--- Log opened Tue Oct 25 00:00:58 2022
00:19 -!- darkdarsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 276 seconds]
00:41 -!- alexbfi [~alexbfi@193.209.237.36] has quit [Ping timeout: 240 seconds]
00:43 -!- alexbfi [~alexbfi@193.209.237.36] has joined #hplusroadmap
01:28 -!- Molly_Lucy [~Molly_Luc@user/Molly-Lucy/x-8688804] has quit [Quit: Textual IRC Client: www.textualapp.com]
01:43 < maaku> juri_: most of the people here are coders
01:43 < maaku> and yet we are still short coders imho
01:44 < maaku> and btw I want to recruit a team to work on better CAD/CAM software, particularly molecular CAD tools
01:44 < maaku> seems like something you'd be interested in
02:08 -!- hellleshin [~talinck@108-225-123-172.lightspeed.cntmoh.sbcglobal.net] has quit [Read error: Connection reset by peer]
02:08 -!- hellleshin [~talinck@108-225-123-172.lightspeed.cntmoh.sbcglobal.net] has joined #hplusroadmap
02:13 -!- alexbfi [~alexbfi@193.209.237.36] has quit [Remote host closed the connection]
02:18 -!- alexbfi [~alexbfi@193.209.237.36] has joined #hplusroadmap
02:56 < juri_> maaku: its something i've stared at a bit, but closing the loop on that (EG: building the machines at home and playing with them) looks hard. i stay in the FDM field, because there is a lot of space there for improvement.
02:57 < juri_> fenn: glad to hear it. i stuck by regardless. :)
02:59 < juri_> for what it's worth, projective geometric algebra is the algebra to be doing this all in, if for no other reason than many of the operations occur on N+1 datum, for a representation of a coordinate in N space, meaning if you're using FPUs, you get to keep using FPUs, and get to store a bit more precision in your representation.
03:13 < juri_> Yes, the answers my slicer returns are wrong, but they're wrong to 54 more bits of mantissa! </not_serious>
05:41 -!- juri__ [~juri@79.140.114.58] has joined #hplusroadmap
05:43 -!- alexbfi [~alexbfi@193.209.237.36] has quit [Remote host closed the connection]
05:46 -!- juri_ [~juri@84-19-175-179.pool.ovpn.com] has quit [Ping timeout: 272 seconds]
06:53 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has left #hplusroadmap []
07:10 -!- juri__ [~juri@79.140.114.58] has quit [Read error: Connection reset by peer]
07:16 -!- juri_ [~juri@79.140.120.117] has joined #hplusroadmap
07:18 -!- darkdarsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has joined #hplusroadmap
07:18 -!- juri_ [~juri@79.140.120.117] has quit [Read error: Connection reset by peer]
07:21 < muurkha> maaku: you definitely want to shoot quickly, but you also want to shoot in the right direction.  thinking is what allows you to shoot at the right person with the right elevation.  you don't want less shooting in the right person; you want the necessary amount, faster
07:21 -!- juri_ [~juri@79.140.120.117] has joined #hplusroadmap
07:22 < muurkha> *at
07:22 -!- juri_ [~juri@79.140.120.117] has quit [Read error: Connection reset by peer]
07:22 < muurkha> various theorists have generalized this to all kinds of competitive situations, applying John Boyd's "OODA loop" with various degrees of success to them
07:24 < muurkha> there are other situations where thinking produces no economic results
07:26 -!- juri_ [~juri@84-19-175-179.pool.ovpn.com] has joined #hplusroadmap
07:37 < maaku> FDM?
07:39 < maaku> muurka: hence my assumption about a Boltzmann distribution, which would integrate out to an S curve of total value per unit time spent thinking.
07:40 < maaku> There is great value to taking a moment to assess a situation before shooting. There is less value (perhaps even negative in this case) to spending an hour contemplating before shooting.
