--- Log opened Tue Feb 28 00:00:56 2023
02:03 -!- adlai [~adlai@80.244.243.194] has quit [Ping timeout: 255 seconds]
04:00 < L29Ah> people, how hard is it to get metformin in your $country without having diabetes?
04:02 -!- AMG [ghebo@user/amg] has quit [Quit: Quit]
04:23 < nsh> fenn, i have neither (regular, non-rehearsing/recollecting) internal monologue nor (non-dreamstate) inner imagery
04:23 < nsh> (i am however a proud philosophical zombie with nary a soul nor conscience)
04:24 < nsh> well, it's a matter of quiddity/nuance/subtlety as when i'm imagining mathematical concepts there is indeed often a rigorous application of geometry but it's 'felt' more than imaged
04:27 < nsh> and in psychedelic, hypnogogic/hypnopompic altered states of consciousness there is often vivid internal imagery, but not subject as readily to the conjuring by imagination that others report in day-to-day mundane consciousness
04:28 < nsh> also residual bright internal phosphene-like activations when waking from REM-stage sleep more often at the tail end of a long sleep cycle period
04:52 < muurkha> nsh: do you mean you can't imagine dialogue, or just that you don't do so habitually?
04:53 < nsh> as in, when i rehearse or recollect dialogue/monologue, the thought process is verbal
04:53 < nsh> otherwise, and by default, it is not
04:53 < nsh> it can be rendered verbal but that is a translation and generally a lossy one
04:54 < nsh> the mind is greater than the stories that can be written about its movements
04:55 < nsh> it might rhyme in persian but not in polish
04:55 < nsh> something is always added, something lost
04:58 < muurkha> docl: aside from the question of whether it's a counterproductive idea, it's a pretty small project compared to the scale of the Earth's climate.  if we suppose the dust is made of plagioclase with density 2.7 g/cc we're talking about pulverizing a chunk of the moon 190 m in diameter
05:00 < muurkha> L₁ is a stable equilibrium, so dust that only gets pushed a little way toward earth would return back to it; there's an energy barrier.  over some period of time, inelastic collisions between dust grains would cause them to accrete (presumably the paper discusses this) which you could accelerate greatly by injecting some dense gas into the area
05:02 < muurkha> nsh: when you say "rehearse" are you talking about imagining possible conversations that haven't happened yet?
05:10 -!- AMG [ghebo@2605:6400:c847:1449::9441] has joined #hplusroadmap
05:11 < L29Ah> 03:43:33]<fenn> https://github.com/ggerganov/whisper.cpp
05:11 < L29Ah> grabbed the "base.bin" model, and it says it's not multilingual T_T
05:11 < L29Ah> still, consumes a LOT of cpu of my laptop and does a passable job at deciphering my english
05:13 < nsh> muurkha, aye
05:14 < nsh> or if i think of a poem 
05:14 < nsh> i'd call it subvocalisation rather than inner monologue though
05:15 < nsh> (which is potentially differentiable by sensitive electric potential detection apparatus of the laryngeal muscle areas)
05:18 < muurkha> you just don't find it useful to engage in that kind of verbal thinking for purposes of logic or planning or whatever
05:25 < muurkha> L29Ah: how much CPU does your laptop have?
05:26 < L29Ah> muurkha: about 20W worth across 4 cores
05:28 < muurkha> haha, I mean, like, how many flops
05:29 < L29Ah> "it depends"?
05:30 < L29Ah> never seen a benchmark that utilizes all the fancy cpu extensions and tells a comprehensive report
05:34 < muurkha> benchmarking gets very complex when you're struggling for 1% improvements, but I mean, within a factor of 2 or 3?
05:38 < muurkha> because I don't know if you're using a 1.2 GHz quad-core fourth-generation Xeon with 128-bit SSE from 15 years ago or, like, a 3 GHz i9-10885H from last year
05:38 < muurkha> with AVX-512
05:40 < L29Ah> muurkha: i7-8xxx
05:51 < muurkha> so probably with AVX-512, but maybe only using AVX-256 for power considerations?
06:05 < hprmbridge> kanzure> welcome back poppingtonic
06:05 < hprmbridge> poppingtonic> Hey hey @kanzure
06:06 < hprmbridge> poppingtonic> I have YALLM  startup now where you can use a LLM to talk to or ask questions of pangenomics papers from open research project GeneNetwork (https://genenetwork.org) here:
06:06 < hprmbridge> poppingtonic> https://genenetwork.fahamuai.com
06:07 < muurkha> sweet, thanks for helping us joots
06:08 -!- luna_ [~luna@user/luna/x-4729771] has quit [Ping timeout: 248 seconds]
06:08 < hprmbridge> poppingtonic> Question for OGs: Is paperbot alive?
