log☇︎
900+ entries in 0.327s
asciilifeform: ( incidentally astute students of FUCKGOATSology will notice that the principle is applied INSIDE the device as well - FG has nfi what is plugged into rng jacks, actual analogue rng boards or tester jig )
asciilifeform: not to mention a proggy that eats rng oughta be periodically tested using ~known~ bits
asciilifeform: but that the ~correct~ move is proggy that takes a /dev/foo1,/dev/foo2... as command line params and eats rng from ~there~
asciilifeform: gotta luvvv all the nitwits who think that pc has any business being able to talk TO an rng box
asciilifeform: in other noose, the (microscopic) ra-226 ( for MB/s 'gold standard' rng test ) came. asciilifeform was half-convinced that it'd get swallowed by the post
a111: Logged on 2017-06-08 16:43 mircea_popescu: much like you know, all sorts of "rng experts" not merely do not wish, but specifically wish NOT TO examine FG. andf so on.
mircea_popescu: much like you know, all sorts of "rng experts" not merely do not wish, but specifically wish NOT TO examine FG. andf so on. ☟︎
a111: Logged on 2017-05-23 03:18 asciilifeform: the 'formulate as predictor' came from asciilifeform's ongoing regular stabs at breaking shitoshi's rng.
asciilifeform: if you can 'learn' from rng -- it's a poor rng
mircea_popescu: it'd seem to me the model as described does in fact generate reals properly from rng.
asciilifeform: ( nobody in fact knows how to generate 'uniform float' from rng )
asciilifeform: floats suck mightily, incidentally, especially for 'model reality' or whichever involving rng -- because your uniform rng now becomes a skewed shitsoup because the number line is now unevenly populated
asciilifeform: it was tricky enough to make 'republican rng', an item closer in order of complexity, to a knife or a fork, than to a comp
asciilifeform: another tip : rng shield is an ok place for the gnd clip
asciilifeform: they're why a misbauded FG is nogood as an rng
asciilifeform: mod6: possibly earliest properly logical unmasking of sham rng
asciilifeform: the perhaps worst 'test' ~of a working rng in particular~ is to look at the hex with naked eye
asciilifeform: phf: d00d was always a bit... odd. recall his paper re 'make as little use of rng as possible, because ohnoez it could be subverted' ?
asciilifeform: (problem was , analogue rng's waveform, while meeting the signal voltage constraint, and even the rise/fall time constraints, sometimes does not meet the hold time constraint! and results in metastable state in digital chip it is connected to
asciilifeform: this is also a good time to remind fols, THIS ^^ is the kind of thing 'rng testing' shines in, picking up misconfiguration/catastrophic os retardation/similar
asciilifeform: i encountered this nonsense when i built my very first prototype rng, iirc mircea_popescu even turned the resulting dialogue into a whole post on his www
asciilifeform: 'One application of the KNO is for testing of random number generators (RNGs). The idea is that it is extremely unlikely that a properly working RNG will generate the same random 128-bit number twice...'
asciilifeform: in other unshielded rng, https://www.araneus.fi/products/alea2/en/
a111: Logged on 2017-05-25 22:47 asciilifeform still finds it interesting that g_l did not ask re the rng sample .
asciilifeform still finds it interesting that g_l did not ask re the rng sample . ☟︎
asciilifeform: anyway gabriel_laddel_p go to nosuchlabs.com , you will find rng bitz
a111: Logged on 2017-05-24 22:16 asciilifeform: previously i thought that you could not have an auditable single-crystal rng, but possibly this is not so : if you make it so that only co-incident (on 2 detectors) gamma is picked up, you can 'yoke'-audit 2 units by placing back to back around test source.
asciilifeform: the reproducible ~t~rng is a contradiction in terms.
asciilifeform: honest rng maker has 0 to be afraid of from posting arbitrarily large sample.
a111: Logged on 2017-05-25 17:42 asciilifeform: while we're on subj, i'll say that there is probably a good reason why most commercial faux-rng makers don't offer a sample download.
mircea_popescu: rng is not === crypto
mircea_popescu: i can conceive that i'd trust x to rng more than i'd trust myself, in some conditions.
asciilifeform: but will observe that both sellers of rng bits, and esp. buyers, are a little 'off their rockers' imho.
mircea_popescu: so does signed and encrypted rng sample.
asciilifeform: montecarlo-grade rng bits -- inf.
