 Hey folks, it's T Tuesday 31 19 The here's the goals for this time here are the results. It's win some lose some I've got a lot of stuff to show you a lot of stuff to talk about so it's gonna take a while So let's just dive right into it. I was cooking these little videos and still right up until this this morning So I apologize in advance. They're a little bit long and a little bit rough, but let's get into it The bigger more immediate issue is it's time to move on that So I ended up going with this AC infinity stuff because it seemed the most plug-and-play and It was supposed to get mounted in the cabinet the bookcase doesn't have a cabinet front So I started making little bits to on the 3d printer We get power temperature sensor switch over to See the temperatures Tidal is not supposed to be there Temperatures are awful The speeds are awful See if anything happens here So if I'm pulling air out the back then I need to close off the holes in the front so that air gets forced through Actually see this happen So I went and got a second Power supply, this is a very aggravating thing. This is a USB format power supply, I believe Yeah, USB a connector on it but output six and a half volts great I know why they do it So So they call it the s7 it's the triple and This box is beautiful Whereas the original box got had a ding here and the plate was was bent This corner of the plate was bent back and USB with a daisy chain There's also a speed controller on this one in case you want to use it separate but my plan is to plug this into The thermostat that we've got which has two outlets and we should oh And we have another one of these power supplies. This is actually It's gonna be extra so I didn't really need to buy an extra one because we're Alright, we'll have an extra one, but assuming we managed to Build, you know Lotus four five six seven And we want to continue this Cooling approach which maybe we do maybe we don't I mean I think It's the early returns are it's it's plausible No, theoretically you're supposed to use this template to put it on your cabinet or your door instead I've got these 3d printed things that I whipped up That are using it on the other one of these seems to work. Okay, the idea is It goes up in behind here and the holes Line up and that supports it so we get the height that we want to get so That's what we've got Get us up above Lotus one where we're going to use some foam core just to kind of lay it over Of the top of this thing Right side grid fan for Lotus one two Let's go see if we can hook it up to the Thermostat the theory is able to start this up Six fans running in 97.6 This one is now 800. It's already gone up So That's encouraging this one is now a thousand and this one is now the thousand All right everybody two little fans So maybe we'll just save that up for now I want to get some more experience with these I have to anchor them somehow He just piece of foam core running all the way along the top From edge to edge and just you know tape it on or something like that So that they can't fall outwards or inwards for that matter but Okay, well Most the heavy lifting was was getting the house cool and going for the summer of the swamp cooler, but Now we've got what passes for spot cooling here as well step by step I made this plate to connect pieces of foam core to the end because I got a bunch of phone Something by 20 by 30 and I need something closer to 50 So I'm going to take a couple of pieces a couple of strips of this and try to connect the end to end using this Clip technology that I made on a 3d printer. I got no idea if it's really going to work This is the test According to YouTube the key to cutting foam core with an exacto is to have the knife the blade way out there and Come at it at a low angle So we'll give it a shot Okay Well, I Mean could be worse quite a bit worse and yet for my purposes Usable could be better. Could be worse That is it first test of foam core Cutting not so good drilling Come on for sure a bit But fall very possible It's working not working But it is working as well as it's actually sucking it in pretty good right there so, okay Sorry about that The the smoke test at the end It was hard to figure out a way to actually video to get the smoke in there But it is the case that if you get close enough to the actual tiles a lot of times near the serial port You can see it actually is getting sucked in so it's something I mean, there's lots of leaks all around that could all you know This was just a first demo and you know it works. Okay, and I just wanted to add one thing. Well, so so number one Really the bottom line was that you know We had to get the the cooling going for the summer for the house And that made the main difference as far as cooling off the grid which made the difference in allowing the CPU speeds to Rebound but the fans. I think they do help and we now have a pretty good regulation and This is the live temperature graph Using the USB temperature sensor that I got a couple of weeks ago This detach bit over here is from before the last update This low part down here is when the grid was turned off And this is daily cycling in temperatures This is when the grid got turned on to start simulations that I'll show you in a little bit And then this was temperature cycling