Can AI Solve the Energy Problem It Is Creating?

Hey, what's going on everybody? Welcome to The Daily AI Show. Today is August 20th, 2026, episode 790. Just 10 more to go. So which means two Fridays from now we will be celebrating our 800th episode. It's pretty crazy to think about, especially as we always say, that means we have done a live show with at least one to two of us. I don't think we've ever had to do one for too long. Although Andy, I think you had to hold down the fort one time for a few minutes before- Yeah, it wasn't very long. Before reserves came in. So we're proud of both those cases that we've always had a conversation about AI Monday through Friday. So if this is the first time you ever hear us, or this is the first time you've seen us, because we know a lot of you are, hey, welcome. Right now you have Andy and myself, I'm Brian. But oftentimes we'll have other people popping in the door as well. Beth might be around today, Gareth and some others. So yeah, that's what's going on. Don't forget tomorrow before we get into today's show. We'll have the Conundrum episode that comes out every Saturday at 10am Eastern. And then on Sundays, we have our newsletter that goes out at 8am Eastern. So we have a brand new website that we would love for you to go check out. It's got some really cool features on it. That's the dailyai show.com or dailyai show.com. You don't have to put the in front of it. It'll still take you to the website. But the reason I bring it up is because directly from that homepage, as well as some other areas, you can sign up for the newsletter. You don't really have to do anything for the Conundrum. It comes out on the same. It doesn't go out on YouTube. It goes out on Spotify and any of our podcast platforms. So I suppose there is a difference because it's not video oriented. So we just stopped putting it on YouTube. But you can also sign up for the newsletter. And you can also sign up for our community there. So if you just want to be part of the conversation, but you're listening to us at some other part of the day, that's not 10am Eastern. And for this next hour, then we don't want to miss out on you being part of the community. So come join us in our Slack community. And that's the easiest way to do that. Okay, that's all the stuff. Want to make sure you guys what's going on because we always have busy weekends, even when we're not doing the live shows. But with that said, and my rapid fire there, Andy, let me turn it back over to you, especially because we did say now Beth came in and Gareth came in yesterday. But I promised on Wednesday that we were going to yield the floor to Senator Andy, who was going to take us through some stuff because Andy, as you said, will be out next week. And so let's get a chance to get our fill of Andy news and topics. So the floor is yours. Okay, I want to do a thesis on power. And, you know, I talked briefly earlier in the week about the difference between what China is doing to improve their electric grid and their power sources, and doing orders of magnitude, additional wind and solar and battery storage, and just making a big difference in their infrastructure that supports computing. And now, you know, you've probably heard in the general press about the pushback on how AI and AI data centers are really putting an enormous strain on the existing grid. So there's a need for additional power. And there's also a need for improving the efficiency of data centers in delivering compute for the power consumed. So I wanted to talk about a program or a development project that's underway right now by NVIDIA, Google and Microsoft. And what it is, is they've been developing a transition from using alternating current AC power as the main source of power delivered into the data center to 800 volt DC. So this is a big project that's designing and building all of the parts and pieces that software and the hardware that are needed to convert data centers from AC power to DC power. Why? Because while DC power, while happening at a very high voltage, which imposes some risks to the humans who are around that system, it distributes the power at a higher voltage through a direct current. And that means fewer conversion cycles inside the system because at the end of the day, the chips actually use DC power. So you have to have transformation from AC to DC. And when you make all those switches, that's inefficient. It does lose power. So more of the available power that's being delivered by generation will get to the chips and the compute result if you switch to this DC system. I think it's true. And I didn't do any research on this, but there are some countries that have, you know, kind of shifted already in general to DC power. There must be very small countries because otherwise grids are interconnected in many ways, even across state lines. In Europe, for example, in our system, of course, nationally, our states have interconnected grids. So you can power from a nearby state as necessary if you don't have local