--- Log opened Wed Jan 06 00:00:22 2016
--- Day changed Wed Jan 06 2016
00:00 -!- justanotheruser [~Justan@unaffiliated/justanotheruser] has joined ##hplusroadmap
00:31 -!- poppingtonic1 [~Thunderbi@unaffiliated/poppingtonic] has joined ##hplusroadmap
00:32 -!- poppingtonic [~Thunderbi@unaffiliated/poppingtonic] has quit [Ping timeout: 260 seconds]
00:32 -!- poppingtonic1 is now known as poppingtonic
00:56 < FourFire> kanzure, re tDCS: https://www.uni-bamberg.de/fileadmin/uni/fakultaeten/ppp_professuren/physiologische_psychologie/Publikation/Electrified_minds_Transcranial_direct_current_stimulation.pdf
00:58 < FourFire> I'm ordering this thing, for initial tests: http://www.foc.us/focus-go-flow-tdcs-brain-stimulator
01:10 -!- justanotheruser [~Justan@unaffiliated/justanotheruser] has quit [Ping timeout: 265 seconds]
01:12 -!- justanotheruser [~Justan@unaffiliated/justanotheruser] has joined ##hplusroadmap
01:18 < poppingtonic> .title http://www.foc.us/v2-tdcs-developer-edition
01:18 < yoleaux> foc.us v2 tdcs - tDCS Devices Available to Buy
01:26 < justanotheruser> http://gfycat.com/TemptingCourageousAfricanmolesnake
01:31 -!- jtimon [~quassel@103.red-80-26-235.adsl.dynamic.ccgg.telefonica.net] has joined ##hplusroadmap
01:43 < archels> "The more extreme version is known as transhumanism (h+ for short)."
01:43 < archels> extreme, yo
01:52 < poppingtonic> archels: extreme version of what?
01:52 < poppingtonic> .title http://harpers.org/archive/2016/01/everything-that-rises/
01:52 < yoleaux> [Easy Chair] | Everything That Rises, by John Crowley | Harper's Magazine
01:55 < archels> .ttle https://www.sciencemag.org/content/339/6121/759.summary
01:56 < archels> grm. short Christof Koch review piece
01:56 < archels> with all due respect to Dr. Koch, not even worth the click
02:27 -!- Madplatypus [uid19957@gateway/web/irccloud.com/x-giulvmcnlluixbqa] has quit [Quit: Connection closed for inactivity]
02:59 -!- chris_99 [~chris_99@unaffiliated/chris-99/x-3062929] has joined ##hplusroadmap
03:27 -!- jtimon [~quassel@103.red-80-26-235.adsl.dynamic.ccgg.telefonica.net] has quit [Ping timeout: 260 seconds]
03:40 -!- Gurkenglas [Gurkenglas@dslb-188-103-077-131.188.103.pools.vodafone-ip.de] has joined ##hplusroadmap
03:54 -!- hehelleshin [~talinck@66-161-138-110.ubr1.dyn.lebanon-oh.fuse.net] has joined ##hplusroadmap
03:56 -!- helleshin [~talinck@66-161-138-110.ubr1.dyn.lebanon-oh.fuse.net] has quit [Ping timeout: 246 seconds]
04:13 -!- Gurkenglas [Gurkenglas@dslb-188-103-077-131.188.103.pools.vodafone-ip.de] has quit [Ping timeout: 250 seconds]
04:45 -!- c0rw|away is now known as c0rw1n
05:01 -!- poppingtonic [~Thunderbi@unaffiliated/poppingtonic] has quit [Ping timeout: 255 seconds]
05:05 -!- justanotheruser [~Justan@unaffiliated/justanotheruser] has quit [Ping timeout: 264 seconds]
05:09 -!- justanotheruser [~Justan@unaffiliated/justanotheruser] has joined ##hplusroadmap
05:18 -!- poppingtonic [~Thunderbi@unaffiliated/poppingtonic] has joined ##hplusroadmap
05:30 -!- PatrickRobotham [uid18270@gateway/web/irccloud.com/x-nzuopecnyugpnkzs] has quit [Quit: Connection closed for inactivity]
05:51 -!- poppingtonic [~Thunderbi@unaffiliated/poppingtonic] has quit [Read error: Connection reset by peer]
05:52 -!- poppingtonic [~Thunderbi@unaffiliated/poppingtonic] has joined ##hplusroadmap
06:10 -!- atomical [~atomical@wl2142.cin.net] has joined ##hplusroadmap
06:15 -!- Gurkenglas [Gurkenglas@dslb-188-103-077-131.188.103.pools.vodafone-ip.de] has joined ##hplusroadmap
06:26 < kanzure> .title
06:26 < yoleaux> The End of the Beginning for the Brain
06:32 < JayDugger> poppingtonic, I have been very pleased with foc.us and their customer service.
06:33 < JayDugger> Whether their new US$9 model is any good--I don't know.
06:48 -!- CheckDavid [uid14990@gateway/web/irccloud.com/x-udlcldiexxovkyyk] has joined ##hplusroadmap
07:12 -!- |nodux [uid125132@gateway/web/irccloud.com/x-zyhjdjluuyhssfhb] has joined ##hplusroadmap
07:22 -!- erasmus [~erasmus@unaffiliated/erasmus] has quit [Quit: Namaste]
07:50 -!- c0rw1n is now known as c0rw|away
07:56 < kanzure> oh noes.... george church is an x-risker.
07:56 < kanzure> anti-x-risker.
07:57 < kanzure> look at al these replies, damn http://edge.org/conversation/jaron_lanier-the-myth-of-ai
07:58 < kanzure> jaron lanier is wasting literally everyone's time
07:58 < kanzure> gershenfeld, smolin, church... wtf.
07:58 < kanzure> kevin kelly too..... why do we let jaron lanier out of his cage at all?
08:03 -!- Houshalter [~Houshalte@oh-71-50-56-224.dhcp.embarqhsd.net] has joined ##hplusroadmap
08:11 -!- erasmus [~erasmus@unaffiliated/erasmus] has joined ##hplusroadmap
08:18 -!- jtimon [~quassel@103.red-80-26-235.adsl.dynamic.ccgg.telefonica.net] has joined ##hplusroadmap
08:32 < JayDugger> Because he's a good punching bag?
08:32 < kanzure> jaron lanier? no, he's not a good punching bag. he never leaves. he's terrible.
08:33 < JayDugger> Ah...he keeps most people from getting worried?
08:33 < JayDugger> "It'll never happen and you don't have to worry about it because I say so" and so on?
08:34 < xentrac> is this a new jaron thing or an old jaron thing?
08:34 < JayDugger> Old in my case. I stopped reading him long ago.
08:34 < JayDugger> New if kanzure's talking about something recent.
08:35 < JayDugger> FourFire, you might want to get the developer's edition instead.
08:36 -!- zadock [~outsider@81.180.209.1] has joined ##hplusroadmap
08:37 < kanzure> jaron lanier stuff is always old
08:38 < JayDugger> Has Lanier done anything since his early telepresence work?
08:39 < kanzure> wasted my time
08:47 < kanzure> https://upload.wikimedia.org/wikipedia/commons/c/c5/Comparison_optical_telescope_primary_mirrors.svg
08:52 -!- andares is now known as esmerelda
08:55 < poppingtonic> he did some cool vr stuff back before the first bust, and his tech was included in a movie
08:56 < poppingtonic> xentrac: he's the same as before. look at his book "you are not a gadget" and you'll see arguments from the same train of thought
08:59 < kanzure> only 75 scholar.google.com results for '"kolmogorov complexity" cerebellum'. yeesh.
09:02 -!- ArturShaik [~ArturShai@195.114.246.88] has quit [Ping timeout: 255 seconds]
09:03 < kanzure> and one of them is eliezer hrm.
09:03 -!- erasmus [~erasmus@unaffiliated/erasmus] has quit [Quit: Namaste]
09:06 < esmerelda> Cerebellum in particular?
09:07 < esmerelda> JayDugger: most recently Lanier is working on project Commodore within Hololens iirc
09:07 < kanzure> "Testing a neural coding hypothesis using surrogate data" http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2529148/
09:08 < esmerelda> Whoa. His website is classic 90s
09:13 < JayDugger> Thank you esmerelda.
09:14 < kanzure> "The importance of lateral connections in the parietal cortex for generating motor plans" http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0134669
09:15 -!- ArturShaik [~ArturShai@37.218.166.45] has joined ##hplusroadmap
09:16 -!- poppingtonic [~Thunderbi@unaffiliated/poppingtonic] has quit [Ping timeout: 264 seconds]
09:21 < kanzure> "The role of the cerebellum in perceptual processes" http://link.springer.com/article/10.1007/s12311-014-0627-7/fulltext.html
09:22 < kanzure> "Thus, in electroreception, the cerebellum is involved in sensing external targets by exploiting distortions in signals generated by the animal’s own activity, while cerebellar-like circuits are involved in sensing external targets by eliminating distortions of target signals caused by the animal’s own activity." eeeeeee
09:23 < kanzure> "The cerebellar cortical circuit common to the cerebellum and cerebellar-like circuits has apparently evolved independently in at least five groups of animals: vertebrates, cephalopod molluscs, arthropods, onychophorans, and polychaete annelids. All species in which cerebellar and/or cerebellar-like circuits have been reported are motile and sufficiently large that their kinematics is influenced by inertia, and they interact with other ...