and then you know here So right there Is where the the swamp cooler we got the swamp cooler started up, so it's all kicking in And Now that the regulation is actually fairly good I think part of these things is due to the thermostat on the fans turning on and off I increased the set point on the thermostat by one degree Just as a test Right at this spot here Right here and it did for a while seem like it was actually reacting made a difference hard to tell for sure, but bottom line all now 57 tiles are running at 1 gigahertz, which is as good as they can do so What I wanted to talk about in the the bigger picture of this is the forgivens of indefinite scalability and you know This is this goes back to the original paper talking about indefinite scalability The the whole idea of doing an indefinite scalable computer architecture Well, the question is you know, what do you get to work with? What what are you allowed to assume you've got and what you're allowed to assume is the forgivens of indefinite scalability power cooling real estate and money to buy tiles and You know that kind of goes without saying but it's actually important and as you know Indefinite scalability is slowly getting traction out in the world I hear it echoing a little bit more from places from time to time But as it gets from you know being ignored to being laughed at men fall The forgivens become more important because it actually makes a difference So the fact that I'm worrying about heat here is officially speaking not the fault of the indefinite scalable architecture, you know On the one hand saying, you know, you can assume power and cooling you can assume the power and cool you need that sounds like you know an invitation to climate destruction, you know And you know to some degree I will take that hit that You know, we would like to have our indefinitely scalable machinery be run very cool and be Power-efficient and so on and so forth But in order to get enough room to do research We say let we're gonna put that off the table as a requirement and instead just turn it into a Desirable feature a quality rather than a requirement and so we'll compare data sheets, you know How much power and cooling does this particular approach to indefinite scale we take versus that one same thing for how big they are How much real estate they need and how much they cost? So okay, so that's just a reminder that We a definite scalability is built on assuming those four givens and those are an IOU But they all have to be paid off in order to get a completely embodied system as we just saw in dealing with heat on the t2 grid, okay So that's the development story. The research story has two angles There's the once again the new engine that are supposed to be more robust and flexible and so forth again I did not actually work on it I'm gonna put it off the table for now and and that means this We're gonna continue to live with the existing code base which has these explicit known robustness failures in particular Once a tile is taking part in a computation and it has any Interactions other than just all empty spots all empty sites with its intertile connectors You cannot shut that tile down without causing failures on the neighboring tiles and you get these crash storms and so forth That happened that is all gonna have to be revisited but at the moment I am more attracted more drawn to wanting to explore higher up And that's the other direction that I did go in this last two weeks. It was also about having a little bit of fun Ringo off the plate the Ringo the ring oscillator that talked about a lot that was sort of you know kind of key innovation of Last year as far as I was concerned So let's talk about that a little bit more so Ringo Ringo dot oolam is a particular set of code to do this ring oscillator thing the general idea is GDRO GDRO the generalized distributed ring oscillator and the question is or a question is is how? General is the generalized distributed ring oscillator, you know and the stuff that we did Last year was all pretty much built on top of the plates the L1 plates and the L2 plates And it did show a certain amount of generality in the sense that the same thing that did the L1 plate I don't know exactly how that actually got to be half green and half blue But you know it's all rigid within a plate and if this generalized idea really is more general Can we take it off the plate? And that's what I went after this week this time and you know fundamentally the way that the GDRO looks at the universe it doesn't need to de-plate all it needs is a notion of upstream and downstream and Derivative from that that if there's no upstream then you're the root and if there's no downstream then you're the tail and That's it and the rest of the rules are built into the Ringo the generalized distributed ring oscillator approach So if we can come up with a notion of upstream and downstream then we could try to play this game So I said why can't we do something that instead of being a 2d grid with the root in the upper left and the tail in The lower right, what if we have sort of a amorphous cloud where there's a root in the center? And the tail is the entire periphery the memo, you know a Molecular cloud a membrane of sorts and have that Do Ringo on that? So that's what this is about