 If folks welcome to the artificial life advanced class, it's t-minus four in the artificial life creation countdown Here is the master spreadsheet. We are way behind But I haven't given up There's a slack built into the system the stretch goal. We can dump that. There's the entire month in November That's not booked We'll see how it goes. So the goals for this time Diamond sequencer, that's the sequential machine that we're trying to make that'll control the old Reducted process supposed to have programs running not even close Spike assembler and loader the prototype of something that's going to actually take the program code That's gonna exist and pull it in and get it just deployed onto a diamond Made some progress on that take a look in the simulator so just gonna try to set this up get it going and Color commentary. All right. So first it's gonna spread out there It is we have our diamond sequencer. This Is our loader? This is The new thing here. It's one of the aspects to the loader that we have to deal with is That as it injects instructions, it has to be careful not to use up the last empty space Let's see this first bit get as far as All right, so there we are okay boom and now we're now we're cooking So The loader is now down here and that's our primary aggravation bug at the moment is that The loader doesn't stay in place and it's the worst excuse deterministic serial computation that what it is likely to encounter today, but we have deployed a Marker that has its special label that is in honor of me For having gotten this far and we've also deployed a song just to take up some space the program that we're doing here Have we got the program is this? So there is our marker and it gets sucked into the label here as part of the first instruction Here's a song another song and then finally we're gonna put in a reset instruction So so far we have a marker and one song all right now we got a shape shifter it needs to find Available space That was just about the first time that I got this work in this far And you can see this this white one here and this black one here. Those those are loop killers They're actually part of the execution of the reset instruction So the things started to happen. I'll come back to that a little later All right The bugs that that thing is full of Are not just because I'm old and forgetful and not as good a programmers I used to be they also reflect a fundamental Tension between trying to keep total control of everything and trying to just be all completely loosey-goosey and You know on the one hand we normally think of control versus chaos as being like a binary distinction So traditional computing to trim this computing is Top-down rigid control nobody can do anything except when they're told to win then they must do it Authoritarian social Societies try to do the same thing and it never works And of course the problem is is that top-down control and like deterministic computers it works for small systems So people think they build a small system and say hey it works But the whole point is top-down control doesn't scale up on the other hand people think that if you don't keep everybody Completely in line that if you let them have any freedom at all, you'll have chaos. You'll have anarchy turbulence But what we're looking for is the space in between where we have resilient systems that consist of Robust agents that can survive on their own and then bottom up they can make Cooperatives to do things bigger than they could do on themselves And so part of the essence of that is having this idea of slip joints that things can interact without being rigidly connected So one can slip and the other one won't go and hopefully our diamond sequencer will partake of some of this results We shall see So, you know the control aspect of diamond sequencer, you know, so here it is It's a modified form of the hg3 Grid and it's got these four pockets around it and the access to these four pockets that represents Input output temporary space in the instruction stream There's a lot of rules that you have to follow about going in and out of these pockets in order to get this Reliable sequential stepping that we're trying to produce on the other hand The pocket chains that are the things that are going to go in there and like one of the rules is you can only have one Codon one of these pocket chains in there at a time They are loosely coupled to their neighbors like this ec here It's not even lined up with the ec that is the next link in the chain because they don't have to be they just have To point at the next pocket and the pocket then points back at them and hopefully when we have Some gigantic mess like this so you can barely find the diamond sequencer buried in the middle of it That will be able to have this combination of control and quietness and reliability in the Diamond sequence pockets the diamond sequencer pockets the DSP While allowing whatever happens out further in the grid Out further in the diamond. We'll see how it goes To recap the actual grid we just saw the simulator on the grid a couple of weeks ago We had a crash storm the underlying Linux stuff got messed up And it took the whole thing down looking at it There was a hypothesis that possibly in some extreme cases it might have run out of room in the Linux kernel buffers We said let's up them by 400% make them four times bigger and start it up again And that's been running now for several weeks and it hasn't happened again that proves nothing But it's a little bit encouraging and the question was how is it going to end? I mean how is a run on the grid gonna end? Is it gonna be another crash storm? Is it gonna be a power outage if the if I lose power to the house for a half an hour or so? The UPS is gonna die and the grid is gone. You know, there's no backup for this This is this is live computation that's happening. It's not even clear what backing it up would mean Or the third possibility is that I will step in and end it and that's what happened I ended the previous run