06:10 < muurkha> it doesn't seem to be
06:12 < hprmbridge> poppingtonic> 🤘🏿
06:53 -!- lsneff [~lsneff@2001:470:69fc:105::1eaf] has quit [Ping timeout: 264 seconds]
06:53 -!- AMG [ghebo@2605:6400:c847:1449::9441] has quit [Ping timeout: 260 seconds]
06:55 -!- Mabel [~Malvolio@idlerpg/player/Malvolio] has quit [Ping timeout: 268 seconds]
07:07 -!- AMG [ghebo@2605:6400:c847:1449::9441] has joined #hplusroadmap
07:18 -!- lsneff [~lsneff@2001:470:69fc:105::1eaf] has joined #hplusroadmap
07:36 < hprmbridge> kanzure> organoid intelligence https://www.frontiersin.org/journals/science/articles/10.3389/fsci.2023.1017235  https://twitter.com/ANeuroExplorer/status/1630521202799411201
08:14 -!- Llamamoe [~Llamamoe@46.204.76.79.nat.umts.dynamic.t-mobile.pl] has joined #hplusroadmap
08:36 -!- Netsplit *.net <-> *.split quits: nsh, mlaga97
08:37 -!- Netsplit over, joins: nsh, mlaga97
09:03 < hprmbridge> harmoniq.punk> I think molecular computing and analog computing will explode. You can compute crazy amount of data with opamps and cells interfaced on a high bandwidth interface like Infiniband
09:09 < kanzure> why hasn't it been done yet? most in vitro neural cultures have been used to do trivial things like play pong.
09:12 -!- cthlolo [~lorogue@77.33.23.154.dhcp.fibianet.dk] has joined #hplusroadmap
09:22 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap
09:29 < nsh> possibly you end up gaining efficiency at the cost of complexity in algorithmic/architectural terms
09:29 < nsh> as with QC 
09:29 < nsh> (vaguely)
09:47 < hprmbridge> nmz787> https://arxiv.org/abs/2202.11142
09:47 < hprmbridge> nmz787> An LLVM-based C++ Compiler Toolchain for Variational Hybrid Quantum-Classical Algorithms and Quantum Accelerators
09:47 < muurkha> cells aren't very high bandwidth, maybe a few kHz, and they're pretty noisy
09:48 < hprmbridge> nmz787> https://www.intel.com/content/www/us/en/forms/developer/quantum/quantum-sdk-beta.html
09:49 < L29Ah> > You can compute crazy amount of data with opamps and cells interfaced on a high bandwidth interface like Infiniband
09:49 < L29Ah> crazier than on ASICs on the same footprint given the current technology for both ASICs and biological circuit?
09:49 < hprmbridge> nmz787> depends how you define bandwidth too... biophotons are using THz radiation, for example.... but possibly are sparse in terms of data transmission https://www.technologyreview.com/2012/05/22/185994/biophoton-communication-can-cells-talk-using-light/
09:53 < muurkha> L29Ah: ASICs can probably compute a lot more in the same footprint, but they have other disadvantages
09:54 < muurkha> neurons seem to be pretty efficient, and plausibly for some applications training neurons is cheaper than designing ASICs
10:18 < L29Ah> but it would take a LOT of research to pack neurons into a competitive footprint
10:20 < L29Ah> and the efficiency of silicon-based artificial neural network may be more than meat-based, since floating mosfet gates and wires are much smaller than the corresponding neural structures
10:21 < L29Ah> it's just that we didn't yet explore the multitude of artificial neural configurations and the mapping of architectures to the task space
10:25 < kanzure> brains already pack neurons
10:38 < L29Ah> no plugging brains into InfiniBand any time soon
12:29 < kanzure> robot hand https://republic.com/clone
13:06 -!- cthlolo [~lorogue@77.33.23.154.dhcp.fibianet.dk] has quit [Read error: Connection reset by peer]
13:46 < hprmbridge> nmz787> also mosfet gates aren't a great comparison to a neuron
13:46 < hprmbridge> nmz787> there's a lot of local interconnects, and QC structures that need considered (like reference voltage generators)
13:46 < hprmbridge> kanzure> agree
13:47 < hprmbridge> nmz787> signal boosters for longer paths
13:47 < hprmbridge> nmz787> (buffers)
13:47 < hprmbridge> nmz787> I'm sure there's a better equivocation
14:32 -!- Llamamoe [~Llamamoe@46.204.76.79.nat.umts.dynamic.t-mobile.pl] has quit [Quit: Leaving.]