09:23 < kanzure> ... such animals. Inertia constrains how the kinematic state (position, configuration, and rates of change) of an object changes as a function of applied force, such that, if an object has inertia, then information about its kinematic state can be used to predict its future position and configuration at least in the short term. This is not true of animals (or indeed objects of any kind) whose mass is small or drag is large relative to ...
09:23 < kanzure> ... applied forces [68]."
09:23 < kanzure> "Animals that have evolved cerebellar(-like) circuits are, therefore, animals for which probabilistic inference about the kinematic states of self and others is both possible and useful. The fact that this group includes disparate, unrelated species indicates that the genetic and developmental capacity for cerebellar(-like) circuits may be shared by all animals with nervous systems and that it has been co-opted by evolution whenever ...
09:23 < kanzure> ... there has been an ecological opportunity for animals capable of dynamic motion prediction and control [69]. More generally, the ability to predict state trajectories of dynamical systems from observations provides a core capability that may underpin a wide variety of perceptual, cognitive, and motor tasks [70]."
09:24 < kanzure> "Until a few years ago, the Kalman filter was the only known practical algorithm for dynamical state estimation [71]. It assumes linear target dynamics, an assumption that does not hold for mechanical linkages like human and animal bodies. Newer algorithms based on drawing random samples from probability distributions defined by observations are able to track states of high-dimensional nonlinear systems [72]. These algorithms can be ...
09:24 < kanzure> ... implemented using spiking neurons, in which a spike at a particular location in a network represents a sample at a particular location in the state space of the system tracked by the network [73, 74]. There is growing evidence that neurons use Bayesian Monte-Carlo algorithms of this kind to implement decisions and actions [75–83]."
09:32 < esmerelda> This also sounds quite a lot like reinforcement learning
09:32 < esmerelda> Like, it's a Markov decision process if neurons represent states of a system right?
09:33 < kanzure> reinforcement learning does not impose restrictions on implementation like "one neuron per state space point"
09:34 < esmerelda> So it could be reinforcement learning, then?
09:34 < esmerelda> One particular implementation thereof
09:48 < Diablo-D3> .title http://www.ncbi.nlm.nih.gov/pubmed/26641877
09:48 < yoleaux> A Culture-Brain Link: Negative Age Stereotypes Predict Alzheimer's Disease Biomarkers. - PubMed - NCBI
09:56 -!- jtimon [~quassel@103.red-80-26-235.adsl.dynamic.ccgg.telefonica.net] has quit [Ping timeout: 250 seconds]
09:56 -!- zadock [~outsider@81.180.209.1] has quit [Quit: Leaving]
10:16 -!- nmz787_i [~ntmccork@134.134.139.76] has joined ##hplusroadmap
10:18 -!- nmz787_i [~ntmccork@134.134.139.76] has left ##hplusroadmap []
10:18 -!- chris_99 [~chris_99@unaffiliated/chris-99/x-3062929] has quit [Ping timeout: 250 seconds]
10:26 -!- chris_99 [~chris_99@unaffiliated/chris-99/x-3062929] has joined ##hplusroadmap
10:49 -!- sandeepkr [~sandeepkr@111.235.64.4] has quit [Ping timeout: 245 seconds]
10:57 -!- sandeepkr [~sandeepkr@111.235.64.4] has joined ##hplusroadmap
10:58 -!- ArturShaik [~ArturShai@37.218.166.45] has quit [Ping timeout: 240 seconds]
11:04 < kanzure> "the real problem is a super-charming machine, not super-intelligent machine" https://news.ycombinator.com/item?id=10851674
11:06 < esmerelda> Super-charming people tend to be problems too, I've found.
11:11 < Diablo-D3> [10:57:58] <kanzure> look at al these replies, damn http://edge.org/conversation/jaron_lanier-the-myth-of-ai
11:11 < Diablo-D3> [10:58:03] <kanzure> jaron lanier is wasting literally everyone's time
11:12 < Diablo-D3> kanzure: btw, I agree, that entire article is a waste of time
11:12 -!- erasmus [~erasmus@unaffiliated/erasmus] has joined ##hplusroadmap
11:13 < Diablo-D3> I don't even know why everyone thinks AI is the end of the world anyways
11:13 < Diablo-D3> AI is just software
11:14 -!- sandeepkr [~sandeepkr@111.235.64.4] has quit [Ping timeout: 276 seconds]
11:14 < Diablo-D3> can my computer kill me? no? it won't if I install an AI on it either.
11:14 < Houshalter> are humans just software?
11:14 < Diablo-D3> _anything_ further discussing that line of thought is just stupid, no matter what side you're on
11:15 < Diablo-D3> but that article in-particular? its not even coherent
11:15 < Houshalter> intelligence isn't harmless
11:15 < Diablo-D3> Houshalter: yeah, but its like, ZOMG A 4 YEAR OLD MIGHT KILL SOMEONE
11:15 < Diablo-D3> usually, 4 year olds don't kill people
11:16 < esmerelda> Humans are legacy code - undocumented, and very difficult to maintain
11:16 < esmerelda> It took the dev team 4 billion years and it looks like they ran obfuscators
11:16 < Diablo-D3> hah
11:17 < Diablo-D3> seriously though, we can't even produce intelligent humans most of the time
11:18 < Diablo-D3> whats the chances we'll make a software intelligent enough that can actually run on any sort of hardware that isn't just locked up at google or microsoft or w/e somewhere
11:19 < esmerelda> Will we discover general-purpose AI all at once (I.E. comparable generality to humans in learning and achieving goals)
11:19 < esmerelda> or will AIs slowly get smarter and more general?
11:19 < Diablo-D3> esmerelda: yes, to both questions.
11:19 < Diablo-D3> its not something that you can just answer like that.
11:19 < esmerelda> I was mostly looking for arguments for and against
11:20 < Diablo-D3> what I don't like is how every AI discussion is just philosophical bullshit strung together
11:21 < Diablo-D3> tl;dr of jaron lanier's article? "we're just worshipping a new God, enriching those behind the curtain pulling the levers and ropes"
11:21 < Houshalter> Diablo-D3, do you believe superintelligent AI is impossible
11:21 < Diablo-D3> Houshalter: I can't answer that.
11:22 < Diablo-D3> I can tell you it can exist given enough horsepower.
11:22 < Houshalter> like there is osme law of nature that says that humans are the peak of intelligence and no improvement is possible
11:22 < Houshalter> well if superintellligent AI is possible, then it is incredibly dangerous.
11:22 < Diablo-D3> okay so
11:22 -!- sandeepkr [~sandeepkr@111.235.64.4] has joined ##hplusroadmap
11:23 < Diablo-D3> lets say you have the smartest brain possible
11:23 < esmerelda> I saw an article a while back describing a theoretical "optimal AI" using a simple (but extremely slow) algorithm
11:23 < Diablo-D3> but its in a jar
11:23 < Diablo-D3> and its interaction with the outside world is heavily regulated
11:23 < Diablo-D3> not an AI, a brain
11:23 < esmerelda> Assume one feed of sensory input data, Sk
11:23 < Diablo-D3> 1960s scifi and everything
11:24 < Diablo-D3> unless you're putting that brain in an atomic powered robot body
11:24 < Diablo-D3> I can't really see anything going wrong
11:24 < esmerelda> And an infinitely enumerable set of models Mk
11:24 < Diablo-D3> its not like, ghost in the shell, where military AIs escape to the internet and copy itself to mainframes everywhere
11:24 < Houshalter> Diablo-D3, why are you assuming AI will exist in a jar? once the technology exists, people will put it everywhere
11:24 < Diablo-D3> Houshalter: not in my lifetime unless there is a massive increase in computational power
11:25 < Diablo-D3> the most powerful "computer" in the world is owned by Google
11:25 < Diablo-D3> its a massive gigantic super-cluster, that doesn't meet the requirements of being on the top500 list
11:25 < Houshalter> Diablo-D3, massive increases in computational power happen every other year
11:25 < Diablo-D3> Houshalter: except they don't
11:25 < Diablo-D3> there may have not been a massive increase in my lifetime.
11:25 < Diablo-D3> it was faster when I was a kid
11:26 < Houshalter> therre has been a massive increase in the past 5 years, let alone your lifetime
11:26 < Diablo-D3> but all the really huge jumps happened in the late 70s
11:26 < Diablo-D3> we have different definitions of massive increase, I think
11:26 < Houshalter> https://en.wikipedia.org/wiki/FLOPS#Cost_of_computing
11:26 < esmerelda> It would find the j such that Sum_k[Abs(Sk-Mj(Sk))]/KolgomorovComplexity(Mj) is maximized
11:26 < Diablo-D3> Houshalter: _cost_ of computing isn't the issue
11:26 < Diablo-D3> and flops is not a useful measurement of computational power
11:27 < Houshalter> yes it is, and yes it is
11:27 < Diablo-D3> I'm not saying computers aren't getting faster
11:27 < Diablo-D3> but there has been no real major jumps
11:27 < esmerelda> Ie. Aply Occam's razor by enumerating models starting with low-complexity going to high-complexity, and choosing the least complex model that explains the sensory input
11:27 < Diablo-D3> the biggest thing in my entire lifetime? home computers getting multiple cores in an SMP arrangement
11:28 < Houshalter> AI is powered by FLOPS, and the cost is the fairest unit to measure that
11:28 < Diablo-D3> I used to own a P3 550, it was the most powerful computer a home owner could have at the time
11:28 < Diablo-D3> I now carry it in my pocket.
11:28 < Diablo-D3> Houshalter: flops are generally measured by uselessly spamming MADs.