07:40 < muurkha> fused deposition modeling
07:41 < muurkha> juri_: I very much hope that I haven't given you the idea that *I* think you don't belong here.  I think you belong here more than just about anyone
07:41 < muurkha> FDM is the technique used by the standard sort of RepRap-derived 3-D printer
07:42 < maaku> thanks
07:53 -!- luna__ [~luna@82-132-231-143.dab.02.net] has joined #hplusroadmap
07:55 -!- luna_ [~luna@user/luna/x-4729771] has quit [Ping timeout: 246 seconds]
08:03 -!- luna__ [~luna@82-132-231-143.dab.02.net] has quit [Read error: Connection reset by peer]
08:03 -!- luna_ [~luna@2a01:4c8:46:b1ff:49bc:c8ec:b694:b857] has joined #hplusroadmap
08:03 -!- luna_ [~luna@2a01:4c8:46:b1ff:49bc:c8ec:b694:b857] has quit [Changing host]
08:03 -!- luna_ [~luna@user/luna/x-4729771] has joined #hplusroadmap
08:38 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0::1909] has joined #hplusroadmap
08:48 < docl> the rapid apm scenario from agi would be that the agi gets very good at inferring stuff from first principles given even limited (but accurate) information about physics. for example, it could derive the speed of light based on stuff we might not think of. easier example: watching objects fall on youtube videos and deriving gravitational acceleration. so maybe it's able to model atomic physics very 
08:48 < docl> accurately and do enough experiments in pure simulation to produce a working apm model on the first try
08:50 < docl> picture a bored, highly obsessive engineer/scientist, very curious about the outside world and trying to infer it as accurately as possible based on a few youtube videos
09:35 < nmz787> docl: focused ion beams exist with few-nm spot sizes (helium, neon, etc). I'm not sure what sort of precision exists out there in being able to control the number of ions shot out
09:37 < nmz787> https://discuss.python.org/t/int-str-conversions-broken-in-latest-python-bugfix-releases/18889/34
09:38  nmz787> printing/parsing ints larger than 4300 is not disabled by default in some very recent python release
09:38  nmz787> because of DOS attacks
09:39 (*) L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
09:40 (*) jrayhawk [~jrayhawk@user/jrayhawk] has quit [Ping timeout: 252 seconds]
09:40 (*) heath [~heath@user/heath] has quit [Ping timeout: 252 seconds]
09:41 (*) sivoais [~zaki@199.19.225.239] has quit [Ping timeout: 252 seconds]
09:41 (*) gnusha [~gnusha@user/gnusha] has quit [Ping timeout: 252 seconds]
09:41 (*) andytoshi [~apoelstra@user/andytoshi] has quit [Ping timeout: 252 seconds]
09:41 (*) soundandfury [~soundandf@user/soundandfury] has quit [Ping timeout: 252 seconds]
09:41 (*) deltab [~deltab@user/deltab] has quit [Ping timeout: 252 seconds]
09:41 (*) Hoolooboo [~Hooloovoo@hax0rbana.org] has quit [Ping timeout: 252 seconds]
09:42 (*) jrayhawk [~jrayhawk@user/jrayhawk] has joined #hplusroadmap
09:42 (*) heath [~heath@user/heath] has joined #hplusroadmap
09:42 (*) andytoshi [~apoelstra@user/andytoshi] has joined #hplusroadmap
09:42 (*) sivoais [~zaki@199.19.225.239] has joined #hplusroadmap
09:42 (*) Hooloovoo [~Hooloovoo@hax0rbana.org] has joined #hplusroadmap
09:42 (*) soundandfury [~soundandf@user/soundandfury] has joined #hplusroadmap
09:43 (*) deltab [~deltab@user/deltab] has joined #hplusroadmap
09:43 (*) Malvolio [~Malvolio@idlerpg/player/Malvolio] has quit [Ping timeout: 252 seconds]
09:48 (*) Mabel [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap
09:53  muurkha> docl: sure, that's Eliezer's "That Alien Message" scenario
09:53  docl> nmz787: does it seem plausible to have a single atom wide aperture that you use to control whether the beam hits/removes a given atom of diamond (checking and retrying each time it fails)?
