Hi, this is Ted. And this is Brandon, and we are on the concerning AI podcast. Because we're concerned. More or less concerned.
More and more. More and more. I would say more and more. As in Moore's law.
More and more. We will definitely be talking about Moore's law today. All right, well, before we jump into our podcast about Exponential's big announcement. Big announcement this week of the Open AI, what is it?
Is it an initiative? Is it a company? What is it? It's a nonprofit company.
It's a nonprofit company funded by Elon Musk and all the tech luminaries, basically. A lot of them, yes, for sure. And I saw the number $1 billion over the course of around the years. Yeah, it's hard to know exactly what that means.
Because there are services being donated by Amazon Web Services, like computing resources and stuff like that. They're going to be factored into it. I think it's safe to say there's a massive amount of support for it. Yeah, and so the idea is that these folks are concerned about the future AI.
They can easily call it their initiative concerning AI rather than a product. That's not a business. So thank goodness that they didn't do that. That's why they didn't do that.
Everybody listen to them. We probably do anyway. But thank you for the three of you that are listening. Thank you.
Thanks, Bob. And so they're hoping to be able to make an open source initiative for AI as a way to ensure that the good AI wins. What's the plan here? I read an interview of Sam and Elon.
And I try to remember if there's anybody else in there. Anyway, the gist of it is they don't know. But they think it's probably better if it's open and democratized so that there's a lot of AI coming up than if a single company or individual controls that company or government or individual controls that government has all the AI that's like a billion times more than humans. So whether this is a right solution or wrong solution or whatever, it's interesting that they're putting that much resource into it.
They're concerned about it. They're doing something about it. I actually would love to do so. We're meeting on this and have a show about it.
Whether we think this could actually work. I think that'd be fun too. I want to add one more thing as they're trying to emphasize creating tools that assist people. So they're kind of a little bit cyborgist in their approach.
All right. All right. Well, very good. So welcome back to that.
But if you're interested, go read about open AI. Big deal. Very big deal. Yeah.
For sure. All right. Well, we'll talk a few more things about Chaken and Focus on the Chromeium. Well, hey, great.
Thank you Ross. We're going to talk a little about how combat is- It's about HOveryanna, buzzing? Yeah. You drive something into your hand.
Boom. So uh, well, let's start with some examples of 500 Hey, you've already been posted about or restaurant news that were given. And we've been posted about how well- what was it made about Rich? It's about what was it, right?
for middle schoolers. Alright, so this is common in France, this lilypad exercise, the truth is around the world. But the idea is thinking about, you got your got a pond and there's some lilypads and they double every day in terms of the area of the surface that they're covering. And so the theory here is that if, and we want the kids to be able to kind of watch this happen but we got to be careful because if we're not careful the lilypads will cover the entire pond and kill everything else.
That's right. So we got to be careful but we still want to watch them double. So we watch them. So when do we need to be concerned?
Right? This is what they have to get. And kids, so when do we need to worry and they might say something like, you know, let's if the pond is half full we should be worried. Yeah.
And so maybe they watch it for a month and it takes 29 days and now it's half full and they're gonna try to do something about it. And then they ask the kids, well how long do you have to do something about it? And it's very unintuitive to think that you've spent a whole month and you've got one day one day left. You're done.
It's the same as a full pond right now, basically. Right. Yeah. Alright, that's a good example.
I think it's a good example. I think that could really work. You can kind of, I could imagine making a pond in our in a classroom and doing it with pieces of paper and you know, getting a kind of a visceral appreciation for that. So I dig the lily pad.
Yeah. I might want to do it with my kids. Yeah. That'd be fun.
That'd be fun. And then we there's another one that's similar but more biological in the sense of microscopic bacteria. Yep. Okay.
I think that one is yeah, that's a good one. Okay. So we've got a bottle. A bottle.
And we've got some bacteria, obviously, microscopic. Microsopic. How many bacteria? We start with one.
One. Let's start with one. That's not very many. No, not very many.
And let's just, let's just, that is a really clean bottle. It is a really clean bottle. And let's just assume that food is no object that they can reproduce at the same rate no matter what. You know, we're just going to do that.
Okay. So that's a bit of a deviation. It is a bit of deviation from actual, you know, real life for the point of illustrating this. Okay.
We didn't talk about where the lily pad see their food either. No, it's just not a stress on it. Yeah, there's lots of water. Okay.
But I mean, bacteria has to eat, but we're just going to track that out. Okay. And so, so let's say we start with the one bacteria in the bottle at 11 o'clock. 11 11 o'clock.
