All About AI Data Centers ~ Conversational Normal Voice ASMR with Greg

[00:00:00] Linda: Hi, I'm Linda, and this is No Frills ASMR, and I have a special guest today. [00:00:12] Greg: Hello. [00:00:12] Linda: It's Greg. [00:00:13] Greg: Hey. [00:00:14] Linda: Back to talk about a subject that kind of overwhelms my brain a little bit. Yeah. So I said, how about we both go off and kind of learn a little bit and come back. Right. And that discussion will be about a, I don't know, maybe I shouldn't have capitalized that. [00:00:37] Greg: That's good. Yeah. [00:00:41] Linda: Data centers. [00:00:44] Greg: Yes. [00:00:45] Linda: So we're really talking more about, and we kind of decided this beforehand, not about AI itself and the AI models and how people can use them for good or bad. That's not really so much what we're getting into, right? [00:00:59] Greg: Maybe touching on it just a hair, but. [00:01:03] Linda: But really we're talking about the data centers they're putting in to run the AI. Right. [00:01:10] Greg: Yeah. Because there's a lot of controversy with an AI data center and where they're going and the environmental impact that they have on communities. That's a big thing. [00:01:20] Linda: Yeah. Well, I just said we wouldn't talk about it, but just one thing about the use of AI. Right. For my studying, a lot of times I go into the podcast apps. And this is true of like when I do Black Death or, um, what did I do recently? Well, the Donner Party watched a documentary, but I like go look up podcasts. So I put in to look up podcasts about the subject and I can't tell you how many podcasts now about any subject you start listening to. And you're like, there are people talking sometimes a female and a male person. And, but it just sounds funky. And then I'll go in and you, you can't even look at the episode. You have to go into the podcast itself, um, description. And if you go all the way down to the bottom, sometimes it'll say developed with AI or AI. I don't know. It'll give you some kind of hint that it's just AI generated podcast. And so I looked it up, I looked it up and 35 to 40, I think, percent of new podcasts coming out right now are AI. [00:02:35] Greg: AI generated. [00:02:36] Linda: I mean, not, not like somebody studied with AI. No, like the voices are like, it's, they can do them so fast. They just on subjects, but you can hear it. You can hear it right now, but this has only been happening for what? The last six months or so. I bet in a year I won't be able to tell. [00:02:57] Greg: Well, there's kind of two things with that. One is as they get better and better, you're not going to be able to distinguish between real people talking and AI. And the second thing is, is that how many people out there are not going to care? [00:03:12] Linda: Well, sure. [00:03:13] Greg: And I don't care if it's AI. [00:03:14] Linda: I mean, especially if it's really well done and just, I mean, yeah, I guess. I don't know. But that's what we're not getting into. We decided that's not ours. [00:03:24] Greg: But these, all these data centers that are popping up around the country and all around the world, they're huge. They're huge buildings. Oh, yeah. And they use an enormous amount of electricity and water and resources. And yeah, but do we really understand why? Why are they so big? [00:03:44] Linda: Right. Yeah. And a lot of them they're putting into old factories that have gone out of use, but they're using that footprint. And so kind of, I don't know if I'm saying this right, but kind of part of the selling point is like, hey, it's basically opening up that old factory and using it. But, and it'll create these jobs while they're remaking this factory. [00:04:15] Greg: While they're building it, right. [00:04:16] Linda: But the problem is it creates short-term jobs while they put in all the processors and get that all set up. But to run the thing, even though it's huge, might not take hardly any employees at all. [00:04:31] Greg: Right. It, it can take hundreds or thousands of workers to build an AI factory. [00:04:39] Linda: Yeah. [00:04:39] Greg: But once it's all done, it's only like, you know, 50 or 100 people it takes to operate it. [00:04:46] Linda: I mean. [00:04:47] Greg: But you're talking about AI factories or AI centers, data centers going into. Old factories. Yeah. I mean, it's an obvious reason why they do that. The, the land is there. Yeah. Good size land. And it's the infrastructure that's already there. All right. So because the factory was there, you had this huge infrastructure of available electricity, water, fiber optic connections. It's already there. [00:05:14] Linda: And I'm sure they get it cheap if nobody's using it. Yeah. [00:05:18] Greg: But a lot of times these factories are in, you know, urban areas or near residential areas. So that's part of it. [00:05:27] Linda: Yeah. Well, how about if we, I don't know if you have a plan, but one thing I was looking up is what, okay, what is all this energy actually going towards with the AI? [00:05:41] Greg: Right. [00:05:41] Linda: And it looks like the main, number one thing, most of the electricity is going towards is training the models. [00:05:52] Greg: That's true. You have two things. You have training and inference. Training. What does that mean? Training is when you're training the model, you're feeding it just tons of data and it's learning how to use it. Inference is after the AI model has learned everything. [00:06:12] Linda: Well, when I saw it broken down, it gave one, training, and two, running it, which is what you're saying, inference. Yeah, they call it inference. [00:06:22] Greg: It's basically that's, you know, you asking a question and it, doing the calculations and sending it back after it's already been trained. But, you know, 95% of AI is processing mathematical calculations. [00:06:40] Linda: Really? Like what? Engineering things? Like why? I don't ever ask it to do math. [00:06:46] Greg: No, no. It's calculating mathematical calculations. The other 5% is like moving data around and writing data to storage and managing memory and stuff like that. I don't know. I don't know. Well, I can explain that. You know, a modern AI processor, they call them GPUs. Okay, yes, yes. Which stands for graphic processing units. Right, right, right, yes. And that actually came from video games. So, back in the 1990s and 2000s, you had companies like NVIDIA and AMD. Right. They were designing these computer chips to run graphics for video games. So, their goal was to make them as realistic as possible and to run faster. [00:07:39] Linda: Right. [00:07:40] Greg: They realized that these processors, which were really good at creating realistic images, they were also extremely good at doing large amounts of mathematical calculations simultaneously. So, you know, the development of video games did not, you know, it did invent AI, but it did allow AI to advance much faster. [00:08:14] Linda: Well, there's a lot of money in video games, I'm sure. Right. So, it probably wouldn't have happened without video games. Interesting. [00:08:21] Greg: So, the name stuck. Right. GPU, graphic processing units, even though it's mostly just doing mathematical calculations. Right. There are other types. A GPU is, kind of does general calculations. There's a number of other AI computer chips that do more specialized calculations. So, you're wondering, why is it just, why is it math? Why is AI all math? Well, if you ask a simple question, like, well, what we think is a simple question, like, why is the sky blue? You know, what actually happens to produce that answer? Well, the first thing it does is it takes the question and it converts it into digital data, binary code, which is ones and zeros. So, as an example, why is this guy blue? The word blue, lowercase b, it turns into, like, an eight-digit binary number. So, actually, for the lowercase b, I looked this up, the number is 01100010. [00:09:43] Speaker ?: Okay. [00:09:45] Greg: That is the code for the lowercase b. [00:09:48] Linda: Okay. [00:09:49] Greg: And so, every letter, every character, every number has its own code. [00:09:55] Linda: Interesting. Okay. [00:09:57] Greg: So, then, it sends this coded message through the internet, mostly fiber optic cables. And it goes to the nearest AI data center. [00:10:13] Linda: Huh. [00:10:14] Greg: Okay. So, let's say somebody in Boston. [00:10:17] Linda: Hmm. Clark. It could be Clark. Yeah. [00:10:21] Greg: Asking their phone, why is the sky blue? Yeah. It takes the question, converts it into binary code, sends it through mostly fiber optic cables to the nearest data center. Sometimes a little, it could