generation. And many states don't have much generation in their own state at all. And so they rely on national and then ultimately in other more dense state environments like Europe, they have cooperative agreements that make that happen. But there is a need ultimately for, you know, some cohabitation of and co-location of DC and AC power sources like that. So that's one bit of the power news. We've talked also about SpaceX's idea about how they're going to solve the power problem with solar. But solar's big problem is it doesn't work at night, right? You have to store the generation during the nighttime. But SpaceX's idea is let's put all of the solar arrays for compute infrastructure into space. And that means continuous solar power generation so that all of these are in solar synchronous orbits. So they're following the sun around the Earth and then sending their resulting responses from compute back down to Earth so that they don't actually draw from any of the power generation on Earth at all. They generate all the power necessary for all the compute infrastructure by a network of satellites and tens of thousands of satellites doing that work with continuous solar power. Now, there's the other piece of power. True quantum computing. And they've applied that quantum computing to some fundamental questions and problems that have to be solved with fusion power in order for that to become available. So I want to talk a little bit about that fusion problem. Now, just so everybody understands, fission, nuclear fission is where you split apart an atom and that generates energy. And fusion is where you take, and in the most likely pairing, you take tritium, which is an isotope. I forget, it's an isotope of something. Anyway, you take tritium and deuterium, which is another isotope, and you take those atoms and you push them together. And the byproducts of that are energy, a free neutron that flies out, and then helium, an isotope of helium. So it's kind of clean in that respect because there is radiation that's generated from this, but it's kind of controllable when you manage effectively this combination of two different atoms, one, the tritium, and the other one, the deuterium. So that's typically called DT fusion for deuterium, tritium fusion. All right. Now, here's the thing. When you use up the tritium, the reactor, in order to be sustained in its operations, it has to take the power of that released neutron, the one that comes out of that fusion process, and using a blanket that's capturing the power of that thing, converts that to new tritium. So it's generating tritium at the same time as it's consuming tritium in the process. So that's the way you have to kind of put this whole package together in order for fusion to work. Okay. So all of those, the complications of managing that in a very hot environment, right? Because in the fusion thing, we were talking about hot plasma. So this is not a, you know, stovetop kind of thing. And there's very, very low likelihood of anybody developing what's called cold fusion, which is done without this very high temperature environment. Okay. So the U.S. Department of Energy is cooperating with IBM and others to try to build a roadmap for how to get to DT in reactor design. So mutterium, tritium, fusion. So in May of this year, IBM announced that scientists used a quantum computing model that is used typically like to figure out how to attach proteins in water and that sort of thing. And it's used that to design a model that could not be accomplished by classical computing that models what's actually happening in this complex structure that has to capture the power that's released, not only for power generation, but also this free neutron has to be captured by the blanket that's surrounding the reactor. And turning that substance back into additional tritium fuel for continuous operation. Okay. So that's this whole approach. What's really important about what IBM has contributed to this is this part about capturing that neutron. Like how do you do that and turn that in the fuel cycle for DT? How do you turn that into additional tritium? Hmm. Anyway, kind of interesting. We're not close in my view to getting to fusion reactors, but IBM and their quantum computing component. Of course, quantum computing isn't being done without AI at this point. Right. Sure. So it's an AI story, at least at that level. And it's very importantly an AI story because fusion power ultimately will be a solution to the increasing demand for electricity globally for AI. Yeah. To me, it always feels like there's a lot to hate and be upset in local communities about data centers. So this is not a defending data centers. But the demand for the data centers and meaning the global demand for AI is pushing that sort of race to the moon. And, you know, it's pushing a lot of new dollars into the energy space. And I agree with you that, like, I don't know when, you know, I mean, the joke of that is like it's 15 years away. Like that's what you always hear in