what else" contenders in single file.
Silly response: Who's ever even heard of network x not equal to Facebook?
That's actually a huge part of the problem under discussion here, about what makes the existing landscape quite so toxic and evil.
You need to use completely independant words to actually get a good passphrase, and if someone doesn't understand the information entropy theory behind it...
It's not just the information, it's also the systems theory behind the information theory.
The underlying problem is that so many passwords in the wild get cracked back to plain text. Any paradigm you come up with is vulnerable to machine learning, which can ultimately identify and extract almost any pattern.
The pattern you describe is this: pick a word somewhat randomly (but not one too long or too hard to type), then use that as a seed word to free associate. Your entropy estimate is good and to my eye occupies the Fermi-estimation sweet spot (which is rather large, in a good way).
But the entropy you report is the conditional entropy on having already decided on the password paradigm (in some cases, this isn't even a guess, if you can associate previously cracked passwords to the same user, or specific password policies to the institution).
With enough randomly chosen paradigms to pick among, you could add maybe another 10 bits of true entropy. But people being people, paradigms are about as randomly chosen as social media networks.
Plus, because of all the cracked passwords, we have strong statistical models about paradigm evolution (these are cultural artifacts, for the most part)—at least for the masses who associate themselves with 2nd-rate IT (my own strong password paradigm is only used on sites highly likely to number among the scrypt enlightened).
Not that I'm including simple plug-board scrambling stages in my paradigm model (such as reflecting keystrokes between the left and right hands, moving home position one keyboard position to the right, or one row up).
A lot depends on the yield model of the attacker. If the goal is to crack as many passwords as possible, then you start with the worst of the worst and increment upwards. You would probably never even rise to plug-board stages.
If you have a value model over the accounts, when some accounts are judged to be a thousand or a million times more valuable once cracked, then the high value targets had better not be depending on their password paradigm adding any true security (unless so invented out of whole cloth, you're reluctant to update your password in less than a decade).
I tend to use apg to seed my passwords. I click generate five or ten times until something grabs my eye, and then I tweak it slightly to make it easier to store in short-term memory as I transcribe it from my password keeper to the passbox by hand. My "throw away" passwords are 11 characters, and more important passwords are usually more like 16. I generally estimate my passwords to have at least 50 true bits of entropy, assuming a very efficient search through password paradigm probability space (possibly by an adversary who has already cracked one of my other passwords into plain text). If the password contains spans of alphabetic dictionary fragments, I rarely estimate the fragment as supplying more than about 13 bits regardless of the letter count (or how close it remains to the original apg output).
Fully directed attacks based on a comprehensive model of human entropy management is an awesome superpower.
I'm sure the NSA has been doing this for decades already: nearly every plaintext password they've ever recovered (more than few) has been melded into some giant statistical model. They've likely identified millions of cognitive paradigms by now (from the ones with billions of breaks, down to some with only ten or so exemplars). There's no doubt in my mind they have explored the use of machine learning to squeeze even more out of this heap (though I suspect it was pretty squeezed out, even before machine learning). I also suspect that as this model is refined, they deliberated re-target recalcitrant breaks from ages past, so as to feed these breaks back in
Can an AI spam filter distinguish this output from your least-gifted regular correspondent?
Can you tell the difference without actually rubbing two brain cells together (never mind that it doesn't take twenty)?
Because this rubbish generator scales like Tribbles evolved into a Borg empire diaspora. And remember: this is day one. Like cracks in cryptography, it only improves from here.
Furthermore, it won't just be your email feed, but nearly anywhere one potentially encounters text (ingredient lists on your groceries are somewhat immune, with their thirty different synonyms for sugar and spice—aka sucrose and MSG—already alive and well, and 2 g +/- 0.5 out of a 15 g "suggested serving" as an established level of numerical precision). Hint: for the sugar and spice line items (there could be many), freely substitute the top of the bracket. Exhausting way to shop? Glad you noticed.
So there's at least one happy thought: it isn't going to break what's already broken much worse than it's already broken. It's just everything else that's now in play.
The first twenty posts on this thread didn't display the vaguest clue about the actual threat vector of concern.
Possible reasons to do things are more common than jelly beans under sofa cushions. Dislodge the sofa cushion of your imagination, then stoutly hold forth your outstretched apron skirt to receive a giant payload.