with no failures outside and now it's running the spike loader So let's just take a look at it for a minute So that's just getting started. I'll run it for several more days anyway Unless you know the code improves enough that I want to bump it off Yes, I'm not going to recap all the previous t2 project worries, but I figure if there's new worries I need to report them There's some kind of issue with the camera and the USB connection to the laptop that's recording the grid. This is a Manolta Tx4 that is recording all this stuff and It flaked out and it flaked out when I was away from the grid and it lost five days of data Now it was just let's play Guess which diamond grid is gonna pop next so it's okay that we lost it But even after I got back and I unplugged the thing and plugged it in again. It happened again I had to power cycle the Fuji Completely and mess around with it. So now I'm scared now I'm anxious that something's gonna happen and when we actually matter like, you know, the grandchild of the ancestors about to be born We shall see All right, I want to take a couple of minutes just before I'm not gonna go along To talk a little bit more about sequential computing and so, you know last time I was showing this Picture from Wikipedia. I call it a traditional processor really should call it a sequential processor because that's absolutely the key to it You know, yes, it computes a function It's it's got some it takes some input it produces some output But the real key to it is this circular connection between the registers that go down to the Combinational logic the thing that actually can you know make nonlinear functions and add things and do whatever and then writes it back Into the registers erasing what used to be there. That's the key so that on the next step We can bring in all the power of combinational logic to apply to the next step And yes, you know, we take some input and whenever we take it we produce map But whenever we produce it but that key of using the same piece of hardware over and over again on each of multiple steps Is the absolute key to the power and the risk of sequential processing now? I'm gonna take a lot of time on this But you know these large language models that have these you know millions and billions of parameters That they're fundamentally by their nature They're essentially Combinational mostly that they just sort of feed forward it depends on the exact model But you can make them into sequential by taking their own output and feeding it back in as more input and people do that Routinely and people somebody found out that adding the phrase let's think step by step Made the pre-trained language model wasn't trained for this particular task at all go from a clear fail to a you know a solid C plus on some of these models and you know My interpretation is the reason it does that is because it encourages the network To expect to do multiple passes through the whole thing Well first I need to do this and the answers blah and then I need to do this in the answers Blah rather than trying to just intuit the answer in one pass through and that is the key right there That's the sequential computing again power and risk And you know the risk is is okay What happens if you lose your place if you just have a feed-forward network if you just have combination on logic It's inherently Self-stabilizing as soon as the inputs stop changing well then the stuff next to it stops changing and the stuff next to it It flows through until the output stops changing and that's the answer that you want Sequential logic cannot do that because it erased its intermediate results So when you have a sequential computer Somewhere in it you have to build a reset button and so that's our title card for this time You know you look at all the devices you have in your life Some of them have a reset button some of them don't that's the hallmark of the sequential machine Reset what's it doing? It's saying forget all the steps and go back to something where it considers a step one Where now as long as we can do all these steps again, we'll get the answer that we had in mind The diamond sequencer is a sequential computer So we need the equivalent of a reset instruction for that as well So that was the LX reset LX. I'm saying is a loop executable I'm kind of using it to put at the front of my instructions And I was working on the LX reset as much as I was working on the loader in the last period And there's a big question a design question You can't reset the entire universe because the universe is indefinitely scalable and you know So in the end I decided that what reset means is flush the output loop and flush the temporary loop and Leave the input loop alone because presumably somebody else put something useful there And then once the loops are gone output and reg Recede them to a new one and then move on to the next instruction. So We really don't have time to look at it We can look I got a demo. Well, what the hell we'll try to do it really quick We do not have enough time, but I'm just gonna do it anyway So here it is here's the reset instruction and I'm just gonna give it an event And it didn't work. Okay Well, I guess we won't do it. We'll do it during the during the live stream. All right, so That is it The next update is gonna be t-minus three My goal my primary goal is to have some kind of program loading and running that copies itself And you know, that's a big chunk of the missing hunk of The project that we're trying to get to so we'll see if we can get that far But hopefully I feel like we're getting close and we shall see how it goes Have some fun as always and again After the next update, there's gonna be a pre-programmed hold in the countdown because the month of November is booked out Have other stuff to happen no updates in November and then we'll be back in December for t-minus two one zero And hopefully we'll have grandchildren for the ancestor on January 3rd. Thank you so much for coming and taking a look I hope to see you next time