log☇︎
700+ entries in 1.117s
asciilifeform: mircea_popescu: problem is, 'database' as commonly seen in sqlworld is not 'collection of btrees, rtrees, etc.' in encyclopaedic algorithmic form, but a misbegotten attempt at ai.
asciilifeform: at the ai lab.
asciilifeform: 1970s mit ai lab != modern mit
asciilifeform: again it wasn't 'mit', it was 'mit's ai lab', wholly - at the time - unique assemblage.
asciilifeform: ai lab strictly.
phf: mircea_popescu: of course not predict. likewise no concept of AI is involved anywhere
asciilifeform: take my old example, 'boobytrap an fpga.' elementarily you WILL need to somehow fit an ai in there, to create any serious problem for UNKNOWN bitstream
a111: Logged on 2017-03-29 23:32 ben_vulpes: and in other wonders of "ai": http://boingboing.net/2017/03/29/countermeasures-are-a-thing.html
ben_vulpes: and in other wonders of "ai": http://boingboing.net/2017/03/29/countermeasures-are-a-thing.html ☟︎
mircea_popescu: about half of what droive the whole "image recognition ai" wank
mircea_popescu: ie, "this sounds great and explains nothing" ; much like the void thinking (void) ai program.
ben_vulpes: oooh and make a bunch of work for 'ai' image recognition programmers
asciilifeform: the ai lab d00d also had nfi that the source existed.
asciilifeform: made friends with d00d who worked in ai lab, he gave me a working alpha (i was a student, and had $0) and FULL set of bolix docs. which i still have.
mircea_popescu: http://btcbase.org/log/2017-03-10#1624395 << your emulator stil lsucks what can i tell you. ai winter! ☝︎
mircea_popescu: meanwhile in ai news, http://68.media.tumblr.com/4277bdf9e99091bd0846649df6201385/tumblr_o6dkfiaXG61umjwnoo1_400.gif
mircea_popescu: well... "ai" at any rate.
mircea_popescu: otherwise this too much resembles the ai problem of thought();
mircea_popescu: there's nothing andressen has to contribute to any discussion. if you want copy/paste "ugc" you can just as well use google "ai", for the exact same result.
trinque: that the character which embodies this is "AI" is not important; it's behavior is entirely human
a111: Logged on 2017-02-22 07:26 mircea_popescu: anyway, there's a common thread going through the google go ai, attempts to "secure the banking system against risk", the surveillance state / internet of things / smart cars and so on.
ben_vulpes: mhm "even google ai is better at driving than most people"
ben_vulpes: aah. hey is the fed ai yet?
ben_vulpes: 'ai' to handle the unexpected situation/'unknowns problem'?
mircea_popescu: anyway, there's a common thread going through the google go ai, attempts to "secure the banking system against risk", the surveillance state / internet of things / smart cars and so on. ☟︎
ben_vulpes: to use it against other 'ai'?
mircea_popescu: yes yes, except it's as far from a fact as it could possibly get. it is, in nucet, the HOPE of usg "ai" work.
mircea_popescu: now, SOMETIMES, it's like in the case of the lisp ai, where i was lazy and didn't actually provision properly.
thestringpuller: "After further investigation, it turned out that the two AIs were communicating using a novel symmetric key cipher, and the key that ended up being uncrackable by the third AI was simply '12345', the kind of password an idiot would have on his luggage."
thestringpuller: https://arstechnica.co.uk/information-technology/2016/10/google-ai-neural-network-cryptography/
asciilifeform: mircea_popescu: i don't do ai. i do ia.