Here's a starting point: people add search terms to narrow the field of search when seeking something you dimly recall, but where the precise contents (such as personal names) have slipped from mind.
If you're trying to recall the design of some cute top, "female friends and gay friends with flamboyant, effeminate style" is your winning ticket. But we're still five years away from machine learning being able to navigate "gay friends with flamboyant, effeminate style" so for now you leave that portion out.
We're all susceptible to advertising (if we see it).
What's the evidence for this oft-repeated claim? That we behave differently in a culture saturation-bombed with advertising than when wearing a saffron robe in a remote retreat hidden away in the inaccessible hills of northern India? I've never had a clear picture how the A/B groups are factored, here.
I know that advertising affects me. This is because when I notice advertising, I make a conscious mental note that the brand is overpriced, due to a high cost structure, where the CMO earns more than the CTO, and has a budget to match.
For every hundreds small nudges I don't notice, there's a giant black mark somewhere else. Revenue vector per ad dollar invested, as measured by my own spending habits: pennies at most, possibly even negative.
I have basically configured my browser so that I go hours between advertisements. I have images and figures and call-outs disabled on almost every site I visit through user CSS scripts, which by now number in the hundreds. 90% of my browsing is pure text mode down the central corridor (if the side flow is too garish, I disable the masthead and footer, too, which means I can only navigate the site through Google search, and not by any internal click-bait).
I control advertising in my personal environment the way a marathon runner controls junk calories.
21st Century Psalm: My mind is my temple, I shall not be browbeat in my own home.
Of course, you can't exactly cancel any pervasive signal to zero by any reasonable degree of human effort. But you can reduce the advertisement signal on the order of 15 dB if you're a technically astute ornery mule.
Your badge just stops working in the door, and your stuff on your desk is either packed up and at the front desk... or is likely at a local pawn shop.
Those in the full time business of making shit flow downhill (like it needed any additional help) typically colour within the lines very carefully.
It's not good business to wind up in small claims court playing the visible role of a heartless Ebenezer. Colouring carefully between the legal lines is the best way to remain in the black in the shitflow business over the long haul. (The at-will American employment regime poses no real incentive to a ruthless management culture in the arena of gratuitous abuse; 90% of what you want to do is entirely legal.)
Thus I take these kinds of sharecropping stories with a giant grain of salt. Yeah, there's always one asshole in the barrel (possibly compensated as badly as you are) who goes above and beyond, but this is hardly the norm.
The Joyent cloud features a second layer of isolation. Sometimes you see this described as "double-hulled virtualization". The OS performance penalty to achieve this is low to non-existent due to the nature of BSD zones (hardened jails).
This is precisely the scenario that Joyent's technology exists to mitigate.
You think you're running Linux containers, but under the hood you've also got zones and ZFS snapshots.
There is a resource penalty involved in using a high-integrity file system like ZFS, (efficient copy-on-write requires extensive write-buffer coalescing) but it's often not a large one compared to the many gains in administrative security and ease.
Slashdot Mirror
That's a dog whistle for anal rape.
Dog whistles are themselves the spittle rain bird of all hat, no cowboy.
Because this kind of lip-licking line-item assuredly never comes to pass IRL.
Not according to the rectal boost regime of one of Dyson's infamous astrochickens.
[*] I decided to split the difference on the two instances of the twin a-acutes.
Today some other Slashdot story had "zebra cosplay" in the story summary. Sheesh. Our collective nerd mojo is a quart down.
Yes, of course. The good, old TLA infinite-budget porn.
Your position in the security food chain determines how much they are willing to spend. Even well-healed Q-class spooks answer to an ROI at scale.
Of course, part of the signal about your rung on the security food chain is determined by how effectively you armour yourself with effective prophylaxis.
This is why security culture can only work as a public good, wherein everyone on principle uses the highest caliber of security practical. When security is practiced exclusively on an as-needed basis, it only helps to paint a more accurate bull's eye on your backside.
All the TLAs must surely love the useful idiots who distract from the economic model that prevails here, by ranting at high pitch about naked capabilities, as there are no endemic constraints on their side of the fence.
I guess I do more cut and paste than your average keyboard monkey, but here's one more tip.