Okay. And then every minute, it doubles. Okay. So the one bacteria becomes two at 11.
Oh, I got two right there. I can't see any of them. Can't see them. No, they're microscopic.
11.02. Now we've got four. Four. Can't see them.
So three. Eight of them. Even with the microsoft. No, it's even.
11.04 11.05. They keep doubling and doubling and doubling. Okay. And when is the bottle get full?
The bottle gets full at noon. One hour. One hour later. So they're very precisely sized back here.
Very just so that it works out this way. Well, the bottle is but it's actually something. Oh, yeah. Okay.
So the bottle might be the size of the skyscraper or it might be the size of a thimble, right? Yeah. Depending on which bacteria it is. Yeah.
Okay. But when you think about 11.30, something bad happens when the bottle is full. Right. That's the key thing.
Yeah. When they're all going to die. Oh, they get all the back. No, there's supposed to be some connection to all the people.
Sorry. Okay. So that is bad. Yeah.
So think about what happens at 11.30. What do you see? 11.30. You look in the bottle at 11.30.
What do you see? Yeah. Microscopic. You can't see them yet.
We can't even see them. No. We have to use a microsoft. Halfway through.
Halfway through. Halfway through. Yeah. That's right.
At 11.45, we can barely see them. Like a little thin film. Yeah. That's all there is.
11.45. Like a little shoot. In the middle of the time. We have a little time.
Look in the bottle. There's nothing in there. More of a machine than a pilot. Nothing going on.
At 11.45. So we've got let's see. We have 45 minutes. We have 15 minutes.
That's right. I can do I can do simple simple math. So at what point is the bottle half way full? See now that is the that's the mind-blower right there.
Yeah. Right. Unless you know the answer. Right.
Because you've had it's like a trick question. It is a trick question. And if you just do the look at experiment, you might know the answer. You might know the answer.
Yeah. The answer is 11.59. And so when do you start to get worried? Let's say when is the bottle quarter full?
11.58. 11.58. So it just seems like plenty of room right at 11.58. They'll back here.
They haven't even like started to experience overcrowding. No. Definitely. They're like I got lots of arm room.
I can go for walks, take my dog around the park. That's right. That's the type of bacteria they are. So at 11.59 now we start to get concerned because that was a pretty big joke.
That was a big quarter to a half. And we can see the handwriting on the wall. There's not a lot of space in this bottle. Yeah.
Yeah. We'll get some more bottles. More bottles. All right.
I'll go. I'll go grab. How many? I'll grab three.
Pretty much. That would be plenty. That I'll bring them over. We'll bring it over.
That'll quadruple the yeah. That'll quadruple the space. So then at noon. All right.
One bottle. We fold with no problem. We got three more. I like it.
1201. What do we got? We got two full bottles. Oh.
1102 or 1202. That whole bad thing. Except it's more bacteria. Right.
Okay. I see where that's going. So yeah. So to sum that up.
So some of what we're saying there is that things are invisible until it's too late. That's right. That's right. And if you think of numbers, you know, point zero zero one point zero zero two point zero zero four point zero zero eight.
They all look like zero point zero one six point zero three two. It still looks like zero. So mathematically if you think about anything that is growing exponentially and then suddenly you've got billions. So millions.
Where they come from. Yeah. It's my blown. Yeah.
So you know, I the one I want to get to. Well, maybe we'll skip it first and say it for last because I love it. My dad told me when we were laying, you know, kind of like, what's it called? The Walla Yagen bed?
It's like snoozing. Don't be in bed one morning. He told us. Let's get to that.
Let's do the classic chessboard one and all right. We'll see if it gets us anywhere. Maybe it won't. I like that one because we get into what quadrillion's?
Yes. Alright. So the idea was, I don't know, there's some king or something and some smart guy brings some chessboard. The chessboard like brings up the game of chess or something.
I love you. This is so this is mind blowing. I give you anything you want. Well, I'll be modest.
Emperor, please put one grain of rice on the first square. That's easy. On day two. Put two grains of rice on the second square.
Oh, that's nothing. Are you impressed with my modesty? Your demands are two two months on the third day put simply four grains of rice and so on for every square on the board. Well, that's all it's all it will be done.
Oh, thank you, your honor. So yeah, no big deal, right? After 10 days, it's only like a thousand grains of rice, right? It's not a big deal.
Not a problem. Yeah, after 20 days, we're talking in what like a million ish grains of rice? Yeah, it's starting to start to become a number. It's starting to become actually rice, you know, like a pile.
