311900+ entries in 0.201s
mircea_popescu: this'd make
the "march lords in a wot -
take your fief and guard it" naive approach
to date actually very well grounded both factually and now philosophically ; and also offer a ready explanation of "why all
this shit everywhere!"
mircea_popescu: and
there's no way
to specify him out of
the machine or vice-versa. for reasons
that perhaps go all
the way
to godel
mircea_popescu: or
to put it another way :
the reason software houses denegerate into makework facilities is much more fundamental
than any sort of policy. in point of fact, a numeric computer of
the sort we're using is NOT useful in either logic or math but as a shorhand for an actual logician or mathematician
that
takes
the abortive nonsense
the machine spits out, and enchants it into actual usable
truth.
mircea_popescu: when i say "therefore - we should all eat an onion" you can bring
that obhection
mircea_popescu: about which
the only
thing
that can be rightly said is
that - it will crash
the next larger system.
mircea_popescu: so yea, i would say
this is very much equivalent. we CAN'T actually make computer systems. and every
time we
try
to make a larger one, we're stuck re-calculating a sort of "pi"
mircea_popescu: except
the larger
the
things
they
tried
to build,
the worse it got. but
then
they fixed it. because, as alf
the greek beedog says, "it IS possioble
to make wall
that won't leak!!11"|
mircea_popescu: phf
the greeks notably
thought
they can build
things, because
they know "the actual value of pi"
phf: except opposite? all
the math, until you find out
that immediately available reality is finite
jurov: isn't
this parallel with old greeks
thinking "all numbers are rational"..
till
they found out
they can't?
mircea_popescu: i agree
the crisis may only be in my head, but in point of fact i'm getting a whole paradigm reallignment
thing, slightly queasy atm.
mircea_popescu: basically, all
the architecture systems architects have managed yet
to produce is
the equivalent of "we need 10.1k builders - 100
to build and 10k
to hold up
the walls once we're done".
☟︎☟︎ phf: mircea_popescu: fwiw a lot of vlsi research in
the 70s and 80s was about
that, how do we make bedrock a lot less engineery, but
that all died with microcomputer revolution(sic)
mircea_popescu: asciilifeform in your
terms : it IS NOT possible
to make abstraction
that does not leak until [condition about machine and only machine]. ALL
the abstractions you make WILL leak once you run out of BRAIN.
mircea_popescu: phf worse
that
that, utterly undermines
the fundamental reason against many
things we supposedly argue against fundamentally.
mircea_popescu: let's work a simple example. suppose
the case is
that your machine is required
to behave coherently with
the rule
that " among even numbers 2 - 8 is as much of a range as 1-4 is among natural numbers".
the MOMENT your solution
to
this was "simple, just
take $i*2", you have in fact c'd it.
phf: it's a philosophical discussion, and mp, not being a programmer, is realizing
that ada or c or .. is not actually it from where he stands, doesn't mean can't be used, just not magick.
☟︎ jurov: all we have is some deryp prrof
that "we
think of something
that we don't know how
to map
to N"
mircea_popescu: asciilifeform you can't define jack shit for a cpu without actually lying somewhere about what you did. is
the problem.
phf: which is entirely
true,
this is like intro
to von neumann 101
phf: mp keeps saying
that you don't have a way
to solve it as an abstract
jurov: mircea_popescu:
there's "unum" numeric system
that does support ranges, and is improved replacement of ieee floats
phf: alf keeps saying
that
there are localized solutions
to various classes of numerical problems
mircea_popescu: and for
that matter, among even numbers 2 - 8 is as much of a range as 1-4 is among natural numbers.
phf: i don't understand why
the point mircea_popescu would be so controversial among compscis here, it's sort of
the whole point of numerical analysis
mircea_popescu: there is no such
thing as contiguity or ranges
that machines can grok.
mircea_popescu: it is, in places such as ada, "must use contiguity"
the EXACT equivalent of "my ai program
thinks because
the procedure is called <<understanding>>"
mircea_popescu: jurov ranges are a sort of abstraction unavailable
to computers.