mircea_popescu: asciilifeform you keep trying to design-for-ai. there's no need and no call for that ; moreover no practical way to ever do it. wot permitted me to ascertain enough information to eg behead that adlai schmuck most recently. there's no way "for you to explain how i reasoned to a machine", or even in general, and that doesn't hurt or stop me. moreover if there was such a way, then he'd simply act differently next time. and ther
mircea_popescu: asciilifeform nothing is easier than http://trilema.com/2017/my-ai-problems-a-humble-confession/#comment-121016 "I didn't ask the girl if she thinks I'm any good or not. This is neither her business nor something within her purview. I asked the girl if she thinks she has it in her to try and survive once the world (WHICH I OWN) is no longer friendly to her. I didn't ask the DA for his evaluation of me. I couldn't give less of
mircea_popescu: v links add value ; which couldn't have been made by "an ai"
mircea_popescu: ben_vulpes first mentioned (ever! in history!) here : http://trilema.com/2017/my-ai-problems-a-humble-confession/#comment-121007 then referenced in http://trilema.com/2017/dupa-dealuri/#comment-121112 and http://trilema.com/2017/how-to-deal-with-little-girls/#comment-121192
trinque: ben_vulpes: http://trilema.com/2017/my-ai-problems-a-humble-confession/#comment-121007
mircea_popescu: AI to power unauthorized access against foreign state and non-state actors, then both code and design could be interpreted as defense items. Then it could fall under Title 50 of US Code, appear on the United States Munitions List, Category XVII (Classified Articles, Technical Data, and Defense Services Not Otherwise Enumerated) and therefore be subject the Arms Export Control Act. Once under DoD control those examples of QC c
mircea_popescu: really not worth doing ai for this.
mircea_popescu: the problem with "string" isn't there just to fuck with mp's ai bot.
ben_vulpes: ai!
mircea_popescu: dang these ai lisp bots are getting better by the day
ben_vulpes: and in other ai herpderps: https://www.inverse.com/article/26054-google-home-bots-falling-in-love-on-twitch-is-the-best-romance-of-2017
mircea_popescu: "we created ai that can pass any college exam. it... counts letters."
mircea_popescu: "evil ai" or "genetically modified" or you know, something.
mircea_popescu: as good a statement as any re why moore law can't ever produce ai.
ben_vulpes: mircea_popescu: http://trilema.com/2017/my-three-days-of-ai/#identifier_10_71072 << busted
ben_vulpes: also http://trilema.com/2017/my-three-days-of-ai/#selection-3947.0-4087.77 would be worth preserving whitespace in to make the point
ben_vulpes: mircea_popescu: many of your parens escapes did not work in ai post
deedbot: http://trilema.com/2017/my-three-days-of-ai/ << Trilema - My three days of AI
phf: typically people recommend http://www.gigamonkeys.com/book/ and it's a sort of "rails for beginners" kind of book, but i think best cl option is http://www.norvig.com/paip.html. norvig's paradigms of artificial intelligence programming. it's not so much about "ai", but about some very useful symbolic algorithms, written in ~very elegant~ lisp code
asciilifeform: http://btcbase.org/log/2017-01-11#1601428 << poker 'solving' ai per se is ~uninteresting. the genuinely nifty thing would be a 'realistic humanlike' (e.g., 'loose') player. ☝︎
mircea_popescu: pete_dushenski ftr, poker needs no such thing as "ai". poker is a deterministic game if you wish, can program machine to play extremely tight, will beat most humans.
pete_dushenski: speaking of better than average ai, looks like some boys from the u of a cracked heads-up no-limit texas https://arxiv.org/pdf/1701.01724v2.pdf
ben_vulpes: ai is impossible and bugs are invariant
mircea_popescu: i thought lisp makes ai possible and bugs impossible.
mircea_popescu: http://trilema.com/2017/my-ai-problems-a-humble-confession/#comment-121016 << convo re "qualia" ended up oddly related.
a111: Logged on 2017-01-07 18:04 mircea_popescu: ah. the way i was thinking this'd work would be : the bot answers to any lines where its name is mentioned ; and i can update its "brain" with a !^ url style command. whereby it replaces its "ai" code with the content of the file.
deedbot: http://trilema.com/2017/my-ai-problems-a-humble-confession/ << Trilema - My AI problems, a humble confession.