11:29 < esmerelda> Computing power is useless without an algorithm
11:29 < Diablo-D3> you have to measure many things, including various interconnect bandwidths and latencies, RAM speeds, storage IO speeds, etc
11:29 < Diablo-D3> esmerelda: I agree
11:29 < Diablo-D3> but whatever algorithm we end up creating, it will require massive compute power
11:30 < Diablo-D3> and its not a problem you can easily parallelize either
11:30 < esmerelda> Discussions of computing power are useless as a corollary. Without knowing the algorithm, we don't know what computing hardware will work it efficiently
11:30 < esmerelda> Uh citation needed
11:30 < Diablo-D3> by "easily parallelize" I mean we can't merely throw GPGPUs at it.
11:31 < Houshalter> Diablo-D3, why not? that's literally how AI has been progressing
11:31 < esmerelda> Why not? Do you know how the algorithm has to be structured then?
11:31 < TMA> the most advance we got in computing power in recent years is from parallelism, not from enhancing the raw single-thread power
11:31 < Diablo-D3> Houshalter: AI research has been using it when it makes sense
11:31 < Diablo-D3> TMA: yes.
11:31 < Houshalter> I think AI will probably move to FPGAs and ASICs instead of general purpose computers, which gets it an order of magnitude improvement
11:31 < Diablo-D3> I mean, I'm not sure why Houshalter is arguing with me
11:31 < Diablo-D3> this is all well known computer history 101
11:31 < esmerelda> Doesn't your brain literally process millions of simultaneous firings simultaneously?
11:31 < Diablo-D3> Houshalter: yes/no.
11:31 < Diablo-D3> esmerelda: yes, but that doesn't mean that is a useful model for AI
11:32 < Diablo-D3> esmerelda: remember, our brain runs at like 100hz too
11:32 < TMA> there is some promise in HP/HPE "The Machine" provided it turns out real and not a vaporware
11:32 < esmerelda> 100hz because it is massively parallel...
11:32 < Houshalter> Diablo-D3, i'm arguing with you saying that dangerous superintelligence isn't likely in our lifetimes. I strongly believe that it is. You are betting everything on technology stagnating which is jsut unreasonable
11:32 < esmerelda> 100hz, and like a quadrillion parallel instructions per cycle
11:32 < kanzure> most dangerous things that superintelligence could do are things that non-super intelligence can do, it's a non-concern in my opinion
11:32 < Diablo-D3> Houshalter: okay so, I once wrote a bitcoin miner, it ended up becoming the world's most popular miner for gpus.
11:33 < kanzure> humans are perfectly capable of nuking the earth or igniting the atmosphere, thanks. now get the hell out of my channel.
11:33 < Diablo-D3> Houshalter: no one uses it anymore, everyone uses ASICs now.
11:33 < Diablo-D3> I suspect more computation power is being spent on bitcoin than anything else.
11:33 < kanzure> "computation power"
11:33 < kanzure> you are ambiguous and i hate you
11:33 -!- sandeepkr [~sandeepkr@111.235.64.4] has quit [Ping timeout: 250 seconds]
11:33 < esmerelda> kanzure: are there any groups working on general purpose AI?
11:34 < Diablo-D3> kanzure: I actually used to run a graph that'd list hash power in 7970 equivalents.
11:34 < kanzure> esmerelda: yes, why do you ask?
11:34 < esmerelda> or were they killed off in the last AI winter
11:34 < Diablo-D3> and also CPU equivalent
11:34 < kanzure> which one was the last winter?
11:34 < esmerelda> I'm curious to see architectures being considered
11:34 < Diablo-D3> esmerelda: Im not even sure the AI winter ever existed
11:34 < esmerelda> The one before CNNs
11:34 < kanzure> esmerelda: well there's boring stuff like opencog, and then exciting things like markram blue brain human brain project thingy.
11:34 < Houshalter> kanzure, well yes humans could destroy the earth easily if we wanted to. thank god we don't want to. The problem with AI is that it would want to do that
11:34 < Diablo-D3> the problem is not AI winters happen, but soon as a technology starts working and everyone integrates into their products, it stops being AI
11:34 < esmerelda> I thought HBP was a giant boondoggle
11:34 < kanzure> Houshalter: humans want to blow up the earth. i'm one of them.
11:35 < esmerelda> Diablo-D3: tell that to the expert system programmers
11:35 < Houshalter> kanzure, well they've been really bad it at so far
11:35 < kanzure> Diablo-D3: that's fucking absurd. how about instead of claiming "AI stops being AI", how about "people were wrong to call these things AI". duh?
11:35 < kanzure> oh my god i hate all of you so much
11:35 < kanzure> this is why the planet needs to be blown the fuck up
11:35 < Diablo-D3> kanzure: well, thats honestly what Ive been saying
11:35 < Diablo-D3> but I get yelled at when I do that
11:36 < kanzure> Diablo-D3: no, you just said the opposite. don't lie to me.
11:36 < Diablo-D3> kanzure: yes, because people don't bitch when I say that
11:36 < esmerelda> kanzure: do you want to blow up the earth before we save a backup?
11:36 < kanzure> you were hanging out with the wrong people, and it shows negatively on you that you didn't recognize this
11:36 < Houshalter> kanzure, wtf
11:36 < Diablo-D3> kanzure: yeah, but somehow I'm wrong if I say everyones the wrong people.
11:37 < Diablo-D3> at least 90% of the human population may, in fact, be too stupid to live.
11:37 < kanzure> Houshalter: i have no idea why you are less interested in humans blowing up the planet.
11:37 < Diablo-D3> I'm the bad guy if I kill off 90% of the human population.
11:37 < kanzure> esmerelda: no i don't think so. there's good science in there somewhere.
11:37 < Diablo-D3> So, w/e
11:37 < Diablo-D3> if I want social interaction with normal people, I phrase it as "it stops being called AI" instead of "it probably wasn't AI to begin with"
11:37 < kanzure> Diablo-D3: "AI stops being AI when it is integrated" has nothing to do with killing off 90% of the population. thanks.
11:38 < Diablo-D3> like, people call google now and siri AI
11:38 < Diablo-D3> I don't know why
11:38 < Houshalter> kanzure, because i'm not a comic book villian?
11:38 < Houshalter> and i want to live
11:38 < Diablo-D3> they're just highly advanced natural language parsers
11:38 < esmerelda> kanzure: also if you could use something like radiation instead..  the rock formations and rivers are pretty
11:38 < Diablo-D3> using an incredibly huge database of knowledge
11:38 < kanzure> Houshalter: i didn't say "you personally interested in acting to blow up the planet". you are clearly interested in AI blowing up the planet. so why not humans? this is absurdly hypocritical of you.
11:39 < kanzure> "they are bad at it so far" applies to both humans and ai, anyway.
11:39 < esmerelda> The kolgomorov complexity of this conversation is rather low
11:39 < kanzure> esmerelda: perhaps you could clarify what you mean by boondoggies then...
11:39 < Houshalter> kanzure, ok that was worded confusingly. but yes I agree humans could blow it up, but they don't want to. And even if we had a coordinated program to put a nuke in every city, there would still be survivors
11:40 < kanzure> Houshalter: yeah i was thinking of doing a paper about the surprising difficulty of killing all humans. thankfully, blowing up the planet does not require killing all humans.
11:41 < esmerelda> kanzure: I read criticisms of the HBP by neuroscientists who claimed the hardware they were building was not necessarily well suited to meaningful simulation of a brain
11:41 < xentrac> .g the energy cost of evaluating earth's human population kragen-tol
11:41 < yoleaux> http://library.sandiegozoo.org/news/2015_news/2015_04cd.html
11:41 < Houshalter> nuclear war hasn't happened so far, and no one wants to start one. Even a nuclear war only destroys the countries involved, not the whole planet.
11:41 < xentrac> .g the energy cost of evaluating earth's human population "kragen-tol"
11:41 < kanzure> esmerelda: did you read the rebuttals by any chance?
11:41 < yoleaux> No results found.
11:41 < kanzure> stop stealing that kragen guy's work, man
11:41 < esmerelda> Kanzure: no, only one article. I said that I thought it was a boondoggle, not that it is in fact a boondoggle.
11:42 < kanzure> Houshalter: i think that nuclear blasts in cities would not be enough to blow up the planet
11:42 -!- sandeepkr [~sandeepkr@111.235.64.4] has joined ##hplusroadmap
11:42 < Houshalter> I believe AI is by far the greatest risk to the world
11:42 < Diablo-D3> sure thing Elon.
11:42 < esmerelda> I meant to imply this was my default viewpoint given limited data, and I don't hold it strongly
11:42 < kanzure> esmerelda: then i would suggest your current understanding is highly biased by detractors :-) it was a $1 billion project and a bunch of people were angry about not getting a slice of the pie.
11:42 < kanzure> esmerelda: there's a bunch of interesting neurophysiology that markram's group incorporates into their work
11:43 < kanzure> esmerelda: http://diyhpl.us/wiki/transcripts/markram-2006/
11:43 < kanzure> xentrac: what does "evaluating" mean here?
11:44 < esmerelda> Kanzure: this is interesting, thank you. I should probably use more clear language when discussing opinions I've heard but don't espouse strongly, to avoid misunderstanding
11:45 < kanzure> there was no misunderstanding afaik
11:45 < kanzure> opencog, blue brain, then you mentioned you had read some criticism, something about boondoggies, then i gave you links or something
11:46 < kanzure> oh, boondoggies came first
11:46 < kanzure> yeah i didn't misunderstand that, i just mentioned there was good science in there somewhere
11:47 < esmerelda> Ahh OK. I thought you were talking about the totality of human existence
11:47 < kanzure> good science in human existence? perhaps :-)
11:47 < esmerelda> Hrm, I had a thought unrelated to all of this.
