Hello, this is Ted and this is Brandon. Welcome to Concerning AI. Yay, yay, yay. We're back again.
Unless you're your first time listening, which is great, if you are welcome, you already know what the show's about. We have asked two questions. One is AI, a Supreme AI in existential threat to humanity. Yes.
And if so, what do we do about it? Oh man, we suck at the second one. We really suck at it, but we're trying. Yeah.
I think when I say we, I mean the royal we here. Is the royal we like really you talking about yourself? So actually I think I meant the non-royal. You meant the peasant.
You meant humanity. Yeah. Yeah. All right.
So let's just dive right in. I also meant the royal we do. We don't. You and I.
Yeah. That second one's hard. Yeah. So we'll probably pump on that.
Just spoiler alert, we'll probably pump on that again. Yeah. Yeah. Well, you know, we'll try.
Yeah. All right. So we got to follow up from last episode and I wanted to. Super.
Yeah. I know. It was cool. Yeah.
So we've got a comment from, I'm going to put your name. So I'm just going to be like that now. Hack on. Hack on.
Hack on. Hack on. I apologize for terrible American English. Just butchering everybody's names in the room.
Sorry about that. Sorry. I think he was referring to Eckertoli's levels and then I asked a question about Dolores. I misunderstood.
I thought he was suggesting Dolores was at level three and he clarified. I don't know. He didn't mean that. He just was saying maybe Dolores is moving through these stages.
And I think level three is this idea that I've read about and never really experienced that through meditation or other ways of examining your own self that you can come to not identify with that voice in your head as being yourself anymore. Yeah. If you've done that, you know, that's why I started the religion that I started. Yeah.
Yeah. That's because of that. Very, very, very. We're just going to leave that as a tease for a future show.
We'll talk about, we'll talk about Salsid. It's a legitimate thing. No one would say such a thing. And mean it.
It's awesome. I love it. Okay. Okay.
So thank you for that comment about Tolle. So that's great. It's about consciousness. And all right.
So then the other point that Mark makes is about how we talked about Brandon was really struggling with this idea that the AIs, the Superhuman AIs could be our children and wanting to be wanted an emotional connection to them. I want to. Yeah. I want to convince.
But it seems like Mark is not convinced. Even less so. Yeah. And articulate about it.
And he talks about, let me just skip a little bit, but do you think home erectus would consider us their children? They are our creators. Their choices created Homo sapiens. If they had never started using fire and cooking, we would probably not exist today.
Thought experiment. Imagine a male Homo erectus runs into a female Homo sapiens while gathering berries. Do you think he sees her as his descendant and wishes for a prosperous future? Maybe he even leaves her his berries to ensure her survival over his?
Very unlikely. Yeah. He calls it Mark and the sapiens. He won't be doing us any such favors either.
Yeah. So, you know, I think that's probably true. Yeah. Probably true.
Who knows? I have some ideas for a startup that the Roddy Brooks article spurred that I think is sort of enmeshed with this. So maybe next week we'll talk about that. Okay.
So, I'm going to talk about the Roddy Brooks article if you haven't seen that in the Facebook group. I posted that this week as well. Yeah. I saw it super good.
Yeah. All right. Cool. And then Matthew also comments on that about AI and children and...
Oh, and the numbers. Let's see. Yeah. The numbers.
So Matthew, I think was the first one that gave us the numbers. So last time we proposed a way of labeling ourselves with how soon or far away we think Superhuman AI is. Yeah. So we had a few people respond.
So Brandon, you said you were at 530-70. And then you said you were at 15-30-50 because you're a lot more confident on the narrowness of the animal. And narrowness. Yeah.
And we're really in the same ballpark as both of us I think. Yeah. We can be friends, Matthew. So, you know, even if we do call them by the wrong names.
Yeah. And then Saskia? Saskia, zero to 20. I was wondering if that was a joke.
I could have been a joke. It was attached to something that was... That was... It sort of had a humorous element.
