Show Idle (>14 d.) Chans

← 2023-01-30 | 2023-02-02 →
15:16 cgra signpost, went and tried to change ocpy arrays' word size to 64bits (from 8bit) only to achieve 0 change in encoding speed
~ 2 hours 40 minutes ~
17:57 signpost cool, was worth trying anyway.
17:58 signpost I wager bigger speedups would be had by removing unnecessary array copying. I don't have my brain-caches filled atm, but I recall there still being a few in there in iirc the decoder.
~ 2 hours 16 minutes ~
20:14 awt $ticker btc usd
20:16 awt Last update in my at from shinohai is 1/16
20:17 awt Anyone had a look at nostr?
~ 1 hours 2 minutes ~
21:19 asciilifeform awt: not heard prev. but lulzy. approximately 'irc w/ pubkey signatures', but mega-marketing push coming from 'nowhere'
21:20 asciilifeform ... endorsements from 'luminaries', e.g. dorsey, 'bitcoin magazine'.
21:20 asciilifeform choice verbiage from author : 'it does not rely on P2P techniques, therefore it works'
21:21 asciilifeform spec. schnorr signature algo.
21:21 dulapbot (trilema) 2017-03-02 asciilifeform: oooh hey hey hey lbj. i finally found out why prb folx like schnorr.
21:22 asciilifeform by all appearances, this is reich's latest attempt at 'vaccinating' folx against pest ( which the perps do not necessarily know about, but imho idea is 'obvious' enuff that 'people need vaccinating' against it )
21:23 asciilifeform anyffin, anyffin that might delay / poison well against p2p wot nets.
21:25 asciilifeform apparently, there's a client in crapple 'app store'. which itself imho quite revealing.
21:25 asciilifeform 'protocol that allows building truly censorship resistant and decentralized social media...' etc etc
21:28 signpost just another client-server turd, and on it most people are "talking" about nostr itself, if anything.
21:29 asciilifeform the latter by itself aint damning
21:34 asciilifeform signpost: from surface reading, ~= irc with (shitcoinistic) pubkey sigs. author's answer to 'what to keep relays from dropping msgs signed by keys they don't like' seems to be 'shuddup'
21:35 awt Jack Dorsey donated a sum of btc to the project and there's some sort of nostr conf happening down here in CR next month. Weird.
21:35 asciilifeform (not even to mention e.g. ddos; or the fact that crapple always has yer current ip & privkey.)
21:36 asciilifeform awt: imho not least bit weird. entirely predictable.
21:36 asciilifeform ( why does a 'decentralized...' need btc? or a dorsey? or a piece in shitcoin mag? imho answer is obv )
21:38 asciilifeform 'tmsr dead' but mp's 'siberian river' apparently alive & well.
21:38 dulapbot (trilema) 2017-03-26 mp_en_viaje: basically a novel vector of imperial attack seems to be this "let's take republican items and ~EXPAND~ the downstream so that siberian river attack is then feasible".
21:43 asciilifeform reich will keep twatter (which apparently essential component of the edifice) alive at whatever cost; incl. aggressively peddling 'alternatives' sprayed with faux-decentralization perfume 'until sumthing sticks'
21:44 awt Can't find an explanation of the siberian river attack in the logs or elsewhere.
21:44 asciilifeform awt: what ethereum was to btc
21:44 asciilifeform ( or what 'keybase' tried to be in re pgp, etc )
21:45 asciilifeform 'friendlier' and glossy-magazined 'modernization', 'use on ipnoje!', etc
21:45 awt I get that but how related to a siberian river?
21:46 asciilifeform see link
21:46 awt ah was the next line:
21:46 dulapbot (trilema) 2017-03-26 mp_en_viaje: (for the unfamiliar, siberian rivers flow the wrong way, melt at source before they melt downstream, causes some problems.)
21:47 asciilifeform aka 'embrace & extinguish'(tm)(r)(microshit)
21:47 asciilifeform (made famous by their 'open document format')
~ 20 minutes ~
22:07 signpost << musk says out loud that he intends to turn twitter into the "west's WeChat". one side "sounds good!" other, "we're so upset about this we're going to tweet about it a lot"
22:07 bitbot Logged on 2023-02-01 21:43:59 asciilifeform[jonsykkel|deedbot|awt]: reich will keep twatter (which apparently essential component of the edifice) alive at whatever cost; incl. aggressively peddling 'alternatives' sprayed with faux-decentralization perfume 'until sumthing sticks'
~ 23 minutes ~
22:31 phf << the hardest part so far is figuring out how Generator(…).choice(…) relates to xorshift1024 algo
22:31 bitbot Logged on 2023-01-29 13:57:20 signpost: << latest src
22:33 phf i assume "xorshift1024" that they use is from this paper, but i'm not quite sure what sort of magic scipy's generator use to e.g. produce Generator(…).random() or choice() or any other results
22:34 signpost worked with any of the provided prngs. I only put that one in because asciilifeform recommended it as one that's cheap to reseed.
22:35 signpost iirc numpy uses alias method to generate choices from a biased distribution. cribbed some guys impl of same for the lisp one I wrote.
22:36 phf well, the point is more that the message block edges generator should probably be explicitly spelled out in python code
22:36 signpost yeah, totally agree.
22:37 * signpost back in a few
22:38 phf on one hand i can probably figure out how to match whatever scipy does, but on the other hand it seems like lowkey waste of time. so far my strategy was to just call out from lisp into a python script to provide me with correct random numbers :D
~ 28 minutes ~
23:06 * asciilifeform ftr not has atm even faintest clue re what kinda prng would work best for lubyism. ( aside from obv., e.g. 'need reasonably long period' and 'not ruinously cpu-costly' )
23:16 phf it's a separate question from pure "which prng", signpost in ocpy is using a numpy mechanism around underlying xorshift1024 implementation
23:17 phf numpy separates "bit_generator" which is an object that managers underlying prng state from higher level "generator" which gives you a variety of methods for e.g. doing choice() or getting integer within a range
23:18 phf which makes producing a predictable and specific sequence out of the same prng but in multiple different implementations a non-trivial exercise in reversing
~ 33 minutes ~
23:52 signpost yeah, and there's some magic involved in massaging seeds I haven't scrutinized yet.
23:55 signpost I oughta possibly pull all the usages of numpy's generator into one place so they're easy to reimplement, possibly lopping off the reliance on numpy for that.
23:56 phf yeah, i got the seeds out of initialized Xorshift1024, and at this point i should be getting same number out of random_raw() as my lisp version, but no such luck
← 2023-01-30 | 2023-02-02 →