09:54  docl> I'm picturing using a biomimetic mechanism (spiroligomer based) to control the positioning of the aperture
09:54  muurkha> but carrying out empirical experimentation seems to enable significantly faster inference, and that may even still be the case after scientific epistemology finishes assimilating the results of Pearl's work on causality
09:55  muurkha> nmz787: "few-nm" falls far short of what is needed fo rAPM
09:57  docl> (I'm trying to build a defensible case for bootstrapping from spiroligomer/solution phase tech to diamondoid/machine phase, and milling diamond nanocomponents seems like an interesting approach)
10:01  nmz787> a shadow mask is possible, but incorrectly aimed ions will mill the aperture and make it wider
10:01  nmz787> shadow mask == stencil
10:02  nmz787> there's possibly some combination of ions, acceleration voltage, and stencil materials that all play well together so the stencil doesn't etch away
10:02  nmz787> "sputter rate" is the term I'd probably start searching with
10:04  docl> stencil might be self regenerating in this context. or maybe you could make it thick enough not to be an issue
10:06  nmz787> thick enough then becomes a tunnel, not an aperture, and may suffer from (mis)alignment issues if the beam and tunnel are not colinear 
10:07  nmz787> that could possibly be self-adjusting by just turning up the voltage and blasting through the tunnel... making it co-linear
10:09 (*) mirage335155998 [~mirage335@2a01:4f8:120:2361::1] has quit [Quit: Client closed]
10:09 (*) mirage335155998 [~mirage335@2a01:4f8:120:2361::1] has joined #hplusroadmap
10:10  docl> ah, and too much voltage probably has unintended effects like removing multiple atoms
10:23  maaku> docl my picture of a bored, highly obsessive engineer/physicist learning from YouTube videos involves the resulting theories being very, very wrong
10:23  maaku> That’s how you get green letter writing crackpots.
10:24  maaku> And as mentioned above, simulations are very often wrong. They have to be carefully calibrated against experiment because of the multitude of ways they can magnify artifacts of the simulation to get non physical results.
10:26  maaku> Something as trivial as using the wrong basis vectors for tessellation can result in magnifying magnifying numerical inprecision and nonsensical results.
10:26  muurkha> green letter?
10:27  kanzure> perpetual motion cranks, i assume.
10:27  muurkha> I don't think there's a meaningful difference here between "simulation" and "theory"
10:27  maaku> All physics professors get these letters from crackpots about their world upending theories. They’re often written in green ink. Why, idk.
10:28  maaku> Maybe a logic-based AI designing from first principles would be able to design simulations free of errors, but those aren’t the kind of AIs we’re making rapid (or any) progress on.
10:29  muurkha> no, simulations (theories) are always simplifications of reality
10:30  muurkha> simplifications necessarily have errors
10:32  maaku> Yeah sorry I was drawing a distinction between errors in the model vs computational errors introduced by the simulation itself
10:32  muurkha> that distinction may or may not exist, depending on how the simulation is structured
10:33  maaku> Both are relevant and reasons why AI virtual scientist won’t FOOM the way some people seem to think
10:34  muurkha> when a human simulates the trajectory of a baseball in order to catch it, it may be hard to figure out whether an error in their simulation is "in the model", "computational error", or even in the motor control system
10:34  muurkha> nmz787: if your 100-nm-thick stencil has an 0.2 nm hole in it that tapers out at 10%, it will be 20 nm wide on the other side
10:36  muurkha> from the point of view of pigs, chickens, or humpback whales, humans are already fooming, in large part because of how they started simulating reality with digital computation with paper and pencil largely about 250 years ago
10:37  muurkha> 250 years is a very short period of time in the lifespan of the chicken species (though not individual chickens) or the CO₂ balance
10:38  muurkha> and yeah, it has resulted in a lot of theories being very, very wrong.  remember Newton's extensive exploration of alchemy (though with limited ability to experiment, since it was outlawed in England) and Biblical numerology
10:40  muurkha> but to a very great extent, modern civilization is the result of applying philosophy to engineering
10:43 (*) mirage335155998 [~mirage335@2a01:4f8:120:2361::1] has quit [Quit: Client closed]
10:43 (*) mirage335155998 [~mirage335@2a01:4f8:120:2361::1] has joined #hplusroadmap
10:49  muurkha> in Lucio Russo's lucid prose:
10:50  muurkha> > Such models, of course, allow one to describe and predict natural phenomena, by translating them to the theoretical level via correspondence rules, then solving the “exercises” thus obtained and translating the solutions obtained back to the real world. There is, however, another possibility, much more interesting: moving freely within the theory, and so reaching points not associated to
10:50  muurkha> anything concrete by the correspondence rules. From such a point in the theoretical model one can often construct the corresponding reality, thus modifying the existing world.