11:48 < kanzure> esmerelda: http://diyhpl.us/~bryan/papers2/neuro/
11:48 < esmerelda> Say that we can store the data required to emulate a brain, but evaluation is very slow (E.g. 1 second per week)
11:48 < xentrac> um, evacuating
11:49 < xentrac> .g the energy cost of evacuating earth's human population kragen-tol
11:49 < yoleaux> https://www.mail-archive.com/kragen-tol@canonical.org/
11:49 < xentrac> https://www.mail-archive.com/kragen-tol@canonical.org/msg00138.html
11:49 < kanzure> oh that makes more sense
11:49 < esmerelda> If you put that on a satellite and shoved it out into space, it would not feel slowed down and thus feel probably pretty happy
11:49 < kanzure> yes but it would have massive latency
11:50 < xentrac> that describes how you can blow up the planet without killing any humans.  not that I think this is a good idea
11:50 < esmerelda> Sure, but maybe it's a hermit! Or you put 10 brains in there
11:50 < kanzure> right, there might be value in having a somewhat working biosphere
11:50 < TMA> Sometimes, I wonder whether we don't have a superintelligence already present. -- the world economy directs our decisions. moreover, in a sense, it has its own will, even though it is not an entity in the traditional sense
11:51 < kanzure> TMA: no, please see http://slatestarcodex.com/2015/12/27/things-that-are-not-superintelligences/
11:52 < xentrac> the economy, corporations, states, etc., are superhuman intelligences, but they are not superintelligences
11:52 < esmerelda> Kanzure: why do you want to blow up the world though?
11:52 < xentrac> they are superhuman in other aspects like longevity and physical force, not in intelligence
11:54 < esmerelda> Given a collection of processors, I think a superintelligence must demonstrate that it makes decisions better than how good all the individual decisions of the processors are.
11:55 < kanzure> blowing up the world would be simpler than tolerating the low-level crap that AdrianG and Diablo-D3 inflict upon me in here
11:55  * esmerelda suggests /mute, or drugs
11:55 < kanzure> well i can also ban them
11:56  * Diablo-D3 rolls eyes.
11:57 < esmerelda> You could make this channel a more complicated variant of #xkcd-signal
11:57 < Diablo-D3> esmerelda: hows that work?
11:57 < esmerelda> so that similar statements or lines of reasoning are detected and cause muting
11:57 < kanzure> right, i would have to do that manually though
11:57 < kanzure> and i would rather that people voluntarily don't suck at all
11:59 < kanzure> xentrac: btw, evacuation is much easier if you assume working and useful brain scanning
12:01 < esmerelda> kanzure: if you trained an HMM on the chat logs here, and then for an input M computed the probability of producing M with the model
12:01 < kanzure> train the model on all the blocked/banned/kicked users?
12:01 < esmerelda> Then you divide that probability by the probability of producing any random sentence (mean probability)
12:01 < esmerelda> Sure, or even on all users if you want to avoid repeat topics
12:02 < kanzure> repeat topics are fine as long as users do due diligence or make an active effort to not be lame
12:02 < esmerelda> Then you'll need a training set of lame sentences
12:03 < xentrac> kanzure: yeah, if you assume the brain emulation hardware is lighter or more robust to space conditions than human bodies
12:03 < xentrac> and I guess cheaper
12:04 < xentrac> right now it's much heavier, much more expensive, and significantly less robust
12:04 < xentrac> and Moore's Law, which we were counting on to fix that, has ended
12:04 < kanzure> well..... you could bury the emulation hardware, and then blow up the planet, and then recover the hardware?
12:04 < xentrac> we'll probably come up with a different fix but it's historically contingent
12:05 < kanzure> we should bury satellites that get deployed when the planet inevitably breaks up
12:05 < esmerelda> Take the simple case: say that Diablo-D3 says "AI is dangerous". Let P(AI, is) = 0.7, and P(is, dangerous) = 0.9 in the model. Then he had a 0.63 probability of saying that statement. Then say that the probability is 0.1 for an average user, that a second model was trained against
12:05 < xentrac> what, when the sun cooks it?
12:05 < kanzure> well perhaps something happens sooner
12:05 < kanzure> maybe
12:05 < Diablo-D3> esmerelda: I wasn't the one saying it was dangerous though
12:05 < Diablo-D3> I was agreeing with kanzure that the article is trash and its wasting people's time
12:05 < esmerelda> I'm talking about signal detection
12:05 < xentrac> here's a sample non-stupid statement from Diablo-D3: 19:32 < Diablo-D3> Houshalter: okay so, I once wrote a bitcoin miner, it ended up becoming the world's most popular miner for gpus.
12:06 < kanzure> what, cgminer?
12:06 < xentrac> here's a stupid thing: 19:33 < Diablo-D3> I suspect more computation power is being spent on bitcoin than anything else.
12:06 < xentrac> dunno, you'd have to ask him
12:07 < Diablo-D3> kanzure: cgminer is a port of diablominer to C, with later added fpga and asic support
12:07 < esmerelda> Right, xentrac. But the system would compare the conditional probability of each transition given that Diablo-D3 said it, vs. the average probability over all users
12:07 < esmerelda> That tells you how "Diablo" a statement is
12:07 < Diablo-D3> ck undertook porting it to C to make fpga and asic support possible
12:08 < esmerelda> So the system wouldn't flag "Hi", but it would flag "bitcoin miner" (problematic)
12:08 < xentrac> but the bitcoin miner statement was actually a point in Diablo-D3's favor
12:09 < esmerelda> Yeah. I was just interested to see if I could design a system to censor certain topics or views
12:09 < xentrac> that shouldn't be too hard
12:09 < esmerelda> Diablo was just a hapless example (sorry, man!)
12:09 < esmerelda> xentrac: without much human intervention? Clearly you and I should make a startup for this
12:09 < xentrac> unfortunately a system to censor stupid or content-free statements about a topic but not insightful ones is much more difficult, probably AI-hard
12:09 < Houshalter> esmerelda, that just rules out all short sentences
12:10 < xentrac> Houshalter: that in itself sounds like a good idea
12:10 < Houshalter> how do you answer a yes or no question?
12:10 < esmerelda> P(yes|Diablo)/P(yes) ~= 1
12:11 < esmerelda> So it wouldn't be censored
12:11 < Diablo-D3> Okay so, what is considered the largest use of computational power of a single project?
12:11 < esmerelda> You know, maybe I should just go build this thing. It seems simple and fun
12:11 < kanzure> probably a hydrodam would count as a computation
12:12 < kanzure> *would qualify
12:12 < Diablo-D3> current hashrate is 9862 exaflops
12:14 < Diablo-D3> 9,862,475 petaflops, folding@home hit 25 petaflops in sep 2015
12:15 < xentrac> Diablo-D3: maybe the black hole at the galactic core?
12:16 < Diablo-D3> so what other projects out there can be measured at single digit zettaflops
12:16 < xentrac> Houshalter: you could say something more interesting than "yes" in response, although as esmerelda is pointing out, that wasn't actually what she was suggesting
12:17 < Diablo-D3> xentrac: thats not a computation project though =P
12:18 < xentrac> the #xkcd-whatever channel does in fact ban responses like "yes"
12:18 < esmerelda> Yeah. My system is different though
12:18 < xentrac> Diablo-D3: just because you don't understand what it's computing doesn't mean it's not computation.  most people don't understand what their remote control is computing either
12:19 < xentrac> esmerelda: yes, I didn't mean to imply otherwise
12:19 < Diablo-D3> xentrac: I don't know what its computing, but I do know its an i.MX
12:20 < xentrac> Diablo-D3: are you giving me the model number for the black hole at the galactic core? you are a very silly sentient being.
12:20 < Diablo-D3> xentrac: sigh, no, my remote control.
12:21 < xentrac> maybe you might get confused and think that your remote control is also not a computer, as you apparently think about the black hole and about kanzure's example of hydroelectric dams
12:22 < Diablo-D3> xentrac: no, its very much a computer, it has a usb port and everything.
12:23 < xentrac> are you suggesting that the black hole at the galactic core is primarily lacking a USB port for you to recognize it as a computer?
12:23 < kanzure> oops it's not called a hydrodam is it
12:23 < kanzure> xentrac: to be fair, lots of bitcoin mining equipment is also lacking usb ports
12:23 < Diablo-D3> I'm not sure how either of those are answers to the question "what computing project uses the most power, or is in the range of single digit zettaflops"
12:23 < xentrac> kanzure: btw I had an interesting chat with Vinay on the tweeter the other night.  i tweet little enough that you should be able to find it easily
12:23 < kanzure> vinay gupta?
12:24 < xentrac> natch
12:24 < kanzure> notch??
12:24 < xentrac> naturally
12:24 < xentrac> it will probably infuriate you
12:24 < kanzure> most things tend to do that
12:24 < xentrac> yes, that's why I said that
12:24 < kanzure> i am mobile at the moment adn can't go searching. maybe later.