And then Simon is a 25-150. Yeah. And that's a longer range and that's more of a future thing. And then I was just talking to Ben today and he said he was a 10-5200.
When I pressed him, he didn't just volunteer that. Yeah. Okay. Okay.
So we're excited about this idea of this concept. It is. And better yet, give us your number. Yeah.
And actually started a thread in the Facebook group that's just like I think going to be devoted just to that number and talking about how to come up with better terms and stuff. Awesome. So if I remember, I'll put that in the show as well. Yeah.
You know, so the Eckhart Tolle. Yeah. Tolle Eckhart Tolle is a good segue. Segue?
Yeah. King of segue. Yeah. Yeah.
Because it's all about, you know, like meditation and being present with your mind. And today we're going to break new ground into turning AI. All right. I can't wait.
Yeah. We're going to do a thought exercise. Oh. My brain is so excited.
Yeah. Okay. So I practice this with that once. We'll see how it goes.
Okay. All right. So imagine that you are, you're looking for the low point around where you live. The physical low point in the ground.
Yeah. Down low. You're looking for the down low. Right.
So you go down the front steps of your house. Yeah. Except you don't, you know, the thing is you don't live on this world. You live on a different place that has no water.
No water. Yeah. So it's, you're going to pretend like your water, but it has no water. It's not actually true that it has no water because all the water is in the atmosphere.
Okay. So we will call this place fog world. Fog world. All right.
So there's no water on the ground. So you can only see, you know, maybe a couple feet in any direction. Oh, I see. So, so when I just said you go down the steps out of your house, that's assuming that you know anything about this place.
But if you're in your own living room, you can see only a couple feet away from you. No, just because you know the area you live, you're going to be able to imagine what it would be like to be in fog world in the area you live. Okay. So, so you're in your house.
You know that your house. Yeah. We're going to start you at the front door. At the front door.
Okay. And you're going to find the low point. So everybody's got steps down. Nobody has steps up unless you live in the basement.
Yeah. What do you do in the basement? I just end up where we're going a little earlier than the rest of us. Okay.
So Ted, let's actually walk through your front steps. Okay. So I've got, so you're there. I've got there on the point.
You're trying to go down. Okay. So I'll go down. So you look, you look around like two feet in any direction.
It is, is there a clear down direction? Yeah. I've got to step down. Okay.
Boom. Down is that. All right. We have another step.
All right. Hey, another step. Where you are. Hey, you're making good progress.
I'm still in these wood steps on my foot course. Okay. All right. I'm going to step down.
Okay. And one more. I'm not to go. I just came down.
But, but, but if the fog only lets me see two feet in front of me or say, you know, three quarters of a meter, right? I'm a little uncertain at this point. We'll say a meter just because that's a nice round number. Okay.
Well, if it's a meter, I think I can now see the edge of a step. Because the step is a meter. I think I've got another step. Yeah.
This is a concrete. This is a concrete. So you're still cruising. Yeah.
So you're like, Hey, the low point that I can see from where I am straight ahead. So I go straight on. I go down another concrete step. Okay.
And now it's a small step. So I can see it very easily. And one more step. That's a little bit more flexible.
Maybe I think it goes a little bit to the left. Maybe the left. The left. Okay.
All right. So let's just sort of follow the signal and start heading down the sidewalk to the left. Yeah. If I go to the left, if I go to the left, it means very shallow.
I don't know if I even noticed it. I mean, what happens if you get off the sidewalk and kind of move too far to the left? Would you run into anything? Does it go up?
Does it go up? If I'm going down the sidewalk on the left, there's a hill to go up. So you're just going to bounce off of that, right? You want to go up that.
Okay. What happens if you get going kind of the other way I might fall off the curb onto the stream? All right, so let's say that happens. So now you're lower.
Are you ever going to get back up on the sidewalk? No, probably not. It's always going to be up. I will hit a driveway or a curb cut, but that still goes up.