I'm not even sure I'm right about that. Is that right? Wait, you want to look at it? Got the numbers here.
Well, I even did this little thing here. We can look. Let's see. Yeah, here, 20 is a, yeah, it's a million.
A total of a million. It's 500,000 on that square. Yeah, when we get to 30, we're talking a total of, it's not a million on that square. It's a total of a million if you take all this, a million on the next one, from 20.
Yeah, right. So if you take the total on 30, we're up to a billion. A billion grains of rice is a million grains of rice and 40 is that's a really big number. What is that?
Is that a trillion? It's actually a thousand trillion. A thousand trillion? No, I'm not wrong.
No, it's a thousand billion. You're right. It's a trillion. Okay.
And then to get to the let's just jump to the 64. The payoff. All right. So it's 18 quadrillion, which is 18, you said it was 18 billion billion.
Yeah. So 18 times a billion times a billion, or 18 times a million times a million times a million. Yeah, which would be how much race? It's a thousand times the world production of rice in 2010.
Yeah. And if you pile it all up, it would be about the size of the box car. More than a box car. Yeah, not a box car.
Yeah, no, it's a lot, right? Yeah. So the emperor, it depends on the story you read, right? But the emperor either rewards the guy for being, you know, for getting him or cuts his head off.
Yeah. Well, either way, somebody's going to die. He was clever. He was clever guy, maybe too clever for his own good.
Yeah. All right. So, so there's your chessboard. There's the chessboard, which means we can get to the paper folder.
All right. Let's do the paper folder one. We'll do it quick because I like it a lot. Yeah.
So if I fold a piece of paper in half, I'll think it's not a twice two pieces of paper. It's two pieces of paper. Right. And you know, I actually sort of accepted a amount for a sort of standardish piece of paper is about a tenth of a millimeter thick.
A tenth of a millimeter for one piece of paper? One piece of paper. Okay. So now we're, so let's fold that thing in half again.
Yeah. So we doubled the number of layers of paper. Now we're four tenths of a little more. Yeah.
Four tenths of a millimeter. How many times, well, let me just ask this. Yeah, you already know the answer. So it's maybe not the right way.
Yeah. How many times you have to double it? How many times you have to fold it, the end up with a tower that is as tall as the distance from earth to the moon? A piece of paper tower to the distance from here to the moon.
It's a, it turns out, which this blows my mind. It turns out 40 times. 45 times. 45 times.
Yeah. Which is pretty similar. But yeah. Yeah.
But if you only folded it 40 times, I mean, you probably hardly, yeah, you wouldn't be on the other one. Yeah. The moon would be like not close. Then, not even anywhere near close.
Yeah. 45 times. So how many times do you think you could fold a piece of paper? Oh, 40.
No, I couldn't fold it 45 times. I don't think so. I couldn't even fold it 40 times. Probably not.
I don't know. I don't know. I don't think I've ever tried to fold it. You know, I could probably fold it, you know, I don't know.
15 or 20 times. No, I don't think so. You don't think so? No, the world record is 12 times.
12 times. That's a really big piece of paper. So most people like with a piece of copy paper, see, like that eight times is kind of like the eight times. It's sort of like an ordinary mortal convenience.
Wait a minute. So the difference, you're saying the difference between a normal person and the world record is only four folds. Yeah. Well, there's the power of the experiment.
There's the exponential right there. So, okay. So why can we put all these examples out here? Well, because Moore's law.
Sorry. I'm concerned. We're getting to Moore's law. Yeah.
And the sort of derivatives thereof. Yeah. Moore's law is that we are doubling our computing power every, you know, whatever it is. What's the actual law?
All right. We got to say this. We got to say the actual. Which actual law we talk about?
Yeah. People, everybody is kind of almost everybody wish you watch about that. The actual law is that the number of transistors on a dense integrated circuit doubles every two years. All right.
So it's every two years, sort of another way of thinking about it is kind of about the density of a double z. So that means the transistors have half the size every two years? More or less. More or less.
Yeah. And so the reason that we get wishy-washy about it is because we tend to see that means that the computing power that's right packed into well, whatever it is. Like the amount of computing power you could say by four thousand dollars. Yeah.
Like the number of operations per second that that could do of non-trivial operations per second. That doubles like every 18 months. Okay. And that probably is a combination of hardware and software.
I'm guessing. Yeah. It's mostly hardware. Mostly hardware.
Yeah. Because software actually uses up a lot of that because software gets less and less efficient because you don't need to be efficient. So our algorithm is better. So we bloat.