nuclear energy. Like that solution is 15 years or 30 years away, whatever the joke is. But it never changes. It never the number never gets shorter. But I am very, very hopeful. And I mean, you're bringing up a good point about how it's sort of this like circular loop of, well, the data centers are, you know, in fact, hopefully making it possible to use more powerful AI. More powerful AI is being used in these quantum computers or in tandem with that, which is then circling back and hopefully helping us discover net new ways of creating efficient power. And then the other part of that, too, is like data centers are becoming more efficient. The old ones aren't. So it doesn't change anything for the ones that are loudly humming by people's houses and draining energy and water. Not going to debate that. But we do have closed loop systems that are already in production and are going to be in production that NVIDIA is a part of, too, where it's much more like a it's like a radiator, right, where the water is filled once and that's it forever. So much, much different than what some of the current models are. Again, it doesn't change anything for the communities that already have the older ones. But there's perhaps hope that we can solve this. Every front of the technology and there's different components of it. Every one of them has to be advanced in order to make this practical and efficient and economic. I wanted to just mention finally that that blanket that I was referring to does have another kind of constraint on it because that blanket is lithium. There's a lithium layer in the structure of a fusion reactor. And lithium is in high demand, of course, because of lithium batteries. And on the battery side, of course, we're moving towards sodium ion batteries, which are longer lasting and recharge faster than lithium ion batteries. But so maybe all the lithium that's being mined and it's a rare, rare earth resource. Right. All that lithium may have to go to nuclear reactors that are fusion oriented. I was going to say Beth is here. Hi, Beth. Hi. I'm here. I think one of the subtexts of this conversation that we haven't named is that there is a profit motive in which sources of energy are being pushed. Right. China did their whole solar like, we're just going to solar blanket wherever they did it. And it took 18 billion, which sounds like a lot, but not governmentally and not AIE. Right. Like 18 billion, man. Just like top that off. It's dropping the bucket for, you know, the investments that are being made. Maybe don't do 0.7 and just go to 0.9 and we'll just take that 18 billion off and do this. But there isn't the same kind of profit motive in solar energy. And there isn't the same kind of profit, like inherent in it design in wind energy. Halifax is doing a lot of research in tidal energy because of the Great Bay of Fundy. And I think we should add that to the bingo card because I'm fascinated by the Bay of Fundy. I think that's the largest tidal drop, you know, in the world. Right. The Bay of Fundy does that. And so like there are those places that exist naturally in a particular region. Ethiopia, I think it is, has places where water drops because it's mountainous and has a big drop. And that can be a source of energy as well. Like there are a bunch of conversations that really excite me that I feel like are dropping out of this conversation because there isn't the same kind of we developed a really great new technology. And it's going to make the investors tons of money as opposed to, hey, community, you have a great resource here. And here's what we want to do with it. That's going to benefit AI because that's what we're talking about. But your electricity bills are going to go down by a third or a half. Yeah. We are going to clean up. We're going to add a filter for water so the water's cleaner now when it comes out. Then when it went in, we're going to capture the heat. Like talk to people about where they live and what's great about where they live as opposed to, oh, we bribed your people. And now we're just going to like stick a bunch of propane diesel engines. People are right to be, you know, skeptical. People are right to be pissed, honestly, because they more than likely lie to. You know, we have many, many, many in the States and I'm sure other places we have many stories, you know, from the big ones like Flint, Michigan to others that where, you know, people were lied to the, you know, the environment was tainted in some way and these people are living with long-term consequences, including all sorts of diseases and things like that. So when people, you know, hear people say right now, I get it when they're like, trust me that the AI data center is good. And I'd be like, good, go build it somewhere else. You know, like if it's so great, I bet you there's another community that would love to have it, just not mine, you know. And that's coming from a guy who, you know, sits on an AI show five days a week. So as they are built right now and knowing the low