be, usually, probably in Massachusetts, although it could go to a data center in Pennsylvania, or there's a lot in Northern Virginia. [00:10:42] Linda: I mean, if it's just dealing with code, does it, does it matter? Like, you know what I'm saying? Like, does that take time? Or is it instantaneous? [00:10:52] Greg: It's very, very fast. Right. But typically, it's locally where it goes. It could go thousands of miles away. It could go halfway around the world. [00:11:02] Linda: I mean, I ask this because it's actually, all of a sudden, to me, became an interesting question. Because you may have a city, like New York City, that's using way more AI than out in the middle of Wyoming. But their data centers, well, actually, Oregon, they're also some data center. So, like, are they using their electricity to run their AI? I hadn't thought about that until just this minute. [00:11:26] Greg: I think. [00:11:27] Linda: Or if you start doing it. [00:11:29] Greg: Wherever the AI data center is located, in that specific town, that's where the electricity and resources are being used. But typically, the question goes to the most closest data center. It could go all the way across the country, like I said. [00:11:49] Linda: Or across the world. It could. It's very fast. [00:11:52] Greg: But it's faster because there's latency the further it goes away. So, it's even faster to go. I see. [00:11:59] Linda: Okay. [00:11:59] Greg: Now, depending on things like the availability of that data center, it may reroute it to a different one. You know? [00:12:13] Linda: Mm-hmm. [00:12:14] Greg: So, then it, depending on the capacity or the network routing, it then performs these billions of calculations to figure out what the answer is. And it determines the most likely response. And then it rearranges them back into a language that we understand and sends it back. [00:12:39] Linda: It's unbelievable. So. And it does that in two seconds. [00:12:42] Greg: It can be two seconds. It can be a fraction of a second. [00:12:45] Linda: I mean, yeah. That's wild. [00:12:47] Greg: Yeah. [00:12:48] Linda: Interesting. [00:12:49] Greg: It's figuring out, like, probability. Yeah. Like, when it's coming up with a sentence that we can understand, it takes one word and says, okay, what options is the next word? What could it possibly be? And it runs all these mathematical calculations and says, well, this is probably the most likely next word. What's the next word after that? So on and so on and so on. And then when it comes up with a whole sentence or paragraph, it runs the whole thing and says, does this make sense? It double checks its work a couple times. [00:13:22] Linda: Doesn't mean it's right all the time. [00:13:24] Greg: No, it doesn't. Sends it back to you. And this is all within a fraction of a second. [00:13:30] Linda: All right. [00:13:32] Greg: So. [00:13:33] Linda: All right. So that's the training and the running and how it does that. So the running is like any time you ask it a question. That's whatever word you used. I wasn't really familiar with. What did she say? Inference. Inference. Okay. And then they said the third thing that the data centers are drawing. That they need the power for is cloud computing. Right. Which is probably what you're saying when it sends it off. Right. Or is that something else? I don't know. [00:14:03] Greg: Yeah. You know, even though these modern AI processors are really fast. And I mean, they can do trillions of calculations per second in a chip that you can hold in your hand. It's still limited. Even though it's trillions of calculations per second. It's still limited. That's why we need so many of them. Now, all of their computer chips were CPUs, which stand for central processing unit. And they typically work by, they do a calculation and then do another calculation after that. So it's sequential, one after another. [00:14:46] Linda: Right. [00:14:46] Greg: But what makes AI so fast and these GPUs is what's called parallel processing. So it happens inside the chip and also outside the chip in the whole data center. So it takes the question, it breaks it down into smaller bits. And thousands of computers will work on the different parts of the question at the same time, put it back together, make sense, and send it back to you. So it's parallel processing. It's doing it all at the same time. So you have, and they're all linked together. So you've got thousands of chips working together as one big chip almost. And that's why it's so fast. Yeah, that's. [00:15:40] Linda: All right, let me finish my list. Oh, go ahead. [00:15:42] Greg: Go ahead. You have a list. [00:15:44] Linda: I'm still working on the. [00:15:48] Speaker ?: Okay. [00:15:48] Linda: The list of what these AI centers need all the power for. So the first thing is training them, then running them. Cloud computing. And then they said scientific and engineering, which I would have thought would have been first. And they said that's like drug discoveries, weather predictions, engineering simulations, which seems pretty useful to me. Research computing. So that's the next thing. And then the last one is government and military. But the percentage that they're actually using, they don't publicly tell you. Right. So we're not really sure about that one. It could be way more. Right. So that's all I have to say about that. [00:16:37] Greg: You know, most of the Internet information is moved by fiber optics. Right. Only like the last mile is Wi-Fi. [00:16:49] Linda: Right. [00:16:49] Greg: When you're talking to your phone. But most Internet information goes around the world, the fiber optics. I find fiber optics kind of interesting. And real briefly, the idea of fiber optics came out in the 1960s. It became more practical in the 1970s when phone companies started replacing copper wires with fiber optics. Fiber optics is basically a, it's transmitting digital data through thin strands of glass with pulses of light. So every little pulse of light is, you know, part of this code, ones and zeros. And these, you know, strands of glass, they're only like the size of a human hair. And there can be dozens or hundreds or even thousands of them bundled together. So they move these pulses of light, which is the digital data, through these fiber optic cables. And we're talking about the speed of light. [00:17:57] Linda: Right. [00:17:58] Greg: Now, as we know, the speed of light in a vacuum is about 186,000 miles per second. As we know. Okay. [00:18:10] Speaker ?: I'm a nerd. [00:18:12] Greg: 186,000 miles per second. But when it goes through a fiber optic cable, the glass slows down the light. It's only about two-thirds the speed of light. So it's going through there about 125,000 miles per second. [00:18:31] Linda: Wow. [00:18:31] Greg: Which, by the way, at that speed, it will go around the earth five times in one second. [00:18:38] Linda: Oh, my God. As we knew. [00:18:41] Greg: And another interesting thing is they have these fiber optic cables that are undersea cables. Right. Like between the U.S. and, like, say, Europe. [00:18:51] Linda: Mm-hmm. [00:18:52] Greg: There are these, they're laying on the ocean floor. And I think, like, right now there's about 20 active undersea cables. [00:19:01] Linda: Okay. [00:19:03] Greg: And it's such a long distance that, of course, the signal gets weakened over time. So over long distances, every so often they have these things called repeaters or boosters. Mm-hmm. So they take the signal, they boost it, and then send it out. [00:19:17] Linda: Oh, I see. Okay. [00:19:18] Greg: So it's interesting how that's part of the reason why the internet is so fast is because of fiber optics. [00:19:27] Linda: Yeah. Interesting. They used to have commercials about that, and they'd show you the thing with the glass tubes in it. Right. When they were trying to push fiber optics years ago, you know, to every home. [00:19:39] Greg: Right. Anyway. You know, the internet is built on redundancy. So if a data center were to go down or a fiber optic cables were to get damaged, they just reroute the information. [00:19:52] Linda: Mm-hmm. Okay. [00:19:53] Greg: It's kind of like a traffic jam. Yeah. You might take an alternate route, take you a little bit longer. [00:19:58] Linda: Yeah. [00:19:58] Greg: But it is redundant because you kind of wonder what would happen if a undersea fiber optics cable got damaged or if a data center went down. They would just reroute the information to somewhere else. [00:20:12] Linda: Mm-hmm. [00:20:13] Greg: So it's redundant. [00:20:17] Linda: Right. Yeah. Yeah. So it's not, yeah, it's like not like electricity where once it's used, it's used. You know