Yeah. No, we're not using things more efficiently. We're just doing doing things more easily. So using a programmer's time more efficiently.
All right. Very good. So yeah. So every 18 months or so.
And that has held true for decades, decades and decades. And when more actually wrote that or said that before he found it until, was that right? Around the same time it was early. I don't know.
I was in this. I want to say it was in this. I don't know. I don't want to.
I don't actually want to talk and say wrong things. All right. So you guys can look at it. So we'll leave that as an exercise to the reader as any good college professor would do.
Yeah. I think sometime in the 70s and it's held true up until just recently. Yeah. And so maybe it's not holding true now.
You know, I'm not 100% sure on that. But it seems like it's sort of getting a little longer, like a little longer than every two years. Okay. So that we will be running up against some fundamental limits in how we do things.
Sure. And that's, you know, the problem with the vector in the bottle, we abstract it away the fact that they need to eat in a physical system, a biological system. I mean, you don't see deer overrunning the surface of the earth. No, right.
Right. Because they run out of things to eat. And so they're feedback. Their exponential growth doesn't hold true.
And so could that be true for Moore's law as well? Could there be a way that there would be some limit to it? I think it's certainly true for Moore's law. That doesn't mean it's true for the computing power law.
Because Moore's law is just about transistors on a semiconductor. So we may be going past silicon. We may be going to something else. We may be going past them.
Yeah. So my sisters, we might be changing how we do things. So, you know, vacuum tubes, you pointed this out as we were talking earlier. They had a similar trajectory.
They did. And then whoa, the integrated circuits. That's right. Right.
So even if so, we talk about this as like the S curve, you know, a thing, a thing kind of starts out slowly and then it reaches kind of an elbow and it looks like it's going up. Oh, yeah. One other thing that Moore's second law, yeah, that the cost of a semiconductor fab grows exponentially as well. Every time you got to build a new factor for the next generation, it's just, it's way, way harder.
Like it's, it's not, I don't know what it's doubling time is for the cost. That might be why we have so few companies that are building fabs. That is, yeah, it's only got three or four in the world. It's massive.
I mean, it's billions of dollars to build a fab now. I don't know how many, but it's like not something you do in that you do on the small, even though people are doing everything else on the small, people are doing biotech and I have basement. I mean, people like everything seems to be getting so cheap. So, you know, I mean, I think a question that we have to ask is, is where are we in the bottle?
And is, you know, is it, yeah, so if Moore's law fails, do we have any reason, the two questions I'm asking at once, let me try to ask one instead. Okay. Is there any reason in your mind to think that we are going to hit the ceiling on computational power before we get to the human brain level of computation? I mean, the way we're going right now, if it doesn't slow down, we're going to get there in the next decade or so.
We're going to get on the order of a decade and if we're off by five years, so what, whatever is that like? Right. We're going to get there. Now, if, I don't know, so that seems really close to me.
It seems close for sure. And I can't imagine that things would slow down fast enough to stop us from getting there in the next, in the next 10 or 15 years. So just based on momentum alone, it feels like we're headed there. Yeah, because the research already is always several generations ahead.
People are working on those next generations. Yeah. So another example is that we have an existence proof that human level intelligence can be packed into something the size of a human brain. Well, and that I think is a strong argument to say that there is potentially something really important about $1,000 of computing buying you more or less the computing power of a human brain, right?
Because you could say, well, so what? So we get out, so $1,000 is where it buys you a human, a human brain amount of computation, but we're not just computation. Humans aren't just computation. So that would say, well, it's still going to be a long time in the future.
But our brains are pretty inefficient as you pointed out. Yeah, I don't think that's the most efficient layout of atoms that one could imagine. Yeah, for sure, for computational purposes. So I don't think there's any way we could say that, you know, the singularity or the point beyond which we can't see will happen magically at the time when $1,000 is computing power.
No, we are. We is the only order of a human brain. Yeah. However, I think it's fascinating that the connection to models that had the breakthroughs they've had recently, like the deep learning is basically neural networks rebranded recently.
They've been doing better. And I don't know, I haven't looked into it far enough. But it seems to me that they're just kind of an inefficient way of doing things. But when you have enough computational power, suddenly this thing that was taking way too long to do doesn't take too long to do anymore.
Well, right. Right. And we've shown over and over again that we don't understand which problems are easy and which problems are hard. So that's the whole another thing.