10:52  muurkha> that is, precisely what has allowed modern engineering to accelerate so dramatically is highly obsessive engineer/physicists coming up with theories ("models", "simulations"), imagining theoretical entities and characterizing their behavior within a simulation, and then putting them into practice
10:54  muurkha> that is what we do whenever we build a circuit, a model airplane, or a computer-running-a-new-program
10:57 (*) mirage335155998 [~mirage335@2a01:4f8:120:2361::1] has quit [Quit: Client closed]
10:57 (*) mirage335155998 [~mirage335@2a01:4f8:120:2361::1] has joined #hplusroadmap
10:58  muurkha> you say "simulations are very often wrong", but this understates the difficulty; these theoretical models are *always* wrong, and how wrong is just a matter of degree.  but by virtue of carefully calibrating them against experiment, we can churn through a dozen new candidate bridge designs in an hour instead of a century, roughly a 1,000,000× speedup.  sometimes when we build the full-scale bridges 
10:58  muurkha> they fall down anyway, because the way the models were wrong turned out to be consequential
10:59  muurkha> the Tacoma Narrows bridge in the case of suspension bridges, for example, and the Ponte Morandi in the case of cable-stayed bridges
11:21  docl> I wonder if the AI could learn to recognize when it's going into green ink letter territory, then selectively breed itself not to. periodically spin up 100 copies and delete the 99 that go craziest, repeat ad nauseum
11:22  muurkha> well, that would also exclude Principia Mathematica
11:22  docl> hmm, point
11:23  docl> I'm putting it in terms of deleting a person-like critter but really it's a question of whether it can develop a reliable noise filter that doesn't exclude valid inferences
11:23  muurkha> experiment
11:41  muurkha> I'm not just saying you'd prune Newton, I'm saying you'd also prune the Principia
11:53  muurkha> it's full of crazy assertions: not just heliocentrism (not original to Newton but still prohibited — Newton was born the year Galileo died under house arrest for, among other things, teaching heliocentrism) but inertia, universal gravitation, the weights of the planets, the motion of the Sun around the barycenter, the non-sphericity of the Earth, and so on
11:56  muurkha> certainly the notion that massive objects in motion tend to remain in motion is thoroughly repugnant to common sense; who has ever seen such a thing happening?  Even cannonballs flying through the air can be observed to cease their motion and come to rest
11:56  muurkha> (the theory at the time was that air rushing into the vacuum behind the cannonball pushed it forward, which you can observe happening with a party balloon or inflated bladder if you try)
11:58  muurkha> Newton's notion that the Earth wasn't spherical clashed not only with aesthetics but with well-known teachings of the Ancients; why, Eratosthenes had even measured the size of the sphere, and gotten within 10% of the correct answer
11:59  darkdarsie> I guess a cannonball would stop much sooner if there was no air behind it. Supersonic projectiles are different, though.
12:00  darkdarsie> 100 kN/m2 is quite a force.
