287300+ entries in 0.191s
a111: Logged on 2016-06-28 11:47 jurov: Framedragger, imo we should mix few interesting failures into
the data. otherwise we risk getting kicked out from
tmsr :D
a111: Logged on 2016-06-28 10:05 Framedragger: asciilifeform: but have you considered
that
there are fewer implementations for ssh (and
the better part of
the ssh servers in
the wild run openssh), and more implementations for all kinds of broken pgp? so it may simply be less likely
to spot a badly generated ssh key.
thestringpuller: mats: nah. if it's
truly believed miners are well enough gripped by
the balls, it's not a big deal.
a111: Logged on 2016-06-28 10:01 Framedragger: asciilifeform: re. keys - interesting. i don't
think i was mitm'd, 13 machines and all, not
that it's not possible or anything, and
the machines were on
the same provider, so, lol, of course possible in principle. but specific mitm's which give working unique keys? hm. but, interesting, i don't have an answer here.
mircea_popescu: anyway. yes dingledine & other sleepers (reilly, whatever) push for chasteen and whoever
the fuck else
to "be hired", because
that's how
their job goes.
mircea_popescu: asciilifeform no patience
to read, but by now, "at MIT" is damning.
☟︎ mircea_popescu: so apparently reddit has selected for
this large-ish population of chicks
that want
to post selfies "anonymously". because
they're
THAt desperate for male attention and
THAT incapable of handling male presence.
☟︎☟︎ mats: as i understand it,
the heterogeneity of clients in
the wild naturally prefers
the status quo
☟︎ mats: anyway, i
think its premature
to assume we need
to influence miners one way or
the other
shinohai: We got a Roger Ver in
the making here!
mats: so you've moved
the goalpost from 'our pool'
to 'bribe miners'
thestringpuller: mats: well from a financial perspective, not a political perspective. at
the very least bribe miners
to vote one way or
the other...
☟︎☟︎ a111: Logged on 2016-06-28 03:47 phf: i have
that cockli box by
the way. i'm probably going
to spin up an instance on it, but if anybody wants
to attempt
that
task, feel free
to
take over
Framedragger: or labour union websites but
that may just cause a stroke and best
to avoid stress induced fatalities
Framedragger: and maybe make sure some of
those ip addresses are in /24's which also house chinese bitcoin miners ;)
jurov: Framedragger, imo we should mix few interesting failures into
the data. otherwise we risk getting kicked out from
tmsr :D
☟︎ jurov: lol i was
then prolly mitm'd
too.. aside from ~10 cases of obvious corruption (openssh won't even read
them) i did not find any key with factor in 10000primes.txt
Framedragger: asciilifeform: but have you considered
that
there are fewer implementations for ssh (and
the better part of
the ssh servers in
the wild run openssh), and more implementations for all kinds of broken pgp? so it may simply be less likely
to spot a badly generated ssh key.
☟︎ a111: Logged on 2016-06-27 17:15 asciilifeform:
the historic ~0.1% popping rate of ssh keys has nothing
to do with flips, and everything
to do with embedded gadgets with no rng
Framedragger: asciilifeform: maybe i did fuck up in some very unique way. don't know. oh and
the ip addresses are in random order, so
there's
that
a111: Logged on 2016-06-28 00:28 mircea_popescu: Framedragger is
this one of
those guys you're friends with,
then ?
Framedragger: i'm prolly mostly afk /
travelling
this week but would be interested
to discuss.
Framedragger: asciilifeform: re. keys - interesting. i don't
think i was mitm'd, 13 machines and all, not
that it's not possible or anything, and
the machines were on
the same provider, so, lol, of course possible in principle. but specific mitm's which give working unique keys? hm. but, interesting, i don't have an answer here.
☟︎ mircea_popescu: then again it's rather common for people
to do something smart early and
then be shockingly stolidly dumb for
the rest of
their lives.
a111: Logged on 2014-09-26 03:11 asciilifeform:
there are
two versions of
the lenat story
a111: Logged on 2014-02-01 21:28 asciilifeform: dollars
to doughnuts
that lenat did not write eurisko.
mircea_popescu: phf anyway, it's a very reasonable way
to waste one's life. it can live on a server(s), connect via chatbot, solicit payment which'll work once
trinque gets it added
to deedbot... it's an actual biznis.
midnightmagic: asciilifeform: Again,
thanks.
Too few people care about
that.. :(
mircea_popescu: yeah, agreed, unlike cyc bs,
that is a credible starting point
mircea_popescu: apparently only some people outgrow being 9, and lenat isn't one of
them.
mircea_popescu: distinctly reminescent of 9yo me writing "utility programs" consisting of pre-selected arbitrary aditions and
their results.
mircea_popescu: you do not. and more improtantly : you need no human anything in
there. you wanna succeed in your own
terms, go, succeed.
mircea_popescu: seriously,
they're going
to put
the rules in by hand arbitrarily and expect
the machine
to do something with
that ?
mircea_popescu: (#$implies (#$and (#$isa ?OBJ ?SUBSET) (#$genls ?SUBSET ?SUPERSET)) (#$isa ?OBJ ?SUPERSET)) << wouldja look at
this inept shit.
mircea_popescu: asciilifeform it largely did work, out of
thousands of
tries a few successes. machine has space for many
thousands.
mircea_popescu: "Dead reckoning dun work without gps". see,
this is why cyc fails.
this mindset.
mircea_popescu: the machine played dead reckoning not some sort of absolute evaluation of
the eternal positions
mircea_popescu: sedol lost
to a machine
that doesn't even know what beans are.
mircea_popescu: phf and
the
third is
that it absolutely should not make any concession
to
the finite nature of
the universe.
three billion
trillion parameters ? OK!
mircea_popescu: now 2 and 3 are iffy, for instance english-chemistry vs english-genre-fiction and english-trilema may be a more practical split
than english vs french.
mircea_popescu: the other is,
that it should keep context of "this
text" ; separately context of "everything i ever saw in
this language" and yet separately "everything i ever saw".
mircea_popescu: phf
there's
two (or perhaps
three) aspects here. one is,
that it should first pass
through
the entire item (book or w/e),
to adjust itself, and
then go once more
through it, on
the basis of
that, and
then again. similar
to how "page rank" works, once
the result is stable it
thinks it's done.
phf: i like
the idea of using hidden markov models on ocr-ed
text
mircea_popescu: ftr, i wouldn't
take cyc for anything other
than a fine guide of what not
to do.
mircea_popescu: use
the fact
that
the machine as is, runnign whatever crap it runs, can still do millions of cycles a second, and can store billions and
trillions of bytes.
phf: i believe
that was described in
the "Sussman attains enlightenment"
mircea_popescu: anyway -
this is a correct wiki. a proper notation engine backed by a self-training, probabilistic model of language.
mircea_popescu: that we don't have
this in a box is simply because nobody wants
to do any work.
mircea_popescu: THIS is what competent (not great. competent, patient, insistent) programmers + 20nm process means.
that we can have
this in a box.
mircea_popescu: i was stuck publishing a pdf in my latest article because fuck me
there's no sane ocr ;/
mircea_popescu: asciilifeform 1 error per 500 pages or less. how's
that.
mircea_popescu: there doesn't exist currently any ocr package
that uses a proper statistical model, and is self-trained on
the
text, and on all
text.