12:24 < xentrac> one of the disadvantages of amphetamine
12:24 < xentrac> Diablo-D3: well, that black hole is certainly computing a lot more than a measly few zetaflops
12:25 < kanzure> even if amphetamines do cause infuriation, i think it's wrong of you to assume it's the sole source of my infuriation. perhaps i am a naturally prone to finding exellent reasons to be absolutely unfuriated by mundane bullshit
12:25 < Diablo-D3> xentrac: but its not computing anything, its not a computer
12:25 < kanzure> *excellent
12:25 < kanzure> +?
12:26 < kanzure> *infuriated
12:26 < kanzure> mobile typing sucks
12:26 < kanzure> Diablo-D3: perhaps you should look into what computation happens to be :-)
12:27 < Diablo-D3> "A computation can be seen as a purely physical phenomenon occurring inside a closedÊphysical systemÊcalled aÊcomputer. Examples of such physical systems includeÊdigital computers,Êmechanical computers,Êquantum computers,ÊDNA computers,Êmolecular computers,Êanalog computersÊorÊwetware computers. This point of view is the one adopted by the branch of theoretical physics called theÊphysics of computation."
12:27 < Diablo-D3> How the hell did that paste like that.
12:27  * Diablo-D3 stares at both chrome and his irc client
12:27 < xentrac> kanzure: you are of course correct, and I didn't mean to imply it was the *sole* source of your infuriation
12:28 < kanzure> xentrac: what is the name of the physicist that i am thinking of? his name starts with an s and he has a wildly famous paper about computation and physics.
12:28 < kanzure> his name has an L in it somewhere
12:28 < xentrac> related to feynman, landauer, dyson?
12:28 < kanzure> oh probably, who isn't?
12:29 < xentrac> computation and physics is a very broad field
12:29 < kanzure> perhaps i should work on remembering authors other than by the initial letters of their names......
12:31 < kanzure> hmm maybe this is more of an nsh paper than a xentrac paper, although i would be surprised if it turns out you have never seen this
12:31 < kanzure> he is somehow related to d'hoof't for some reason
12:31 < nsh> what are those space-invaders blockshapes for visual recognition of username/number hashes?
12:32 < nsh> maybe all the sciencepals should have one of them each
12:32 < xentrac> then you could draw it to look them up?
12:34 < kanzure> gerald sussman
12:35 < kanzure> http://link.springer.com/article/10.1007/s00422-015-0662-6
12:35 < kanzure> this is not the article i'm thinking of, but it's close enough
12:35 < kanzure> oh fooey
12:36 < kanzure> that's not it.... grr.
12:36 < kanzure> http://lambda-the-ultimate.org/node/4389
12:36 < kanzure> .title
12:36 < yoleaux> "We Really Don't Know How to Compute!" by Gerald Sussman | Lambda the Ultimate
12:38 < kanzure> dunno why i had so many apostraphes in d'hooft's name
12:39 -!- eudoxia [~eudoxia@r186-55-160-144.dialup.adsl.anteldata.net.uy] has joined ##hplusroadmap
12:39 < kanzure> yesss maybe eudoxia will avenge me. surely he knows which sussman paper i'm talking about?
12:39 < eudoxia> let me read the logs
12:40 < eudoxia> kanzure: Structure and Interpretation of Classical Mechanics, although it's a book
12:40 < eudoxia> you might be thinking of functional differential geometry which is related to the topic
12:41 -!- metathrone [~NemeSiS@85.239.195.89] has joined ##hplusroadmap
12:41 < xentrac> that's not really about the physical phenomenon of computation
12:41 < kanzure> no not that book
12:41 < xentrac> it's more about formalizing Lagrangian classsical physics algorithmically
12:41 < kanzure> come on he probably worked on lisp machine this is straight up your ally, eudoxia....
12:41 -!- sandeepkr [~sandeepkr@111.235.64.4] has quit [Ping timeout: 260 seconds]
12:42 < xentrac> as it happens, classical physics has nothing interesting to say about computation
12:42 < eudoxia> i interpreted "computation and physics" as "computational physics"
12:43 < xentrac> Diablo-D3: it's not clear to me what chain of misguided character encodings turned non-breaking spaces into Ê
12:43 < eudoxia> http://groups.csail.mit.edu/mac/projects/amorphous/papers/aim1665.pdf ?
12:46 < kanzure> that's an interesting paper, but the one i'm thinking about is much more general and probably older
12:46 < Diablo-D3> xentrac: yeah, I can't figure out why that did that
12:47 < kanzure> i am probably thinking of an essay, not a paper
12:47 < kanzure> and also, there was at most two authors, probably only one
12:49 -!- sandeepkr [~sandeepkr@111.235.64.4] has joined ##hplusroadmap
12:49 < eudoxia> ftp://publications.ai.mit.edu/ai-publications/1000-1499/AIM-1039.ps has two authors
12:50 < eudoxia> oh here it is probably http://groups.csail.mit.edu/mac/users/gjs/cellgates/cellgates.pdf
12:51 < xentrac> that isn't about physics
12:52 < kanzure> nope, sorry, wrong one
12:52 < kanzure> also funny to see tom knight
12:53 < eudoxia> hmm
12:53 < eudoxia> it's probably not sussman
12:54 < kanzure> how about "physicists that have a name starting with L and have published to arxiv or seem like they should have, and also wrote about computation and physics"?
12:54 < kanzure> the problem is that sussman is such a close match
12:54 < kanzure> how can it not be sussman?
12:54 < eudoxia> specifically what was the paper about
12:55 < eudoxia> hypothetical modes of computation?
12:55 < kanzure> principles of computation apply to physics in general way
12:55 < eudoxia> ahh
12:55 < eudoxia> "atoms are bits, collisions are ops" sort of thing?
12:55 < kanzure> might have included hypothetical computation stuff, but was not main point i think
12:55 < kanzure> yea something like that
12:55 -!- jtimon [~quassel@103.red-80-26-235.adsl.dynamic.ccgg.telefonica.net] has joined ##hplusroadmap
12:56 < kanzure> and it was definitely absolutely not gershenfeld
12:57 < eudoxia> how about seth lloyd
12:57 < kanzure> hey that has both the letters i wanted
12:57 < xentrac> haha gershenfeld
12:57 < kanzure> gershenfeld talks about the same concepts, but he wasn't the author
12:57 < xentrac> gershenfeld is awesome but I wouldn't expect him to publish papers on either physics or the fundamentals of computation, let alone both at once
12:57 < kanzure> well, i mean, he also doesn't focus on physics either
12:58 < kanzure> indeed
12:58 < kanzure> yes you have stated what i should have said, except much more elegantly
12:58 < xentrac> he's more a "let's hack stuff and get people involved" kinda guy
12:58 < xentrac> I ran into him by chance on the T platform
12:58 < kanzure> he's a twooter?
12:59 < xentrac> ?
12:59 < eudoxia> found it http://arxiv.org/pdf/quant-ph/9908043.pdf
12:59 < kanzure> twitter tweeter
12:59 < xentrac> not that I know of
13:00 < xentrac> I mean the platform where you wait to get on the train
13:00 < kanzure> eudoxia: fucking awesome dude, thank you. amazing work.
13:00 < eudoxia> yw kanz
13:00 < kanzure> now why wasn't this written by sussman? >:(
13:01 < xentrac> oh, yeah, I have seen this paper
13:01 < kanzure> haha
13:01 -!- |nodux [uid125132@gateway/web/irccloud.com/x-zyhjdjluuyhssfhb] has quit [Quit: Connection closed for inactivity]
13:02 < xentrac> https://queue.acm.org/detail.cfm?id=2874238 says we need to totally rearchitect our computer systems in order to take advantage of how fast Flash is now
13:03 < xentrac> because Flash is a thousand times faster at random access than spinning rust, but only five or seven times faster at running MySQL
13:03 < kanzure>  think seth lloyd has a slightly different version of this paper somewhere, where he talks about the general aspects of computation and physics, not just the ultimate limits of computation
13:04 < kanzure> *i thnk
13:05 < kanzure> .title http://arxiv.org/abs/quant-ph/0110141
13:05 < yoleaux> [quant-ph/0110141] Computational capacity of the universe
13:07 < kanzure> "Is the universe roughly-tuned for computing?" http://arxiv.org/pdf/1501.01754.pdf
13:08 < kanzure> "Uncomputability and physical law" http://arxiv.org/abs/1312.4456
13:14 < kanzure> oh weird, i should have guessed tegmark. i definitely didn't mean tegmark, but he wouldn't be a bad place to start from.
13:15 < kanzure> or john smart, for that matter....
13:35 -!- PatrickRobotham [uid18270@gateway/web/irccloud.com/x-vlqebucoeqmjptkq] has joined ##hplusroadmap
13:38 < xentrac> http://status451.com/2016/01/06/splain-it-to-me/ may be interesting to you, kanzure, particularly with regard to the "why are you being so rude to me?" comments you get.  although you don't get to claim to not be playing pecking-order games :)
13:39 -!- sandeepkr [~sandeepkr@111.235.64.4] has quit [Ping timeout: 265 seconds]
13:41 < kanzure> xentrac: you mean, for explaining to others how it is rude to inject low-quality shit into this channel? not sure this is a good link for that.
13:42 < kanzure> as for pecking order, perhaps you mean moderating
13:45 < kanzure> this article got very weird in a way that i would only expect to find on autistics.org
13:45 < kanzure> (i mean autistics.org quite literally, it has a good library of stuffs)
13:46 < pompolic> around which part?
