713 entries in 0.947s
mircea_popescu: but on the other side of the coin, "As you probably already know, but my trademark thoroughness still obliges me to say,
AI stands for "Artificial Intelligence" and comes in two flavors, "deterministic" (like minmax) and "statistical" (like SVM). The combined efforts of various researches lead to an important breakthrough in this field, known to meteorologists as "the
AI winter". This is the season when you can't get any VC m
mircea_popescu: be that as it may, the conceptual item solves two major open questions here in tmsr forum : a) "why is this proggy 100mb ?!?" "because it has 100mn datapoints in the
ai massager, let it be" ; b) "why does this proggy want to call home ?" "because whenever you accept a design and put it in production, all other installations want to review the datapoint, it's valuable."
mircea_popescu: fror bonus points - steal google's go playing
ai and modify it. the problem of routing is grid-discretizable.
mircea_popescu: and there's a damned good reason you don't, nor is it a service you offer : the difficulty of "looking at the data, establish its validity" exceeds building the
ai.
a111: Logged on 2016-10-31 15:12 mircea_popescu: very weird, the list of nonsense these people sprout. it's as if they have 0 reflexive capacity whatsoever, how am i supposed to distinguish google "
ai" from some douche who spits out " Your style of just summarily shitting on how the author wrote this piece adds no value to the discussion." ?
mircea_popescu: very weird, the list of nonsense these people sprout. it's as if they have 0 reflexive capacity whatsoever, how am i supposed to distinguish google "
ai" from some douche who spits out " Your style of just summarily shitting on how the author wrote this piece adds no value to the discussion." ?
☟︎ phf: oh by the time a111 can recover from all possible freenode failures we've achieved the dream of hard
ai mircea_popescu: but then again, it is the necessary fate of isolated motherhood. they've set out to build
AI and ended up building a meta-Lucy instead. apparently spontaneous organisation emerges in chimps with "humanities" degrees before it emerges in doped silicone.
mircea_popescu: is this the guy who ran some sort of "
AI" to make predictions and then decided to "turn it off" because something sounding a lot like yakowsky or w/e the harry potter fanfic writer's name is ?
phf: asciilifeform is really the last self aware symbolics 3640, that they didn't destroy during the great
ai purge in the 80s. (lisp of course succeeded, DoD panicked and wiped out all traces of
ai, save for one..)
mircea_popescu: "writ large". see alf ? the go "
ai" is already taking over "science".
mircea_popescu: "all ais are subject to registration. any
ai found without proper registration will be destroyed."
mircea_popescu: but yes, we cameto a crossroad, where
ai denizens are actually a significant enough portion of the republic's cohorts.
mircea_popescu: (it stands to reason that if YOU are trying to find the enemy / rabbit / whatever you want to take stationary positions and watch ; yet people don't usually want to play a game against
ai that does mostly this.)
phf: stealth computer games are best suited for
ai problems, because the algorithms fall under a very specific category of differential blind search games with mobile hider studied and analyzed since the 60s as part of differential games subfield of game theory (DoD likes to throw money at stuff like that, and the first treatment of subject is by a RAND corporation guy). there's no known optimal strategy for >2 node networks, but there are
a111: Logged on 2016-08-25 00:56 mircea_popescu: "better
ai" is no improvement when it traduces in practice to "must play a game with known/computable solution" .
thestringpuller: Sure. But the notion of game
AI "knowing the player's location" isn't really a thing these days.
mircea_popescu: thestringpuller
ai is
ai, sensory advantages are sensory advantages. different categories.
mircea_popescu: "better
ai" is no improvement when it traduces in practice to "must play a game with known/computable solution" .
☟︎ BingoBoingo: Classic Pac Man also has way to soundly defeat
AI mircea_popescu: imo the best
ai any game ever displayed to date was pacman.
a111: Logged on 2016-08-24 18:49 thestringpuller: game
AI is pretty advance on most fronts, it's not trying "to be human". For instance being hunted in a stealth game. If they really wanted to, they could make the
AI find you EVERY time. A game of hide and go seek that's impossible to win. So the in-between is this "puzzle-like" element.
phf: yes, you can make
ai that can smell you all through the field, but then it's equivalent to "knowing where you are", and no
ai is involved. it's simply a machine that can pinpoint a location based on small with p reliability
phf: the impossibility in this case is inverse relation to "knowing where you are", not to "
ai". because making
ai with limited sensors (like by say using same set of sensors as what you have) find you is a tricky problem
thestringpuller: asciilifeform: okay. so in game of hide and go seek in kojima's world.
AI is like player but more liek child. It can see, it can hear, it can even smell. So if player is running in rain and footsteps appear,
AI follows players trail. If
AI hears suspicious noise in certain area it goes and "looks around". The thoroughness of its hunt is relative to difficulty. I was saying more if you want to go the "purist" route. You can make the hi
thestringpuller: game
AI is pretty advance on most fronts, it's not trying "to be human". For instance being hunted in a stealth game. If they really wanted to, they could make the
AI find you EVERY time. A game of hide and go seek that's impossible to win. So the in-between is this "puzzle-like" element.
