log☇︎
143900+ entries in 0.085s
a111: Logged on 2017-11-14 20:43 mircea_popescu: no conception of who the enemy is and why exactly, etcetera.
mircea_popescu: i could reinterpret the objection in the terms of "mit ai lab had no common ideology", a copacetic statement of "they failed to republic" which is exactly what i meant by http://btcbase.org/log/2017-11-14#1738294 and so on. ☝︎
asciilifeform: re (b) neither of'em was pictured as in same market at all, any moar than pepsicola
mircea_popescu: given that a) someone was going to make a billion dollars and b) it'd have been preferable it not be gates & jobs.
mircea_popescu: not altogether clear how self-serving that was.
asciilifeform: near as i can tell, he was the proverbial nursing home patient described on trilema, who says 'died, whaddayamean, died, where is jane?! we were USED to her' re every colleague who went off to bolix or lmi
mircea_popescu: phf in the sense that they were trying to sell the machines, but avoid as much of the dev work as humanly possible ?
phf: mircea_popescu: i was taking a poetic liberty there. it's the new appearance of a self serving strategy where there wasn't necessarily one (or the strategy changes drastically). the core of activity shifts elsewhere, but the original points of contact are maintained for the sole purpose of taking any new ideas that might've been introduced.
mircea_popescu: control as the psychological construct is complicated.
asciilifeform: he wasn't the director of lispm group or anything of the kind
asciilifeform: d00d was in no sense in control tho
mircea_popescu: his political activism later on entirely predicable on this.
mircea_popescu: no more than this.
mircea_popescu: i suspect he was a bureaucrat (intellectually insufficient ; aware of this ; politically ambitious) and he resented the lack of control over smarter people.
asciilifeform: i suspect that he simply did not even think about it.
asciilifeform: mit was never in the business of selling comps
asciilifeform: it was never clear to me from rms's surviving writing what he thought mit ought to have done with the lispm, if not to license to manufacturers
mircea_popescu: plurious agents can own the same idea in the mannger they can have the same girlfriend. knowing about it not required.
mircea_popescu: it is particularily difficult to explain what one means by theft of a) ideal objects by b) parties who actually own them intellectually and c) when the objects were produced within their lifetimes.
a111: Logged on 2017-11-14 20:10 asciilifeform: apeloyee: the other thing to consider, is that the preserved bolix material has ~unspeakably~ rich ( and quite high snr ) collection of artifacts, perhaps 1000 asciilifeform-years of work. take the ns vlsi compiler alone. i have the binaries, but not the src. and ~someone~ will have to make a sane (i.e. fully lispified and zero-externals) vlsitron.
asciilifeform: rms was not ,in so far as i can tell, involved in the http://btcbase.org/log/2017-11-14#1738171 aspects ) ☝︎
asciilifeform: none of the code in question -- at least in so far as it is accessible to modern archaeologist's eye -- was particularly deep
mircea_popescu: asciilifeform i had the same impression. social not technological issue.
a111: Logged on 2017-11-14 20:43 mircea_popescu: it was involved as mental poison in these people's heads.
asciilifeform: iirc rms was most incensed re the luring away of ~people~, rather than lifted code
mircea_popescu: im not even sure what theft means in this context.
phf: well, rms's nervous break down makes some sense when you actually read the code. there's a very obvious clear "steal everything that's not bolted" going on. cadr's SYSTEM sources were being developed in parallel with symbolics and LMI (to the point that there are conditionals in the cadr's code to compile os both on lmi and genera), with interactions slowly decreasing over time
asciilifeform: ( died with the sunken atlantis, nothing afaik heard of it since 1990 )
mircea_popescu: phf i had the idea cadr more of a common ancestor, to smbx too.
mircea_popescu: amusingly bucharest had nfi idea at the time it COULKD have done such a thing. literally 0 conception of what any of the items we're discussing here WERE. at all. not even the priors present.
phf: it's funny that 36xx series is basically an improved cadr. ivory on the other hand? literally scheme86: they poached both the main guy who worked on the cpu ~and the entire toolset~. ivory was still designed on CADR (rather than smbx), because that's where scheme team designed theirs ☟︎☟︎
mircea_popescu: asciilifeform something deeply wrong with response to mother shrieking about her daugher being fucked diverging from beating mother up / fucking her too. but... we digress.
asciilifeform: same reason for ~90% of closedsourcedriverism today, say.
asciilifeform: 'must hide my theft'
mircea_popescu: "worst thing that can happen to smart people is an easy money source early on"
asciilifeform: and yes if you publish , patent troll could easily use it against you, so culture of seekrecy
phf: well, folx in question plundered as much as they could from mit and carefully guarded whatever they themselves produce. i don't know about there not being "logical reason to publish"
asciilifeform: they were ~same as su academics , in that respect
mircea_popescu: no conception of who the enemy is and why exactly, etcetera. ☟︎
mircea_popescu: it was involved as mental poison in these people's heads. ☟︎
asciilifeform: mit was not involved other than as patent licensor/troll at that point
asciilifeform: ( can't even say i have any outrage over this, there was no logical reason to publish )
asciilifeform: these folx had no notion of making any of the details public.
mircea_popescu: esp as this apparently was ephemera.
asciilifeform: plenty of unanswered q's re the arch.
asciilifeform: mircea_popescu: wish i knew ! sorta why i wanted the thing microscopied.
mircea_popescu: (lulz of all time, 3600 came with built-in bitcoin miner)
asciilifeform: but modern silicon instead gets 2million transistors for motherfucking tlb cache so that winblowz can page less.
mircea_popescu: asciilifeform what did they use as hardware hash ?