got two demos to look at the first one Is I was just building up the infrastructure to do this the first one doesn't use Ringo the second one V11 does Let's take a look at it here. Which one is it this one? Yes, we have to start explicitly, okay, so One that will get one two I'm multiple I'm putting in multiple seeds there And each one is just this little thing. Okay, and the fourth one triggers an interaction and we get this No, and you know when I saw this I said wow, this looks really cool. Let's put it on the grid and see what happens This is full of bugs. I don't even know what's going on here I eventually figured out where the that big cell membrane the big orange rim came from But essentially what's going on here is those original yellow things they declared themselves to be a root with their Their hop count is zero that means they're the root and they spawned more gas cloud molecules near them That looked around and said oh, I see a zero that means I'm a one and then they spawned more and they say oh I see a one that means I'm a two and so forth until they got up to 14 Which is the limit of how many hops we said this experiment you were allowed to go And if you were the 15th one then you were supposed to decay because that's the special trigger Now here we switched to all right, so let's hold this up for a second and So that was pretty cool And the point was was that these the gas cloud molecules could move around and when they would wake up or the neighbors would wake up They would once again They would just search their environment and take whatever the minimum hop count they saw and make themselves Hop count one more than that So I tried to say okay Well, why don't we start with that gas cloud and now say do wringo that if you're the root if you're if you're hop count zero you're the root and if you're hop count 14 if you have no Downstream if you have no things that are further away than you if you look around and you're the biggest hop count You can see that means you're the tail even though that means you now the entire periphery of the cloud is tail And I tried to get the thing to sink as well as bop around like that stuff was doing couldn't get it to go So I said, okay, let's try to do something more simple and that's what this one is So what we have here is We put out a root that spawns a cloud the same way growing out the same distance but then it starts to cycle and Now you see they're not actually moving and that's the difference These now just grow like plants and then stay rooted But it when you're a periphery when you're one of the tails you have you throw a low odd a rare die And if it comes up heads you say oh, I'm gonna spawn another seed which rattles around tries to find an empty space And then creates another root So what we're seeing here is a whole bunch of roots that are cycling together And I didn't even know that this was gonna work So that they're not completely in sync and there look it locked up because there's some configuration That it got into a deadlock, you know, I don't know whatever it was But I had a background watchdog timer that the root says I want to be a jerk Remember the root is the one that keeps making things different the tails all say I'm gonna follow you so the root has an additional timer saying if I've gone this long and Failed to be a jerk Then I erased myself and so the all the hop count ones look around and they can't see a hop count zero anymore So they increase only they basically when the root disappears in this particular approach The whole cloud just goes with it. So what happens is is now once again, this thing is cycling It's coordinated, but it's not slave lock. They look at that. I really love it You know, it's like you got sort of like different organs or different pieces of a larger thing that have their own local centers Their own local business, but in addition they Coordinate with the things that they touch. We've increased the the time-lapse speed now. So you see it going quicker And this is the way it goes the as long as there's as long as the tails are cycling They may toss off new seed seeds And so here we have a detached thing where we've got two bits that are cycling independently But eventually they throw a seed in between in which then get some coordinated Maybe that caused them to lock up and so forth. This is a fascinating dynamics This is multi-scale structure just you know almost for free. It's almost just what Ringo wants to do with the root being the center and the tail the hop maximum hop count Whatever we're doing for size being the The tail and cycling back and forth with that. So I have some additional ideas about how we might be able to You know all the things that I did so far trying to get this cycling going while also having them move around Like what's happening in gas clouds be one that it didn't cycle that they would just lock up But I have some additional stuff that I haven't had a chance to implement No, and even as it stands, you know, there's actually that part in the lower left corner there that It has been stuck for a long time and I'm not entirely sure what's going on there. It's still running It's still stuck in fact as far as we know So I'm hoping I suspect what's happening is that the root that's locking it up and managed to get the