It's still up a bit, right? So you're kind of like the crown of the road. Yeah, I can't go into the middle of the street. No, the crown of the road intersects.
I mean, it kind of goes down to intersects. There's like a valley there. I mean, the gutter. Right?
You're in the gutter. Boom. All right, so you keep walking. Let's say fast forward a little ways here.
Tell you coming up to where that road crosses the next street. Yeah. Well, now what happens there? The road that intersects goes down further, steeper, down to the left.
I turn left. I turn left onto another street. Ah, see, I think this is where maybe your intuition is your schematic topological rather than topographical. Oh, you think I might be skipping a subject?
I might be skipping some stuff because what do you run into when you're walking down the gutter, regularly, run into? Oh, like dirt? Leaves. You run into that well.
Cars. What if there was a big, flushing rainstorm? Trains. Boom.
There is a drain in the corner. Okay, so is the drain at the low point? Are you going to get stuck at the drain? That's my question.
I guess I am. Yeah, because you're actually coming up to another slot where everything kind of funnels down into a little slot. That's right. So from a three, from a one meter radius, I can't see that that road goes down on the other side of the drain.
It really goes down that way. And in fact, it goes down another. There's several different directions that it goes down, right? It does.
So you are stuck at what we call a local minima. So let's say that the height you are above sea level is your error signal. My error signal? Yeah.
Yeah. So you've got, you've got, you've got several things, right? You can see it there. You can look around.
You can see what direction the ground slopes for a meter around you. Yeah. You've got that. But you also have this uncanny ability to take a measurement because you've got these surfaces and your cell phone is telling you how high you are.
So I'm 50 meters above sea level. Yeah. Okay. But I can't tell every direction you go, it goes up, it goes up like two, three, five, ten centimeters.
So you're stuck in a local minima. Oh. From gradient descent. No.
It's not a local minima. No. It turns out that the storm sewer inlet is not that big of a deal. Do I have to crawl down in there?
Oh, yeah. I had been thought about, like actually losing through. That would be interesting. But that's very different.
We're going to stay. We're going to stay, we're just going to forget about the top above here. So let's say you you know you kind of heat things up and jiggle around and take some bigger steps and finally you get back in the groove again Headed down the hill again anyway like I just like randomly go off in another direction hoping I might yeah It was local minimal. Yeah, so what's gonna happen is you're gonna do the kind of essentially the same thing because you're on streets Yeah for now you're on streets.
You're gonna end up following those down. Yep until you get to something What are you gonna? What are you gonna run into? I'm gonna run into the Willamette River You're gonna want to run into the Willamette River right and how high above sea level is that would you say that's probably you know?
I don't know 25 meters make if my high is 50 meters. Yeah, maybe I think since there's no rapid Between there and the ocean and it's a big slow river I'm I think probably 75 25 meters is 25 meters is probably a distance of more than a hundred kilometers Yeah, so there's no water in the Willamette River right so you can actually get down probably even below the surface Which is what you were probably talking about right probably down I can get into the into where the bed Yes, once you're in the bed You're in this you're in this this thing where there's gonna be like potholes and all kinds of stuff where it's gonna go down and up And it's gonna be pretty flat. This is wide river. Yeah, so within the banks of the river It's really hard to know where to go.
Yeah, there is a signal there, but it's 25 meters over the course of you know a hundred and sixty kilometers, man. I'm just pulling that out. I have no idea. It's more like a hundred hundred miles So is it close?
Yeah, I'll see you close. Yeah, okay. My school system here in the United States is totally family But anyway, that's a very stone shallow gradient. Yeah, so so here's the deal Like we are we this is very similar to gradient descent Which is one of the algorithms that we use to try to find these these local minimar of the air function or local maximum of this gradient Ascent what you're doing is you're taking the stock of where you are and you're looking at the gradient in all directions And based on which gradient is the strongest you're gonna walk that direction.