When you need to select more text than fits into your window, don't do that horrible slither-drag where you take your mouse outside of the text window, causing your window to continuously smooth scroll in the desired direction.
At least on my system, the smooth scroll is never smooth enough, and if you're the least bit impatient, you never see the target coming until you overshoot.
For long selections, I almost always start by positioning my target at the bottom of my 23" portrait screen, then begin the selection by double-clicking the center of the last word of the last sentence, thus to begin an upward word-mode selection drag. Go up an inch or two to establish a visible region, then pause.
Maintaining your drag operation (with the mouse button held down), press Page Up until your target scrolls onto your screen, then complete the mouse drag with the page at rest. I begin at the bottom of my monitor so that ending the selection is almost always an upward motion into my primary field of view (the very bottom edge of my display is barely an inch above desk height, which is not my favourite sight line). Of course, Page Down will correct an overshoot.
[*] In my Firefox, there's a bug where scrolling under your mouse with the keyboard does not update your selection region if the mouse remains motionless relative to your display (rather than your document). If you experience this problem too, jostle the mouse a bit while doing this to the selection region visible.
[**] There's a second bug in my Firefox: I can't do Pg Up select in a text input box without into disjoint, multi-select mode, and for incomprehensible regions, too.
One more thing. My security mode in Firefox means that many scripts don't run, and my layout is often fubar.
It's not unusual for one end of my desired selection to wind up underneath some other text object, where the other text region grabs the initial double-click.
Again, start at the other end, if possible, and extend your selection into the overlap (your selection is now securely anchored to the right text layer). For some reason, it's often hard to get the mouse to end such a selection precisely where you mean to.
Once again, the shift arrows to extend your selection in the text-flow domain instead of layout domain are your eternal friends. But here you'll find many good reasons to use ctrl-shift arrow to advance your selection in full word units.
By starting your drag at an internal letter boundary within a word, you can get half a word at the starting end, and still advance your selection region in speedy word units. And then you can release the ctrl key at the end, and fine tune the end selection in character unit, too.
New discovery
Actually, I already knew that you could resume mouse drag by pressing the shift key before clicking down, and that if your original selection was in word mode (or line/para mode) it would resume your original mode as soon as the mouse began to move.
What I hadn't noticed is that right at the point of click, it selects exactly where you click (middle of a word, no problem).
You can actually start your selection with the shift key depressed (one to three clicks), release the mouse button, travel to your destination as a letter boundary, then single-click to extend your selection to exactly that point.
In fact, you can single-click to drop an invisible selection anchor, then shift click at the other end, and the whole selection appears all at once (but you don't get any feedback on the accuracy of your initial click if you do this, not until the full selection is complete).
I suppose I could monkey around with visible text carets in my Firefox settings. But that doesn't seem necessary for now.
It would actually be nice in some ways to double click on the first word alone (release the mouse), then double click on the final word (this
I've been using a Logitech G5 Laser as my primary desktop mouse for close to a decade now, on an indigo Func Industries Surface 1030 mouse pad of the same vintage (smooth side).
Both remain optimal for desktop work.
Apparently there were some G5 mice that lacked the thumb buttons. Mine has two thumb buttons, and a tilt wheel, and the DPI controls.
I'm right handed, but I switched my mouse to my left hand a long, long time ago to reduce back pain. By doing so I position my mouse hand closer to my midline: my ancient Compaq keyboard has a full numeric keypad on the right hand side, which adds an extra seven full inches to my hand travel distance (6" home-to-mouse on the left side, 13" from home-to-mouse on the right side).
Irony: this makes the "thumb" buttons relatively useless.
Logitech G5 Laser Mouse: When an update is not worthy of a new name — July 2007
Reading more closely, I discover that the my G5 is the second edition of the G5, which is basically an MX518 upgrade product, with the same internal engine.
———
My experience
I'm using maximal ballast weights which for me greatly improves proprioceptive feedback. Because this is a super sensitive mouse, I have my mouse response cranked up almost to ludicrous speed.
