Profile cover photo
Profile photo
Geoffrey Transom
122 followers -
Maybe I'm too smart to sleep. Yes... that's it: Stimpy's sleeping like a baby, and he's an IDIOT.
Maybe I'm too smart to sleep. Yes... that's it: Stimpy's sleeping like a baby, and he's an IDIOT.

122 followers
About
Communities and Collections
View all
Posts

Post has attachment
Turns out that fuckwit high-school under-performers whose only option is to join the retard "Border Farce", get treated close to what they deserve (I.e., scorn and contempt) rather than the hagiography they expect.

One thing about being a fucking idiot whose best alternative use is saying "Your papers, please" like a good Little Eichmann... They undererestimated how much the average punter fucking hates them, and also they failed.to realise that the fuckbags above them in the hierarchy are even worse than the dummies at the coal face.

I mean take one look at their overall boss and tell me that his parents weren't cousins or siblings.

Fuck 'em - they're worse than folks on the dole, because they cost twice as much and are good for nothing.

Post has attachment
You might want to check whether or not that (very low-lying) land would be subject to LSIO planning overlays.

Hands up who has the VicMap Planning overlays in their database? I have, and the entire site is lower than existing land in the area that's already LSIO.

Oh... and LSIO stands for "Land Subject To Inundation".

Good luck living on a chunk of dirt that has had lewisite and other WWI-era nasties stored on it: better hope that folks whose best-economic-use was 'soldier who looks after barrels of nasties', did their job with more alacrity than the average soldier.

(inb4 stupid .mil hagiography: I was a soldier once, and nobody really wants to know how much soldiers skive off: destroys the whole Americanised 'heeeee-ro' narrative that is only believed by people who think that 'The Sullivans' is a history documentary)

Post has attachment
This 457 bullshit proves, to anyone who has a fucking brain[1], that the Australian unemployment data is bullshit. Why would you be doing shit to stop labour mobility if the economy is supposedly close to zero non-frictional unemployment?

And is it just me, or would others not be surprised if it was discovered that Dutton had tastes that tended towards "&i"? Fucker looks like a dead-set paedo (the only missing hint is a cassock).


[1] does not include anyone who watches Q&A, Lateline, The Drum, or anything else where teleprompter-readers pretend to know anything about anything. Those fuckers (the viewers) are as retarded as the iCult (the journos themselves are probably aware that they know fuck-all, and are just trying to make a living),

Post has attachment
This is fucking pathetic - worse than AQ tests or the worse-than-horoscopy Briggs-Myer classification. Like this piece of shit test, I can test 'wherever I want' on BM and AQ tests (I once tested an AQ of 56 - on purpose - to win a bet. If I was genuinely that much of an Ass-burger I would be rocking backwards and forwards like a mental patient).

If I can see the obvious answer to result in a lower score on all metrics than the score I probably deserve, then this test is fucking useless.

This is par for the course with this sort of bullshit: tests that rely on 'honest' answering are the product of halfwits. (For a very concrete counter-example, think about trying to fake an 'upside surprise' to your IQ test - it pretty much can't be done. However it's simple to test 'below' your actual for garbage psychometrics like this).

Psychology is the last professional refuge of the true charlatans in society - beta-level pseudo-academics who do literally no rigorous statistical testing of their 'competing essays' hand-waving bullshit.

Only politicians are more dishonest than psychology.

Who else thinks it's bizarre to have a Math group on a platform that has no support for math rendering (if user ∉ {unicoders}).

Post has attachment
What's worrisome to me is that the audience is (notionally) a group of current and potential ML/DL practitioners, and yet at 14:30-ish Coates takes excessive pains to avoid discussing why \theta is chosen to minimise the Lagrangean in a bog-standard logistic regression. FFS: back when I was a tutor, any second-year econometrics student who didn't grok that shit would have been a washout.

Call me old-fashioned, but anybody for whom a logit model is the equivalent of Jedi-level sorcery is a half-decade away from the sort of competence that I would consider the minimum for a ML/DL practitioner: half the problem with the entire globe is that we have semi-smart folks who implement pedagogic tools without genuinely understanding them.

Having done Ng's Coursera version of CS229 ('the' Stanford ML course, perhaps watered-down for general consumption) - and gotten 100% for every assessment task (penalised to 92.6% overall because I enrolled in Week 6) - I can say that this stuff is very much like quant finance: interesting, only mildly technically-challenging, and dangerous if implemented by people who have no fucking idea of the statistical theory underlying the parameter-estimation methods they employ.

When is the last time a ML practitioner looked at a t-test (for individual \theta_i) or a Wald test (for subsets of \theta), or did a Chow test for parameter stability?


