--- Log opened Sat Jan 21 00:00:20 2023
01:01 -!- Llamamoe [~Llamamoe@46.204.76.114.nat.umts.dynamic.t-mobile.pl] has joined #hplusroadmap
03:17 -!- TMM_ [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
03:18 -!- TMM_ [hp@amanda.tmm.cx] has joined #hplusroadmap
03:43 -!- Malvolio is now known as Guest5327
03:43 -!- Guest5327 [~Malvolio@idlerpg/player/Malvolio] has quit [Killed (sodium.libera.chat (Nickname regained by services))]
03:44 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap
05:19 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0::4249] has joined #hplusroadmap
07:28 < kanzure> "Embryo-mediated genome editing for accelerated genetic improvement of livestock" https://pdfs.semanticscholar.org/7db0/c12c65b112d09f72814bd662603e2d0c0b2a.pdf
07:34 < kanzure> http://rmebrk.kz/journals/6283/94586.pdf#page=11   "The Cogent (UK) was the first company in the world, which began to use the method of separating the servicing bulls' semen by sex under production conditions (1999). The method of dividing semen by sex was developed by X&Y Inc. (USA). It is based on the fact that gamete (germinal cells) of bulls contain a haploid set of chromosomes. ...
07:34 < kanzure> ...Consequently, some sex cells contain X chromosomes, and others Y chromosomes. Gametes with the X chromosome contain 4% more of DNA than sperm with the Y chromosome. While staining chromosomes of germinal cells, it was found that gametes with X chromosomes absorb 4-5% more of stain than gametes with Y chromosomes [4]."
07:36 < kanzure> their website seems to be gone http://www.xyinc.com/
07:37 < kanzure> "... However, the only proven method for separating male- and female-bearing sperm is the fluorescence activated cell sorting approach of Johnson et al. (1987a, 1987b, 1989, and 1999). Also known as the “Beltsville Sperm Sexing Technology,” this method was subsequently licensed to XY, Inc. (www.xyinc.com) for commercial development." ( ...
07:37 < kanzure> ...https://www.sciencedirect.com/science/article/pii/S0022030204700673 )
07:38 < kanzure> http://web.archive.org/web/20050829201230/http://www.xyinc.com/
07:42 -!- TMM_ [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
07:42 -!- TMM_ [hp@amanda.tmm.cx] has joined #hplusroadmap
07:42 < kanzure> apparently there were lots of legal battles involved? https://www.cbc.ca/news/canada/british-columbia/bull-semen-xy-sexing-1.3644165
07:45 -!- Llamamoe [~Llamamoe@46.204.76.114.nat.umts.dynamic.t-mobile.pl] has quit [Quit: Leaving.]
07:45 < kanzure> there was also some securities fraud going on? https://www.theage.com.au/business/exec-in-asic-probe-sells-off-us-holding-20070518-ge4x4l.html
07:47 < kanzure> a modified cytometer can't be that valuable, what was it just a camera attached to it?
07:50 < kanzure> https://patents.google.com/?q=%22mervyn+jacobson%22&oq=%22mervyn+jacobson%22
07:52 < kanzure> https://patents.google.com/patent/US7723116B2/en
08:00 -!- test_ is now known as _flood
08:06 -!- Llamamoe [~Llamamoe@46.204.76.114.nat.umts.dynamic.t-mobile.pl] has joined #hplusroadmap
08:26 -!- stipa_ [~stipa@user/stipa] has joined #hplusroadmap
08:29 -!- stipa [~stipa@user/stipa] has quit [Ping timeout: 255 seconds]
08:29 -!- stipa_ is now known as stipa
08:42 -!- ia3orn [~ia3orn@2600:1700:97b8:a050:6444:eae2:172e:5930] has joined #hplusroadmap
09:18 < kanzure> i was not aware that alex zhavoronkov actually has funding for his longevity work (his drug development company last raised $200m?)- how did that happen? 
09:30 < docl> woah that's amazing
09:37 < juri_> bravo. now deliver results.
09:37 < kanzure> "Vascularization and engraftment of transplanted human cerebral organoids in mouse cortex" https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6243198/
09:38 < kanzure> "Human cerebral organoids establish subcortical projections in the mouse brain after transplantation" https://www.nature.com/articles/s41380-020-00910-4
09:39 < kanzure> "After transplantation into the mouse medial prefrontal cortex, the grafted human cerebral organoids survived and extended projections over 4.5 mm in length to basal brain regions within 1 month. The transplanted cerebral organoids generated human glutamatergic neurons that acquired electrophysiological maturity in the mouse brain. Importantly, the grafted human cerebral organoids ...
09:39 < kanzure> ...functionally integrated into pre-existing neural circuits by forming bidirectional synaptic connections with the mouse host neurons. Furthermore, compared to control mice, the mice transplanted with cerebral organoids showed an increase in freezing time in response to auditory conditioned stimuli, suggesting the potentiation of the startle fear response."
09:44 < kanzure> combined with micro-patterning to control architecture, that could be an interesting technique https://www.med.upenn.edu/songlab/assets/user-content/documents/CurrOpinNeurobio_Apr2022.pdf
09:44 < kanzure> (page 7)
09:48 < docl> .title https://insilico.com/
09:48 < hplusbot> docl: Main
09:50 < kanzure> or figure 2 on page 4
09:52 < hprmbridge> kanzure>  https://cdn.discordapp.com/attachments/1064664282450628710/1066414924940710018/image.png
10:03 -!- cthlolo [~lorogue@77.33.23.154.dhcp.fibianet.dk] has joined #hplusroadmap
10:09 < docl> . During this process, Sonic hedgehog (SHH)
10:09 < docl> is critical for ventral region patterning, while BMP and
10:09 < docl> WNTs are important for dorsal fate patterning.