13:46 -!- metathrone [~NemeSiS@85.239.195.89] has quit [Ping timeout: 246 seconds]
13:47 < xentrac> kanzure: moderation is one possible application of social dominance hierarchies
13:50 < kanzure> also i think this article ends poorly. it fails to point out that the social status game is much cheaper to play, and the other one is more difficult and requires more effort, which is a useful explanation for why it's not a universal default.
13:51 < kanzure> also, to be fair, "randos bringing up random additional factual correctness statements" is in fact somewhat socially maladjusted (for the most part, social agreement throughout history was based on leadership and individuals and cooperation, not about abstract ideation), so it makes sense for people to be socially accustomed to dismissing inline email replies (which while not explicitly stated in the article, i'm fairly certain the ...
13:51 < kanzure> ... article's author would agree is a relevant topical example).
13:53 < kanzure> also i think that kindness should be extended to potential apprentices or learners who show some kind of promise, but there is very low utility in fixing e.g. all of AdrianG's broken beliefs.
13:56 < maaku> Kick people that are disruptive, ban them if they make a big deal out of it
13:56 < kanzure> i used to think that i could convince people into becoming high ability (because it seemed to work on a number of occassions) but looking back perhaps i was just talking with young people (i was also young at the time) who hadn't yet bothered to check if they were high-ability.
13:57 -!- Madplatypus [uid19957@gateway/web/irccloud.com/x-bchrqwyaazdutvhq] has joined ##hplusroadmap
13:58 -!- eof [~NemeSiS@5.52.144.118] has joined ##hplusroadmap
14:00 -!- eof [~NemeSiS@5.52.144.118] has quit [Read error: Connection reset by peer]
14:00 < maaku> It really is rare that you can teach an old dog new tricks.
14:00 -!- metathrone [~NemeSiS@5.52.144.118] has joined ##hplusroadmap
14:01 < kanzure> sounds too pessimistic for my taste. have you tried spiking the dog's water with amphetamines?
14:02 < xentrac> Yeah, I don't think you can convince people of much of anything by talking to them, let alone support them in a rigorous and grueling program of mentally empowering themselves.
14:02 < kanzure> i'm... pretty sure i did that in the past. long before this channel existed.
14:02 < kanzure> i wish i would have collected more evidence or measurements or something
14:02 < xentrac> You really need some kind of cult or something, like the 1980s BBS scene, MIT, or Aum Shinri Kyo
14:03 < kanzure> i mean, what about brainwashing and religious indoctrination?
14:03 < xentrac> Yeah, those totally work
14:03 < kanzure> ah good.
14:03 -!- metathrone [~NemeSiS@5.52.144.118] has quit [Read error: Connection reset by peer]
14:03 < xentrac> Although they have their own failure modes at a social level
14:03 -!- eof [~NemeSiS@5.52.144.118] has joined ##hplusroadmap
14:03 < xentrac> But they're very different from just talking to people
14:03 < maaku> Oh stubborn people are capable of learning. They just don't want to.
14:03 -!- eof is now known as metathrone
14:04 < xentrac> Yeah, people do mostly what they want to do
14:04 -!- erasmus [~erasmus@unaffiliated/erasmus] has quit [Quit: Namaste]
14:05 -!- Guest83875 [~blueskin@unaffiliated/blueskin] has quit [Ping timeout: 240 seconds]
14:05 -!- PatrickRobotham [uid18270@gateway/web/irccloud.com/x-vlqebucoeqmjptkq] has quit [Ping timeout: 240 seconds]
14:05 -!- Guest83113 [~abe@2601:184:4002:2290:f1a8:a452:1c0f:93f8] has quit [Ping timeout: 240 seconds]
14:06 -!- PatrickRobotham [uid18270@gateway/web/irccloud.com/x-pafvrylohqlqppjv] has joined ##hplusroadmap
14:10 -!- blueskin [~blueskin@unaffiliated/blueskin] has joined ##hplusroadmap
14:11 < archels> .g when does a physical system compute
14:11 < yoleaux> http://arxiv.org/abs/1309.7979
14:11 < archels> didn't Anders write a bit about the physical limits of computation as well?
14:12 < maaku> Young people are looking for something to do with their lives. They have an itch, a desire to do something important. You just insert your own vision like an ointment and they'll latch on.
14:12 -!- metathrone [~NemeSiS@5.52.144.118] has quit [Quit: Leaving]
14:12 < maaku> I think tengmark as well.
14:15 < maaku> I need to improve my own productivity before acquiring minions however...
14:16 < kanzure> well, i wasn't looking for an article about limits to computation
14:16 < kanzure> anders did write an article on that subject, yes
14:16 < kanzure> "When does a physical system compute?" is not what i was thinking of, but yes this looks highly relevant
14:19 -!- Guest83113 [~abe@2601:184:4002:2290:f1a8:a452:1c0f:93f8] has joined ##hplusroadmap
14:19 < archels> a black hole probably doesn't compute if it's just sitting there, being chaotic and complex
14:19 < kanzure> anders sandberg's paper is http://diyhpl.us/~bryan/papers2/The%20physics%20of%20information%20processing%20superobjects%20-%20Anders%20Sandberg.pdf
14:19 < archels> if it's at a sort of criticality, things might begin to get interesting, but that just as an aside ;)
14:19 < kanzure> and he cites seth lloyd, but only the papers on limitations of physical computation
14:20 -!- c0rw|away is now known as c0rw|scrollback
14:20 < kanzure> hm his bibliography is nice
14:21 < kanzure> i definitely saw seth lloyd's "Ultimate physical limits to computation" paper because i had a copy here http://diyhpl.us/~bryan/papers2/physics/Ultimate%20physical%20limits%20to%20computation.pdf
14:22 < kanzure> ... and the "Amorphous computing" paper, http://diyhpl.us/~bryan/papers2/Amorphous%20computing.pdf
14:22 < kanzure> http://diyhpl.us/~bryan/papers2/Computational%20analysis%20of%20galactic%20exploration%20with%20space%20probes%20-%20implications%20for%20the%20Fermi%20paradox.pdf
14:23 -!- metathrone [~NemeSiS@5.52.144.118] has joined ##hplusroadmap
14:24 < kanzure> "NP-complete problems and physical reality" http://diyhpl.us/~bryan/papers2/NP-complete%20problems%20and%20physical%20reality%20-%20Scott%20Aaronson.pdf
14:24 < kanzure> this is clearly not the paper i was thinking of, but highly topical still.
14:25 < kanzure> "On the maximal quantity of processed information in the physical eschatological context" http://diyhpl.us/~bryan/papers2/On%20the%20maximal%20quantity%20of%20processed%20information%20in%20the%20physical%20eschatological%20context.pdf
14:25 < kanzure> cc nsh
14:27 < kanzure> xentrac: seth lloyd cites neil gershenfeld in his 2000 paper on ultimate limits. heh.
14:29 < nsh> more interesting than the physical limits of computation is perhaps the computational limits of physics
14:29 < nsh> where some actual progress is being made
14:29 < nsh> scott aaronson has followed some of this recently
14:29 < nsh> in a few talks
14:30 < nsh> the gradual conquest of all sciences by theoretical computer science has a long way to go yet
14:33 -!- jtimon [~quassel@103.red-80-26-235.adsl.dynamic.ccgg.telefonica.net] has quit [Ping timeout: 245 seconds]
14:33 < kanzure> s/conquest of all sciences/conquest of everything
14:38 -!- chris_99 [~chris_99@unaffiliated/chris-99/x-3062929] has quit [Quit: Leaving]
14:38 -!- c0rw|scrollback is now known as c0rw1n
14:39 -!- Diablo-D3 [~diablo@exelion.net] has quit [Quit: This computer has gone to sleep]
14:39 -!- Diablo-D3 [~diablo@exelion.net] has joined ##hplusroadmap
14:41 -!- metathrone [~NemeSiS@5.52.144.118] has quit [Quit: Leaving]
14:42 -!- eudoxia [~eudoxia@r186-55-160-144.dialup.adsl.anteldata.net.uy] has quit [Quit: Leaving]
14:48 -!- drewbot_ [~cinch@ec2-54-145-0-108.compute-1.amazonaws.com] has quit [Remote host closed the connection]
14:49 -!- drewbot [~cinch@ec2-54-211-26-108.compute-1.amazonaws.com] has joined ##hplusroadmap
15:23 -!- arc3v1l [~arc3v1l@46.21.154.83] has joined ##hplusroadmap
15:44 -!- Houshalter [~Houshalte@oh-71-50-56-224.dhcp.embarqhsd.net] has quit [Ping timeout: 276 seconds]
15:52 -!- Gurkenglas [Gurkenglas@dslb-188-103-077-131.188.103.pools.vodafone-ip.de] has quit [Ping timeout: 276 seconds]
16:10 < xentrac> kanzure: really? Maybe Neal's got a history I don't know about
16:11 < kanzure> well, i mean, mit media lab people have a sneaky way of going about stuff, they tend to show up everywhere
16:30 < kanzure> docl: i found this in my paper archive, http://www.bnglifecasting.com/diy/chamber/
16:38 -!- |nodux [uid125132@gateway/web/irccloud.com/x-oeneagiewjaquakg] has joined ##hplusroadmap
16:47 < kanzure> prompted by archels, i have cleaned up some of the directory structure of my collection and also here's an index http://diyhpl.us/~bryan/papers2/index.2016-01-06.txt
16:56 -!- nmz787_i [~ntmccork@134.134.139.70] has joined ##hplusroadmap
16:56 < nmz787_i> sup
16:56 < kanzure> 8448 papers
16:57 < kanzure> that's only 2 per day?