what I mean? You can keep sending it. It doesn't, I don't know what I'm saying. Like if you send water through a tube and the tube got broken, that water is gone. But this you can just keep sending, right? Okay, so look, I've traced five states. You know why? Ohio. Yeah. Georgia. California. Texas. Virginia. I believe that they're the five states with the most data centers right now. Virginia in particular has a ton. I was listening to a podcast with, what's her name from the 1980s movie, Erin Brockovich. And she's interesting. I'll show you guys a website in a minute that she set up that I thought was cool. But she was talking about how Virginia is like a hotspot for data centers right now. [00:21:21] Greg: I've heard that, yeah. Northern Virginia, yeah, especially. [00:21:24] Linda: Yeah. [00:21:25] Speaker ?: Yeah. Go ahead. [00:21:28] Greg: When we talk about just the physical size of these data centers. First off, an AI processor, it's only a few inches square. I mean, you can hold it, one in your hand. Right. [00:21:42] Linda: So you wonder, why can't you put a GPU right on your telephone? Good God, what am I a hundred years old? On your phone. You know what I mean? You can. [00:21:50] Greg: You can. They do have, they're localized AI computers. They can put on your personal computer. And in the future, they'll be on your mobile device. Right. [00:22:01] Linda: That's the future. [00:22:03] Greg: Yes. And some people argue, well, we won't need so many data centers in the future. [00:22:08] Linda: Right, that's right. [00:22:09] Greg: Because we're going to have all these individual AI computer chips on our devices. The issue is, is that the chips on your personal devices are only operating, doing inference. So they're just using the information that they already have learned. So you still have to train. Right. You still need data centers to train new models and to improve existing models. [00:22:41] Linda: Interesting. Okay. And that's the biggest energy drain. And the biggest buildings they need is for training, from what I understood. [00:22:50] Speaker ?: So. [00:22:50] Greg: Right. [00:22:50] Linda: Interesting. [00:22:51] Greg: So also, when we talk about, like, just the physical size. So we talk about an AI chip you can hold in your hand. Yeah. They're put into a circuit board, which is roughly the size of a large textbook. [00:23:05] Linda: Yeah. [00:23:06] Greg: But besides the computer chip, you have got in there, you've got high-speed memory and the power hardware to run it, cooling systems and network connections are on there. Yeah. And then you take the circuit board and you put it in to, or actually, yeah, you took the circuit board and you put it into a server. [00:23:26] Linda: Mm-hmm. [00:23:27] Greg: So a server may have, like, four or eight circuit boards. Then you put the server in a rack. And a rack could have dozens of servers. And the size of a rack, they could be there seven feet tall and two feet wide and about three or four feet deep. And then each data center has thousands of racks. Hmm. Okay. That's, they're so big because we need so many chips working in parallel processing. We need thousands of them. [00:24:08] Linda: Right. Okay. [00:24:10] Greg: And that's why they end up becoming so. It's starting to make sense. That's why it takes up so much physical space. Right. The chip is small. Right. But it's got to go into a circuit board. Circuit board goes. [00:24:18] Linda: They have a little chip. It goes into a circuit board. The circuit boards, they put tons of them into a rack. [00:24:24] Greg: A server. Into a server. [00:24:25] Linda: Oh, into a server. [00:24:27] Greg: Right. There's four or eight circuit boards in a server. [00:24:32] Linda: Okay. [00:24:32] Greg: And then there's dozens of servers in a rack. [00:24:37] Speaker ?: Okay. [00:24:38] Linda: So then you put these, like, all of these things. And by this point, how big are we? [00:24:42] Greg: The rack? Mm-hmm. It's, like, about seven feet tall, two feet wide, four feet deep. That's one, right. So then you have a huge data center, which is a massive building, that has thousands of these. I see. And they're all connected together. [00:24:59] Linda: Right. [00:25:00] Greg: And you mentioned earlier about AI data center complexes. Yes. That's where you have multiple buildings. Right. Instead of just having one building, you have, like, four or six buildings. But they're all connected together, too. They just happen to be separate buildings. [00:25:19] Linda: Now, well... Okay. No, I don't... What part of this is using up so much of the... What? And I know the answer to this, just so you guys are. But the electricity, and then there can... Okay. Let's... Should we talk about what the concerns are a little bit? Well... And then you can tell us how it works. Or do you have a thought? [00:25:41] Greg: I'm not... Well, we can first talk about why they use so much electricity. [00:25:44] Linda: Okay. Okay. [00:25:45] Greg: Okay. Well, inside a AI processor, you've got billions of transistors. [00:25:53] Linda: This is this part. The processor is this little dot in our little thing. The little chip. Yeah, that's the GPU. [00:25:58] Greg: That's the GPU. Okay. Which you can hold in the palm of your hand. All right. [00:26:02] Linda: Okay. [00:26:02] Greg: Inside of there, there's billions of transistors. [00:26:05] Linda: Okay. [00:26:06] Greg: Which basically turn on and turn off billions of times per second. [00:26:13] Linda: Wow. [00:26:13] Greg: And this is what's doing the mathematical calculations. And that requires electricity. Okay. To turn every one of those transistors off and on requires electricity. That's an electric bulb right there, guys. So, a typical modern AI processing chip uses about 700 to 1,200 watts of electricity. [00:26:42] Linda: Okay. Give me a comp for that. Like what else? A hair dryer. Okay. [00:26:46] Speaker ?: Like a... [00:26:47] Linda: Per what, though? Per... [00:26:48] Greg: One chip. [00:26:49] Linda: Per minute? Per second? Per day? Like what is that? [00:26:52] Greg: It's using that much watt. It's running. It's like plugging in your hair dryer. But leaving it running constantly. Or... Right. Leaving it running. Oh, wow. Okay. So, imagine a building where you've got thousands of hair dryers... Oh, my God. ...plugged in, just running full glass... Running all the time. ...24-7. [00:27:08] Linda: Wow. Okay. That's actually pretty good. Okay. I mean, not pretty good. [00:27:12] Greg: They use different amounts of electricity depending on the model of the chip and also the workload. Right. So, they just use... Each individual chip uses so much electricity. [00:27:25] Linda: Right. Okay. [00:27:27] Greg: So, now, then the second part of the using electricity... Yeah. ...or using it is cooling it. Right. Yes. Because these things get hot. Right. Because... Enough heat. Because, right. Because some of this energy that's used to turn these transistors on and off turns into heat. Right. So, these AI chips heat up. They heat up... That's heat. One chip can heat up to like 140 degrees or 180 degrees or more. Okay. You can't hold it in your hand. It's too hot. And that is, of course, too hot for the AI chip to operate. It's too high of a temperature. So, you have to reduce the heat. So, you have to cool it. Right. So, now we get into the cooling aspect of it. Okay. [00:28:20] Linda: Yeah. [00:28:22] Speaker ?: Hold on. [00:28:24] Linda: Let me get a board. Is this something you can draw? [00:28:30] Greg: Sure. You can draw. [00:28:32] Linda: You're better at art. [00:28:34] Greg: So, every AI chip is connected to what's called a heat sink. [00:28:45] Speaker ?: Sorry. Hold on. I'm trying to get it. [00:28:48] Greg: Trying to get a board out. [00:28:51] Linda: That was really out. Sorry. People don't know. People are waking up on that. Sorry about that. Okay. [00:28:59] Speaker ?: Oh, my God. I'm sorry. [00:29:01] Linda: Okay. So, what do you think? [00:29:03] Greg: A AI computer chip is connected to what's called a heat sink. [00:29:07] Linda: Okay. Wait a minute. Now, you're saying each chip is connected. [00:29:10] Greg: Yes. Each. [00:29:11] Linda: Not. [00:29:11] Greg: Each individual chip. [00:29:12] Linda: So, within all this, each one of these has its own cooling thing. [00:29:17] Greg: Right. [00:29:17] Linda: Okay. [00:29:18] Greg: It's a heat sink. And what it is, it's a metal block that's attached to the chip, which absorbs and spreads out the heat. [00:29:28] Linda: What do we imagine that looks like? A square around it? [00:29:31] Greg: Um, they're like, kind of like metal fins and it's about this big and they've got, you