10:14 -!- pfui [~jwd@47.189.1.51] has joined #hplusroadmap
10:14 -!- pfui [~jwd@47.189.1.51] has left #hplusroadmap []
10:15 -!- pfui [~jwd@47.189.1.51] has joined #hplusroadmap
10:15 -!- pfui is now known as Jay_Dugger
10:16 < Jay_Dugger> Hello, everyone.
10:16 < docl> hello Jay_Dugger 
10:16 < Jay_Dugger> Has all action moved to Discord?
10:17 < docl> not really, there's just a bridge now
10:17 < Jay_Dugger> Thank you.
10:18 < docl> kanzure: making mice closer to human scares the heck out of me, to be honest. I already feel like a murderer for using mouse traps. I want as safely non morally relevant agi as I can get
10:24 < hprmbridge> kanzure> even if you had all the human brain cells added, i don't think the mouse would have a human brain.
10:24 < L29Ah> is it a murder if the victim is a deathist anyway? philosoraptor.png
10:34 < hprmbridge> kanzure> and, i don't think it would work anyway
11:54 < hprmbridge> BrickOnKeyboard> again we don't mean a literal single robot is self replicating.  we mean a set of machines, with automation for all tasks.  Today we have automation for *most* tasks, human workers are performing only a limited percentage (it depends on how you measure percentage.  if by *energy* input into a process, human workers do almost nothing.  If you mean by complexity, human workers do *some* of the
11:54 < hprmbridge> BrickOnKeyboard> complex steps for medium value products.  For high value products like semiconductor chips, they do NONE of the complex steps - the production is 100% automated.  The machine maintenance is done by humans though.)
11:54 < hprmbridge> BrickOnKeyboard> the last remaining piece needed to satisfy automation is from cutting edge AI, using techniques that have only existed for months or a few years
11:55 < hprmbridge> BrickOnKeyboard> so it's not a very interesting claim to make - most humans are not working on cutting edge AI, and so far little effort has been made to this, but Deepmind already has found techniques that right now, as they exist, would scale to full self replication
11:56 < hprmbridge> BrickOnKeyboard> it has to do with a machine able to generalize to a solver for all tasks in task space.
11:57 < L29Ah> task space, hmmm
11:57 < hprmbridge> BrickOnKeyboard> or in short, with demonstrated robotics it is technically possible to automate all steps to self replicating robots.  Using techniques that have existed for decades.  It isn't *economically feasible* because there are many complex things that are only done a small number of times that are involved
11:58 < hprmbridge> BrickOnKeyboard> example: changing the struts on a particular type of mining machine
11:58 < hprmbridge> BrickOnKeyboard> we could, without any new AI techniques, automate doing that
11:58 < hprmbridge> BrickOnKeyboard> but it's not worth the *cost* to build all the setup and the strut-swapping plant that would have to be used
11:58 < hprmbridge> BrickOnKeyboard> instead human workers will do it with basic tools like lifts and wrenches and so on
11:59 < hprmbridge> BrickOnKeyboard> it's because paying the workers is cheaper than designing and maintaining the strut swapping plant
11:59 < hprmbridge> BrickOnKeyboard> however, these workers are geniuses.  They do a set of simple tasks
11:59 < hprmbridge> BrickOnKeyboard> like "maneuver your body until you can reach this specific bolt with a wrench"
12:00 < hprmbridge> BrickOnKeyboard> or "unbolt that bolt without too much damage"
12:00 < hprmbridge> BrickOnKeyboard> or "connect the lift to this part"
12:00 < hprmbridge> kanzure> have you ever fabricated a chip? or studied the process of nanolithography in fabrication facilities?
12:00 < hprmbridge> BrickOnKeyboard> I worked for Intel so kinda?
12:00 < hprmbridge> BrickOnKeyboard> didn't work in that area though
12:00 < hprmbridge> kanzure> intel does not have self-replicating fabs
12:00 < hprmbridge> BrickOnKeyboard> I explained this to you
12:01 < hprmbridge> BrickOnKeyboard> above.  see above.  it already has been answered.
12:01 < hprmbridge> kanzure> is there anything that would possibly convince you (like with evidence) that you are wrong about deepmind having solved self-replication and microprocessor fabrication?
12:02 < hprmbridge> BrickOnKeyboard> if you could find a hole in my reasoning, yes.  but you have to actually look at my reasoning and internalize it in a way that I know you understand it
12:02 < hprmbridge> BrickOnKeyboard> otherwise I would not bother to listen to any claimed holes
12:03 < hprmbridge> BrickOnKeyboard> so far you do not appear to understand what my reasoning is
12:03 < hprmbridge> kanzure> correct, most likely i don't understand your reasoning
12:04 < hprmbridge> kanzure> a lot of factory automation is surprisingly rigid and fragile, and nobody knows how it works because of lock-ups and closely held secrets
12:04 < hprmbridge> BrickOnKeyboard> I agree
12:04 < hprmbridge> kanzure> even an AI wouldn't know how it works because the companies don't tell anyone how it works; otherwise why would ASMR have a monopoly on nanosteppers
12:04 < hprmbridge> nmz787> Where? I used to work in JF, now in RA
12:05 < hprmbridge> BrickOnKeyboard> that isn't necessary
12:05 < hprmbridge> nmz787> I agree it's often cheaper to use humans vs cost of automation development
12:06 < hprmbridge> BrickOnKeyboard> JF also.  dumped them for Qualcomm, but still are in Hillsboro
12:06 < hprmbridge> kanzure> have you met nmz787 before?