So give us a I mean we're talking about Yeah, I know we're still talking about surface of the earth. So we are still we're still trying to develop our sense for how Hard it is to find good a maxima. So if if I'm in the if I'm in the river Yeah, with no water But I'm in the river I'm in the river bed I would know that I had succeeded if I'd walked that 160 kilometers and I went down that 25 meters, right and I got to zero My my phone tells me I'm at zero and then you've got this great thing that really starts heading down the floor of the ocean Yeah, so I get a bigger gradient at that point You get a pretty nice gradient for the most part there Yeah, for a while and you know you got the con shell and all this sort of stuff Well, you can you can get a lot better than the 25 that you're at yeah But that's a long ways to find that right and the earth is big and it's gonna take a long time to walk that and I wouldn't know that It was a hundred and you have no idea that it's 160 kilometers, right? If I lived another hundred kilometers, you know to the south of here it'd be the same problem only works Alright, so here's what we're gonna do we're going to say so this is right now This is a two-dimensional search problem, okay two-dimensional.
There's it's southeast west or my two axis It's surface. You're on the surface and you can pick a direction along the surface So this is a two-dimensional search sort of problem, okay? So what we're gonna do and it's really it really maps the two-dimensional space that we pick which is the surface of the earth Really maps well to the actual problem domain of you know Kind of where the water has worn away like that really does those dimensions have a lot of signal in them Yeah, so let's say we add a couple more dimensions because they can't possibly hurt right you need more information Yeah, there's gonna be some information in those signals I should have thought more carefully about what is in those but I don't I don't know what the information source is for these dimensions But what I do know that their effect is is that they create a much higher frequency signal of its complete noise Let's say it's almost completely noise, but it's a high frequency signal. So what does that mean?
So that means it's like you're walking on ripples, but let's make it even bigger than you're walking through a You know how glaciers have crevasses. Yeah, yeah So you've got all these crevasses and like ridges and there's not actually any useful information in those dimensions that we added But every time I get to a crevasse it seems like I have to go down You gotta go down and you're stuck you really stuck Yeah, right and you like how I found it was this a strong? I don't know this is so suddenly that signal of 25 meters over the space of 160 kilometers that signal that was already tenuous at best It was already very difficult to follow is completely washed out by these useless dimensions that are creating the corrupt crevasses And when they're projected into this space they're creating these crevasses and ridges and things Yeah, it's like hopeless. They're bigger than the banks of the river are it'd be easy to In that out of the river bank out of the river bed on actually because my my getting out of the crevasse algorithm would also Lead me to get out of the river get get out of the river Yeah, well, and maybe you get into a crevasse that's like cuts through the bank of the river Yeah, and then you're like all right now.
I've come to the wall here in this crevasse What do I do? I'm gonna have this is still the best I found I'm gonna have to kind of am I close? Yeah, do I turn around go back? I mean do I try climbing around here?
Let me try a little around here I climb out and I'm like this is not that much better. Yeah, so it pretty much ruins any any possibility of getting from here to the ocean So this is this is this is one of the kind of sideways kind of a sideways Direction in which the curse of dimensionality It's kind of a curse of dimensionality sort of problem that we run into here where the search space is so huge And it feels like you can it can never hurt to add information right even if there was no information How could that possibly hurt your learning? I can because it can obscure like the true structure because the true structure is so hard fun because the signal is so small So the thought exercise like what's the purpose of this whole body? Yeah, what are we trying to do?
We're trying to understand I guess we what it makes me think of is remember last last episode I talked about the terrible 1972 version of Westworld. Mm-hmm. Well, the man of black there Mm-hmm had visual sensors which were completely pixelated and absurd like they showed him through us sometimes Yeah, yeah, we using our human eyes couldn't tell what the heck he was looking at But it was enough of a signal for him to know what he was supposed to shoot at And and so you know his visual system gave him superpowers Well, he not only had he not only had like infrared and sound and vibration and things like that But like the visual information, you know how iPhones went from how they were pixelated at one point and then I had four came out It was not no. Yeah, well, he had the opposite of that He had the non-retin version he had the the Atari you know the Commodore 64 version of that would ride low-res graphics Remember low-res graphics.