mircea_popescu: kinda how this works. by the timpe peopple pay bots here for rng, we're simply owning the world and no more. which is why i said maybe premature.
asciilifeform: while we're on subj, i'll say that there is probably a good reason why most commercial faux-rng makers don't offer a sample download. ☟︎
asciilifeform: microshit's rng has (as of xp3) 0 hardware component.
asciilifeform: iirc also mircea_popescu's technical objections were, imho, intellectually lazy ('pipe fittings will cost their weight in gold, because boat and they can' -- by same token rng MUST cost 10k $, because raytheon's does; 'must drag it to shore to paint bottom' -- also fixable, make the bottom from inconel or hastelloy ; etc )
asciilifeform: fwiw d00d actually works (as academic) on optical rng
asciilifeform: previously i thought that you could not have an auditable single-crystal rng, but possibly this is not so : if you make it so that only co-incident (on 2 detectors) gamma is picked up, you can 'yoke'-audit 2 units by placing back to back around test source. ☟︎
asciilifeform: single-crystal MB/s rng.
asciilifeform: so far it's pretty hilarious, i build a box that actually solves $problem, and not even close to 'megabestseller', all the meanwhile 'people' paying, supposedly, multi-$k for usg rng.
asciilifeform: the 'formulate as predictor' came from asciilifeform's ongoing regular stabs at breaking shitoshi's rng. ☟︎
asciilifeform: however there is a number of tests-for-broken-rng. my idea was strictly re an algo for how to measure their actual relevance .
asciilifeform: but function of rng AND enemy. (see ancient thread, 2014ish iirc)
asciilifeform: and yes, this is still same old tests, nothing magically different about them, there will ~never~ be a mechanical test for 'good rng'
asciilifeform: right. output ought be a 'hey schmuck, if you use this rng, i can predict, like-so...'
asciilifeform: say an rng fails the 'birthdays' test.
asciilifeform: let's take concrete example -- if i make a rubbish rng that leads 'ent' to conclude that pi == 4
asciilifeform: http://btcbase.org/log/2017-05-23#1660248 << this ain't the problem -- already right off the bat , sane operator knows that it is physically impossible to test for ~good~ rng, only for ~bad rng~ ☝︎
asciilifeform: 'purpose' in this sense is not an attribute of the object, but of the object-and-maker system. see also the thread re 'rng entropy'
asciilifeform: gamma rng has certain advantages which cannot be had any other way
asciilifeform: because rng is NOT concerned with accurate photon counts
asciilifeform: (which is no good for rng, for reasons explored in older thread)
asciilifeform: ( any other rng , available for love or money, where this is true ?? )
asciilifeform: same problem with intel's ( probably the most ubiquitous hardware rng )
asciilifeform: Framedragger: re earlier thread : i looked far and wide to possibly buy rng to compare with -- and found nothing worth buying ( no, i won't compare with whitened; i won't pay 20,000 $+ for a box of schematicless ??? ; etc )
asciilifeform: mod6: the thing to compare against would be another UNWHITENED rng.
asciilifeform: mircea_popescu: i ain't particularly interested in testing whitened 't'rng , also
mircea_popescu: asciilifeform you still have to pick an X. if machine has no RNG, its picking of X will be weak.
mircea_popescu: incidentally, FG could plug directly into raspberry give it true rng, huh ?
asciilifeform: any rng test can be rewritten as a producer of a predictor. or the test is rubbish.
asciilifeform: mircea_popescu: so, to take example, if you found a rng that turns out to emit digits of pi -- the proof-of-weakness will be plouffe's algo.
asciilifeform: http://btcbase.org/log/2017-04-26#1647917 << imho it's the only sane standard. rng 'weakness' can only be expressed properly as a 'predictor'. anything else is 'suggestively named token' ☝︎
BingoBoingo: Or did Mexico try the cocaine and toggle switch RNG idea and it turns out cocaine is deterministic
a111: Logged on 2017-04-19 17:48 asciilifeform: mircea_popescu: at one point we had a thread with d00d who sold 'sram as rng' iirc.
asciilifeform: bonus: your rng also doubles as a radio-shenanigans detector!111
asciilifeform: mircea_popescu: at one point we had a thread with d00d who sold 'sram as rng' iirc. ☟︎
asciilifeform: mircea_popescu: ~whole point of gamma rng is to divorce from electronic effects.