I have 45" of horizontal display travel (three 23" monitors, one in landscape, two in portrait). My full-bore mouse flick (45" bezel-to-bezel) measures just under 2" on my mouse pad. My 23" vertical throw on my portrait monitor measures just under an inch. (These are consistent numbers.) When I creep across, my horizontal throw measures closer to 3" instead of 2". (Turns out, I have far more speed and far less acceleration than I believed, prior to making this measurement just now.)
Interesting algebra: 100 dpi screen resolution * 20:1 fast-movement mouse response ratio = 2000 dpi mouse resolution requirement to address single pixels. Three resolutions available on mouse: 400/800/2000. For the 13:1 slow-movement response ratio (usual speed at the landing site), I'd need 1300 dpi for single pixel address. (Meaning that I do have to switch down to slow speed to access individual pixels, but once I do, I have a decent margin.)
I'm using roughly 5 square inches of mouse mobility to manage three 23" screens in text-selection mode, and ranging over all of 2 square inches for window and focus management.
Super important tip: use double-click drag to select text in full-word mode as often as possible. I always aim for the middle of my target word, double-click to select the full word (do not release the double-click) and then drag to select a word region, aiming for the middle of the final word, where I finally release the mouse button.
Sometimes one end of your text selection contains a weird punctuation mark, which makes for a narrow target. There are three solutions for this. The first solution is painfully precise initial aim (down to a single letter). This is bad. Try reversing the selection by starting at the other end (it's fairly rare that both ends are problematic). Painfully-precise final aim is still better than painfully-precise initial aim, because you're usually coming in much slower after starting the selection, and
For some reason, I dumped a clause in a weird place. It should have read:
Also, I kind of ignored integer floor thinking there might be a trick as simple as a walking bit mask (perhaps not).
However, the resultant store should be fully parallel, although we might be exceeding L1 store bandwidth hammering these wide registers out on nearly every cycle).
Turns out, proper truncation is god awful:
Harder than it looks: rounding float to nearest integer, part 1 — May 2013
Also, there's horrific precision loss here on my assumed double precision, and it actually turns out everything much past this row is integer per force:
[76] 3.416455e+15 5.527940e+15 8.944394e+15 1.447233e+16 2.341673e+16
Beyond this point you don't even need the add 0.5 part (you might as well add 0).
Anyways, forget the algorithm, and hire the guy or gal who posts the "actual correct rounding is god awful" link as part of his or her solution.
Before getting to your comment, I wrote three lines of R code, and determined that the pow(-0.618034,n)/sqrt(5) term is below rounding error for all n.
If you scroll down in the link you gave, above, to the "rounding" section, turns out my "amazing" discovery was previously known :-)
If some current implementation of AVX-512 has two issue pipelines, you basically need one vector multiply by constant register (rep(phi^k)) and one vector add with constant register (rep(0.5)) to compute eight terms (k=8) double precision.
A minor amount of cleverness to achieve roughly a 4-way interleave accounts for pipeline latencies (one way to slice the cat, you can think of this as k=32 over four working registers concatenated).
Close to eight double precision terms per clock cycle, if you really maxed this out. Which is sad, because prior to term 1500, your values are in well in excess of e+307.
[1471] 1.178511e+307 1.906872e+307 3.085383e+307 4.992255e+307 Inf
[1476] Inf Inf Inf Inf Inf
Nice, in one way, because now you don't have to analyze for numeric instability.
But damn. The fully optimized AVX-512 implementation is a thirty hard minutes of brow sweat to achieve those 200 gloriously productive clock cycles (storing the resultant vector in memory, definitely not printing the values out).
Assuming AVX-512 at 2 GHz (in practical silicon, this mode is likely to be deturboed), the 30 minute programming time to 100 ns execution time ratio comes in at a hefty 18 billion to one (just using one core).
You are so FIRED!
With a price to match: ten of thousands of dollars per workstation at that point in time (likely in excess of $100,000 each in 2019 USD). Moreover, you couldn't run this without a dedicated support staff, because it was extremely raw technology.
It wasn't just a decade ahead of its time: it was a full decade ahead of any viable market. Commercializing this beast was a creative act of the first magnitude, all by itself. Xerox had very little expertise to offer in pinching pennies to hit consumer price points.
If they form a joint venture with some Steve Jobs figure down the road, Xerox probably turns into Daddy Bigbucks as the project goes over budget time and time again. You can't license this to a young, upstart, thrifty company like Apple Inc., because Apple certainly did not have the cash on hand at that time to pay hefty licensing fees.