14:47 < L29Ah> ... and my laptop CPU can't handle the large™ model
16:40 < kanzure> "The nation that builds the most semiconductors in the next decade is going to be the next superpower. Silicon is the new atom bomb."
17:01 < L29Ah> china is already a superpower, ikr
17:30 < kanzure> multi-modal large language model https://arxiv.org/abs/2302.14045 https://twitter.com/AlphaSignalAI/status/1630651280019292161
17:36 < kanzure> "Moore's law has been looking weaker and weaker. Clock speeds paused a long time ago. Die size is hitting physical limits. Cost per compute is still falling but it lost the exponential it was on. Without some major changes (new architecture/paradigm?) this looks played out" https://twitter.com/WilliamAEden/status/1630690010520498176
17:37 < kanzure> ehh i think ASICs are still significantly faster
17:37 < kanzure> "Existing AI applications are already taking a significant fraction of global compute power, I find it implausible that global AI efforts could scale more than another 2 orders of magnitude, and that crowds out a ton of other compute to do so. Unlikely IMO" yeah maybe not everything will be priced out
17:37 < kanzure> "In addition to physical limits on Moore's Law, I will note that the semiconductor industry is highly centralized, with ASML being the only company making cutting edge fabs, and 1-3 companies variously on the cutting edge of chip manufacture, the major one being in Taiwan..."
17:38 < kanzure> "For one thing, this belief in imminent AGI seemed to spread like a social contagion, with the epicenter beginning as a media/PR blitz by DeepMind. This is great for them as a company, but there's no reason to trust corporate PR in particular"
17:38 < kanzure> "There is an undeniable social incentive for AGI people to hype AGI for more investment, and for safety people to hype danger for more investment. And the arguments themselves are perfect: if hard takeoff is instant death it's crazy not to stop or slow AI progress somewhat!"
17:38 < kanzure> "Especially now that everyone (in my circles and on Twitter) has turned AGI-imminent and certain-doom, it's pretty socially costly to say that maybe AGI is still further away and maybe we won't all die and maybe it's okay we develop AI a little further and see what we learn"
17:40 < L29Ah> nice demo
17:40 < L29Ah> the author list looks very asian
17:40 < kanzure> does ASML also have a monopoly on the EUV lithography stuff? 
17:40 < kanzure> "Before EUV, chipmakers could buy DUV lithography machines from three companies: ASML, Nikon and Canon"
17:41 < kanzure> hmm "ASML has a monopoly on the fabrication of EUV lithography machines, the most advanced type of lithography equipment"
20:25 < fenn> clone robotics is doing exactly what i was trying to build in 2007
20:26 < fenn> i wonder if they have anything special for the valve problem
20:29 < fenn> is organoid intelligence / biocomputing really a new idea? i mean it seems pretty obvious. even star trek had "bio-neural gel packs"
20:29 < fenn> mostly so they could get infected with colorful alien goop
20:41 < fenn> hehe Kosmos-1 describes "the story of a small robot named WALL-E and his quest to build the ultimate life-form."
20:43 < fenn> the name sounds like an early soviet satellite
20:46 < fenn> has there actually been a lot of progress in AI or is it just being revealed now?
20:56 < fenn> re will eden, it's totally plausible that switching to analog computing for neural networks will yield >2 orders of magnitude in performance and efficiency
21:06 < fenn> he also misrepresents the slow vs fast takeoff scenario entirely, by saying "we haven't yet closed the loop on self improvement"
21:06 < fenn> that just means we don't have any evidence at all
21:09 < fenn> "takeoff" is what happens AFTER the AGI becomes self-improving
21:09 < fenn> it could also happen with narrow AI
21:57 < fenn> "Apple’s approach to glucose monitoring is said to combine silicon photonics and optical absorption spectroscopy: It beams specific wavelengths of light into the interstitial fluid below the skin, and all light not absorbed by glucose bounces back to the sensor. From there, an algorithm calculates the wearer's blood sugar levels. "
--- Log closed Wed Mar 01 00:00:57 2023