Yeah, he had the low-res graphics version and you couldn't tell what he was looking at It was I don't know how he identified what a person was versus something else Yes He's got almost entirely noise dimensions to try to try to learn from and yeah And in fact when he was seeing heat the way that the guy got away from him was he went and picked up a torch And then he couldn't tell what he was looking at it was no so we He just wandered away. Okay, so that was that was like a I don't know if that was like a two-dimensional if he was He his visual system was like a two-dimensional world. How would he know what you know what he should be looking for? How yeah, well noise.
I mean you're right was all noise Yes, so probably in this case like adding that visual noise sort of thing would would hurt his performance in every way When it would instead of just relying on the heat and vibration or whatever the other ones that actually messes up So so what are the things to take away from this gradient descent little thought exercise? So one of the things is the one that I mentioned for which is it feels like more is better But more is usually worse because you really want to have your your learning algorithm really needs to map on the problem Well, how do you know how to do that? Wow? That's part of the black art still so most of the work There's several several phases of machine learning that are a ton of work one of the phases is getting all of the data into the right Shape it's just a lot of like so for supervised learning algorithms You got to like make sure all the labels are right on all the places and that for visual algorithms For example that your masks actually match up with objects that you care about and all this kind of stuff is a lot of groundwork there Yeah, so that's one area But the place where you put a lot of effort into I think that is where you win or lose is on the Feature engineering because the learning problem can be well the learning problem is completely intractable in many many many in most in almost all the feature spaces What you need is you need so an example is we're talking about a couple dimensions here Like let's talk about an image for a minute Yeah, so if you take the space of all possible that favicons knows those little tiny logo things that show up in your Browser tab to indicate that this is you know, Google's Gmail There's 32 by 32 now they used to be 16 by 16 Yeah, and there's there are bigger ones now But but the kind of standard is like a standard one size is 32 by 32.
Okay. Yeah, so that if you just take the space of all favicons That is a 10 24 dimensional space because each pixel is dimension because each pixel is dimension They can vary in the pixel if I have whatever the values are that we allow a favicon to have in terms of color Yeah, so each of those 1024 pixels can have any number of values Yes, but we know that there's some structure to that that there is a much much lower dimensional manifold in that space of that 1024 spaces It's not three-dimensional It's not the sort of size that we can actually reason about intuitively But it's much much smaller than the 1024 and that space has things like like lines and stuff in it It's not that we have to search exhaustively the space of you mean because favicons tend to have orders on them Or they tend to have the information in the middle and the background and the signs or what do you mean by that? It's because we make favicons that have structure that is meaningful. Oh, okay So is this also why you can Compress an image like that?
Well, it's why track learns the world is tractable in the first place Why people can even learn it is because you know that the space of all possible of all possible inputs and outputs is just beyond just beyond belief in its size But there is actual there is that yeah, right? But there is actual structure that makes the actual problem be a much lower Like I said much much lower dimensional sort of subspace embedded in so okay So let's try to let's try to bring this back to our three to our two-dimensional our two-dimensional search that we get earlier Okay, so let's say that we're we know some structure of the problem and what we know is that Really the lowest point on the earth is on a particular latitude line or a particular longitude line one of these things that goes all the way around The area yeah, so now we just our search to find that low point is a lot easier, right? Yeah, yeah, we're just gonna walk along that we can just search the whole thing we can search the whole thing Yeah, so that so that that for an individual person searching for the low point Yeah, if we know the structure is that it's on one of these lines So now it's turned it into a one-dimensional search problem. Yep.