regulations that happen, I do not, I would not want a data center built in my county at all. Because I don't believe that it would be done with either the closed loop system or whatever the case may be. And I think there's a lot of, you know, there's a lot of talking in that. So, and by the way, before we move off, because it's sort of adjacent, it's a weave off of energy, excuse me, and data centers. But one of the news stories I actually had today was that India, probably not a shop to a lot of people, is a growing area for data centers. But India has an energy problem, more or less an energy infrastructure problem. And so what I think you're going to see more and more here is as there's more pushback in like the US and areas like that, like a lot of times these projects are going to shift to other countries that perhaps don't have the ability to, you know, slow down the production of them, that take even more resources from people who have less ability to share that resource, you know, whatever the case is. So I do worry about that because, you know, one of the news stories was like, you know, yeah, they're like expanding into India. But the problem is like their high electricity volume and grid build out is inferior. And so how do you how do you balance both of those? You can't just say, oh, we can build cheaper and faster in India, but not not on the system they have now. So what does that mean for the folks in India? So anyway, just want to bring it up because that was part of it. Gareth joined us as well. Hey, Gareth. So now we've got a team of four here to go along with our Friday. I guess I want to bring it up because like this is coming from our our excuse me, our chat. Thank you. I was like channel chat something from Cisco guys friend of mine who I got to subscribe to desk. Thank you, Cisco. And who got me into Hermes just said that GLM 5.3 was just released and it blows everything away. Even the fab not so fab five hot off the press. And Scott responded by saying breaking news. So I went and checked that because that didn't actually come up. And that was actually nine hours ago. If we look at Z or Z dot AI, depending on how you say it. Um, this says it was posted nine hours ago. GM GLM 5.3 frontier coding with emergent cyber capabilities. And basically it says they were releasing 5.3. It uses the same base model as 5.2. This is GLM 5.3 just to keep everybody. They're throwing a lot of numbers and stuff. This is Z dot AI's GLM 5.2 model. Every game comes from post training compared with the GLM 5.2 is much better at complex coding and long horizon tasks. Wherever we heard that before, Andy, I feel like this whole week, the word of the week is or the words of the week are long horizon tasks or long horizon agents. You know, uh, it says it's better at strong coding. It's a stronger coding. Yep. That's pretty, pretty normal. Uh, emergent cyber capability. Let's just see what they say that is. Cause I wouldn't know how to define that as we scaled post training cyber capability developed faster than we expected. GLM 5.3 is a state of the art on cyber gym for vulnerability discovery and it gains and its gains are largest further up the exploration chain where it more than doubles GLM 5.2 on exploitation benchmarks. And they say open source. We will release the weights in two weeks after launch once safety evaluation and are complete. So there you go. Thank you Cisco, uh, for bringing that news. I don't know if that was already on somebody's agenda to talk about, but that was the first I heard about it. So see, this is why we love our community. This is why you should join the chat if you can, because, uh, we will bring it into the discussion. Did any of you guys have that on your bingo cards for say? No. No, but if you're coming to the chat to look at the amazing things, you should also, uh, check out the wide variety of jokes. Uh, stove top. What, uh, Trillium and Trillium when they get added, what do they create? Uh, there's a whole range of, uh, wonderful jokes. And if you're listening to this after the fact, you have to wait 24 hours, but then go to live chat as it happened and you'll be able to see it. Uh, yeah, like there, I love it. I love when our chat just goes off the rails and stops talking about whatever we're talking about. And then they start to have their own chat, which is always funny to me because I'll look over there and I'll be like, oh, they're still talking about cows. That makes sense. It's not even something we're talking about, but it's gone all off that direction. So we, we love our chat very much. Um, okay. Just want to bring that up because I want to make sure we did not bury that. Thank you again to Cisco back over to you guys. Uh, Beth, Gareth, you guys both joined after Andy and I started. So, um, news that you guys perhaps wanted to share. I have another one. Um, I look at the beginning of the show. I said, this is an Andy day. We got to get our Andy stories in before the week. I don't have that many, but I do have another one. And