12:06 < hprmbridge> BrickOnKeyboard> right.  but the Intel example specifically shows how the most difficult part - actually processing the wafers - is 100% automated
12:06 < hprmbridge> BrickOnKeyboard> if you can automate wafer processing with robots there is no task a human hand can do that you couldn't automate
12:06 < hprmbridge> nmz787> Not all the steps tho
12:06 < hprmbridge> kanzure> yes, if you spend enough time developing it, you can automate most things
12:06 < hprmbridge> BrickOnKeyboard> it's not a matter of whether you *can* automate something
12:07 < hprmbridge> BrickOnKeyboard> it's whether or not the task is done often enough that you can *afford* to automate it
12:07 < hprmbridge> BrickOnKeyboard> that is the core claim I have made
12:07 < hprmbridge> BrickOnKeyboard> if you see a fault, let me know
12:07 < hprmbridge> kanzure> btw have you read freitas KSMR http://www.molecularassembler.com/KSRM.htm
12:07 < hprmbridge> BrickOnKeyboard> because then explaining what deepmind has done is simple
12:07 < hprmbridge> BrickOnKeyboard> I know about nanotech, they aren't necessary though they do make self replication simpler
12:07 < hprmbridge> BrickOnKeyboard> they won't be how we achieve
12:07 < hprmbridge> kanzure> this is not nanotech
12:08 < hprmbridge> BrickOnKeyboard> nanoassemblers, fine
12:08 < hprmbridge> BrickOnKeyboard> I know what they are
12:08 < hprmbridge> kanzure> although freitas does write a lot about nanotech, his book on kinematic self-replicating machines is not mainly about nanotech
12:08 < hprmbridge> kanzure> no this is like meso-scale replication systems
12:08 < hprmbridge> BrickOnKeyboard> well we don't need that *either*
12:08 < hprmbridge> BrickOnKeyboard> literally i am saying "exactly how we make everything right now, but a robot arm used wherever a human arm was needed"
12:09 < hprmbridge> nmz787> A lot of that is including driving trucks full of chemicals around
12:09 < hprmbridge> BrickOnKeyboard> the limiting factor is simply that there are easily hundreds of thousands, probably millions of discrete 'tasks'. and complex conditional trees of what to do when the 'normal' way to do something doesn't work
12:09 < hprmbridge> kanzure> there's a lot of knowledge that has never been written down and we would not be able to replicate intel's facility with a robot or even a human for that matter
12:09 < hprmbridge> nmz787> Not nano at all
12:09 < hprmbridge> BrickOnKeyboard> does everyone agree that :
12:09 < hprmbridge> BrickOnKeyboard>
12:09 < hprmbridge> BrickOnKeyboard> 1.  robotic hardware can *do* any task
12:09 < hprmbridge> BrickOnKeyboard> 2.  the limiting factor is simply that there are easily hundreds of thousands, probably millions of discrete 'tasks'. and complex conditional trees of what to do when the 'normal' way to do something doesn't work, where we use humans to fill in the gap
12:09 < hprmbridge> nmz787> Eh a most is academic, secrets about scale up for the most part
12:10 < hprmbridge> BrickOnKeyboard> because the reasoning for why deepmind has solved it drops right in #3
12:10 < hprmbridge> BrickOnKeyboard> (and by solved I mean they have demonstrated algorithms that solve this problem when scaled, not that the current implementation is good enough)
12:11 < hprmbridge> kanzure> i'm not really sure what you are trying to argue with self-replication; it almost looks like you're more trying to make a point about AI not self-replication
12:12 < hprmbridge> BrickOnKeyboard> I'm an engineer, I'm describing a way to use the tools we have today to achieve self replicating in the near future
12:12 < hprmbridge> BrickOnKeyboard> I don't care what you call the tools
12:13 < hprmbridge> kanzure> are you proposing that we use AI to bruteforce self-replication of industrial factories/fabrication facilities/etc?
12:13 < hprmbridge> BrickOnKeyboard> no
12:15 < hprmbridge> nmz787> What is your area of expertise?
12:16 < hprmbridge> BrickOnKeyboard> embedded ML systems
12:16 < hprmbridge> nmz787> RTL or firmware?
12:17 < hprmbridge> nmz787> Circuit vs programming i mean
12:17 < hprmbridge> BrickOnKeyboard> firmware and our middleware stack that goes to the application layer
12:17 < hprmbridge> BrickOnKeyboard> so it's all programming
12:18 < hprmbridge> nmz787> I started in validation, now in sort of fab process integration
12:18 < hprmbridge> nmz787> PDK development
12:18 < hprmbridge> nmz787> Also a programmer by nature
12:21 < hprmbridge> BrickOnKeyboard> anyways RL solvers have some goal they get reward to accomplish, so they propose actions or action sequences to achieve those goals.  A "task space" RL solver would be to generalize your goals into where any task a robot can do that is a simple step can be described as a task.  For example you could have a task space solely of "MOVE <source> <dest> <information about fragility>"
12:22 < hprmbridge> BrickOnKeyboard> so the *space* is every possible combination of the variables source, dest, fragility_info
12:22 < hprmbridge> BrickOnKeyboard> and it turns out it's possible to train a neural network on a finite number of permutations of that space, and reach a solution that statistically solves *all* the rest of it
12:22 < hprmbridge> BrickOnKeyboard> (they call this 'generality')
12:23 < hprmbridge> BrickOnKeyboard> in the real world a robot running an agent with this solver could thus move any object it's actuator can grasp that fits into the general classes of things it was taught on in the training examples
12:24 < hprmbridge> BrickOnKeyboard> and this actually works.  anyways self replication is just a larger set than that
12:27 < hprmbridge> kanzure> originally in 2008-2011 this channel mainly focused on the problem of self-replication; our concept was a sort of library of manufacturing processes that refer to other manufacturing processes, and once you have a suitable library you could do a graph search to find a closed cycle.