16:58 < kanzure> somehow i suspect that this is not anywhere near a reasonable limit of scientific input
16:58 -!- arc3v1l [~arc3v1l@46.21.154.83] has quit [Quit: Leaving]
16:59 < xentrac> yeah, you can probably only read one or two a week
16:59 < kanzure> i was thinking more than 2/day
16:59 < kanzure> i don't think the goal should be to memorize every single mundane detail of every paper, so it seems sensible to not optimize for that
17:00 < kanzure> wondering if a "review of everything i have read in the past ~year" would be a productive thing to write
17:01 < kanzure> xentrac for one of your crazy hackdays project things would you consider stuff like "things to help kanzure do better pdf management of his collection" (like pdfparanoia similar stuff?), or even "things to help kanzure when he manually reads all the hplusroadmap irc logs sometime soon"?
17:06 < streety> is PVC the way to go for vacuum chambers? I was under the impression it had a tendency to violently fail
17:06 < streety> although that might have only applied to overpressure
17:06 < nmz787_i> PVC is very strudy
17:06 < nmz787_i> vacuum is not very stressfull
17:07 < nmz787_i> the max pressure differential you will ever have at sea level is 15 PSI
17:07 < superkuh> Fine for casting stuff and the like. But for actual physics stuff the outgassing would probably be horrible.
17:07 < nmz787_i> this is much much less than say, using PVC for a potato cannon
17:07 < xentrac> 01:00 < kanzure> wondering if a "review of everything i have read in the past ~year" would be a productive thing to write
17:07 < nmz787_i> what superkuh said
17:07 < xentrac> YES YES YES YES YES
17:08 < xentrac> actually even the past week would be useful
17:09 < nmz787_i> so I'm not sure why I wanted to do a slicer in BRLCAD, when I already know how to export my models as STL, which slicers exist for (though I don't know how well they work to generate g-code for a laser)
17:09 < kanzure> recurrent neural network effectiveness http://karpathy.github.io/2015/05/21/rnn-effectiveness/
17:09 < kanzure> for some reason i have http://karpathy.github.io/2015/05/21/rnn-effectiveness/ in my head as "most interesting explanation of recurrent neural network stuff" but i don't know if this is true
17:09 < nmz787_i> but I made decent progress last night, implemented getting a bounding box for objects, and returning opposite points on the box
17:09 < xentrac> nmz787_i: also writing a slicer from STL is super easy
17:10 < xentrac> kanzure: my notes are
17:10 < nmz787_i> I can only surmise that I had some idea that the STL to g-code would be sloppy or something
17:10 < xentrac> links from 2015-08-15:
17:10 < xentrac> http://karpathy.github.io/2015/05/21/rnn-effectiveness/ “The Unreasonable Effectiveness of Recurrent #Neural-Networks” for #image-recognition, #NLP language modeling (producing what looks an awful lot like Markov-chain text with slightly better context-free properties), sequential processing for directing attention over an image, etc.  Lots of animations and visualizations of RNNs doing their thing.  The models are written in #Lua with so
17:10 < nmz787_i> though, slicing in BRLCAD, then exporting an STL of that slice, and generating G-code for a single slice might be nicer in some way
17:11 < kanzure> enkiv2: i think you have cutoff syndrome ("Lua with so"
17:11 < kanzure> )
17:11 < xentrac> well, the STL step does inevitably convert curves into straight lines
17:11 < kanzure> .... i mean xentrac, not enkiv2.
17:11 < xentrac> in #Lua with something called “Torch 7” and run with #GPGPU.  Lots of great comments!  "unreasonable RNNs"
17:11 -!- Houshalter [~Houshalte@oh-71-50-56-224.dhcp.embarqhsd.net] has joined ##hplusroadmap
17:11 < xentrac> related:
17:11 < xentrac> https://github.com/karpathy/char-rnn “Multi-layer recurrent #neural-networks (LSTM, GRU, RNN) for character-level language models in Torch” in #Lua, which generate pretty spectacular random text.  This is (some of) the code for #unreasonable-RNNs.
17:11 < kanzure> k well it says it's unreasonable, so i guess that means it's a dead end
17:11 < xentrac> https://gist.github.com/karpathy/d4dee566867f8291f086 A hundred-line example of recurrent #neural-networks for text generation, by the author of #unreasonable-RNNs.
17:11 < nmz787_i> since layers in photolithography need a spin-coat cycle... which in g-code would probably just be a "change Z" command
17:12 < xentrac> nmz787_i: you might have some M code to set the spin-coating parameters
17:13 < kanzure> "STL to gcode" er... CAM should not be direct translation.
17:13 < kanzure> kinematics often plays a role
17:13 < xentrac> yeah, that's what I said yesterday
17:13 < xentrac> although laser cutters in particular are almost immune to kinematic concerns
17:13 < kanzure> btw fenn is our resident linuxcnc kinematics monkey, he wrote some sort of inexplicably-high-dof kinematic solver for linuxcnc
17:14 < xentrac> and 3-D printers very commonly "cheat" so that you don't have to specify limited-absolute-jerk motion curves in your G-code
17:15 < xentrac> but usually laser cutters use HP-PCL with embedded HPGL apparently
17:15 < xentrac> but I think nmz787_i built his own or something?
17:16 < xentrac> the best documentation I've found about COTS ones is this
17:16 < xentrac> http://hci.rwth-aachen.de/materials/publications/oster2011a.pdf Thomas Oster’s bachelor’s #pdf #dissertation on the VisiCut #laser-cutter software.  Mentions he got about an inch per second at 500dpi out of the Epilog Zing cutter he was using, or 8 inches per second with the laser off, and documents the HP-PCL, PJL, HPGL, and LPD interface of the machine in some detail.
17:16 < kanzure> a dissertation about proprietary laser cutter software...?
17:17 < xentrac> no, it's about the software he wrote, which was a printer driver for his laser cutter
17:17 < kanzure> this sounds less awful
17:18 < xentrac> I don't understand how writing a printer driver merits a bachelor's degree but I guess most bachelor's degrees don't require any kind of thesis
17:18 < xentrac> it's free software but it's kind of specific
17:18 < kanzure> yes i think bachelor thesis standards are more nebulous
17:18 < xentrac> I'm starting to get nostalgic for HPGL
17:18 < xentrac> somebody please help me
17:19 < kanzure> imagine you're in a world where people don't produce closed-access firmware, and you can program all the laser cutters to your heart's content
17:19 < kanzure> does that help?
17:19 < nmz787_i> xentrac: no I bought it, but it uses GRBL firmware
17:20 < nmz787_i> I guess G-code is a type of b-rep?
17:20 < nmz787_i> if it uses arcs
17:21 < xentrac> not really
17:21 < xentrac> it doesn't describe solid shapes at all
17:21 < xentrac> it describes toolpaths
17:21 -!- eof [~NemeSiS@178.131.210.142] has joined ##hplusroadmap
17:21 < xentrac> STL is a type of b-rep
17:22 < xentrac> the solid shapes you get out will depend on things like (classically) your cutting tool shape, size, and momentum
17:22 < kanzure> triangles are a very limited subset of all possible b-reps :P
17:22 < xentrac> yes
17:23 < xentrac> I mean "the solid shapes you get out of a given piece of G-code will depend on things like (classically) your cutting tool shape, size, and momentum"
17:23 < xentrac> also there are often considerations in generating the G-code other than the shape of the resulting object
17:23 -!- eof [~NemeSiS@178.131.210.142] has quit [Read error: Connection reset by peer]
17:23 -!- eof [~NemeSiS@178.131.210.142] has joined ##hplusroadmap
17:24 < xentrac> surface finish, other aspects of tolerances that aren't just surface finish, heat-affected zones, that kind of thing
17:24 -!- eof [~NemeSiS@178.131.210.142] has quit [Client Quit]
17:27 < kanzure> when i first learned about "c-to-asic" stuff in uh i guess 2004, i wanted something like a c-to-shitty-neural-network generator. do we have that yet?
17:29 < kanzure> actually i would be partly satisfied by just arbitrary-math-function, maybe even with a constraint about no discontinuities, instead of arbitrary c source code input
17:34 < xentrac> I haven't heard of one
17:38 < esmerelda> Hrm, interesting
17:38 < esmerelda> I wonder if you could do it with rewrite rules
17:39 < esmerelda> You can compute any continuous function with a three-layer ANN right?
17:51 -!- nmz787_i [~ntmccork@134.134.139.70] has quit [Quit: Leaving.]
17:52 -!- nmz787_i [ntmccork@nat/intel/x-dbyddymlfgxvoduf] has joined ##hplusroadmap
17:53 < kanzure> recurrent neural network stuff http://www.cs.toronto.edu/~graves/
17:53 < kanzure> and http://www.cs.toronto.edu/~ilya/
17:53 < kanzure> and http://www.rnnlm.org/
18:01 -!- nmz787_i1 [~ntmccork@134.134.139.77] has joined ##hplusroadmap
18:02 -!- nmz787_i1 [~ntmccork@134.134.139.77] has quit [Client Quit]
18:04 -!- nmz787_i [ntmccork@nat/intel/x-dbyddymlfgxvoduf] has quit [Ping timeout: 240 seconds]
18:12 -!- c0rw1n is now known as c0rw|zZz
18:32 -!- jaboja [~jaboja@esb12.neoplus.adsl.tpnet.pl] has joined ##hplusroadmap
18:48 < nmz787> xentrac: so it sounds like you're proclaiming to represent boundaries, the representation needs to be 3D?