know, metal, you know. I don't know. I'd have to, you know. Well, like if your computer chip, you know, it's got like these, these metal, you know, pieces of metal, you know. Okay. I don't know exactly like that. It's, it's called a, you know, a heat sink. So, it, it transfers the heat into this metal thing and spreads it out. [00:30:02] Linda: Okay. [00:30:03] Greg: So, now you've got two different main types of cooling systems. You've got air cooled and liquid cooled. Uh, most older chips, um, use air cooling, which is basically, you run a fan across it. [00:30:19] Linda: Okay. [00:30:20] Greg: So, it just runs air. And air blows across it. Yeah. That's more like a windmill. And that, and that cools it. Okay. Now, this is also making the data center itself warm. So, you need air conditioning to also help keep the building cool. So, that's using electricity. So, these fans are using electricity. The AC to cool the building is using electricity. But, more and more modern data centers are using what's called liquid cooling. Okay. Because liquid, um. It's good at it. It removes heat much more efficiently than air. Yeah. And so, they have this thing where water runs through it. And, there's a couple different ways of, now you have this hot water. [00:31:13] Linda: Mm-hmm. [00:31:13] Greg: So, you run these water through there and you have to cool the water. And there's a couple different ways of doing it. Most commonly is cooling towers, which is used as evaporation. [00:31:26] Linda: Okay. [00:31:26] Greg: So, they take this warm water, they pump it outside, and they put it in these big cooling towers. And it cools by evaporation. [00:31:35] Linda: Okay. [00:31:35] Greg: So, when it evaporates, it cools. And that's where you lose a lot of water. [00:31:42] Linda: Mm-hmm. [00:31:43] Greg: Because you have to replace that water that's evaporated. That's why they start using so much water. [00:31:49] Linda: They don't catch the evaporated water like a distilled water situation. I don't think so. [00:31:54] Greg: I think it just goes. Okay. It just goes in the atmosphere. [00:31:57] Linda: Mm-hmm. Okay. [00:31:58] Greg: It's very efficient, but it uses water. [00:32:01] Linda: Efficient at cooling. [00:32:02] Greg: Efficient at cooling. Right. But you have to replace the water. Right. Now, how much water are they using? [00:32:08] Linda: Mm-hmm. [00:32:08] Greg: Well, a typical large data center would use like 300,000 gallons of water a day. [00:32:14] Linda: Wow. [00:32:15] Greg: That's through evaporation. That's equivalent to about 1,000 homes. [00:32:20] Linda: Okay. [00:32:20] Greg: Would use that. And then an even bigger data center could use 1 to 5 million gallons a day, which is what a small town would use. [00:32:32] Linda: Well, you know, right before we started this video, an AP article came across my desk, which means my phone. Anyway, and they're building a million square foot data center in Southern California. Mm-hmm. And there was discussion in the comments, so I don't know if any of this is true because comment sections could be full of whatever. But they were talking about where the water is coming from, and they said they're going to take it from, what's the river, Colorado River? Yeah. And it'll just dry. Like, they already have water issues. I don't know. So it's like, yeah, I mean, that could be a problem if you're taking out all this water all the time. I mean, I know it stays in the atmosphere, but it moves places. [00:33:21] Greg: It comes down as rain somewhere else. Yeah. Yeah. Anyway. Another type of cooling method is kind of called dry cooling or, like, radiators. [00:33:31] Linda: How many cooling methods are we going to have? I didn't leave myself much space. [00:33:35] Greg: Just mean two main ones. [00:33:37] Linda: Okay. Dry. What's it called? [00:33:39] Greg: Dry coolers or radiators. They're called heat exchangers. Okay. Depending on which, how you want to. It's a closed loop heat exchange. So the water doesn't leave the system. It doesn't evaporate. The water, it's a closed loop system. So the water stays in there. And it goes through these little heat exchangers and fans run across them to cool them. It's, think of a radiator in a car. Yes. Your car cooling system doesn't lose coolant, if it's working correctly. [00:34:11] Linda: Mm-hmm. [00:34:12] Greg: And air blows over those little fins that the coolant is, and it cools it and then sends it back in. [00:34:20] Linda: So it's like in a pipe or something, it goes through, and then the fans blow on it. And when the fans blow on it, it gets... [00:34:31] Greg: It removes the heat. [00:34:32] Linda: It gets cool. Now... And then it goes back over all the stuff, and it gets hot again. So... So it just goes around. [00:34:40] Greg: That system is a closed loop system, so it uses little or no water. [00:34:45] Linda: Mm-hmm. [00:34:45] Greg: But it's less efficient, especially in hot weather. [00:34:49] Linda: Okay. Yeah. [00:34:51] Greg: Places like... [00:34:52] Linda: You couldn't boil the air over and have it be that cool. [00:34:54] Greg: You know, in like California and the desert, plus it. So it's less... It's less efficient. It uses less water, but I think the system uses more electricity. [00:35:12] Linda: Oh, yeah. Well, I guess so, because you'd have to run more fans. Right. And pump it. [00:35:18] Greg: Mm-hmm. [00:35:19] Linda: So there are a few things there. [00:35:21] Greg: And then there's hybrid systems, which uses a combination of both. [00:35:26] Linda: Okay. [00:35:27] Greg: But no matter what system you use, you still have to have air conditioning for these big buildings. Right. So they have to keep the air inside the building at a certain temperature. So whether you use liquid cooling or air cooling, you still need AC. Right. And you know, when people talk about living near a data center and it having noise, like a low hum, this noise, that noise is almost completely the cooling system. [00:35:58] Linda: Mm-hmm. [00:35:59] Greg: Because a computer chip itself is basically quiet. Right. And all the computer processing is quiet. It's the cooling system. It's the fans running. It's these heat exchangers. It's the air conditioning units. That is what makes all the noise and the humming. [00:36:18] Linda: Well, talking about that, you know, the complaints that I saw were, yeah, noise is a big one. Like, people, like, when these things go in, it can be so loud, it's, like, deafening. Which you can imagine, because, like, we got a new air conditioner in our house. And we tried to find a more quiet one. But it's stinking loud. We built, I did a video on it, but we built, like, a wall around it. Mm-hmm. Because it's so loud. But imagine if you had thousands of those running 24-7, too, nonstop. Don't click on and off. [00:36:50] Greg: It's noise pollution. [00:36:51] Linda: It's noise pollution. So I could totally understand that being an issue. [00:36:58] Greg: Now, some people could say, well, why don't we build these? [00:37:02] Linda: Well, the other one is we already talked about water usage. [00:37:04] Greg: Yeah, go ahead. Why don't we build these data centers, like, in a really cold climate? [00:37:08] Linda: Right, right, right, right. [00:37:09] Greg: Because cooling is such a big issue. [00:37:11] Linda: Yeah, okay. [00:37:13] Greg: Because it can reduce the cooling cost if it's in a cold climate. [00:37:17] Linda: Yeah. [00:37:17] Greg: But the issue is infrastructure again. You have to build these data centers where they have access to electricity, water, fiber optics, connections, and all that. So you can't just build them out in the middle of the, you know. Right. Interesting. Same thing without in the desert. [00:37:38] Linda: Yeah. [00:37:39] Greg: You know, you think, well, why don't you just put these things out in the middle of the desert, out somewhere in Arizona or Nevada where nobody lives? [00:37:46] Linda: So they tend to go closer to population, just because that's where the infrastructure is. Because of the infrastructure. Okay. [00:37:51] Greg: That's where the electricity, the water, and the fiber optic connections are. [00:37:55] Linda: Right. [00:37:56] Greg: That's why. [00:37:57] Linda: That makes sense. Well, one thing I read about, and maybe you can tell me about this, is the, not just the water usage, but that they're putting these, and I don't know why, but they put, like, different chemicals and things into the water. And, like, the forever chemicals get in the water, I guess, from running through the air conditioning systems, maybe. I don't know. And then they're releasing the water out. So the water is getting polluted. Did you see