asciilifeform: 'It is my opinion that the fans of this family of hardware/vaporware, consistent advocates of The New Age of Computing, have serious AI problems. Here's a sample quote on cellular automata: "I guess they really are like us." Well, if you want to build a computing device in order to have a relationship with it, maybe a cellular automaton will do the trick. Although I'd recommend to first check the fine selection of Homo Sapiens we have
a111: Logged on 2017-01-08 13:54 mircea_popescu: and the other thing : that entire "it is unclear whether people aiming to make an artificial friend have seriously considered the much cheaper, ordinary kind" which i can't fucking find in the log for some reason goes A LOT deeper than generally realised. it's not "oh, i'm not discussing ai so it's not about me". yes, IT IS about you. if the software you're contemplating aims to take fifty hours of engineer work to replace fi
a111: Logged on 2015-09-25 21:50 mircea_popescu: http://log.bitcoin-assets.com/?date=25-09-2015#1285635 << exactly the wrong kind of fucking ai. a) not actually intelligent ; b) negative - the last thing i fucking want to see is mechanically powered women nagging ; c) not self contained - at least the stupid nag you can beat into the ground.
mircea_popescu: and the other thing : that entire "it is unclear whether people aiming to make an artificial friend have seriously considered the much cheaper, ordinary kind" which i can't fucking find in the log for some reason goes A LOT deeper than generally realised. it's not "oh, i'm not discussing ai so it's not about me". yes, IT IS about you. if the software you're contemplating aims to take fifty hours of engineer work to replace fi ☟︎
a111: Logged on 2017-01-07 18:15 phf: http://btcbase.org/log/2017-01-07#1598258 << you know those guys that periodically stop by lisper venues, and they don't really program, but they want to use LISP to build an AI, because metacircularity of code is data is giving them mystical visions..?
phf: http://btcbase.org/log/2017-01-07#1598258 << you know those guys that periodically stop by lisper venues, and they don't really program, but they want to use LISP to build an AI, because metacircularity of code is data is giving them mystical visions..? ☝︎☟︎
mircea_popescu: ah. the way i was thinking this'd work would be : the bot answers to any lines where its name is mentioned ; and i can update its "brain" with a !^ url style command. whereby it replaces its "ai" code with the content of the file. ☟︎
mircea_popescu kind-of has a hankering to dick around with "ai", but no time to fuck with linux eigenstates.
asciilifeform: 'friendly ai' was thiel.
mircea_popescu: (ftr, the musk "friendly ai" thingee had NINE employees. to medium's 150.)
mircea_popescu: i fail to see this potential. looks entirely like "hey guise, let's agree on definition of ai that allows the pretense of sv technologee to survive a little longer plox "
mircea_popescu: this is obviously true ai problem.
mircea_popescu: for that matter, "ai" wouldn't be confused with a large bank of switches.
mircea_popescu: https://twitter.com/BunnyFuxalot/status/793954128657412097 << in other lulz the twitter hate speech ai detector is working like a charm.
asciilifeform: so that it can look like 'ai'.
mircea_popescu: anyway, the fate of the socialist empire in its eternal historical cyclicity is fascinating. so - nazis put everything on large tanks ; and failed. but then soviets put everything on a particular notion of industrialization (steel-per-capita industrialization, so to speak) and... failed. so the one socialist empire left standing is putting it all on "ai". and...
mircea_popescu: the applications to eg banking are beyond lulzy. but they're willing to destroy the currency (currency ? what currency ? dollars don't pay nothing, it's notes from hitler if you want anything done) to try and prop it up because outside of usg.ai there's nothing
mircea_popescu: basically, the ~one hope of the imperial technologistic church of some sort of practical results is a peculiar form of ai. they're doubling down on it exactly in the manner usg doubled down on "alternative fuels".
mircea_popescu: iction, generalized AI, we wouldn't have to worry about all the messy stuff like politics and society. They think machines will just figure it all out for us."
mircea_popescu: Joi Ito, who runs the MIT Media Lab, said a wonderful thing in a recent conversation with President Obama: "This may upset some of my students at MIT, but one of my concerns is that it's been a predominantly male gang of kids, mostly white, who are building the core computer science around AI, and they're more comfortable talking to computers than to human beings. A lot of them feel that if they could just make that science-f
asciilifeform: 'AI risk is string theory for computer programmers. It's fun to think about, interesting, and completely inaccessible to experiment given our current technology. You can build crystal palaces of thought, working from first principles, then climb up inside them and pull the ladder up behind you.' ☟︎
a111: Logged on 2016-12-11 19:54 mircea_popescu: phf suppose you make an ai expert system to beat us at go. this gives you two practical options : either include 10gb worth of binary flags preset ; or else have us beat it at go for 10 centuries before it gets to where it plays like a freshly fucked 19yo.