mircea_popescu: that'd be the ultimate rng pump. "We have this array of matter -- think memory stick. at every given tick, each either decays or doesn't. tick here has no physical meaning, it can be shorter than measurable irrespective of measuring mechanism"
asciilifeform: Framedragger: almost like somebody thought about rng.
asciilifeform: and, apparently, the same tests weaken/fail in proportion to rewinds, regardless of whose rng you used, or even prng.
a111: Logged on 2017-04-04 04:00 asciilifeform: ( now, good q is 'why do this?' and answer is -- 'you get analogue rng that is VERIFIABLE, just like the digital board in fg.' because you can pick up decay from below/above the board !!)
asciilifeform: in other noose... LYSO ( cerium-doped lutetium yttrium orthosilicate ) scintillator is apparently just the thing for rng.
asciilifeform: they will reliably find what one might call 'coarse errors in pilotage' of rng builder
asciilifeform: uses file as 'rng'
asciilifeform: definitely not a low-tech item tho, quite a bit moar complex than plain old analogue rng
asciilifeform: ( incidentally -- FUCKGOATS worx ok (if very slowly) with toggles instead of analogue rng modules. )
asciilifeform: the funny bit is, in usa you can generally buy, e.g., actual uranium oxide (not by the kg, but can buy, easier than, say, gunpowder.) but it'd be quite useless for ~mass~ rng.
a111: Logged on 2017-04-14 14:57 mod6: with both of my RNG-TWs connected, it was collecting at about 7kB/s ... for that quick 10s test or so to ensure that it was a "full speed"
mod6: with both of my RNG-TWs connected, it was collecting at about 7kB/s ... for that quick 10s test or so to ensure that it was a "full speed" ☟︎
a111: Logged on 2015-01-15 00:45 asciilifeform: (what rng? ring oscillator jitter, with sha1 whitening.)
mircea_popescu: FG is, importantly, an ~auditable~ rng. you can audit the thing, and if you do, i'd be very interested to hear the results.
BenBE: What's the entropy source used in those Cardano RNG?
BenBE: Didn't know that particular project, but have a true RNG based on the WhirlyGig design at warmcat
asciilifeform: asciilifeform finally found a rng-usable beta decay source that is permitted in every country's post office, available worldwide, and for a few bux per kg : ordinary KCl.
asciilifeform: ( now, good q is 'why do this?' and answer is -- 'you get analogue rng that is VERIFIABLE, just like the digital board in fg.' because you can pick up decay from below/above the board !!) ☟︎
asciilifeform: there is, however, a correct part for this : 'PIN diode.' i have an experimental setup with one. at some point we might offer a fg rng module based on it. but not yet.
mircea_popescu: i imagine some of the best electronics for rng would be satellites. measure it straight in the solar panels!
asciilifeform: upstack mircea_popescu had the right notion, though, it is difficult to build a decay rng that does not rot.
asciilifeform: i've given some serious thought to 'what would be highest bit rate honest rng possible with current tech'
asciilifeform: so we wait for EACH rng : http://btcbase.org/patches/fg-genesis#L360 : to pulse, before entering normal operation.
asciilifeform: so that leaves the rng.
asciilifeform: on the other hand, there is no external input into the box other than the 2 rng.
asciilifeform: btw, this is not a bad time to explain WHY i set it up so that fg remains 'sad' until BOTH rng start up
Framedragger: "Note: Both RNG-TW (Analogue) modules must be present and in working order during power-up, or FUCKGOATS will remain in a SAD state (steady RED lamp.) When performing basic tests, start with a powered FUCKGOATS, with BOTH RNG-TW (Analogue) modules installed. "
asciilifeform: speaking of which, asciilifeform finally discovered how to make a decay-based rng that doesn't wear out or need high voltage -- use PIN diode. the q was always what to use for an emitter, that won't be stolen by post office of half the globe. granite, it occurs to me, is probably a good bet.
Framedragger: also, ip packets with custom proto number would (1) stand out more easily to enemy, and could be more easily filtered out (vs. udp header with rng-data within) - see how chinese firewall blocked tor bridges etc etc; and (2) i'm sure quite a few appliances would filter them out by default (like how they filter out icmp, etc.)
asciilifeform: mircea_popescu: and yes, they are shipping as soon as crate of rng lids (due in today) is here.