12:29 < hprmbridge> BrickOnKeyboard> right.  which is unironically the easy part
12:29 < hprmbridge> BrickOnKeyboard> the difficult part is actually completing a step.  most real world processes have a 'happy path' where everything works fine
12:29 < hprmbridge> BrickOnKeyboard> and a 'countably infinite' number of ways things can fail
12:30 < hprmbridge> BrickOnKeyboard> this is why human workers have to stand around, to catch all those countless fail cases
12:30 < hprmbridge> BrickOnKeyboard> like you said, if you wanted to be *one company* doing self repl, yeah, can't be done because you don't own the IP for many steps
12:31 < hprmbridge> BrickOnKeyboard> but you could generalize the above into a software service, where you make it affordable to automate tasks
12:31 < hprmbridge> BrickOnKeyboard> and then companies competing with each other rent robots using your stack (or the maybe 1 competitor, it's a very strong natural monopoly)
12:31 < hprmbridge> kanzure> well, i look forward to seeing it done in practice
12:32 < hprmbridge> BrickOnKeyboard> yes, I expect it to happen with the usual S curve adoption. but all the underlying tech has to exist, the real life tech tree has a node unlock order
12:32 < hprmbridge> BrickOnKeyboard> before these AI improvements it wasn't possible at all
12:33 < hprmbridge> kanzure> oh it wasn't?
12:33 < hprmbridge> BrickOnKeyboard> nope
12:33 < hprmbridge> BrickOnKeyboard> the reason it wasn't is because even if you posit infinite money, very complex software solutions reach an assumptope of reliability
12:34 < hprmbridge> BrickOnKeyboard> so imagine you had to handle a real world task like 'drive a truck' with if statements
12:34 < hprmbridge> BrickOnKeyboard> so you hire 1 million programmers and you hardcode the logic
12:34 < hprmbridge> BrickOnKeyboard> for every possible event that could happen
12:34 < hprmbridge> kanzure> sorry, are we talking about AI again, or are we talking about self-replication of industrial facilities
12:34 < hprmbridge> BrickOnKeyboard> they are the same problem
12:34 < hprmbridge> BrickOnKeyboard> one is impossible without the other
12:34 < hprmbridge> BrickOnKeyboard> like saying "are we talking about electricity or industry"
12:34 < hprmbridge> kanzure> (do you believe biological self-replication requires intelligence...?)
12:34 < hprmbridge> BrickOnKeyboard> no electricity = factories of very limited capability
12:35 < hprmbridge> nmz787> Which assumes perfect knowledge of the current environment/system, as well as perfect knowledge of how things may change in the future
12:35 < hprmbridge> BrickOnKeyboard> sure, that wasn't where I was going with this
12:35 < hprmbridge> BrickOnKeyboard> I was trying to say that if you tried to do it by hand, using say 2008 era techniques
12:36 < hprmbridge> nmz787> But like that clearly isn't the case
12:36 < hprmbridge> nmz787> (feasibly)
12:36 < hprmbridge> BrickOnKeyboard> what would happen is your solution would likely have too many hidden bugs from unexercised cases
12:36 < hprmbridge> nmz787> Right, that's all i meant
12:36 < hprmbridge> BrickOnKeyboard> and would probably crash the truck more often than would be feasible economically
12:36 < hprmbridge> BrickOnKeyboard> cheaper to pay a human than to pay for all the crashed trucks
12:36 < hprmbridge> kanzure> i think that AI can figure out self-replication, if given enough resources and funding to try out a bunch of different pathway combinations, just like a human team could if given enough funding and time- but i don't think this statement is interesting; we already knew it was possible to explore possibility space.
12:37 < hprmbridge> BrickOnKeyboard> I am saying we specifically have near future narrow AIs that exist now in demos
12:37 < hprmbridge> BrickOnKeyboard> that the AI isn't "figuring out" anything
12:37 < hprmbridge> BrickOnKeyboard> but it makes it feasible to automate simple tasks
12:37 < hprmbridge> BrickOnKeyboard> and construct large factories that use many of these
12:37 < hprmbridge> BrickOnKeyboard> and have true near full automation
12:38 < hprmbridge> BrickOnKeyboard> human engineers still define the tasks at least for a while
12:38 < hprmbridge> nmz787> Lol but humans already do that for a dollar a day
12:38 < hprmbridge> BrickOnKeyboard> but the machine is robust, if it drops an item or something else goes wrong
12:38 < hprmbridge> kanzure> a few years ago, in an off-hand comment, anselm levskaya said that making the modern electronics industry required something on the order of ~1 trillion dollars of investment; i don't know how accurate that is- and whether it includes all the dead ends humans had to waste time and resources exploring.