12:00  muurkha> I don't think the difference is large
12:00  muurkha> and Newton's claim that the Sun wobbled around the "true center of the World" contradicted not only the Ptolemaic model but even Copernicus and Kepler
12:02  muurkha> so, I don't think it's an accident that Newton himself didn't have a reliable noise filter and was kind of a paranoid delusional nutcase with an obsession with Biblical eschatology
12:02  muurkha> any reliable noise filter would have reliably filtered out the Principia
12:03  muurkha> if it was based on some kind of anomaly detection or detecting disagreement with common knowledge, rather than on experiment
12:10  muurkha> contemporaries complained that a powerful invisible force that could reach all the way from the Sun to the Earth through empty space was more properly a magical delusion than a scientific theory
12:42  docl> maybe this is a way for AI to develop rationality to a superhuman level. give it a data set about a few obvious things like heavy stuff slowing down less than light stuff, plus the knowledge newton would have had access to, and ask it if the principia is nonsense vs the competing nonsense explanation from the time. train it until it reliably guesses correctly, then repeat with similar examples from 
12:42  docl> history
12:43  docl> I'm not arguing that it could derive scientific truth fully without empirical data, just that it could filter much better and arrive at newton's conclusions more easily with less of it
12:44  docl> better still if it arrives at relativistic physics instead
12:51 (*) spaceangel [~spaceange@ip-94-113-214-149.bb.vodafone.cz] has joined #hplusroadmap
13:09 (*) mirage3351559983 [~mirage335@2a01:4f8:120:2361::1] has joined #hplusroadmap
13:13 (*) mirage335155998 [~mirage335@2a01:4f8:120:2361::1] has quit [Ping timeout: 244 seconds]
14:46  maaku> handsight makes it look easy, but from the other direction there are millinos of possible theories that could fit the data and avoiding overfitting is hard
14:47  maaku> again a from-scratch logical AI could probably do a better job here, but the AI algorithms which are making big strides now are based on machine learning and are really just glorified recursive pattern matchers
14:47  maaku> like people, in that respect
14:47  maaku> but what do you get when you throw people at a bunch of data? generally speaking, get get stuff like religious and animus belief systems
14:49  kanzure> what happened to the "micrometer scale robots making one order of magnitude smaller robots" stuff? was that feynman?
14:49  kanzure> or did nobody actually pursue that
14:55  muurkha> that was feynman, and I don't think anyone has done a lot on that sort of top-down nanotech.  I think it's a promising approach in some ways
14:56  muurkha> very promising if you have very superhuman AI so thinking is no longer the bottleneck
15:31 (*) luna__ [~luna@82-132-230-112.dab.02.net] has joined #hplusroadmap
15:34 (*) luna_ [~luna@user/luna/x-4729771] has quit [Ping timeout: 276 seconds]
15:36 (*) luna__ [~luna@82-132-230-112.dab.02.net] has quit [Read error: Connection reset by peer]
15:38 (*) luna__ [~luna@2a01:4c8:46:b1ff:35fb:707c:fa02:fe62] has joined #hplusroadmap
15:44  maaku> kanzure: I don't think anyone has actually worked towards that, unless you consider MEMS and AFM as a step down that path
15:45  maaku> one problem, though absolutely not a show-stopper, is that micro and submicro equipment operates by different rules, with electrostatic forces being highly dominant, high friction due to rough edges, etc.
15:46  maaku> so it's not a matter of just making the same machine but smaller. you'd have to reinvent major components for each process step down to work by different principles
15:47  maaku> if there was a clear shot to going direct to APM, that would avoid all that unnecessary work
15:52 (*) luna__ is now known as luna_
15:52 (*) luna_ [~luna@2a01:4c8:46:b1ff:35fb:707c:fa02:fe62] has quit [Changing host]
15:52 (*) luna_ [~luna@user/luna/x-4729771] has joined #hplusroadmap
16:01 (*) spaceangel [~spaceange@ip-94-113-214-149.bb.vodafone.cz] has quit [Remote host closed the connection]
16:16  kanzure> yeah, the centimeter factory would be different from the millimeter factory probably looks different from the micrometer factory, and so on
16:18  muurkha> yeah, one of the most tempting things about wet to me is that it avoids all that fucking annoying supercritical drying
16:18  muurkha> you can probably make all the centimeter stuff work at the millimeter scale if you aren't too concerned about wear lifetime
16:34  maaku> muurkha: what's the pathway from wet biochemistry to diamandoid machines?