Chests, we thought was hard. Right. It seems quintessential human. And we're finding more and more that the problems, the problems that humans are, anything that humans invented as a problem in the last, I don't know, 100,000 years seems really easy for computers to do.
And any older than that, it's harder. Much harder. Yeah. The stuff that's subconscious.
That's not our conscious mind. Right. In our brain, we have a visual cortex and that is big and it does a lot of stuff. It does a lot of stuff.
That's hard for computers to do. But things like strategy seem really easy for computers. So, so coming back to the answer to this question, I don't think, so I think what we're saying is neither one of us sees any reason to think that there's a fundamental wall that we're going to hit that's going to going to stop the train headed towards the availability of the human brain, the computational ability of human brain. No.
And we didn't even talk about how there's a positive feedback loop potentially that the more computing power we get, the easier it is to develop the next generation of whatever it is. You know, the next generation of neural networks, we have now the ability to use the computer power we have to develop those. Yeah, possibly. Yeah, there will be a point where that's certain.
You know, the first computers had to be developed with pen and paper. Now we develop computers between computers. Yeah. Right.
And so that process keeps going. The human genome project, you know, they were 1% done of sequencing the human genome after seven years of a 15 year project. And so I think this was one of those things all they'll never finish. It's the same as the bacteria in the bottle.
It's the it's the little pan thing. So where are we coming full circle? Where are we on the bacteria in the bottle thing? I don't know.
Are we 1%? Yeah. Are we at the point where? Okay.
So let's go back to the revisiting. 1130. Nothing's visible. Nothing's visible.
Now I wouldn't say that nothing's visible. We've got lots of press. We've got lots of articles all the time about or maybe 1130 microscopic 1130 still microscopic 1145 is starting to be visible. You can see it with the negative.
I would say we're certainly at 1145 or later. I feel like we are definitely 1145 or later. Yeah. Well, you know that that there's a visualization out there of how many cycles.
What is it? Can you can you tell us that the lake lake Michigan is Lake Superior. And it's the idea that if we were to have the computing power. I can't remember exactly how it is.
So the following of you have the full lake is has the same sort of number of drops in a number of drops of water that was the computing power that yeah operations of the brain can do. Yeah, for a second. So then you can visualize the progression to hey we've we don't at the moment we don't have the computational power of the brain and then there comes the point where we do and at that point the lake is full. That's right.
So in 1901, you know, we see one little drop. Not much happening. Next drop happens, you know, a decade later or something like that. Yeah.
So they have it they have it being completely full of 2025, which is pretty quick here. And at 2018, it's there's like there's just 10% full. It's just a little in the bottom. Just a little bit of this thing you're like, hey, we're not that close.
And so if these if this open AI, you know, those folks are seeing early week signals that could be if that we're 5% full, the bottle is 5% full, the lake is 5% full. Yeah, we I mean, the amount is 5% covered. I think it's it's realistic. It doesn't feel 100% to me because I still think linearly.
It doesn't feel 100% like right on it's what it's to me. But I suppose it's possible that somewhere between 2025 and 2035, yeah, we'll have the computational resources of the brain, brain human brain on like $1,000 computer. That's right. And today's knowledge.
Yeah. And I think anybody that says that that those advances in computing power, that it's impossible with those that it's impossible that superhuman AI will follow based on that level of computing power. I think that's folly. I think that there was an AI researcher who recently I heard him quoted and he said, worried about, worried about AI at this stage, dangerous AI, super intelligence is like we're going to about overpopulation on Mars.
Yeah. Yeah. And he's a PhD researcher in the field, well respected, working on all this stuff. I just I have a hard time really putting credence in that that just seems like head in the sand kind of it seems like linear thinking.
Yeah, I'm going to use that in a majority way. All right. Well, so to recap, sum it up here, take us out with a nice non pejorative, non pejorative sum it up. Yeah, some of the unexponentials, the exponentials are powerful.
And we don't really understand them when we try to think this through. And yet when we see that early signal when it's no longer microscopic, which is what we're seeing with AI, it's too late. And it seems to be too late at this stage to stop it. I would 100% agree that it feels like it's too late barring some other existential crisis occurring.
So we're going to have to explore all right, if it's well given that's not too late, then what do we do about it? Is it open? I can't open AI work. Are there other other things we can think of?
Or given that it's too late? Yeah, that it's coming. What can we do to make it smooth? How can we not turn the universe into paperclips?
That's really what I want. I'm just going to ride on your coattails. I'm going to want that too. All right.
All right. Very good. So we'll see you next time, I guess. We're done.
Yeah, thanks. Bye.