12:38 < hprmbridge> BrickOnKeyboard> it almost always finds a way to succeed
12:39 < hprmbridge> BrickOnKeyboard> so a large *set* of separate companies using separate machines to automate separate tasks, becomes self replication once all humans have been fired
12:39 < hprmbridge> BrickOnKeyboard> that's specifically what I mean
12:40 < hprmbridge> BrickOnKeyboard> maybe that doesn't count until one company then buys all the separate companies
12:42 < hprmbridge> kanzure> could human ingenuity be attributed as responsible for any actual speed-up compared to the time and expense of just trying lots of different processes (eg 100 variations of chemical vapor deposition etc)  to develop the modern electronics industry
12:42 < hprmbridge> kanzure> it would be really interesting if the answer was "no" or "nearly no"
12:47 < hprmbridge> BrickOnKeyboard> I would have to ask my friends who do process optimization, but that's not really the right question.  Given knowledge about previous experiment runs, guessing where the optima is in a possibility space is something AI already does better than humans.  https://www.deepmind.com/blog/accelerating-fusion-science-through-learned-plasma-control
12:47 < hprmbridge> BrickOnKeyboard> it's not just guessing but it's also not something where you can make progress without spending resources on all the dead ends in a possibility space
12:47 < hprmbridge> nmz787> I think the answer is yes
12:48 < hprmbridge> nmz787> But whether AI can also do that, seems like a yes too
12:48 < hprmbridge> kanzure> even hod lipson's eureqa symbolic regression software was faster than some students https://www.creativemachineslab.com/eureqa.html
12:53 < hprmbridge> nmz787> @kanzure ever grow your fingernails long, like 5-8mm? And see how it affected your typing speed?
12:53 < hprmbridge> nmz787> It's pretty significant for me
12:54 < hprmbridge> nmz787> No reason for me to try this experiment other than  I've been lazy to cut my nails
12:54 < hprmbridge> nmz787> It's a MUCH different tactile experience at several levels
12:57 < kanzure> nah, usually interferes too much long before that point
12:58 < hprmbridge> nmz787> It was surprising that my truck's touchscreen didn't work
12:58 < TMA> nmz787: I don't like it when the nails hit the keyboard. It is reason enough for me to cut the nails.
12:58 < hprmbridge> nmz787> I wonder if that isn't the case with like a luxury car
12:58 < kanzure> just put it on the calendar on repeat interval
12:59 < hprmbridge> nmz787> Long nails are useful for stuff like scratching out bee stingers, or small thorns, and prying plastic gadget housings open
12:59 < hprmbridge> nmz787> But yeah mostly I haven't been able to find the clippers at the opportune moment
12:59 < hprmbridge> BrickOnKeyboard> the touchscreen needs the capacitance of your body to detect your touching it.  The nail apparently doesn't apparently expose enough surface area to have sufficient capacitance.  There are gloves that fake this property
12:59 < hprmbridge> nmz787> Not all screens are cap sense
13:00 < hprmbridge> nmz787> I believe I watched a volvo commercial recently that specifically mentioned it was resistive
13:00 < hprmbridge> BrickOnKeyboard> no, but it's the best of the available methods so a luxury car probably uses it
13:00 < hprmbridge> nmz787> Or maybe it was volstar, the electric company
13:01 < hprmbridge> BrickOnKeyboard> mine super annoying hides the seat warmer behind a menu, and you can't reach that screen without both agreeing to their legal agreement and also waiting for their bloated HMI cluster to boot
13:01 < hprmbridge> BrickOnKeyboard> wish I bought a tesla
13:02 < kanzure> i wonder if there are any after market touchscreen to physical knob conversion kits, those would probably be popular these days
13:03 < hprmbridge> nmz787> Yeah apparently volvo, at least in 2017, was using resistive screens due to 'gloves' and being designed in/for cold climate
13:03 < hprmbridge> nmz787> https://www.theverge.com/2017/8/17/16141366/screendrive-volvo-v90-cross-country-touchscreen-sensus
13:03 < hprmbridge> nmz787> You have volvo?
13:06 < hprmbridge> kanzure> https://enhauto.com/ aftermarket buttons for tesla cars haha
13:07 < hprmbridge> BrickOnKeyboard> hyundai ioniq 5
13:07 < hprmbridge> BrickOnKeyboard> seat warmer makes a huge difference to short trip energy efficiency but inconvenient to enable
13:20 < hprmbridge> BrickOnKeyboard> it's not resistive, it's a *hybrid* system that uses IR as well.  plain resistive is awful and barely usable.  cap touch was one of the 'tech nodes' that make smartphones possible
13:27 -!- Llamamoe [~Llamamoe@46.204.76.114.nat.umts.dynamic.t-mobile.pl] has quit [Quit: Leaving.]
13:36 < hprmbridge> nmz787> Afaik that's just a problem for swiping, not pressing
13:36 < hprmbridge> nmz787> Plenty of super cheap electronics used to use resistive matrices
13:44 < fenn> resistive has very poor precision, you end up mis-clicking a lot
13:45 < fenn> also you have to push harder which makes single handed use very difficult
13:45 < fenn> my nook uses IR and it works pretty well; a lot better than a resistive screen
13:47 < fenn> the bezel sticks out a few mm from the screen to fit the IR LEDs and sensors in
13:48 < fenn> cars just shouldn't use touchscreens
13:52 < hprmbridge> BrickOnKeyboard> Maybe.  Depends how long true self driving takes to both become good enough and adopted
13:53 < fenn> well that's not really a car is it
13:53 < hprmbridge> BrickOnKeyboard> But yeah for human driving ideally every control is mechanical
13:53 < hprmbridge> BrickOnKeyboard> That you normally need
13:53 < hprmbridge> BrickOnKeyboard> True for mine EXCEPT the heated seat
13:54 < hprmbridge> BrickOnKeyboard> Everything else is either a switch or the HVAC doesn't normally need adjustment between trips
13:54 < fenn> seems easy enough to hotwire
13:54 < hprmbridge> BrickOnKeyboard> It uses software and relays so kinda
13:55 < hprmbridge> BrickOnKeyboard> Hotwiring it to always be on is possible
13:55 < fenn> uh, don't do that
13:55 < fenn> i just meant a rocker switch to turn it on/off
13:57 < hprmbridge> BrickOnKeyboard> You would bridge the relay for it so it always has power then install a switch that splits the power wire to it drilled into the side of the seat plastic by the other switches
13:57 < hprmbridge> BrickOnKeyboard> Lotta effort
13:58 < fenn> meh
13:59 < fenn> i find it hard to get excited about new products anymore
14:00 < fenn> more shit that doesn't belong to me even though i paid for it
14:35 < hprmbridge> BrickOnKeyboard> there are economic forces and practical forces that may lead to it all being like that
14:35 < hprmbridge> BrickOnKeyboard> it's not necessarily a bad thing.  if all your stuff was rented *but* you owned lots of shares of index funds
14:35 < hprmbridge> BrickOnKeyboard> you're not actually any poorer and there are advantages
14:39 < L29Ah> 22:20:47]<hprmbridge> BrickOnKeyboard> it's not resistive, it's a *hybrid* system that uses IR as well.  plain resistive is awful and barely usable.  cap touch was one of the 'tech nodes' that make smartphones possible
14:39 < L29Ah> i still love my resistive Nokia N900, it can be used with fingernails and pointy stilo, so i can draw precise stuff unlike the modern capacitive touchscreen-based smartphones
14:40 < fenn> "you will own nothing and be happy" - no thanks
14:40 < juri_> owning your own gear is a full time job.