ben_vulpes: http://btcbase.org/log/2016-12-19#1585796 << http://cmu-ai-mirror.bvulpes.com/afs/cs/project/ai-repository/ai/ ☝︎☟︎
phf: i'm pretty sure that ai memos are enough to rebuild computing from scratch. has architectural descriptions, cpu design, fabrication, language designs, text editors ("emacs" before it was taken over by rms is described in one of the ai-memos, both as a standalone thing and as set of TECO marcos), various algorithms
phf: i think "ai research" is a mislabel, because the promise of ai was used to keep the funding going, and there is a lot of failed hacks on the subject, but the bulk of actual work was more about how to do computers
phf: mircea_popescu: csail (mit ai lab then) published "ai memos" from 1959 to sometime in early 2000s where they described a lot of "firsts" in computing in general. scheme spec was published by sussman and steele, lisp 1.5 compiler (first "self-hosting"), lisp machine architecture, etc. most papers are not so much ai as computing in general.
mircea_popescu: phf hey, at least unlike the 15-20 years of "ai research", this ~actually does something~.
mircea_popescu: incidentally, is there such a thing as an "ai code commenter" ? specifically, rather than the hard problem of "human speech" or junk of that nature, is there any machine approach to turning compileable code into literate code (kuhn) ?
mircea_popescu: anyway. for completeness let it be stated that perceived problems of thought-computing mismatch are thoroughly a matter of perception, and in principle can not be fixed (other than fixing the perception). it's the fundamental problem of "ai", as derided often enough here (see the "what if you name the procedures something other than "understanding" etc ; see also chomski's attacks on "ai" centered on the constructed repeating
asciilifeform: and, for instance, also had to outwit the 'ai' in the thing to keep it from 'recognizing state machine' and ADDING GATES wtfomfgh
mircea_popescu: phf suppose you make an ai expert system to beat us at go. this gives you two practical options : either include 10gb worth of binary flags preset ; or else have us beat it at go for 10 centuries before it gets to where it plays like a freshly fucked 19yo. ☟︎
phf: but so that it's not just noise, here's one thing that i used as a jumping point, http://foldr.org/~michaelw/log/programming/lisp/diff-sexp there's also a cmu archive of differs, including a working common lisp diff implementation (with a different output though) http://www-cgi.cs.cmu.edu/afs/cs/project/ai-repository/ai/lang/lisp/code/tools/src_cmp/0.html
trinque: at any rate the AI thing is exactly "discover the meaning of life"
mircea_popescu: the same 60-70 years that failed to produce anything woreth the mention in ai also failed to produce an even vaguely coherent model of the state, or of any sort of identity in any field, or of you pick it. not even their distribution models work.
mircea_popescu: but then again : the most hardcore anti-ai stance, and the one to which i vaguely subscribe, is to observe that the 1 is entirely illusory
mircea_popescu: now then : the issue of arbitrarily hard ai existing / being produced is separate of same being produced ~by you~. for any definition of you.
mircea_popescu: i can grant that ai is a psychiatric issue of plenty of computer programmers ; i can even grant that it is perhaps best to pretend like ai is "impossible" in the sense he means to perhaps cure the idiots.
asciilifeform: there are 2 types of ai skeptic, the 'vitalists' ('immortal soul') and the 'shambling horror of computer', 'we don't even know how to make a spreadshit that dun crash' folk.
mircea_popescu: i guess in his terms this makes me an ai weenie. you know, like hofstadter.
mircea_popescu: http://yosefk.com/blog/ai-problems.html
mircea_popescu: as the idea is to make ai not a"i". THAT we already have.
mircea_popescu: oney if you mention AI anywhere in your business plan."