16:35  muurkha> if I knew that, it would be because I had succeeded in doing it, but I didn't mean wet *bio*chemistry necessarily
16:42  muurkha> I mean HF etching and ECM are wet chemistry but not very bio.  if you need a dry process step after that, or if the resulting device has to operate dry, you need supercritical drying to keep surface tension from wrecking your carefully designed MEMS structures, so if you can just leave it wet forever instead that'd be an improvement
16:54 (*) luna__ [~luna@82-132-230-112.dab.02.net] has joined #hplusroadmap
16:57 (*) luna_ [~luna@user/luna/x-4729771] has quit [Ping timeout: 246 seconds]
16:57  maaku> ECM?
16:57  muurkha> electrochemical machining
16:58  maaku> gas phase CVD wouldn't have those issues, right?
16:59  muurkha> well, neither ECM nor CVD have them.  it's the transition between them that does
16:59  muurkha> in either direction
16:59 (*) luna__ [~luna@82-132-230-112.dab.02.net] has quit [Read error: Connection reset by peer]
17:00  maaku> sorry I was assuming more vaccum machine movements interleved with gas-phase CVD
17:00  maaku> like zyvex's process
17:00 (*) luna__ [~luna@2a01:4c8:46:b1ff:1d8:1264:cdca:2694] has joined #hplusroadmap
17:02  maaku> it's just transition to-from liquid phase that could cause those difficulties no?
17:04  muurkha> yes, exactly
17:06  muurkha> there's an important advantage to staying in liquid phase all the time instead of vacuum/gas phase, which is that it greatly ameliorates the stiction problem
17:11  docl> would diamond a nanostructure tend to lose surface atoms in a typical liquid? I've been sort of assuming it would be safe if there's no UV light
17:13 (*) darkdarsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 272 seconds]
17:14  muurkha> thin film polycrystalline diamond coatings don't dissolve in water, is that what you mean?
17:21  docl> I'm more wondering about whether nanocomponents made in an aqeuous solution would have defects. I think that answers it
17:29  muurkha> I don't really know
18:00 (*) luna__ is now known as luna_
18:01 (*) luna_ [~luna@2a01:4c8:46:b1ff:1d8:1264:cdca:2694] has quit [Changing host]
18:01 (*) luna_ [~luna@user/luna/x-4729771] has joined #hplusroadmap
18:03 (*) yashgaroth [~ffffffff@2601:5c4:c780:6aa0::1909] has quit [Quit: Leaving]
19:30 (*) mirage3351559983 [~mirage335@2a01:4f8:120:2361::1] has quit [Ping timeout: 244 seconds]
19:32 (*) Mabel is now known as Malvolio
19:35 (*) Malvolio is now known as CVDFNSTR
21:14 (*) alexbfi [~alexbfi@dzy9b2yypwzmq353lw3bt-3.rev.dnainternet.fi] has joined #hplusroadmap
21:28  maaku> I don’t think there is a toolset for doing diamond mechanosynthesis in solution. The Merkle Freitas stuff uses highly reactive tooltips that would only work in vacuum.
21:29  maaku> Given the energies involved in hydrogen abstraction on diamond, I don’t know how you could avoid this issue.
21:38  fenn> a local nm scale shroud around the tip to maintain a more or less vacuum environment inside
21:41  fenn> maybe noble gas backfill
21:42  fenn> carbon won't bond with helium at room temperature
22:20 (*) CVDFNSTR is now known as Malvolio
22:32  fenn> some sci-fi concept art http://ivangraphics.com/
23:36 (*) darkdarsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has joined #hplusroadmap
--- Log closed Wed Oct 26 00:00:58 2022