14:40 < L29Ah> i had no resistive touchscreens that were worse than the capacitive ones wrt precision tbh
14:42 < L29Ah> somehow the only capacitive touch thing i've experienced that doesn't insist to be poked with a sausage (that is incompatible with precision) is not a touchscreen
14:42 < fenn> i just want to boot up an old program and read the old file i created, and it just works, because nothing has changed
14:43 < fenn> and so on, for the various domains of praxis
14:44 < fenn> instead it's a never ending desperate race to stay "up to date" with an endless stream of incompatible changes that provide no real useful additional functionality
14:45 < juri_> this will get much worse when industry starts producing the things we want.
14:46 < fenn> i WANT a standardized tupperware lid that fits all other standardized tupperware containers in its class
14:46 < juri_> hense, we really need to develop our own tools and protocols.
14:56 < hprmbridge> docl> I mostly agree with @BrickOnKeyboard. Some nuance though. It should be possible to have "simpler" systems by pushing the complexity into blackboxes, and/or discovery of actually-simpler processes. Like say you have (near term / relatively low difficulty) nanotech or biotech based monomolecular membranes for materials separation by ion channels (which seems to be possible with spiroligomers). That
14:56 < hprmbridge> docl> eliminates a lot of high temperature processes from the necessary stack. Novel catalysts (again, near term for nano/bio) could wipe away swaths of difficulty for chemical synthesis as well. Maybe you use electrochemical methods + 3D printing, like Project Quine is trying to do, for all the parts and motors and so on. Anyway I see no reason to think the full range of current industrial ecology
14:56 < hprmbridge> docl> (which evolved with human-labor-hours as an abundant resource), or even that substantial of a fraction, has to be represented in the self rep system we actually end up using, any more than we needed the big belts and central motors of the steam era to harness electricity. Consequently, I could see it happening with something more like a pile of if statements instead of AI, once the simplifications
14:56 < hprmbridge> docl> are discovered. Main reason it wouldn't is AI being more expedient, but I think on more on the design/discovery side rather than the operations side (i.e. AI engineers, not AI workers). The idea of just going ahead and automating away every step in the current industrial ecology via AI in place of the humans is useful as a proof of concept narrative requiring zero creativity to imagine, but I
14:56 < hprmbridge> docl> think the reality will be a meeting in the middle / selection of easier to automate work chains.
15:03 < fenn> .g project quine
15:03 < hplusbot> fenn: https://www.quine.no/ -- Quine Home: ""The QuineCore extension for Adobe Team Projects has simplified shared editing, review processes and post-integration significantly and brings new ..."
15:03 < fenn> hm probably not that
15:04 < hprmbridge> docl> https://projectquine.substack.com/
15:04 < fenn> isn't this just reprap
15:05 < hprmbridge> docl> yeah plus electroplating
15:05 < kanzure> he seems to talk about reprap https://twitter.com/alexandrosM/status/1366105227607252993
15:05 < hplusbot> kanzure: error: unauthorized
15:05 < kanzure> .tw https://twitter.com/alexandrosM/status/1366105227607252993
15:05 < hplusbot> kanzure: error: unauthorized
15:05 < hplusbot> kanzure: error: unauthorized
15:05 < kanzure> hmph
15:06 < kanzure> "To my knowledge, the most 3d printed 3d printer is the Mulbot. Please let me know if there's a 3d printer goes farther." https://reprap.org/wiki/Mulbot
15:07 < kanzure> their facebook group has 1500 members? https://www.facebook.com/groups/275718849985834
15:08 < fenn> a facebook group -_-
15:08 < fenn> we're doomed
15:08 < hprmbridge> docl> Another way we get there is making energy so cheap you can use low complexity high energy methods like plasma separation http://www.molecularassembler.com/KSRM/3.14.htm
15:14 < fenn> is there some lower limit to FDM print resolution that means circuit boards will have to be big clanking things?
15:15 < L29Ah> fenn: what
15:17 < L29Ah> seems like Mulbot has a smaller operating space than the printer required to print its parts
15:18 < fenn> i just feel that a fully self printing system will remain an academic pursuit as long as it's much larger and more unreliable than an equivalent mass produced machine
15:18 < fenn> extrapolating the demo circuits they're printing, the required circuitry would fill a room
15:19 < kanzure> what was the best self-replicator achieved by the reprap people? in terms of input parts and components i mean.