What actually took the market by storm was the IBM PC shitbox, where tiny amounts of memory were suited to an appalling limited operating system. (We're looking at you, MS-DOS.)
How do you win the installed-base software war of the early 1980s, bootstrapping the world with Smalltalk? You can find a price and performance point for a Smalltalk system that will move hundreds of thousands of boxes per month, as the IBM PC later did? Before something else an order of magnitude less sophisticated (at a quarter to one tenth of the price) gains complete market control?
Normally, in technology, polishing something up for market is the other 90%.
But in this case, Xerox was multiple 90% efforts away from a viable sales model, if there was any such model at all.
Probably their best inroad to the future was to build a line of Xerox LaserWriters spanning desktop to enterprise, while pricing the desktop model so attractively that they rarely ever sold the enterprise model (except to displace a fleet of expensive Xerox copiers).
And then somehow you try to cram your LaserWriter authoring software onto any cheap-ass PC client that comes along. Not that IBM wouldn't change the API underneath your hands if you got too big and powerful as a result. So it's better if you own the cheap PC client hardware, too. But this is not a business Xerox could feasibly have entered. $$$ ran in their blood. Good grief, what other kind of company would have a research center with a $100,000 toy stuffed under every desk, ten years ahead of any viable market strategy?
Sure, Xerox built PARC because they were secretly Walmart at heart.
And I've got an Ethernet bridge to sell you, with 16 glorious switched ports of 10BASE5 coax.
People who spend money merely because it is sitting around rarely have $100 in cash on hand (much less $100 billion).
Ostrich leather ad hominem: attack not on the man, but the man's bulging pocketbook.
Deep neural networks barely made it through a decade-long siege of Leningrad where it became so unfashionable it was almost left to die in the snow. Is that your definition of "instant gratification"?
Humans are equally terrible at articulating many of our fundamental skills. Even grand master chess players only manage to articulate a pedagogical narrative, and not the real thing.
It does bug me sometimes that people forget that 90% of the reason we like our machines is they provide complementary abilities: massive databases with total recall, blinding fast arithmetic, rarely ever making an error, sub-microseconds reaction times rather than tens of milliseconds. Where we're at now is substituting mechanical systems that overlap key human competences, where the mechanical system is nowhere near as good on many dimensions, but nowhere near as erratic as human performance, either.
Finally, wherever did this idea originate that big messy systems were going to have clean analytic decompositions?
Back in the 1950s the excuse for this view was that when you only have a hammer, everything looks like a nail. When you're limited to a few kilobytes of memory, the computer is applicable to a few classes of extremely analytic systems, where no part is giant and messy. But actually, DNN systems for machine translation require hundreds of megabytes. Because human language is extremely messy. In NLP, the GOFAI agenda was only ever aimed at some kind of highly constrained conlang, which encapsulated a dense, proposition nucleus (completely bereft of metaphor) entirely unlike any human language ever spoken.
At no point in the last forty years have I not regarded GOFAI as some kind of adolescent SF fantasy reified.
Do you look at Winograd's work from 1970 and see a glass half full or a glass half empty? It was cool for its day, but as a software engineer, I always thought to myself "this dog doesn't scale". And I was right. There was no era of SHRDLU 2.0 or SHRDLU 3.0. The analytic complexity in this domain scaled far faster than the analytic ingenuity of Terry Winograd's graduate students.
So much for Lisp. Then along came Prolog: another scaling disaster.
Perhaps once we refine the DNN and invent the first DNN rectifier (mapping a messy world onto a clean, orthogonal conceptual world) maybe we'll finally find a good home for the kind of cleverness we once thought of as the whole AI cheese plate.
Boy did you not get the memo on how this thing actually works.
LIGO searches for an extremely precise signal known as a "ring down" which is entirely unlike any kind of dump truck doing anything dump trucks do.
The problem is somewhat different: a dump truck plus exactly the right random noise might produce a nun-bun artifact in the shape of a ring down.
So it certainly helps to corroborate detections by having multiple detectors.