that anywhere, or has that been hyped up? I don't know. [00:38:28] Greg: Well, I do know, they don't just run pure distilled water through these systems. They do add some chemicals, which have anti-corrosion properties to them. Yeah. You know, it helps to, you know, preserve the systems and keep them from corroding. So I don't know what type of chemicals they use. [00:38:53] Linda: Well, one of the things said those PFAs, which are, like, forever chemicals. Right. They've tested, and they've had those in the water. Yeah. And then they mentioned some other, I don't know, minerals and stuff. So. [00:39:07] Greg: But when you have an evaporative system, when water evaporates, it leaves all those minerals and chemicals behind. [00:39:15] Linda: Mm-hmm. [00:39:15] Greg: But then you have to do something with them. [00:39:17] Linda: Right. [00:39:18] Greg: So they build up. [00:39:19] Linda: Mm-hmm. So. [00:39:22] Greg: Hmm. Another thing that these data centers have are backup generators. Oh. They have to have backup power. Yes. I heard about this. Okay. And they're typically diesel generators. [00:39:35] Linda: Right. That's part of the pollution part of it. Right. Yes. [00:39:41] Greg: Well, here's the thing about that is they're not typically used because they only need them if there's a power outage. Sure. Or they're testing them. But when they are testing them and have a power outage, these things are noisy, and they put out pollution because they're running off of diesel. [00:40:04] Linda: Right. [00:40:05] Greg: But they have to have them in case electricity goes down. They can temporarily use these diesel generators. So that's an issue, too. [00:40:12] Linda: Right. Yes. They, on the, I think it was the Erin Brockovich podcast that I was listening to where she was talking at, they said that in polling right now, that has been done multiple times, multiple ways, so it seems to be legit polling. People are more worried about data centers than they are nuclear power. Not more, they're more favorable towards nuclear power, which is interesting because that used to be such a, you know, thing. But now people are realizing we're going to need nuclear in order to power these things. And Microsoft apparently just recently said they want to try to get Three Mile Island, Three Mile Island, back up and running for their own data center. So they're talking more and more about these companies having, not using the power grid, but having their own electric, you know, capabilities next to these. They won't be on the towns. [00:41:11] Greg: They're like a small nuclear reactor. [00:41:15] Linda: Well, some of them are talking small, but he's talking Three Mile Island. But I also saw, yeah, these smaller ones. They're trying to develop. [00:41:21] Greg: They're called small modular reactors. Yes. Yeah. And that's an interesting thing because if you put aside for a second the safety concerns, the public concerns, the nuclear waste concerns, you can put those aside just for a second and think about using nuclear energy to power these data centers. It's extremely clean energy. [00:41:50] Linda: Well, as long as it doesn't get it by her. [00:41:54] Greg: Right. Yeah. It would be an excellent way to power these data centers, but people have memories of Three Mile Island and Chernobyl. [00:42:06] Linda: A nuclear scientist would say Three Mile Island was a success. That it never actually, the core started to melt down, but it worked. They kept it in check. But other people would say, no, it took billions of dollars to clean up after that. So you could, I'll have to do a video on Three Mile Island. I kind of did a mini deep dive, but I never made a video. [00:42:29] Greg: You know, talking about these small nuclear things, you know, the U.S. Navy has been safely operating nuclear-powered submarines. [00:42:38] Linda: That's true. [00:42:38] Greg: Nuclear-powered aircraft carriers for decades. Yeah, that's true. It can't be done. It's a fear. The first nuclear-powered sub was 1954. [00:42:48] Linda: Yeah. [00:42:48] Greg: It was the USS Nautilus. I did not know that. And the first nuclear-powered aircraft carrier was the USS Enterprise. [00:42:57] Linda: No. [00:42:59] Greg: In 1961. [00:42:59] Linda: That's from Star Trek. [00:43:01] Greg: It was called the Enterprise. Okay. In 1961, they're both retired now, but the Enterprise ran for 50 years. [00:43:08] Linda: Okay. [00:43:10] Greg: But, you know, they're getting, you know, nuclear power is getting better and better. Yeah. Do you know, like, France, for instance. France gets a lot of their electricity from nuclear power. As a matter of fact, they generate 60 to 70 percent of their electricity is produced by nuclear power. Yeah. And they've had a very good, excellent safety record for many years. Yeah. Now, compare that to the U.S., we only get about 18 or 20 percent of our power from nuclear power. [00:43:42] Linda: Right. Well, part of that's because, well, I won't even say because that'll sound political. By the way, this is a subject that is truly bipartisan. You know, when people show up at these town hall meetings, it's the most bipartisan group of people who are like, I don't want this thing. You know, and they said that a lot of the city commissioners who are working on building these things have signed NDAs with the companies, these non-disclosures. And they've known about the data centers, some of these city council members, for like two years. But the public didn't find out until two months ago. It's all pretty sketchy. I tell you, when I was looking into it and listening to Aaron Brockovich, it's pretty sketchy. [00:44:30] Greg: Well, here's a couple of things that we haven't really talked too much about is when you put a data center in a city and it starts using so much electricity. It puts a huge drain on the electrical grid of that area. And when there's a higher demand for electricity, electric rates go up. And so the local people are paying more for electricity because the data center next door. [00:44:58] Linda: Right. [00:44:59] Greg: That's a huge thing. [00:45:01] Linda: I mean, think about that for a town. So you're not really getting very much job-wise. You're getting to pay higher electric rates, maybe water. I don't know. Rates. You're, and getting to, I mean, sarcastically. You have noise pollution and maybe water, but we don't know what other pollution, actually. [00:45:22] Greg: Well, you, you know about, you know about the water when, when these centers use more water. Yeah. The, what was it? Water table. The water table can go down. Yeah. And when the water table goes down, groundwater. Yeah. Um, it, what, you can get, um, naturally get pollution in your water. Well, one thing is. Lower water pressure. [00:45:45] Linda: Especially in some, you know, more rural areas where they use well water, their wells are drying up. Completely dried up. Or they're just full of sediment because they've just suddenly lowered the water tables. But for people who live with well water, which by the way, a lot of people do, it can really screw things up for them. And people sometimes use that kind of water for their farms and stuff too. Right. And then you don't have access, you know, it's, it's pretty devastating right there. And do these, I didn't look up whether these guys pay as much taxes as, I don't know. Are they helping out the communities? [00:46:23] Speaker ?: I don't know. [00:46:24] Greg: I think a lot of these big tech companies are getting tax breaks. That's what I would think. On building these centers in certain areas. [00:46:31] Linda: Yeah. I would think. [00:46:33] Greg: There's sort of an ironic, uh, thing that as we move into the future, these data centers will become more efficient. They will, they will use less water. They'll develop more improved cooling systems. But what's the irony is, is that the total electricity will continue to rise because the demand for AI will increase. [00:47:02] Linda: Unless they can somehow make it more, it doesn't heat up as much, more, what's the word I'm looking for? [00:47:11] Greg: Cooling. [00:47:12] Linda: Like just set these GPUs so that if they can create a system, I don't know where they don't tend to get as hot. [00:47:19] Greg: Well, I don't know. I mean, these things run on electricity and that produces heat. So they're going to get hot and you got to cool them. I don't know. You can, you can make more efficient cooling systems. [00:47:32] Linda: Maybe we should ask AI what to do about this situation. [00:47:36] Greg: But here's the thing. As, you know, as AI develops, people will find more ways to use it. [00:47:46] Linda: Oh, absolutely. Yeah. [00:47:48] Greg: You know, so it's going to just continue to grow. [00:47:51] Linda: Yeah. Oh, yeah. [00:47:53] Speaker ?: Yeah. [00:47:55] Greg: And we'll always need these huge data centers because they need them for training new AI models. [00:48:02] Linda: Yeah. [00:48:04] Greg: But it's kind of interesting going all the way back to the beginning of this discussion about how AI works. Yeah. And how it, you know, answers a question that you prompted to. It's all mathematical calculations. [00:48:17] Linda: I did not. Yeah, that's wild. [00:48:19] Greg: It's predicting. [00:48:20] Linda: I mean, I guess that's been true of computers forever, isn't it? I don't know. [00:48:25] Greg: It's figuring out what each word means. Yeah. How they relate to each other. [00:48:31] Linda: Yeah. [00:48:32] Greg: Connecting concepts. And they do it by mathematical calculations. That's crazy. It's all with data. That's how it works. [00:48:42] Linda: You know, before, right before we recorded this, you showed me a quick video of video games that are, like, ultra realistic. Right. And you can choose the directions you want. I mean, Greg and I are not gamers, so forgive me if this is, like, yeah, duh. But, like, that's where AI, like you said at the beginning, comes from. And that's kind of interesting, isn't it? That creating these games that are, like, ultra realistic and you can move around and make decisions and the AI's got to keep up with you, I guess. I don't know. [00:49:13] Greg: It's kind of interesting. I mean, they're incredibly realistic. Yeah. And that's all from those processors who can do all that graphic stuff so fast. [00:49:23] Linda: That's wild. I wanted to show something. [00:49:26] Greg: So, yeah, I don't know. You know, nobody wants an AI data center in their neighborhood. Right. But, like I said, it's the infrastructure. Yeah. It's where the electricity and the water is. [00:49:43] Linda: I was just going to show this. This is BrockovichDataCenter.com. So, Erin Brockovich in the, I think it was the 80s, she, there was a movie made about her with Julia Roberts. She won the Academy Award for it, I think. But it was a good movie. And she was, like, found out there was, like, this area, and I forget where now, but where there were just a lot of people who were getting cancer. I think a lot of kids, wasn't it? And she kind of started putting the clues together. [00:50:15] Greg: It was in Massachusetts, wasn't it? [00:50:16] Linda: I don't remember. I feel like it was. It was, like, one of the very first lawsuits that went after a company for something like that, I think. And the biggest payout at that time ever was, like, $5 million or something. But, um, so that's kind of her claim to fame. And they made a movie about her, and it's good. But she has started really looking into this data center stuff. And she was saying how, because these companies, like, you know, I don't know who, Grok, OpenAI, not Microsoft anymore, although they did. Amazon's a big one. Google's a big one. But they were making people sign NDAs, nondisclosure agreements. It's about, so there was very little information about these data centers coming along and where they're building. And sort of they started looking at who was buying up the most GPUs, and that gives them an idea of who's building data centers. That's part of how they can know who's building data centers. Like, that's how secret it is. Like, they don't have to tell people, I guess. I don't know. This whole story is kind of interesting. But, so anyway, she just started doing this, what do you call it, like, asking people in the world, there's a name for that, crowdsourcing, to understand where they are. So she has people, if you have one by you, you can put it on the map. So there's community reported, proposed, under construction, operational. And then she even put a little thing on here for drought conditions to kind of keep track of weather. Because there are some people who think, and this is not documented, I don't think, at this point, that these things might actually change the weather in an area by three degrees. They have some reporting that shows it increased the air temperature by three degrees in areas where they are. So they're kind of doing these studies on that, too. But you can look at this map and see where they are. It's kind of interesting, if you're interested in that kind of thing. It is interesting. I don't know. [00:52:32] Greg: I don't know how you can reduce the amount of electricity that AI computer chips use. Yeah. Because they use what they use. [00:52:41] Linda: Yeah. [00:52:41] Greg: It's creating the heat and then trying to cool it. That uses a lot of resources. [00:52:47] Linda: Yeah. [00:52:48] Greg: That's a big thing, so. [00:52:51] Linda: I'm not getting enough Wi-Fi. Hold on. Talk about having old school. I don't have AI. I don't know. Let me open the doors. I want to look at this. [00:53:00] Speaker ?: That's great. [00:53:02] Linda: I think if I open the door, I'll get the Wi-Fi. Is that crazy? That's like when I hold my phone up, the boys make fun of me. They're like, Mom, that doesn't work. [00:53:10] Greg: You know, like 75% of most AI data centers are smaller ones. [00:53:19] Linda: Yeah. [00:53:19] Greg: And then like 20% are larger ones. And I think like 5% are like the monster ones. Let's just kind of put that into perspective. [00:53:32] Linda: Well, I was going to look at the key concerns with AI data centers on here, but it won't load. [00:53:39] Greg: But anyway, I'm going to look at it. I think we've kind of, I mean, that's the ones, the noise, the electricity. [00:53:46] Linda: It was the noise, it was the electricity, the water usage, e-waste, I think was one of them. Location risk, which has to do with possibly heating up and drying out an area, like taking all the water if you're in the desert or heating it up anymore. And then the other one was scalability, but I'm not, I was going to look at what that meant because I'm not sure. [00:54:09] Greg: Well, I don't know. I also read something that's kind of interesting. These data centers have really high security. [00:54:19] Linda: Oh, really? [00:54:21] Greg: Oh my gosh. Cameras everywhere. You have to have badges. [00:54:25] Linda: Because they think people will just vandalize them or what's the reason? [00:54:29] Greg: You have to have like, buy, there's a real security. [00:54:32] Linda: Are these things super expensive? People steal them? What's the reason? [00:54:37] Greg: They're because of, you know, terrorism. [00:54:42] Linda: Okay. [00:54:43] Greg: You know, everything. Vandalism, terrorism. Well, that's what it is. They're secure. They're very secure. Extremely secure. [00:54:52] Linda: Well, I guess I'm curious. Are they secure because of terrorism, vandalism? Are they secure because if you were to get a hold of these things, you could sell them for a whole lot of money? [00:55:01] Greg: I don't think. [00:55:01] Linda: Is it the value of the GPU? [00:55:03] Greg: No one's going to walk out of a rack of GPUs in the trunk of their car. Yeah. I don't think that. It's for security reasons. [00:55:16] Linda: Interesting. All right. I give up. That won't load. Anyway, do you feel like we touched on most stuff here? [00:55:26] Greg: You showed that map that showed how many of them there are around. Yeah. You know, we've talked about why we need so many. [00:55:32] Linda: Yeah. [00:55:32] Greg: But it's also, it helps with the speed of the internet and the speed of AI by having them closer. Even though we're talking about extremely fast and data's moved around extremely fast with fiber optics, it's still faster to send the information to a data center that's 20 miles away than to send one that's, you know, 6,000 miles away. [00:56:03] Linda: Right. [00:56:05] Greg: It's just a fraction of a second faster. [00:56:08] Linda: Yeah. Yeah. [00:56:10] Greg: Yeah. But nobody wants one in their backyard. [00:56:16] Linda: No, I wouldn't. [00:56:18] Greg: Well, also, they're ugly, too. Hmm. [00:56:20] Linda: Yes. You know, I have some people think they're ugly. I have seen one, actually. There's one not far from here, and it just looks like a giant, nondescript metal building. [00:56:34] Greg: I guess I shouldn't say they're ugly. They're just not. [00:56:38] Linda: They're ugly. You can say they're ugly. [00:56:40] Greg: They're not beautiful buildings. They're very functional. [00:56:45] Linda: No, they don't. Yeah. [00:56:47] Greg: Yeah. [00:56:48] Speaker ?: Yeah. [00:56:50] Greg: Well, it's interesting where this is going with these. [00:56:56] Linda: Well, there's no stopping in this. [00:56:57] Greg: Well, these tech companies