18:49 < nmz787> I don't see why g-code is seen not to represent boundaries
18:50 < nmz787> i guess I get what you mean about optimizing g-code for physical transfer
18:50 < nmz787> we do that with lithography too, and that sort of thing is something I will probably have to consider at some point
18:51 -!- justanotheruser [~Justan@unaffiliated/justanotheruser] has quit [Ping timeout: 240 seconds]
18:52 < nmz787> but I am not sure I'd want to modify the g-code itself, rather I might just morph the solid model, and use the same g-code generator as before
18:53 -!- justanotheruser [~Justan@unaffiliated/justanotheruser] has joined ##hplusroadmap
18:58 < xentrac> G-code doesn't say where to put the boundary; it says where to put the tool
18:59 < nmz787> yeah but that is reflective of the boundary
18:59 < xentrac> and how fast to move it, what temperature to heat it to, when to turn cutting fluid on and off, all that stuff
18:59 < xentrac> yes, but the boundary is not in the same place as the tool
18:59 < xentrac> often it's nowhere near it in fact
18:59 < kanzure> "The mad man behind aprilzero.com uses a non-monospace font on a light background last time I checked. What an animal."
18:59 < nmz787> hmm, for a laser causing coagulation it seems like they'd be the same
19:00 < xentrac> the coagulation still has a minimal line width
19:00 < xentrac> but I agree that for a laser it's going to be much closer than for FDM or milling or whatever
19:00 < xentrac> whether you're cutting or hardening
19:01 < xentrac> probably — and I'm guessing a bit here — at least 100μm difference
19:01 < xentrac> which is enough to make the difference between "parts fit together tightly" and "parts have play and fall apart when the assembly is turned upside down"
19:02 < xentrac> or between "parts fit together snugly" and "assembly is impossible"
19:02 < xentrac> for most materials anyway
19:02 < xentrac> and you also likely want to cause coagulation inside the shape you're trying to make, not just at the boundary
19:02 < xentrac> maybe in some kind of topopt-generated or crosshatched pattern
19:04 < nmz787> actually I've been thinking that the laser would define the entire object, the width being controlled by focusing/defocusing the beam spot
19:05 < nmz787> but yeah there could be some cross-hatching to do at some point
19:05 < xentrac> so that's almost more like a raster representation than a boundary representation
19:05 < nmz787> also depends on the device i want to make
19:06 < nmz787> well except that I want vector paths, not some flyback raster output
19:06 < kanzure> "Long short-term memory: a tutorial on LSTM recurrent networks" http://people.idsia.ch/~juergen/lstm/
19:17 < xentrac> nmz787: did you see the Laser Origami paper?  I can't try this myself
19:17 < xentrac> http://stefaniemueller.org/laserorigami-lasercutting-3d-objects/ #laser-cutter #manufacturing of non-planar #3D objects by softening acrylic with the defocused laser, allowing gravity to pull it into shape.  Got Best #Paper in CHI 2013
19:17 -!- justanotheruser [~Justan@unaffiliated/justanotheruser] has quit [Ping timeout: 260 seconds]
19:24 < kanzure> http://www.nervanasys.com/demystifying-deep-reinforcement-learning/
19:24 < kanzure> https://github.com/google/prettytensor (for tensorflow)
19:26 < kanzure> https://github.com/samjabrahams/tensorflow-white-pages-notes
19:26 < kanzure> archels: https://www.reddit.com/r/MachineLearning/comments/3x2ueg/nips_2015_overviews_collection/
19:27 < kanzure> http://www.johnglover.net/blog/generating-sound-with-rnns.html
19:30 < kanzure> https://github.com/kjw0612/awesome-rnn
19:30 -!- justanotheruser [~Justan@unaffiliated/justanotheruser] has joined ##hplusroadmap
19:33 < kanzure> "Visualizing and understanding recurrent networks" http://arxiv.org/pdf/1506.02078.pdf
19:45 < kanzure> "Soft materials in neuroengineering for hard problems in neuroscience" http://rogers.matse.illinois.edu/files/2015/neuronreview.pdf
19:45 < kanzure> that is a cool review.
19:50 < kanzure> "neural dust" concept (10-100 micron free-roaming neural sensors) gets remote power transmission over ultrasound "
19:50 < kanzure> "Ultrasonic beamforming system for interrogating multiple implantable sensors" http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=7318942
19:51 < kanzure> neural dust concept was http://arxiv.org/pdf/1307.2196.pdf
19:52 < kanzure> further "neural dust" follow-up, "Model validation of untethered, ultrasonic neural dust motes for cortical recording" http://www.sciencedirect.com/science/article/pii/S0165027014002842
20:00 -!- PatrickRobotham [uid18270@gateway/web/irccloud.com/x-pafvrylohqlqppjv] has quit [Quit: Connection closed for inactivity]
20:06 < kanzure> "pHflourins are pH sensitive fluorescent proteins used as genetically encoded reporters to measure synaptic release. pHfluorins can be targeted to synaptic vesicles by fusing pHflourin to synaptic vesicle proteins such as SNARE, synaptophysin, and the vesicular glutamate transporter. The acidic environments inside synaptic vesicles quench fluorescence while exocytosis exposes the lumen of the vesicle to neutral pH,which reduces quenching ...
20:06 < kanzure> ... and increasing fluorescence 66. The development of mOrange2-pHluorins tagged to VGLUT1 together with a green calcium reporter GCaMP3 fused with synaptophysin-GaMP3 (SyGCaMP3) permits analysis of vesicle release with calcium concentration at the same synapse 67. In addition, glutamate-sensitive fluorescent reporters have also been developed to visualize the release and diffusion of glutamate at the synapse 68."
20:07 < kanzure> from "Molecular neuroanatomy: A generation of progress" (2014) http://europepmc.org/articles/pmc3946666
20:12 -!- atomical [~atomical@wl2142.cin.net] has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
20:18 < kanzure> i was looking up what cited "Three controversial hypotheses concerning computation in the primate cortex" and now i am convinced that marblestone's entire bibliography is essentially "whatever kanzure had in papers2/neuro/"
20:24 -!- ArturShaik [~ArturShai@37.218.166.45] has joined ##hplusroadmap
20:42 -!- Guest83113 is now known as abetusk
21:31 -!- jaboja64 [~jaboja@esf125.neoplus.adsl.tpnet.pl] has joined ##hplusroadmap
21:35 -!- jaboja [~jaboja@esb12.neoplus.adsl.tpnet.pl] has quit [Ping timeout: 245 seconds]
21:37 -!- jaboja64 [~jaboja@esf125.neoplus.adsl.tpnet.pl] has quit [Ping timeout: 260 seconds]
21:37 -!- pompolic [~A@unaffiliated/pompolic] has quit [Ping timeout: 260 seconds]
21:40 -!- ArturShaik [~ArturShai@37.218.166.45] has quit [Ping timeout: 264 seconds]
21:49 -!- CheckDavid [uid14990@gateway/web/irccloud.com/x-udlcldiexxovkyyk] has quit [Quit: Connection closed for inactivity]
21:53 < nmz787> xentrac: I feel like I saw something like that a while ago
21:54 < nmz787> when I read your comment I first thought of micro/mini scale plastic transparency film, folding UP on it self
21:55 -!- ArturShaik [~ArturShai@37.218.156.241] has joined ##hplusroadmap
21:55 < nmz787> inkscape is not a good workflow for me
21:56  * nmz787 is reading about XML namespaces :/
21:56 < nmz787> all I wanted was to convert my SVG to g-code
21:57 -!- Houshalter [~Houshalte@oh-71-50-56-224.dhcp.embarqhsd.net] has quit [Quit: Quit]
22:08 -!- PatrickRobotham [uid18270@gateway/web/irccloud.com/x-huahnwrbjejnweuv] has joined ##hplusroadmap
22:31 -!- |nodux [uid125132@gateway/web/irccloud.com/x-oeneagiewjaquakg] has quit [Quit: Connection closed for inactivity]
22:39 -!- erasmus [~erasmus@unaffiliated/erasmus] has joined ##hplusroadmap
22:54 -!- Diablo-D3 [~diablo@exelion.net] has quit [Ping timeout: 250 seconds]
22:55 -!- Diablo-D3 [~diablo@exelion.net] has joined ##hplusroadmap
23:02 -!- justanot1eruser [~Justan@unaffiliated/justanotheruser] has joined ##hplusroadmap
23:02 -!- justanotheruser [~Justan@unaffiliated/justanotheruser] has quit [Ping timeout: 256 seconds]
23:09 -!- jaboja [~jaboja@esf125.neoplus.adsl.tpnet.pl] has joined ##hplusroadmap
23:16 -!- justanot1eruser [~Justan@unaffiliated/justanotheruser] has quit [Quit: leaving]
23:16 -!- justanotheruser [~Justan@ip70-187-8-107.om.om.cox.net] has joined ##hplusroadmap
23:16 -!- justanotheruser [~Justan@ip70-187-8-107.om.om.cox.net] has quit [Signing in (justanotheruser)]
23:16 -!- justanotheruser [~Justan@unaffiliated/justanotheruser] has joined ##hplusroadmap
23:20 -!- Gurkenglas [Gurkenglas@dslb-188-103-077-131.188.103.pools.vodafone-ip.de] has joined ##hplusroadmap
23:35 -!- sandeepkr [~sandeepkr@111.235.64.4] has joined ##hplusroadmap
--- Log closed Thu Jan 07 00:00:58 2016