Grave doubts over LIGO's discovery of gravitational waves — 31 October 2018
I'm not going to read that article again just now, but as I recall it, the detection algorithm is not detecting objects at the two main LIGO installations independently, so that the detections corroborate each other, but combining both signals into what amounts to a single instrument (basically into a single sigma budget, rather than separate sigma budgets).
Secondly, the search is template-driven, scanning for exactly the kinds of ring-downs they expect (hope) to find.
Between these two things, it's certainly possible into deceiving yourself into thinking you've detected something you haven't detected.
(I haven't followed up on this data analysis challenge recently.)
Finally, the cosmic directionality of the two LIGO machines is terrible. I forget the exact number, but between the two machines, you get something like a giant banana whose length is 20% of the sky.
The Direction of LIGO's Gravitational Waves — 6 March 2016
That provides an introduction, but does not quantify the banana in square degrees that I can see on a quick revisit.
Ideally we would have four machines, and the machines would be partitioned for independent detection. Once the detection is confirmed to the same sigma twice, then all four machines can be combined into a single directional assessment, and then we can get hot onto neutron-star mergers in visible light.
Double funny, because as science writing goes, this is good. Every sentence has a job, and does a job. No mealy mouthed phrasing. Round pegs in round holes, square pegs in square holes. Easily in the top quartile of Slashdot science summaries.
However, one nit: compared to the abstract someone else posted, they err in talking about "the" drug, when the experiment seems to have covered three related compounds, only two of which exhibited the vaunted memory-improvement profile.
Good lord, welcome to the middle ages.
Is it common sense that the fructose half of sucrose is metabolized in the liver by much the same pathway that processes ethanol (which if abused, in either case, contributes to fatty liver disease)?
No, it is not.
Is it common sense that the pancreas contains a melatonin receptor, so that your metabolic response to carbohydrates varies throughout the day?
Is it common sense that ulcers are mainly caused by Heliobacter pylori?
Is the effect of Toxoplasma gondii on motorcyclist and mouse behaviour common sense?
Is it common sense that fecal microbiota transplants would prove more effective in treating C. difficile than vancomycin?
Is it common sense that wholesome fresh fish potentially contains toxic levels of methyl mercury that bio-accumulate in adipose tissues?
Is it common sense that the high-productivity crops introduced during the agricultural revolution (not yet using GMO breeding techniques) remain as nutritious as the original heirloom crops?
(Besides, that was a trick question. There were three separate agricultural revolutions as human population exploded, so there are—logically—three entirely different tiers of heirloom throwbacks; the only reason this hasn't shown up at a Whole Foods near you is that Amazon's marketrons have yet to figure out how to make Silver Heirloom, Gold Heirloom, and Platinum Heirloom sound appetizing—though it does accurately reflect viable price points, given the associated yields.)
Diet is super important. We can't go around making naive assumptions. Neither can we trust failed epidemiology to untangle these incredibly complex signals. However, from the microbiome (and proteomics) much truth shall flow, even if it proves to be slow going.
I know someone very smart, who is currently doing graduate work in logic, who actually bothered to go to a library (perhaps it was Princeton) to take a boo through some of Einstein's original manuscripts.
Somewhere in there, he found something like nine pages of notes over the course of which Einstein essentially taught himself four-dimensional differential geometry. He said it was an extremely efficient self-course, setting a pace he couldn't imagine himself.
Not long ago I audited about five hours of Susskind's introduction to GR. (About 200 hours of Susskind's lectures are available on YouTube.)
Adding the Lorentz transform was pretty straightforward, but then when you add accelerating frames of reference, you're left with a deep problem, which actually stumped Einstein for some while.
Eventually, he wrote down the Einstein metric:
G mu nu = R mu nu - 1/2 R g mu nu
and the rest was history.
Susskind commented that this was quite a bit worse that QED, because gravitation self-interacts far more than EM (I think his analogy was to imagine photons that also carry charge).
Neither of these anecdotes in any way supports the idea of Einstein as a weak mathematician, though clearly his intuition in writing down the right problem greatly exceeded his formal abilities.
My friend concluded that what Einstein really when he commented something to the effect of "if you think you have problems, they're nothing compared to mine" was relative to the task at hand: inventing a whole new metric tensor.