are investing billions and billions of dollars. Yeah. In billions of dollars. They can't keep up with them. [00:57:05] Linda: Yeah. [00:57:10] Greg: Yeah. So, the demand for AI is going to continue. Yeah. So, we were talking about, we mentioned about some of the videos going around of the college graduation ceremony. Oh, yeah, yeah, yeah. They're booing people. Commencement speeches. Yeah. Where you've got these tech people and talking about the greatness of AI and the future of AI, and they're being booed. Yeah. And I had one person come and it's like, oh, that younger people hate AI. They don't hate AI itself because they use AI. [00:57:54] Linda: And how else would they have cheated their way through college? [00:57:57] Greg: That's how they got through college, right. And the video games they play, they love AI. It's all about the threat it is for their careers. Of course. I mean, half those people sitting in that room are probably going, AI is going to take over my career, if not tomorrow, pretty soon. [00:58:14] Linda: Can you imagine if you got like a degree in graphic design? [00:58:17] Greg: Oh, my gosh. [00:58:17] Linda: Forget about it. Even engineering, I don't know, you know. [00:58:24] Greg: It depends on what your degree is, right? [00:58:26] Linda: It's like I showed the cartoon the other day. I was talking about, I saw a cartoon and it was of the Jetsons, which was, you know this, but in case people don't, a 1950s, 60s. [00:58:39] Greg: 1960s, yeah. Cartoon show. [00:58:41] Linda: Early 1960s. That was cute. And George Jetson and his wife, Judy, were living in this future where they had a robot named Rosie who would vacuum and do all the cooking and cleaning and they're living the good life. Well, this cartoon said the Jetsons lied and it was Rosie the robot doing the art and George Jetson vacuuming the rock, which is kind of true. They get to do all the fun art stuff. They get all the fun jobs. And by they, I mean AI. [00:59:11] Greg: I know. What was that quote that I read somewhere that someone said, I don't want AI to do my creative work so I have more time to do the dishes and wash clothes. Right. I want AI to do the dishes and wash clothes so I have more time to do my creative work. Yeah. [00:59:32] Linda: Yeah. Oh my goodness. [00:59:36] Greg: Yeah. [00:59:38] Linda: We'll all be out of business. Well, I tell you, YouTube is constantly telling me to use AI to make shorts. They say, I can just upload, like I could take a picture of this and upload it and they will, I don't know if they use my voice or what, but they'll make shorts for me in AI. Isn't that crazy? I'm like, no thanks. [01:00:03] Greg: Well, let's say, and this is, you know, I don't know if this has happened quite just yet or will happen. If you have a movie that is 100% created by AI, the script, the actors, everything is AI, would you go see it? [01:00:20] Linda: I wouldn't. Just because it rubs me the wrong way. [01:00:24] Greg: I know. [01:00:25] Linda: I just couldn't. It's the same thing with the, like I said, the podcast. As soon as I register their AI, I'm like, oh, turn that off, turn that off. Like, I don't want to give it the time of day. [01:00:35] Greg: I remember the first time we heard about AI DJs. [01:00:39] Linda: Oh yeah. [01:00:39] Greg: A few years ago. Where they, yeah, in between songs, they, they sound like a real DJ. They introduce the next song and, and then these DJs can be tailored to your own music. Yeah. Like you pick your music you like and it's sort of like a personal DJ. Yeah. That was, and we started hearing that actually. [01:00:59] Linda: Well, it was like the day we heard about it, we went into this big antique store that plays country music and it was like, you, you went to the bathroom. I'm sitting there listening. I go, this is an AI DJ just because it made the dumbest jokes that didn't make sense. That's right. But that was five years ago. I'm sure now they're way better than that. [01:01:17] Greg: They say things that just seem slightly odd. It was just so weird. That a human wouldn't say. [01:01:21] Linda: We were like, what? [01:01:22] Greg: But you know what? They'll, they'll get so good that you won't. Well, they probably already are. You won't even tell. That puts DJs out of work. [01:01:29] Speaker ?: Yeah. [01:01:30] Linda: I hope not. DJs have kind of been out of work, I guess. Radio's not. [01:01:35] Greg: Well, what about when the AI can, can write funnier jokes than humans? Oh my God. Do stand, do stand up comedy, stand up comedy. [01:01:43] Linda: Stand up comedy? [01:01:44] Greg: Committee. [01:01:45] Linda: Committee. Well, that is, there's this TV show that I watch, you didn't watch it, but The Comeback with Lisa Goudreau. Yeah. And this latest one was that she was so desperate for work and they hired an AI writer assistant. But like, there'd be a joke that didn't work and she'd go, can you tell the writers to work on that joke? And he'd put it in the computer and literally 30 seconds later, he'd have 50 jokes for her. And she's like, how did you do that that quick? It did it that quick. And he goes, well, I stalled. I didn't want you to think it was. So he like waited a minute instead of 30 seconds. But anyway. [01:02:24] Greg: They're all AI jokes. [01:02:25] Linda: Yeah. And I think there's a lot of reality to that. [01:02:29] Greg: AI can explain jokes. [01:02:32] Linda: Yeah. [01:02:32] Greg: Tell you why they're funny. [01:02:34] Linda: If I have to be told why a joke's funny, maybe it's not that funny. [01:02:39] Greg: Oh, it gets into a real... If I need AI to tell me. It gets into a real detailed, analytical way. [01:02:46] Linda: But you know what? I don't even know how to say this without... It's like tough because if you're a songwriter and you're jammed up, like you've got to have a song ready for the album. [01:02:58] Greg: You've got writer's blog. [01:02:59] Linda: You've got writer's blog. And you use AI at that one time. It's almost like taking a drug probably because you use it. It gives you something. You know, just give me prompts. Don't write the song. Just give me ideas. And it gives you some good ideas. How do you not keep using that in the future? You know what I mean? And is that the worst thing? I don't know. I don't know. [01:03:17] Greg: Well, here's an interesting thing. I think I told you about this. You know, I studied playwriting in college. Yeah. And wrote this play. And I was just curious about... I've never used AI for this. Of course, I'm not writing anymore. But I remember the first scene of the play. And so I went to AI and said, I've got this idea for a scene for a play. These are the character names. This is the location. This is the situation. This is sort of the conflict. This is what they're doing. And just gave them some prompts. I said, could you write some dialogue? And boom, in two seconds, it comes up with this script. And I read it. And I was like, this is like 60 or 70% of what I wrote. What? Yeah. [01:04:12] Linda: That's wild. [01:04:16] Greg: Very similar. And actually, it had a lot of lines in it that were really good. [01:04:21] Linda: You're like, that's better. [01:04:22] Greg: Yes. Like, ooh, that's a good line. [01:04:24] Linda: I mean, why wouldn't it be, though? It has the whole of the world's knowledge to work with. It can look at Sam Shepard and Pinter and who else? I don't know. Shakespeare. Shakespeare. Yeah. It comes back with, no, I don't know. [01:04:39] Speaker ?: I don't know. [01:04:39] Greg: I mean, to me, that was one of the hardest parts about writing was just coming up with ideas. [01:04:47] Linda: Yes, that's what I'm saying. That's really hard. [01:04:49] Greg: And writing a first draft was tough. Yes, yes. So if you have an idea for a play and then you feed it into AI and say, this is my idea. Yes. They spit out a first draft and you go, okay. Then you go through it and say, well, I like this. I like this. And rearrange this. [01:05:03] Linda: But here's the problem. Here's what can become very difficult, I think, is if you get that first draft and all of a sudden you go, well, I can't do better than this. You know? Like, you might think, oh, well, then I'll put my human twist on it. But the truth is, you might read it and go, good God, I can't think of a better line than that. That's a great line. You know what I'm saying? [01:05:23] Greg: I think you would just go through it and pick out, oh, I like this part and I like that part and then rewrite a few things. Or you can just tell AI, I don't like this part here. Or could you make the character a little bit more of this? I don't know. [01:05:35] Linda: I just feel like it's such