Instead of a two-dimensional search problem That's made an enormous difference, right? You agree that for that's right that's something that a person could actually find the actual global maximum in their lifetime Probably in vlog world if they have food and all that kind of stuff that they need just walking Yeah, right, right? But hopeless completely hopeless if you've got the two-dimensional case and that's just one dimension Like that's just the collapse of one dimension, right? So in these really high dimensional spaces There's something like that line and all of the interesting structure and the answers of like good parameters for our models is a fall Somewhere on those space, you know on that lower dimension So maybe it's a surface that we're looking at let's say we're collapsing three dimensions to two Okay So instead of having to look up and down and all through all the space of the solar system We know that it's on the surface of the earth because it's a lot more tractable Yeah, and this kind of generalizes as you go up like that being able to find that that productive Lower dimensional space is part of what is the black art and features features are how you go from the raw Space to a transform space to feature space instead of like an image you've got the raw pixel values You can actually you know run edge detectors and gradient detectors and things like that to start coming up with lines and other things Like that that start to become your features And then you can try to find meaning from the features rather than from the pickles right and so that's a way of compressing the dimension Is that what our brains do our brains our brains work on the features of the pixels?
We'll yeah, actually, you know I don't know if we know exactly how it works, but we do know I mean I certainly don't know exactly how it works But I do know that our visual system for example has there are neurons that are not directly sensing the photons that are coming in But there are a couple layers away you know that code for Light dark patterns that are oriented in a particular direction with a particular sort of extent So we got all these I'm just sort of drawing like little lines in the air with my hands here Yeah, we got one that's like 45 degrees this way light dark Yeah, I want this 45 degrees this way light dark when something comes across our into our eyes boom The suddenly this neuron starts firing because it's detecting that pattern so I don't like a cascade of features that go higher Yes I mean I'm imagining that if you want a few levels deeper neurons you would get face detection maybe there's there's there's this there are spots Yes, that's you get further in I don't know how many layers, but yeah I mean it makes me wonder if you know the whole cat versus dog question You know why is it the babies are so good at that? You know I wonder if it's because they have faces like our faces So somehow the face detection is connected well I you know different people believe different things about this and time will tell but I'm one of the people who believes that are We have a lot that's hardwired into our brains that we start with and babies can't use it because they have to like they have to tweak the parameters Yeah But the connections and the way that it works already has faces has things like cats and dogs and the propensity to be able to distinguish those It just needs to be just fine-tuned a bit through actual experience in the world But the hardware is already evolved yeah, so human brain human brains got All these millions of years of tweaks to it. Yes from experience and so we're starting from scratch with these ais This seems like a really hard impossible. Yeah.
Well you think about the search problem We've explored a lot of space. Yeah through evolution through the search process of evolution a genetic it's a genetic algorithm, right? Right for sure Yeah And there's this part of us that we don't use consciously that's super sophisticated and fantastic that we don't even know is operating It's metabolizes a lot of energy, but it's it's completely unconscious and it's doing all that visual processing and stuff like that for us And so what floats up through our talking mind is? You know these kind of these concepts and ideas and stuff like that we might call intuition and other things like this They're coming up to our attention of our speech center of our symbolic processing not because this symbolic processing is pulling them up But because this amazing machine is working in the background and generating them all right So if we step back a second and go back into the thought experiment, why is this talk experiment interesting?
Why do we or any thought experiment? I mean well, I you know I'm inspired by Albert Einstein Yeah I mean he ruined physics for everybody right because he did like so much incredible groundbreaking work in so many different directions that To ever be a physicist after him is hopeless. You can't possibly live up to Albert Einstein Yeah, but one of the things that he did was he was able to use that very incredibly sophisticated hardware our visual cortex that that part That bubbles things up to our speaking mind like these fully formed beautiful ideas and understandings of the world He was able to use that part to think about very abstract things like the speed of light and what it's like You know that he was able to imagine riding along on a beam of light in a very visceral tangible way So I feel like that if we can kind of do similar things with our machine learning algorithms If we can if we can not just use our symbolic processing to do mathy kinds of things that actually get into kind of a more a deeper visceral Visualization of what's going on like what the curse of dimensionality means why extra data is not necessarily a good thing Why you know what happens with your what gradient descent innocent are I feel like that that spreading that is spreading The wisdom of Albert Einstein No, I don't know I'm hopeful for myself, but I'll be able to understand the problem better and that for other people make the same thing So that's why I Don't know well, okay, so here's not sure about yeah Well, I mean maybe maybe not I don't know but you know I learned I don't remember exactly how I learned numbers and math You know as a kid yeah, but my kids are going to a school where they're learning in a very different way that is unfamiliar to me Okay, and so when they were introduced to the numbers They are the my kids already about account and stuff before they were introduced to this been school They were introduced to the concept of one nest ah so I won this and then to well, you know two sides of your hand is two Yes, right and then so they're in the world exploring Three nests and finding triangles in the world and for this and what are you know? What are where there four petals on a leaf or whatever?