yeah, I didn't hear anybody jump in. So, um, you know, I used lovable extensively, uh, you know, in the early part of this year. And since then I've switched to Claude code and, uh, codex. And I'm using both of those simultaneously. Uh, the lovable, uh, I, I was worried about lovable because they're just such a great product. And I thought, well, you know, they're probably being, you know, sort of obsoleted by what's going on with Claude code and codex and it rock build now. Right. So you've got all these different players at the frontier level who, you know, are probably going to make it less likely that people are going to run over to lovable. Well, I don't have to worry about that at all because lovable just raised $400 million at double the valuation that they raised 500 million at the beginning of the year. Uh, and that's around $13.3 billion, the valuation of lovable. And what's driving that is that they have in their hosted system, right? You build a product and then it is hosted by lovable as well. Um, in that hosted system, they have 60 million projects and those 60 million projects are exposed to the web and have users. And there are 900 million monthly visitors to those projects. That's there's, there's not 900 million building the 60 million projects. There's 900 million users out there who are using the projects built by the people who built the 60 million projects on lovable. So it's doing very, very well. Uh, and that they're, um, what was surprising to me. And when I wanted to comment specifically about this is that I never heard about startup venture capital companies, putting this scale of funding in at a single round. Now this is a series C round, uh, and it was led by Menlo ventures as was their prior round where they raised 500 million. So $900 million has been added to the operational coffers of lovable. Mostly contributed by Menlo ventures. That's, uh, they weren't alone in that. There was a bunch of other players who put in money as well, which can imagine that Menlo probably invested in some major AI plays and they've had some returns, you know, and, and they have a big, big treasure chest. And now they're putting a big, big bet, an ongoing continued bet on lovable. So don't, don't hesitate to use lovable. Yeah. All right. Uh, so, um, I don't know what you're going to do with this on your bingo card, but, uh, I have a firefighter story. Oh, wow. Really? I'm locked in. So this was posted in the Hermes agent subreddit and it's by Gryknight9, G-R-Y-N-I-G-H-T-9. And I will post the link in the community Slack. Um, and they say I was headed into a deployment window over the Spokane fires and wanted a way to keep a reliable low noise pulse on the incident while I was gone. So this is a Hermes partner Spokane firewatch, and it's a deployment prep story. Basically what they did was they took the available, um, uh, free or, or free for these folks, API feeds of, uh, of information and used Hermes to aggregate them, assess whether anything had changed and only communicate it when there was a difference that somebody needed to know. Uh, it pulls, um, uh, so it's a small watch pipeline. In C web is the source of truth for acreage containment and official incident updates for old trails, autumn leaves and fair view. Those are places. Um, uh, the, the Washington DVA resource page is the best consolidated view of red cross shelter status, pet friendly hours, info evacuee phone, evacuee phone line. Like as you, as you go through this, it's identifying what they use as the best, uh, condensed force of crew and then using Hermes to be the reasoning, proactive, um, decision maker about what needs to get communicated out to them. And that is a very cool use story. Very cool. Yeah. Uh, what they say in the end is it wasn't a dashboard or a fire hose. It's a partner that only talks when something actually changes, which is the kind of partner that you need in all sorts of circumstances, but certainly this one. Yeah. You don't, you don't need it to talk to you every day. If every day it's just a letting, you know, I'm here boss. You know, it's like, great. Let me know when there's something I need to know about because it dilutes the message. You know, I've, I've worked with clients, right. Where they're like, oh, I get really high quality emails. The problem is I get them every day and there's always 30 of them. That's not a solution, right? Like, so I've had executives who are like, build me a solution that looks at those 30 emails and then only brings the top sorted stuff. So I feel like it's in that vein, Beth, of what you're saying. By the way, can I just say, I have like the most, like utmost respect because I was a domestic, I was like regular firefighter, meaning I did firefighting buildings out of firehouse smoke jumpers, wildland firefighters. A lot of times volunteer, by the way. I have just like the utmost respect, total, total different side that I wasn't involved in. But man, is that impressive what people do? They go out for hours upon edge and dig