asciilifeform: asciilifeform ( and quite a few other folx, i suspect even earlier ) realized in ~2005 that it was eminently practical with modern silicon
mircea_popescu: he's now at describing the mindbogglingly ugly hacks of how to get 32 bit fixnums out of 36 bit words with 6 bits of tag and 2 of code and so on, "oh actually the lowe order quarter can be repurposed"
asciilifeform: item grew as the iron became available ( i do not even say 'affordable' , thing was a 'cost no object' design, like aston-martin )
asciilifeform: fwiw the 3600 series ( pre-ivory ) were 36b
mircea_popescu: and that meanwhile became quite technologically comfortable.
asciilifeform: whole thing was a valiant effort to make a thing that was just short of technologically impossible.
asciilifeform: as it is 40b broke the back.
asciilifeform: they'd happily have made 512b words, if the iron existed.
mircea_popescu: asciilifeform yes but too small.
asciilifeform: and no such thing as untagged, on whole box. that was the whole strength of the edifice.
mircea_popescu: "oh, couldn't be done on hardware of back then". tell you what, fucking your wife also couldn't be done on your hardware of back when junior high.
mircea_popescu: whatr the fduck is he even talking about, "tagging every word". fucker, small words hardly worth tagging! make your word 4096 bits, tag THAT.
mircea_popescu: if this makes sense...
mircea_popescu: asciilifeform typical pc has the following situation : 64 bit registers, 128 bit memory, 1024 bit disk sectors, 64 mb video buffers, and atop sitting a drunk driver who thinks 8 bits are a byte. ☟︎
asciilifeform: well yes. hence typical pc has 128+b wide drams
mircea_popescu: asciilifeform because you ~never read one bit. that's the issue here, whole fucking hardware is built around nonsense. nobody reads 1 bit ; and for most applications 1 byte (ie, 64 bits!) is actually an underbyte.
mircea_popescu: and this matrix is about... 1mb wide ?
mircea_popescu: asciilifeform how is a 1GB ram chip store the 2^30 bits ?
asciilifeform: what's the use of simultaneously lighting up a whole line, simply to blow the clock cycle on reading 1 bit ??
mircea_popescu: but this is spuriosity. let it feed out... the whole line.
mircea_popescu: now suppose i want to read some of this memory. what happens is that the controller reads A WHOLE LINE, and then selects from it the bit i wanted and spits it out.
asciilifeform: but even with the widest, you will need quite a few to make machine with, e.g., 8192-bit word. ( and your cpu now needs 8192bit regs 0
asciilifeform: at one time they were uncommon, and a stick of memory tended to have as many chips as the bus bitness
asciilifeform: ( e.g. 0000000000000000 input gives you the first of'em )
asciilifeform: so you can readily buy, e.g., 16x1 i.e. 65536 bits of ram, addressable 1 bit at a time
asciilifeform: chip is typically 'x * y' where x is input bitness and y -- output
mircea_popescu: consider if you wrote/read the disk/net/video 1mb at a time.
asciilifeform: let's consider how ram is typically sold
asciilifeform: ( the 1 exception where the cost is nonlinear may be the barrel shifter )
mircea_popescu: rather, keep the regs, gimme "page to page hdd to ram and video" ops.
asciilifeform: with same transistor cost
asciilifeform: i have noted in the past that it is possible to have, e.g., wider regs but fewer of'em
mircea_popescu: asciilifeform this seems sufficient cause to you ?
asciilifeform: there being a relationship at all, comes from the fact that the two types of bucket regularly get emptied into one another
mircea_popescu: no, i am asking why exactly is one item unrelated to another item married to it.
asciilifeform: well orig q was 'why is there a relationship' but seems like mircea_popescu is really asking 'why pc regs so tiny'
asciilifeform: ( they are only reflected in iron at all, on 'optimized for c' abortions where have 'iron pagetable' )
mircea_popescu: in this model they are. memory (ram, disk, whatever -- comes in pages)
asciilifeform: pages are not a fundamental unit tho
mircea_popescu: let's instead say thatr we have a modern machine, with 64 bit bytes and 1mb bit pages.
asciilifeform: mult is necessarily 2b wide ( where b is the reg bitness )
mircea_popescu: how does mult of two regs look ?
asciilifeform: how would memory-to-reg and reg-to-memory MOV look on this iron ?
asciilifeform: let's suppose the contrary
mircea_popescu: for one thing, historical concept of pages seems to suggest that the relationship is exactly of the nature of "byte=octet, we're dumb".
mircea_popescu: why, again, is there a relation between cpu register bit width and memory-allocation-unit bit width ?
mircea_popescu: since we're doing this, here's a fundamental point i don't understand.
asciilifeform: ( src, another d00d who was there, infamous crackpot http://fare.livejournal.com/168016.html )
asciilifeform: ividual bought the assets and Kalman spent 4 years as a "captive consultant" for him. Since 2002, when he ran out of money, I've worked for Ab Initio (doing nothing with lisp except emacs hacks).'
asciilifeform: Kalman was hired after that by Symbolics Technology, Inc., who had acquired the assets of Symbolics from the bankruptcy court, to work on further productizing the emulator for the Alpha. This was used by John Mallery at MIT to run his document distribution system for the White House during the Clinton presidency. When the owners of that venture decided that a financial company had no business owning a computer company, a private ind
asciilifeform: 'Kalman Reti worked for Symbolics from 1982 through 1992, mostly on VLSI tools for Ivory but also writing low-level device support (e.g. a LMFS recovery utility, R/W optical drive support, LZW compressor, etc.) The last few years of that were spent doing customer consulting. After being laid off when Symbolics went into chapter 11, Kalman worked for Apple in Cambridge, first on MCL and later, after MCL was sold to Digitool, on Dylan.