a, it's slippery. You know, I don't know. I don't know. It's tough. I just think AI might be a better writer than most of us. So you're going to have it help you and all of a sudden it's writing it. And now all of a sudden you're just an AI writer. And not you, obviously. You're not writing anymore and you're like, whoops. But you know what I'm saying? It's, I don't know. It's tempting. [01:06:00] Greg: I remember writing, you know, papers and stuff in college and like struggling over writing a paragraph going, how do I word this paragraph and struggling over the wording of it. [01:06:12] Linda: Wait a minute. I remember in college going to the library to look up, you know, stuff for papers and you're reading the thing and you're like, well, I can't phrase that better, but you know, you can't plagiarize. So you're like changing two words in the sentence to try to make it your own words. But really, it's like, I mean, I can't say it better than that. You know what I'm saying? But now it's like a hundred times worse or better. I don't know. I guess it depends on how you look at it. It's tough, man. It's tough. When I first started memberships, they said I needed to make, what do you call them? Like emojis or whatever they are. I don't know. Badges. Badges. And YouTube said, we'll help you with the badges. You know, just put in what you kind of want the badge to look like and we'll help you do that. And I'm not an artist, so I was like, oh, cool. So I just put in, you know, I was talking about owls at the time. So I said, can you do an owl, a sleepy owl with headphones? In two seconds, it made the cutest little sleepy owl with headphones. I was like, that's so cute. So I added that badge and I had like three different ones and within 10 seconds, people are like, how could you do that? Those are AI, shame on you, blah, blah, blah. I was like, oh my God. You know, I didn't even, at that point, understand, I think, that it was AI. I just thought it was like, what did I, like YouTube had this fun little thing you could use. I didn't really think about the fact that by doing that, I'm taking away work from people. Although, as it turns out, I draw little stick figures and that's what they get now because I'm not paying anybody. I don't have that kind of budget. [01:07:58] Greg: When you train an AI model to draw an owl, you feed thousands and thousands of photographs of an owl into the computer and it reads them all and it goes, okay, this is what an owl looks like. What do you want? And it creates his own. [01:08:16] Linda: And I say sleepy owl with headphones and it made me a cartoon with color in two seconds. And it made that all binary, sent it all out, parallel processed it all, put it back together. [01:08:31] Greg: I'm amazed at how well AI is at looking at a photograph. [01:08:38] Linda: Oh, yeah. Oh, yeah. No, it is good. And recognizing it. Can I, okay, this is something I have used AI for myself is fixing things like, you know, I was working on a record player actually just yesterday and I just couldn't quite figure out one of the wiring things. So I took a picture of the bottom of the thing with the wires and it came back and it's like, that's award 1970s and I'm like, [01:09:04] Speaker ?: wow, wow. [01:09:05] Greg: It can't, it can recognize that. [01:09:07] Linda: Wow. And it's like, here's probably what it is that's wrong with it. I'm like, oh my God. It's so crazy. And guess what? It was right. [01:09:16] Greg: Well, I don't know. When we become more and more reliant upon AI, I don't know how you go back. [01:09:22] Linda: You can't, that's the thing. And now, like people say, I don't use AI. Well, if you're Googling things right now, AI is, it gives you a summary of everything that's found. That very first thing, that's AI. So, and, I don't know, I don't know if that there's much you can do right now. Well, [01:09:40] Greg: I mean, people have been like Googling stuff for a long time. Right. so before when you Googled, it would bring up a list of websites and you'd go into each one and read an article about the subject and then you'd have to kind of figure, oh, this is a good one. This is not a good article. This is good information. And you had to kind of figure it out all on your own. AI goes through all of those websites and pulls all the information and summarizes it. [01:10:07] Linda: although sometimes it's pulling from Reddit and it's like in the comments and I'm going, I don't know if that was a great source, but okay. [01:10:16] Greg: Well, you can, it'll show you the sources if you want. [01:10:19] Linda: It does. I mean, mine does. Mine shows the sources on Google, I think. Yeah, yeah. What is the Google one called? Oh, I know this. I know this. Wait. I'm blanking it but I didn't know that. [01:10:35] Greg: If you, Oh, Gemini. I was going to say Gemini. So when you hit the source, does it take you to that source? [01:10:44] Linda: Yeah. [01:10:45] Greg: See? [01:10:45] Linda: Yeah, it does. [01:10:47] Greg: Yeah. [01:10:48] Linda: So the big, the big ones are chat, GPT. That's open AI. Gemini, which is Google. There's Claude, which is the, um, anthropic. There's Grok, which is X. It's Twitter's. I don't know it's owned by Twitter, but it's, you know, Elon Musk. Right. And it has some safeguards taken off or at least it did. So sometimes it'll spew out racist stuff and things like that. I think the Grok one, I think. And then Meta is Facebook's. And then there are others too. But those are like the big. [01:11:27] Greg: If Meta, Facebook is called Meta? [01:11:29] Linda: Well, Facebook, like if you buy stock on Facebook, it's Meta. Right. So Meta's there. It's, well, they were going to do this whole like world. Meta, I forget what they're going to call it, where you'd wear, got the face things and you'd talk to people. Virtual reality. Yes, that's the word. But I think that didn't happen. So Meta is Facebook, but also Meta is their AI, I think. I don't, I'm not exactly sure. I think that's right. Like Facebook uses AI now. I think. Oh, everybody does. Well, I don't know. I mean. [01:12:04] Greg: We haven't even got into the dangers of AI. Well, [01:12:07] Linda: that's a different video, isn't it? [01:12:09] Greg: I think it is. [01:12:10] Linda: That sounded British, [01:12:11] Speaker ?: isn't it? [01:12:13] Linda: I think we have to do, I mean, this whole second half of this turned into the video I said we weren't doing today. [01:12:19] Greg: We're just rambling right now. [01:12:21] Linda: Well, we'll do another video sometime on just. [01:12:24] Greg: The potential dangers of AI. Where it could go. [01:12:27] Linda: Yeah, especially, I mean, there are a lot of things we've got. All right, let's do a different video because I got to look into it more. [01:12:33] Greg: Yeah. Today was just talking about data centers and what they are and why they're so power hungry. [01:12:41] Linda: I wonder how many people like watching have data centers near them even. You know, we live in such a [01:12:49] Greg: built up urban area. There isn't enough room. There's no space. [01:12:54] Linda: Right where we are. But you go out just. [01:12:57] Greg: Yeah. [01:12:57] Linda: You know, where Mark lives. That's where one is. Out in that area. They're open areas and old factories and stuff. They can build them. Oh, yeah. One thing I also read is a lot of these data centers because they're where old factories are. that is a lot of times [01:13:13] Speaker ?: that is a lot of times [01:13:13] Linda: also not the, you know, it's kind of the poor side of town. So the people who kind of need their electric bills to go up and pull any more pollution are the ones getting it. Wait, that was a weird way of saying that, but you know what I'm saying? Like they go into. [01:13:31] Greg: They're the ones who are suffering from it. [01:13:33] Linda: Yeah. Or, yeah. Yeah. Typical. The rich people throw the stuff in the neighborhoods that are the poorest and go suck it up. [01:13:47] Greg: Yep. [01:13:49] Speaker ?: Wow. [01:13:51] Linda: All right, guys. Let us know. Let us know in the comments if you have one near you and if you hear it. And I'm curious. I've never listened to one. I heard there's, there's one in Arizona that's near like a hiking trail and they said it's so loud. You wonder how that affects animals. [01:14:09] Greg: It's all the cooling units. It's going. Can you imagine Arizona? [01:14:12] Linda: How much cooling you'd have to do? Also, they don't have water. How could they do that? Anyway. All right. I'll talk to you guys later. Okay. [01:14:23] Speaker ?: Bye. [01:14:25] Linda: That was an hour and 14 minutes. [01:14:28] Speaker ?: Holy smokes. Bye.