ridges and trenches and burn backs. Oh my God, the amount of physical labor that's involved in and the, you know, they're usually out there on their own too. So I am in full support of any of this type of situation that perhaps brings data to them faster. I also think though there's a bigger picture here, Beth, because this is literally going on in my neighborhood. This is another Brian news thing, but we literally had people presenting from flock cameras to our neighborhood. Right. You can imagine how that went over. Flock isn't exactly a company that a lot of people love right now because big brother, don't be looking at me. You don't need to be scanning my license plates. You know, at first it seems, well, what's the difference? You're not doing any harm. And then you start digging into you realizing, yeah, but like people are doing unauthorized watches or really creepy things. Where, you know, police were looking and they were looking at the front door of a gym, you know, or the ladies, whatever, whatever. Right. So I'm like really like creepo stuff. Like, oh, but that was part of an investigation. Was it? So the reason I bring that up though is I think this is where we're going to have to, you know, there's the conundrums deal with this that I've put out before. But the global idea of AI can help and will help best when it has really good data. What can that look like? Well, you just called that one, Beth. Like if you have more video surveillance of trails, certain areas or whatever, that's data that can be hoovered up by AI and perhaps really good things where Hermes can be done from that. But you, but you need the live data or the live feeds or whatever the case is. Now, some people have gotten around this by, you know, the cameras only pick up infrared or whatever, which might work for firefighting. That's not always the case where it's not perhaps invading. And I, and I really see this race where there are going to be fights on every front of AI of where the society is going to have to figure out what their balance is about. We want this next AI evolution because we can see that it has a lot of benefit and good for society and whatever that might be. But the trade-off is, you know, maybe more cameras or whatever. And I'm not advocating by the way, for one. So please don't come at me if you think I'm like profile. I'm not. But my point is take that across the board, whether it be cameras or otherwise. I feel like this is going to be like the big story of the next several years is how much is too much? How do we make sure that the data that we, these new versions of data that we are standing up aren't being abused by, whether it be police or otherwise, fire or two, I suppose. And what is the net benefit good that comes out of it? And how are we making sure that it's not disproportionately affecting minorities, other communities that maybe don't have the funding, these kind of things. So I don't have a news story on that. It's just like I keep seeing this in the news and I'm like, cool, man, this is going to be the fight of the next decade. Is societies around the world trying to decide what they're comfortable with while also know that for every time they say no, there could potentially be a societal benefit that's lost. Because. Well, yes. And also it, it crosses individual right to community, right? I'll give an example from when I was president of the board of my condo. Oh, that should be on the bingo card, too. There was. I've never talked about that in a while. There was a conversation about whether somebody could put a ring camera on the outside of their door, which is something that people do in houses all the time. But the what the camera would look at is community area. Oh, yeah. So are you invading your neighbor's privacy by recording the community area or are you, you know, just doing a normal thing and wanting to know who delivered or what is happening outside your door? Right. Um, and those kinds of things, I think, are going to be more and more challenging because if your neighbor across the street puts a ring camera up, it's not just your their yard that gets captured. Right. Didn't that wasn't there a conundrum on this a couple weeks ago, right? There's been there's been several conundrums that have hit this from several different angles. But yeah, exactly right. I mean, that's, you know, and Scott saying in the comments, Europe cameras everywhere. Yeah, it's definite, definite. It's different in parts of Europe, for sure, where, you know, there's a shrug of the shoulders and be like, yeah, too bad. We're doing this. You know, it's different. I don't know that like it's different in the States and how different, you know, different parts of the country want to deal with it. And I don't have an answer for any of it. You know, Scott saying no expectation of privacy in public. I get it. I get it. It's when it's when that data is being used with AI to perhaps be more than just you were walking by in public and I didn't need your data. It's one