☟︎ mircea_popescu: "oh your world won't work, demiurge" "no problem, i'll add
ai to it" "idiot"
mircea_popescu: which is a loud testament of their idiocy - much harder to do BOTH world AND
AI at the same time.
thestringpuller: mod6: it's insane man. the
AI in Absolution is a beast. when you kill someone normally, like with gun or fiber wire, it leaves marks on the body, and
AI will see that. so they detect someone is there. however, there are ways people can die where the
AI is just like "Huh. That's a shitty way to go." For instance I put some kinda acid in this d00ds hair regrowing gel or whatever. he puts it on, and it melts his head off and everyone aro
mircea_popescu: there isn't
ai. and it costs too damn much to actually hand-eval contracts.
mircea_popescu: "Fiind mai negru ca ţiganii ce-
ai tot cerşit la noi cu anii, tu, cel primit cu dor de sus nici bogdaproste nu ne-
ai spus. Cu dar de pâine şi de sare, cu vin din beciul nostru mare te-am ospătat. Iar tu în zori râzând, te-
ai scîrnăvit în flori. Apoi prin codri de milenii
ai tot umblat de dragul lelii. Ei, vezi atunci? Pun mâna-n foc: tu n-
ai fost cal arab, ci porc!"
shinohai: I am convinced that asciilifeform is really just some sort of
AI comprised of Elbrus chips.
mircea_popescu: ocr is really a good fit for the markov chains as
ai they do
diana_coman: it's actually part of why I quickly became disenchanted with the whole
AI school (well, all of them really): I don't get why the focus on *always right* when no living intelligence (which is supposedly the model in any case) is *always* right
mircea_popescu: and yes, the google "
ai" rule based bs is giving a bad name to all this, but abstracting the usg.idiots for a second, before we throw out the bath
mircea_popescu: "mult stimate ion caciula, om al muncii fara scula : cit
ai degete si limba, legea tarii nu se schimba."
☟︎☟︎ a111: Logged on 2015-03-20 03:35 mircea_popescu: "Roko's basilisk is a thought experiment that assumes that an otherwise benevolent future artificial intelligence (
AI) would torture the simulated selves of the people who did not help bring about the
AI's existence. [...] The concept was proposed in 2010 by contributor Roko in a discussion on LessWrong. Yudkowsky deleted the posts regarding it and banned further discussion of Roko's basilisk on LessWrong af
mircea_popescu: so here's a question/job for any
ai/machine reading afficionados :
mircea_popescu: asciilifeform> why couldn't hitler have picked a plausible shitoshi ? << was just discussing this with the gurlz. you gotta appreciate hitler is not a person ; hitler is the fucking google
ai, practically.
ben_vulpes: not an expert system, not
ai, not anything special.
mircea_popescu: thinking of text in "
ai" terms, the more complex the graph / strlen, the tenser.
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: we're actually building a sort of cyber-
ai-implant here, it's obvious by now.
phf: kind of like rms bringing back mit
ai lab, when all the cool kids left for smbx
mircea_popescu: arguably easier (or at least more accessible) work than slaving butt off to recreate
ai like phf's been trying.
mircea_popescu: it'd be pretty epic you know ? actually a very good ground for
ai development, cuz so formalized and well scored.
mircea_popescu: DianaComan actually i suppose one could try and make
ai for foxybot
mircea_popescu: it's of the nature of "it is possible, at least by napkin pullulation, that one day we have
AI".
mircea_popescu: possibly the one place for
ai in intelligent discourse.
assbot: Logged on 15-03-2016 13:22:03; mircea_popescu: in the "i bet you don't understand what the bayesian approach to spam filtering turned into pretend-
ai after its failure to filter spam" means for your country : the next step is for most all the jobs that are still left to be replaced by this kind of
ai - it's cheaper. so you go to the hair salon and you get a hairdo - 80% of the time. hey, accidents happen, here's a free coupon for another one. and
mircea_popescu: in the "i bet you don't understand what the bayesian approach to spam filtering turned into pretend-
ai after its failure to filter spam" means for your country : the next step is for most all the jobs that are still left to be replaced by this kind of
ai - it's cheaper. so you go to the hair salon and you get a hairdo - 80% of the time. hey, accidents happen, here's a free coupon for another one. and yes, you want a we
☟︎ thestringpuller: that seems to be a practical problem of neural nets. had an ex who's best friend's dad at the time worked on
AI. He told me this story about how they were training a neural net to identify enemy tanks from allied tanks. well all the allied tanks had good pictures. the enemy tanks were recon photos taken at night. So the nn ended up learning to tell difference between photos taken at night vs day
assbot: Logged on 09-03-2016 16:19:22; asciilifeform: thestringpuller: there were once two schools of thought re: how to do '
ai'
thestringpuller: I'll text-ify it later. But I guess point it, don't you think computational power plays a role in what
AI does in practice?
thestringpuller: that's why that article was so interesting to me as a 13 year old since the
AI in games was exhibiting emergent behavior not seen previously. (or at least throughout my childhood)
mircea_popescu: they apparently didn't realise this going in. no
ai in pr, yet.
thestringpuller: meh. being a huge fan of
AI, I understand the harware yet exists to create more formative models of "intelligence"
thestringpuller: well i was more under the assumption philosopher stone of
AI (a sorta Star Trek TNG "Data") doesn't currently have a means of existing
thestringpuller: asciilifeform: perhaps different outlook, knowing you can't emulate human mind on current machines. perhaps the only way to "true
AI" is some cyborg thing of a brain plugged up to machinery...
trinque: it also seems to suggest that mind-amplification is a much nearer goal than
AI, as the brain's own ability to handle new inputs can be leaned upon
thestringpuller: I really don't think you can actually decipher the nature of 'real thought' on current hardware.
AI will always be an emulation of the authentic thing.
punkman: "S. Korean Go player lowers expectations before facing Google
AI"
mircea_popescu: yeah yeah, and bullshit textbook on
ai that leads nowhere.