15:19 < kanzure> it's been quite a while since i have looked, surely they have a review article by now
15:19 < fenn> the reprap people were never very serious about publishing results
15:20 < fenn> it was a lot of individuals working for a period of time and then falling off the radar because life intervened
15:20 < L29Ah> kanzure: i don't recall any integrative project
15:20 < fenn> and a small mechanical engineering group that didn't provide any direction to the rest
15:21 < L29Ah> i.e. there were successful efforts in designing printable circuits and motors, but i didn't see any printer design incorporating these
15:22 < fenn> metallicarap (electron beam melting printer) seemed the most promising to me, but it didn't get very far into development
15:22 < fenn> metalicarap*
15:26 < fenn> more expensive bootstrap designs, materials, vitamins, supporting infrastructure, and generally required more expertise in what i call "technar" - high vacuum and high voltage electronics
15:26 < fenn> but it should have been able to replicate a larger fraction of its components
15:26 < fenn> as well as being generally more useful
15:27 < fenn> but $5-20k per bootstrap kit is out of reach for most hobbyists
15:27 < L29Ah> i'd work on metalicarap, but i don't have enough faith in the design idea to fund it
15:29 < L29Ah> could be under $1k i think after streamlining and mass production, like FDM printers now cost under $200
15:29 < fenn> sure, but that hasn't happened yet
15:40 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has quit [Ping timeout: 255 seconds]
17:04 -!- cthlolo [~lorogue@77.33.23.154.dhcp.fibianet.dk] has quit [Read error: Connection reset by peer]
17:27 < hprmbridge> BrickOnKeyboard> Things like improved 3d printing methods - if they aren't easy with a huge upfront advantage (like resin printers) - may be many generations of prototypes from being better than the alternative.  Note that with the way I described AI driven robotics working, the alternative are just in time factories located nearby in shipping terms that mill their parts or use molded plastic, etc.  These are
17:27 < hprmbridge> BrickOnKeyboard> superior industrial methods in that they trade off flexibility for tooling costs and a stronger and cheaper product.
17:27 < hprmbridge> BrickOnKeyboard>
17:27 < hprmbridge> BrickOnKeyboard> Another interesting factor that I think, due to the exponential ramp of AI and the fact that we are approaching (if hasn't begun) Singularity Criticality, there are whole *classes* of valid technologies that will likely never be developed.  Like spiroligomers.  The reason is that they will take too many decades to happen pre-Singularity, and once we have AI superintelligence, they are not *good
17:27 < hprmbridge> BrickOnKeyboard> enough*.  Vacuum phase nanotechnology is probably much better, and so it's a waste of time to go for anything but the best when you have a superintelligence to blow through any technical problem in your way.
17:28 < hprmbridge> BrickOnKeyboard> past examples of this : rechargeable alkaline electric cars, or gasoline burning steam cars.  It's the same idea - they were *valid* technologies that worked, and when developed they were even a slight improvement over what they had, but around the same time period there were superior alternatives that sucked away all the effort and money
17:29 < hprmbridge> BrickOnKeyboard> oh, high speed propeller aircraft.  https://en.wikipedia.org/wiki/Republic_XF-84H_Thunderscreech
17:29 < L29Ah> resin printers are a tradeoff compared to FDM as well
17:30 < L29Ah> singularitarianism = doomerism
17:30 < hprmbridge> BrickOnKeyboard> I consider it to be a valid statement of reality, I do not know a *realistic* way it won't happen
17:30 < hprmbridge> BrickOnKeyboard> in the same way it might be 'doomer' to say if you put too much fissionable material together you will get a flash
17:31 < hprmbridge> BrickOnKeyboard> but it will happen...
17:31 < hprmbridge> BrickOnKeyboard> the singularity is even convergent
17:31 < hprmbridge> BrickOnKeyboard> there's many possible forms of AI, the Transformer for instance is just one of a nearly infinite number of versions of a neural network or other ML techniques
17:32 < hprmbridge> BrickOnKeyboard> but see, once you get some form of AI that works well enough, you can use it to discover better ones, and you *converge* on superintelligence
17:32 < L29Ah> well, i've not yet seen a consistent definition of a technological singularity that doesn't state that it already happened
17:32 < hprmbridge> BrickOnKeyboard> well you must not have read anything by the authors of the term
17:33 < hprmbridge> BrickOnKeyboard> upgradable intelligent agent will eventually enter a "runaway reaction" of self-improvement cycles, each new and more intelligent generation appearing more and more rapidly, causing an "explosion" in intelligence and resulting in a powerful superintelligence that qualitatively far surpasses all human intelligence.[4]
17:33 < L29Ah> ok, how do you know if it already happened or not?
17:33 < hprmbridge> BrickOnKeyboard> so this is very simple, obvious, hasn't happened yet, and we can easily determine when it has
17:33 < L29Ah> no it is not obvious
17:33 < L29Ah> no i can not easily determine when it had
17:34 < L29Ah> please provide the algorithm to do so
17:34 < hprmbridge> BrickOnKeyboard> I'm not going to entertain spurious false claims
17:34 < hprmbridge> BrickOnKeyboard> you're not engaging with the statement
17:34 < hprmbridge> BrickOnKeyboard> for an AI to have human intelligence, it needs to score consistently as good as humans on a test of intelligence
17:35 < hprmbridge> BrickOnKeyboard> benchmark right *now*: https://twitter.com/sergeyi49013776/status/1598430479878856737?lang=en
17:35 < hplusbot> hprmbridge: error: unauthorized
17:35 < L29Ah> any decent intelligence won't engage in entertaining you with test-passing
17:35 < hprmbridge> BrickOnKeyboard> that's not correct or consistent with current research
17:35 < L29Ah> esp if it has a self-preservation value
17:35 < hprmbridge> BrickOnKeyboard> that's not required
17:36 < hprmbridge> BrickOnKeyboard> like L29Ah, do you follow current AI research at all?