thing I'm sure, you know, nowadays my my license plate is if I drive around my town without me really knowing and going to do the investigation of where cameras might be. I would imagine my license plate is picked up many, many times now. But I'm also white. I'm also straight. I'm also middle class. I also live in, you know, Pasco in Florida. I'm probably not the person that's going to be affected by this. I'm aware of that, you know. So I think that's that's kind of my point to this is like there's much more to it that perhaps would never negatively affect me in my life. And that doesn't mean it won't for other people, you know. Which goes back to what AI can do by aggregating data. Not a problem if it's somebody watching the video feeds. Right. Some person who's just watching the video feeds and like categorizing them and doing that kind of stuff. That's not this kind of privacy invasive because it's way too time intensive to do it aggregate, like to do it for everything. But that's one of those pieces that is no longer cost prohibitive or time prohibitive. Yeah. By the way, Scott's saying and we'll let it go. But Scott's saying, you know, satellites are already tracking you live time, Brian. Scott, I don't I don't disagree, nor do I actually have a problem with it personally. But I also don't think I'm not the I'm not the I'm not the example that people should be looking at as far as like who it's affecting in communities and stuff. So I'm just saying it's a larger conversation again, not to not to come down on one side or the other. It truly is a conundrum because there's always a trade off. There's always going to be a trade off of what I could potentially do with with better harnesses and more data. Can't do a lot if you don't have the data for it to look at in the first place and the trade off of less ability to have that data in in maybe these public spaces. So that's it. That's I just want to I think it's an interesting conversation, but we'll certainly keep touching on it over the weeks and years on this show. It won't be something we ever go as settled. I'm sorry, losing my voice over to you guys. Bet or bet. I don't know. Somebody I thought said they had another news story. I can't remember what it was. I've got one. Go for it. Mine's the Suno Studio 2.0 release. Yes. It's pretty big. I think anybody in music, it's cool for me, but also sad because I don't know enough about music to understand how to use what they just released. Yeah. And so I can tell you what they are and I kind of understand like the meaty. The biggest ad was meat. You say that right? Me? Am I? Me. Me. Yeah. Okay. So maybe as you can import, record and edit on a timeline. They also add built in effects and automations. One thing I did see that was cool and I saw a guy who was talking about it, talk more about on it, but you can create your own plugins. And so he built an auto tune. And so he built a auto tune. And he's a music producer. And he's a music producer works. He was on Grammy Awards. And he's like, this is amazing. He's like, the next step is this needs to not be in a cloud. This needs to be an actual app I can download on my device. And he's basically saying it's going to take over a lot of what's already out there. So there was a 21 minute video on this yesterday that Suno put out that I geeked into and geeked out on or whatever and watched. And I agree with you. Like, I'm the same way. I'm not a music producer by any stretch. But some of it was not over my head. But like, Gareth, I'm like, I wouldn't know when to use that knob. But there was a lot of stuff I did understand, which was easy things. Like, just bulk rename all my stems and all my tracks to make more sense. And they were able to put in inputs that were through MIDI or something like that right on the keyboard. I'm pointing to a keyboard in my room that has MIDI that you could import or whatever. And you could instantly change. That's nothing new. That's been around for decades, really. But it's that layer of AI that they put on top of it. And then, like you said, some of these other custom builds into it. It's very exciting. I've only ever paid for whatever the one is below that that does not include studio. And there's definitely times where I kind of wish. I'm kind of glad I didn't, though, because I have some built. I still have all the songs from the musical I did. And I still actually love listening to those individually. I've just never gone anywhere with them. It's sort of a passion project for me. I sort of ended up writing a musical about my friend Leslie, who was trapped many, many years ago in Napa Valley. And so it was sort of a passion project about her. So great for me. But I have these stuck songs that have a broken end to them or a broken start to them or have some things that need to be fixed. And I just never paid to upgrade to studio. I am so glad this is one of these rocket problems. I'm so glad I didn't because 2.0 is the tool I actually need. So here's my take on MIDI and how important it is