root into You know a triple corner Which requires three locks and it's the slowest to get events So it just hasn't timed out yet, but I'm not actually sure Okay, this goes on for a while, but let's say we've gotten the idea here And so uh, that is the, uh, generalized Ringo and I'm pretty happy about it. I think You know breaking out of the idea that there has to be one tail or even that there has to be one root Uh, but when there is more than one tail or more than one root then the spatial interactions that follow that fall out of this thing Get resolved in a reasonably natural way. Uh, so okay, so that that was my uh, both Some research stuff and some fun stuff for uh, this time. Okay, so that's the research story Outreach I have two little bits to talk about like I said, this is a long one. I'm sorry Uh, uh, so one of the goals for this time was to increase the rejection count on my little science fiction story Short story search quiet wake sqw Uh, I failed to get an additional rejection because I didn't turn it in yet Uh, but I did actually make some progress. I didn't engage with the text and Started to make a few changes that I had accumulated from talking to people and thought about and so forth So I'm going to repost that as a no fool in uh for next time But in addition right after the last t tuesday update, uh, I went to one I did one of these virtual, uh, santa fe institute workshops that there's been several of them over the last, uh, year uh on embodiment and Uh embodied situated and grounded intelligence implications for ai and I missed the first day april 12th because that was the last t tuesday But I attended virtually for the rest of the week and you know, it was it was really great It was a lot of fun, uh, you know a ton of smart people and a very diverse group So there, you know, there were philosophers neuroscientists social scientists As well as computer scientists and biologists and you know, you name it And you know and the fellow in the middle there, uh, john crack hour He is a neuroscientist and and you know, and he had a lot of thoughts Uh, uh about uh approaches to intelligence and how to understand the brain and the mind and cognition and general and so forth And you know, and he and I kind of got into it a little bit Over the last several days because because you know, he he actually wasn't all that happy with a lot of approaches That people had taken uh including uh, lots of computational approaches like that So so that was a lot of fun and and I really enjoyed interacting with him at one point In the chat that goes with the zoom. He made this this comment Talking about something I've been previously discussed that is the classic move Squeeze interesting cognition until it is crushed by embodiment from one side and culture on the other This is explaining away from these two sides will not get the job done in my view like that And you know, when I looked at that it was like, you know, you know, except for the negative About it that seems really right Um, and so, you know, I made up a a slide or two and when it was, you know, one of the discussion periods You know, I raised my hand and and I I made the pitch that you know, well, you know, what else could it be? Maybe the way we should understand the job of what interesting cognition is is by implementing it by combining Embodiment from one side and culture on the other where the point is is that Embodiment writ large in general computer architecture means the implementation and embodiment is an Implementation which doesn't just mean the registers and the instruction set I mean, it also means the energy and the heat and the deployment and the cost and the entire Uh, uh physical aspect of the machine that is the implementation That is the embodiment that has tremendous consequences And on the other hand, we are not just implementing arbitrary machines We are implementing a programmable machine specifically And that means that we can be reprogrammed by culture We can be reprogrammed by code that we receive while the machine is running after it's been built That's the essence of programmable computing Embodiment is the implementation of the machine culture is the code that we transmit that's running on the machine And it sounds just right. It sounds like a very good way to understand the big picture Where does something in the mind come from it comes from either the implementation as a programmable machine Or it comes from the code that's running on the programmable machine Crush interesting cognition So that's it All right, so this has gone pretty long well half hour. Okay, so be it Next t tuesday will be on may 10th Uh, my goal is you know, I've had this this uh paper intro that I've been supposed to write I'm way late on it. But that now goes to the top I got a ship that overdue paper intro before we talk again and I need to also submit Uh, science fiction short story search quiet wake going to submit it to clark's world And then the thing the fun right what I actually want to work on is so can we put uh gas cloud v1 v10 and v11 together can we get it to do It's distributed cycling while also moving around a little bit jiggling around and going for it We shall see But that's the news for t tuesday 31 19. Thank you so much for stopping by. I hope to see you next time