Furthermore, Einstein probably was Einstein in school, it's just that no teacher ever set a test in writing down the right problems (rather than the right answers). Having such a gift at writing down the right problem, one can imagine why he didn't exert himself in the competition to write down answers to the tired problems of yesterday.
This can be viewed through the economic lens of comparative advantage at an individual scale. You might just be the best person alive on the planet at writing down the right problem (this is not easy). Should you invest your marginal effort in developing that capacity, or in polishing apple's for your teacher, using a skillset where you are definitively ordinary (formalism) as compared to Poincare or Riemann? Where being merely Poincare or Riemann would be a definite step down, as compared to your one true gift.
I don't really care what these companies charge, so long as every point of payment offers a discount to people electing to pay by a competitive method with a lower cost structure (one of these may involve cash).
Cash Might Be King, but They Don't Care
And there's the problem in a nutshell.
I don't agree. There's an question in the air about whether the social media world needs to slap the hand of one giant bad apple (Facebook), or seriously retrench from a failed, life-destroying technology across the board.
If not Facebook, then what else? And so you interview the "what else" contenders in single file.
Silly response: Who's ever even heard of network x not equal to Facebook?
That's actually a huge part of the problem under discussion here, about what makes the existing landscape quite so toxic and evil.
It's not just the information, it's also the systems theory behind the information theory.
The underlying problem is that so many passwords in the wild get cracked back to plain text. Any paradigm you come up with is vulnerable to machine learning, which can ultimately identify and extract almost any pattern.
The pattern you describe is this: pick a word somewhat randomly (but not one too long or too hard to type), then use that as a seed word to free associate. Your entropy estimate is good and to my eye occupies the Fermi-estimation sweet spot (which is rather large, in a good way).
But the entropy you report is the conditional entropy on having already decided on the password paradigm (in some cases, this isn't even a guess, if you can associate previously cracked passwords to the same user, or specific password policies to the institution).
With enough randomly chosen paradigms to pick among, you could add maybe another 10 bits of true entropy. But people being people, paradigms are about as randomly chosen as social media networks.
Plus, because of all the cracked passwords, we have strong statistical models about paradigm evolution (these are cultural artifacts, for the most part)—at least for the masses who associate themselves with 2nd-rate IT (my own strong password paradigm is only used on sites highly likely to number among the scrypt enlightened).
Not that I'm including simple plug-board scrambling stages in my paradigm model (such as reflecting keystrokes between the left and right hands, moving home position one keyboard position to the right, or one row up).
A lot depends on the yield model of the attacker. If the goal is to crack as many passwords as possible, then you start with the worst of the worst and increment upwards. You would probably never even rise to plug-board stages.
If you have a value model over the accounts, when some accounts are judged to be a thousand or a million times more valuable once cracked, then the high value targets had better not be depending on their password paradigm adding any true security (unless so invented out of whole cloth, you're reluctant to update your password in less than a decade).
I tend to use apg to seed my passwords. I click generate five or ten times until something grabs my eye, and then I tweak it slightly to make it easier to store in short-term memory as I transcribe it from my password keeper to the passbox by hand. My "throw away" passwords are 11 characters, and more important passwords are usually more like 16. I generally estimate my passwords to have at least 50 true bits of entropy, assuming a very efficient search through password paradigm probability space (possibly by an adversary who has already cracked one of my other passwords into plain text). If the password contains spans of alphabetic dictionary fragments, I rarely estimate the fragment as supplying more than about 13 bits regardless of the letter count (or how close it remains to the original apg output).
Fully directed attacks based on a comprehensive model of human entropy management is an awesome superpower.
I'm sure the NSA has been doing this for decades already: nearly every plaintext password they've ever recovered (more than few) has been melded into some giant statistical model. They've likely identified millions of cognitive paradigms by now (from the ones with billions of breaks, down to some with only ten or so exemplars). There's no doubt in my mind they have explored the use of machine learning to squeeze even more out of this heap (though I suspect it was pretty squeezed out, even before machine learning). I also suspect that as this model is refined, they deliberated re-target recalcitrant breaks from ages past, so as to feed these breaks back in
Who doesn't like the Android security model—raise your hand.
Well, that's just about everyone in the room with any brains remaining at all.
Motion denied.
Can an AI spam filter distinguish this output from your least-gifted regular correspondent?