Don't get me wrong: I am an absolute fan of this stuff... but I dread the idea that it becomes the domain of people who can code Python/Octave/Matlab, but have no fucking idea what a statistical test is. That's what has happened in quant finance, and it is a fucking disaster in the making.

Post has attachment
OK so when it comes to the Iron Age nonsense, this is where the failure to get story straight succumbs to science, bitchez...

When data-analytics turn their lidded gaze on the gallimaufry of primitive drivel in everybody's favourite story of incest, genital mutilation, invasion and genocide (all demanded by a Sky Maniac), the story's inconsistencies leap out like some sort of thing that leaps.

Data geeks and d3-lovers (a subset): feast your eyes on this. It is, as someone once said, "a wonder to behold".

Post has attachment
Geoffrey Transom commented on a post on Blogger.
This article scratches the surface of the problem - to do justice to the question of state legitimacy (and the tendency of megalomaniacal sociopaths to seek, and achieve, peak political office) requires a good decade's worth of diligent study.

If you began with "The Problem of Political Authority" by Michael Heumer, you would get a distillation of a couple of years' worth of that required study in a week - although you would be left without good explanations of

① why the 'Social Contract' is a lie (from the POV of Contract Law itself);

② why 'Public Goods' explanations are lies (even based on utilitarian externality arguments);

③ why "The General Will" (and 'representativeness') is a lie (based on Arrow's Impossibility Theorem and the Gibbard-Satterthwait Theorem);

④ why voluntaryism would not descend into Hobbe's "War of All Against All" (game-theoretic and network-theoretic arguments);

⑤ why governments deliberately interfere with expectations-formation (through propaganda on the one hand, and deliberate enstupidating of the population on the other - again, network-theoretic... especially DeGroot style 'learning networks');

⑥ The problems of cognitive shortcomings in the electoral base (the PIAAC survey, showing that a median IQ of 100 does not equip people to make long-tailed decisions for themselves, let alone for the whole of society).

That last point is really quite a subtle one, since it - along with the relatively large importance of government in expectations-formation (from ⑤) - ensures that even if it was morally defensible, electoral politics cannot, and would not, result in an unbiased estimate of 'true beliefs' about policy (if you know anything about the De Groot model, you will know that if one 'agent' has disproportionate influence, beliefs will not converge-in-probability to to underlying reality, even asymptotically... and elections don't take place once-every-infinity-years anyhow).

I have had a partially-finished manuscript that covers all that stuff (working title "The Worst Will Rule") for almost a decade, and I go back to working on it from time to time, but I doubt it will never be finished. The hardest part is making sure that it contained all the mathematics required, in a way that does not make the reader's eyes glaze over.

Post has attachment
Geoffrey Transom commented on a post on Blogger.
The primary cause of this, is that datasec is the responsibility of one guy - who got the job as server admin because he was the only guy who applied, and had done a six week course that entitled himself to self-declare as a "sysadmin". (Or worse - because he had 'experience' that dated from MS-DOS 3.3). Nothing against such folks: i've been interested in crypto since the 80s, and there is no chance in hell that I would ever want to sysadmin a secure server.

Also, the majority of corporate database servers have security that is implemented poorly - and/or encryption protocols for stored data that are implemented incompetently.

It's even more worrisome when you consider that web standards had profound flaws that persisted for years - e.g., TLS 1.0, and pretty much any 'roll your own' authenticated-encryption (e.g., CBC-MAC with fixed-IV - the default, which few people bothered to over-write).

Add in that management often wanted to 'own' datasec - and so had encryption systems built in-house (and badly)... just so that corporates could claim the expenditure as an asset.

Crypto is a really important thing; almost nobody knows how to do it well, and when it's done badly the vulnerabilities are almost impossible to spot - until someone undertakes a padding oracle attack on your mal-implemented CBC-MAC and gets hold of your entire database.



And one last thing: jerkwads who work for .gov are the MOST incompetent crypto guys in the world. To get into any position of even modest responsibility in .gov, you have to be a triangulator. That's why the NSA does not have a snowball's chance in hell against the cypherpunks: everybody above GS9 is a triangulating bullshít artist with zero actual skills.

Post has attachment
Geoffrey Transom commented on a post on Blogger.
This works perfectly with the example in question, but more generally it'll go spack if the first bits of ciphertext and key XOR to 0 ... you get left with something that you can't  .decode('hex') to recover the plaintext.\

If you check len(ans) within decrypt() and front- pad it with '0' as required, and pass back the front-padded ans, that will fix it.

I usually pre-pend '0000000' just in case, then take a [-l:] slice of the result (where l is the length of the original hex)... kludgy, but it works.
Wait while more posts are being loaded