I know four petals on a flower, you know five so they like that was one year then the next year Which is this year they came home with a board that they'd made it was ten nails around a circle ah-huh and they had five different colors of yarn So each each nail was numbered one through ten okay, or maybe they were numbered Well, I don't know they were numbered they were numbered yeah And so then you could make the in the yarn the five different colors of yarn went in five different patterns one went from one to two to Three to four to five. Ah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah, yeah Next one went from you know two to four to six to eight to ten the next one went from you know three to six to nine Ah, and then it went to two which represent a twelve Five and so you can do multiplication on this and the kids have learned how to do multiplication up to their five times tables using this thing And it's fascinating well I sat down with it and I started thinking about could you do multiplication by six and of course You can it's just four backwards and seven is three backwards and eight is two backwards and yeah Yeah, because we're because it's a base 10 because it's a base 10 all of our times table I mean, it's just because of the way I mean so this was like a way to like viscerally get something different about what numbers mean in the Relationships with each other now. They've also done a lot of work with like counting by twos and counting by threes And I started to think what if you count by three starting from a number other than I yeah, right? Yeah, you come by three starting from one well the board gives you that too you can do that even starting where you want Now that's different from location doesn't give you multiplication at that point if you start on a number other than three You're gonna buy threes, but one four seven ten is a heck of an interesting yeah Yeah, yeah, yeah, it's not a pattern that I would have ever gotten in my learn the way I learned numbers Right, so you there's like this visceral connection to what that sequence is yes that is difficult to get from just a symbol Yes, yeah So that feels like it ties back to the somebody AI thing you've been talking about and this like my kids are now embodied in a Different way so somehow their brains have access to something different than I had growing up Which I think is really cool and bringing up the embodied AI just have to give a shout out to Rod Brooks again That article is great and he's one of the fathers of the embodied AI whole thing Yeah, so we'll put that in the show notes about some of the problems of self-driving cars But I think that's gonna be a show of it.
Yeah, yeah, I'm excited about that one too We'll get to that yeah, share your numbers and also maybe some of your favorite visualizations for like how to connect with machine learning Concepts from a very visceral intuitive direction rather than the sort of symbolic math It would be fun to have some others to try maybe we have a whole show that's devoted to several of them or something like that So if you think it's fun if not we can go back to our normal Puntificating about whatever that we do that yeah, yeah So see us in the concerning a group on Facebook Yeah, and keep bringing the bring in the good content and ideas and telling us one more office and point us in good directions and fun Yeah, so now we've reached the exciting conclusion, you know, we sound really hip and an idea It's like high-five. Oh, I know what it was we're gonna say at the end we're gonna say don't drive like my brother Yeah So suppose you've seen listen to the car talk that's where that's the reference to but I don't know if you've got ideas for You know a cool way to end the show make us makes us sound really hip and happen in like we know what we're doing Let us know because I promise that we have a lot of people in that way so we do we do We're putting ourselves a good face on yeah, right? We like to feel like we that's right too. Yes for sure all right So next time this is said so you know, it's not terrible What we're doing here yeah, it sounds try to say goodbye, you know, yeah, so good.
Okay. Bye Harmony