17:36 < L29Ah> the current research doesn't demonstrate anything remote to AGI
17:36 < hprmbridge> BrickOnKeyboard> it doesn't sound like you have any knowledge of it
17:36 < hprmbridge> BrickOnKeyboard> incorrect
17:36 < hprmbridge> BrickOnKeyboard> it's actually very, very, very close to it
17:36 < hprmbridge> BrickOnKeyboard> and there is an obvious way to bridge the last steps
17:37 < hprmbridge> BrickOnKeyboard> note that AGI is defined *functionally*
17:37 < L29Ah> i chat with a professional neural network researcher about related stuff a few times a week and trained a couple myself, does that count?
17:37 < hprmbridge> BrickOnKeyboard> no
17:37 < hprmbridge> BrickOnKeyboard> also what you are saying is not consistent with current results
17:37 < L29Ah> how do you tell "very close"?
17:38  * L29Ah feels dangerously close to the no-philosophy rule
17:38 < hprmbridge> BrickOnKeyboard> current systems hit around 83 on an IQ test, and beat humans consistently on many tasks.  A GPT based system with vision and memory modules would likely hit human level intelligence on most tests within 1-2 years.
17:39 < hprmbridge> BrickOnKeyboard> there already are prototypes, transformers do work with vision
17:39 < hprmbridge> BrickOnKeyboard> the metaculus prediction of weak AGI of 2027, of 4 years, is probably a conservative estimate
17:41 < hprmbridge> BrickOnKeyboard> the singularity criticality is amplification -as these AI tools get capable enough, they accelerate progress towards more advanced and capable AI tools.  there is an obvious way to get exponential gain from this
17:41 < hprmbridge> BrickOnKeyboard> I'm sure if you work with a neural network researcher, and this person is up to date on remotely the latest, they will know how to do it
17:42 < hprmbridge> BrickOnKeyboard> it's very, very obvious and already in use
17:43 < hprmbridge> BrickOnKeyboard> in essence, any AI tool we build right now will of course be limited in some areas and flawed.  but past a certain level of capability, it allows you to automatically explore the possibility space of network and cognitive architectures with plausible guesses
17:44 < hprmbridge> BrickOnKeyboard> and find the combinations of architectures that will have strong superintelligence.  you measure this by *performance* - you have a cognitive benchmark of many separate tasks, and you are seeking an architecture that learns well and performs with superhuman ability across as broad a set of tasks as possible
17:45 < hprmbridge> BrickOnKeyboard> it's also self amplifying - if you can find an architecture a little bit better that your current best, the new one is chosen because it's measurably better at general cognitive tasks, one of which is designing cognitive architecutres...
17:45 < hprmbridge> BrickOnKeyboard> it's going to result in you very rapidly hitting the limits of what your compute allows, and you then obviously need a lot of that, and you see where the next phase is
17:46 < hprmbridge> BrickOnKeyboard> that's the Singularity.  for it *not* to be about to happen...I'm not sure what would have to be true, other than an imminent nuclear war
17:51 < hprmbridge> kanzure> what are your favorite manuscripts for transformer, large language or other approaches at the moment
17:53 < hprmbridge> BrickOnKeyboard> https://github.com/EleutherAI/gpt-neox I have just been looking directly at the source for a SOTA model
17:53 < hprmbridge> BrickOnKeyboard> the ironic thing is it's very short.  the SOTA takes almost no actual code
17:55 < hprmbridge> BrickOnKeyboard> https://github.com/EleutherAI/gpt-neox/blob/main/megatron/model/transformer.py
17:55 < hprmbridge> BrickOnKeyboard> https://github.com/EleutherAI/gpt-neox/blob/main/configs/20B.yml
17:55 < hprmbridge> BrickOnKeyboard> here's the config file for the 20B model
17:56 < hprmbridge> BrickOnKeyboard> the one it's based on from OpenAI - this is what powers chatGPT - is 175B
17:56 < hprmbridge> BrickOnKeyboard> the main difference is this single file
17:56 < hprmbridge> BrickOnKeyboard> between "dunce" and "subhuman savaant"
17:56 < hprmbridge> BrickOnKeyboard> (that and ~$500,000 of compute)
18:59 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has joined #hplusroadmap
20:04 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0::4249] has quit [Quit: Leaving]
20:16 -!- TMM_ [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
20:17 -!- TMM_ [hp@amanda.tmm.cx] has joined #hplusroadmap
23:07 -!- Roy_Mustang [A_D@libera/staff/dragon] has joined #hplusroadmap
23:08 -!- A_Dragon [A_D@libera/staff/dragon] has quit [Killed (silver.libera.chat (Nickname regained by services))]
23:08 -!- Roy_Mustang is now known as A_Dragon
23:08 -!- balrog_ [balrog@user/balrog] has joined #hplusroadmap
23:10 -!- soundand1ury [~soundandf@user/soundandfury] has joined #hplusroadmap
23:13 -!- nsh- [~lol@user/nsh] has joined #hplusroadmap
23:14 -!- Netsplit *.net <-> *.split quits: jrayhawk, nsh, Malvolio, soundandfury, balrog
23:15 -!- Netsplit over, joins: jrayhawk
--- Log closed Sun Jan 22 00:00:21 2023