Can you tell the difference without actually rubbing two brain cells together (never mind that it doesn't take twenty)?
Because this rubbish generator scales like Tribbles evolved into a Borg empire diaspora. And remember: this is day one. Like cracks in cryptography, it only improves from here.
Furthermore, it won't just be your email feed, but nearly anywhere one potentially encounters text (ingredient lists on your groceries are somewhat immune, with their thirty different synonyms for sugar and spice—aka sucrose and MSG—already alive and well, and 2 g +/- 0.5 out of a 15 g "suggested serving" as an established level of numerical precision). Hint: for the sugar and spice line items (there could be many), freely substitute the top of the bracket. Exhausting way to shop? Glad you noticed.
So there's at least one happy thought: it isn't going to break what's already broken much worse than it's already broken. It's just everything else that's now in play.
The first twenty posts on this thread didn't display the vaguest clue about the actual threat vector of concern.
Possible reasons to do things are more common than jelly beans under sofa cushions. Dislodge the sofa cushion of your imagination, then stoutly hold forth your outstretched apron skirt to receive a giant payload.
Here's a starting point: people add search terms to narrow the field of search when seeking something you dimly recall, but where the precise contents (such as personal names) have slipped from mind.
If you're trying to recall the design of some cute top, "female friends and gay friends with flamboyant, effeminate style" is your winning ticket. But we're still five years away from machine learning being able to navigate "gay friends with flamboyant, effeminate style" so for now you leave that portion out.
Now to update that for the 21st century:
I'm not exactly filing this glitch under "accidental".
You know, if you track it all the way through, the Wizard of Oz has a surprise ending.
The Man Behind the Curtain
Big Shadow, Little Creature
Greedy people do shitty things behind closed doors. News at 11.
What's the evidence for this oft-repeated claim? That we behave differently in a culture saturation-bombed with advertising than when wearing a saffron robe in a remote retreat hidden away in the inaccessible hills of northern India? I've never had a clear picture how the A/B groups are factored, here.
I know that advertising affects me. This is because when I notice advertising, I make a conscious mental note that the brand is overpriced, due to a high cost structure, where the CMO earns more than the CTO, and has a budget to match.
For every hundreds small nudges I don't notice, there's a giant black mark somewhere else. Revenue vector per ad dollar invested, as measured by my own spending habits: pennies at most, possibly even negative.
I have basically configured my browser so that I go hours between advertisements. I have images and figures and call-outs disabled on almost every site I visit through user CSS scripts, which by now number in the hundreds. 90% of my browsing is pure text mode down the central corridor (if the side flow is too garish, I disable the masthead and footer, too, which means I can only navigate the site through Google search, and not by any internal click-bait).
I control advertising in my personal environment the way a marathon runner controls junk calories.
21st Century Psalm: My mind is my temple, I shall not be browbeat in my own home.
Of course, you can't exactly cancel any pervasive signal to zero by any reasonable degree of human effort. But you can reduce the advertisement signal on the order of 15 dB if you're a technically astute ornery mule.
Those in the full time business of making shit flow downhill (like it needed any additional help) typically colour within the lines very carefully.
It's not good business to wind up in small claims court playing the visible role of a heartless Ebenezer. Colouring carefully between the legal lines is the best way to remain in the black in the shitflow business over the long haul. (The at-will American employment regime poses no real incentive to a ruthless management culture in the arena of gratuitous abuse; 90% of what you want to do is entirely legal.)
Thus I take these kinds of sharecropping stories with a giant grain of salt. Yeah, there's always one asshole in the barrel (possibly compensated as badly as you are) who goes above and beyond, but this is hardly the norm.
The Joyent cloud features a second layer of isolation. Sometimes you see this described as "double-hulled virtualization". The OS performance penalty to achieve this is low to non-existent due to the nature of BSD zones (hardened jails).
Joyent hybrid cloud: Triton Compute Service and Triton DataCenter integration
This is precisely the scenario that Joyent's technology exists to mitigate.
You think you're running Linux containers, but under the hood you've also got zones and ZFS snapshots.
There is a resource penalty involved in using a high-integrity file system like ZFS, (efficient copy-on-write requires extensive write-buffer coalescing) but it's often not a large one compared to the many gains in administrative security and ease.