jurov: mircea_popescu:
there is, using ranges
mircea_popescu: there is no way
to represent "all
the numbers between 1 and 2" in a finite space.
mircea_popescu: nothing imaginary about it, it's factual.
the sum
total of all satoshi in existence once
the bitcoin ran its course will not be 21mn, but 131k satoshi less.
mircea_popescu: and
the problem is not, apparently, resolvable by fixing
the machines. even if you had ideal machines,
they'd still haveto halve blocks, and
the results would still conceivably be...
this.
mircea_popescu: but in
the end, i say
this :
the difference between 20999999.9769 and 21000000.0000 is
the unpayable change. point of fact remains
that we can't escape
this situation where we draw one
thing, and
the machine pops up another
thing.
mircea_popescu: yeah well. my
theoretically-useful example failed
the
test of practice.
mircea_popescu: (that is in practice an equivalent of
the mantissa
trick, it allows you
to get out of all sorts of problems)
mircea_popescu: i picked it as a point because of
the "abstract number"
thing, ie, redefine 4.
mircea_popescu: and
this is why it goes all
the way down
to fortran.
the unresolved problems are actually very deep.
mircea_popescu: no, it sounds like until
the day you can machine-represent an irrational quality without rounding, you're lying
to yourself about having washed anything.
mircea_popescu: has no bearing on
the problem
that hey, math is still not represented, and some people'd like it
to be.
mircea_popescu: but all
this ~fundamentally c discussion~ which is EXACTLY what it is,
mircea_popescu: so yes, i can understand WHY you'd like it,
think you need it, see it as best or better.
mircea_popescu: the fundamental problem here being, obviously enough,
that
the FORMALISM used
to describe math for and in computers is shit. and yes, on
that shitty basis you will never have math, but an engineering-useful hack.
mircea_popescu: to
the question "why did
they not define EVEN correctly, eschewing
this problem
they perceive with mod" you answer
that " ada is a civilized lang like commonlisp and
there is NOT a presumption
that integers are machine words !".
This objection, if accepted as
the correct response, ALSO invalidates using, say, XOR, and for
the same reason.
mircea_popescu: the problem is
this : in ada manual it is said, "So we see
that
the predicate in
the subtype Even cannot be a static predicate because
the operator mod is not permitted with
the current instance. But mod could be used in an inner static expression." ; it is further said "and, in addition, a call of a Boolean logical operator and, or, xor, not whose operands are such static predicate expressions, and, a static predicate expres
mircea_popescu: ie,
that
the
thing as-is is broken, and should be fixed, but other
than
that
the idea is ok ?
mircea_popescu: so
then are you saying ada shouldn't allow either mod or > ?
mircea_popescu: how are you going
to define a sorting rule for a
type you don't know yet.
mircea_popescu: it's
true
that haskell
tries and splendidly fails at being not-this.
mircea_popescu: so
this is what i mean by "it's a c" : whatever
the fuck you do, it's still going
to be an ugly hack where you "can't define
this" but nevertheless define
THAT, which is just as broken.
mircea_popescu: a math machine is one where you define bitcoin max coins and
the sum works properly and
the series converges
to 21mn
phf: asciilifeform: recall
that i spent probably most
time here on
tinyscheme going as far as writing swank integration and unreleased bignums, i'm saying
that you go
through phases of "this is how we solve bitcoin". i grok
the value of ada, and i grok
the value of scheme, but neither are alternative-less. in fact with
the amount of skill available, simply hacking on btc consistently we would've been further along
mircea_popescu: an engineering machine is one where you define bitcoin max coins as it's defined, and you end up with a max equal
to 20999999.9769 btc.
mircea_popescu: " people don't like
them, but it is also possibly
the case
that
they've gotten as close as possible." <<< we're in exact agreement, here. no alternative known, and not likely possible.
phf: asciilifeform: well, same way as no particular alternatives were seen
to
tinyscheme few months ago
mircea_popescu: "mod is not definable here because hey, abstract
types. nevermind, > is defined, fuck abstract6
types".