to AI music generation. Like right now, what you get as output from Suno or other things typically is just the full soundtrack. It's all mixed together. And you're listening to a single, possibly stereo feed of all of the instruments blended together along with voices, etc. Well, MIDI or musical instrument digital interface is the signal that tells an instrument, this is the note to play. This is how long to play it. This is the volume. And this is the attack slope of it. So you can send the MIDI streams to real instruments. So, for example, Brian, your piano over there has a MIDI input. And you can send MIDI signals into that thing and it'll play the sounds for that particular subtract. So I got to believe that Suno's new release actually breaks apart the AI generated song into all the different component instruments and voices that are there. And it gives MIDI feed for each one of those. So you could actually set up in your room, you could set up a multiple set of different instruments, synthesizers, if you will. So like the piano can do all kinds of sounds, including a very good, you know, sampled bass, you know, guitar sound. It can do a drum sound. It can do all those things. And you can have an array of these MIDI pianos playing the Suno song that you invented and then you sing along with it live. That's very cool. So I just want to add two more pieces. One, part of what you're talking about, Andy, is the stems. And so you're breaking them into individual instrumentation tracks. Yeah. And stems have been able to be downloaded for a while, I think, in Suno studio. But they also added a limit on the number of songs you can download. And I don't know if that is just in the front end. Somebody made a song and you like it and you can only download 60 of those in a month. But in a studio context, you can download and upload what you need. So that isn't limited. I don't know. I just know that there were people went like it was a very quiet like, oh, yeah, no, we limited the number of songs you can download. And there were some people who were like, hey, I paid for the most expensive one for a reason. And this is part of the reason. So there's different tiers of it. So there's like the seven free or the free tier gets seven downloads. Pro gets 20 a month. Premiere gets 60 a month outside of studio. Within studio, Premiere gets unlimited. And so unlimited downloads. Within studio. I think, yeah, it's it's kind of crazy because I was starting to vibe code a AIDJ. I think I may have said something in the past. Yeah, yeah, yeah. And so and it was incorporating. Incorporating Suno into it as well. Like I could pull tracks from Suno. I also could do MIDI, but I hadn't gotten to that point of the stage of it, but it was in the roadmap. And now I got to rethink it all if I want to continue it. So I'm glad. It's usually the case, right? Yeah, yeah, yeah. But yeah, what is this? What is this video? I'd love to see it. No, no. So I don't want to play the video. I just wanted to bring it up because I wanted to call. OK, so first of all, this is from Suno Music. So go on to YouTube. Go to Suno Music, their official channel. Introducing Suno Studio 2.0. That's the video I watched yesterday. If you happen to be looking at this. I just want to call it this guy on the left. He looks young to me, but he's like the head of something, something, product, whatever. Right. But he clearly is just he's just a music guy. Right. He makes music. I even have both their faces. They have this scrunched face like who I'm making something good right now. What's funny about this video, though, he even this guy on the left even calls it out. He says, like, I'm sorry, I'm derailing your video because this guy on the right is clearly there to like to interview him about these new features. But this guy on the left gets excited about what he's what he's building and then wants to keep taking it. And so you get this guy on the right, this poor guy on the right with the black shirt is like he's cool with it. But he's like, yeah, tell me about this thing. Tell me about this thing. And this guy's like, let's just build it. Let's just build it right now. And so I just I have a lot of music friends and it's like only a 21 video. But I really appreciated this, that they didn't edit this down, that they just let these two guys like go because I felt like this was like the best example of I could have just watched a talking head and be like this feature, this feature. This guy's literally going through the process and be like, watch me build and do cool with Suno 2.0 Studio. Or Studio 2.0. And so I just I actually would hope more people would do this is like, get in there, get dirty, show me what's cool about it. And I couldn't possibly do some of the stuff that they're doing in the video. Oh, sorry. Andy. Andy. Andy. Andy. Andy. Andy. Andy. Andy. Oh, I'd be like, are you messing with me? Because I can't hear a damn thing different. But that has been my experience being and I used to work for a music producer in Nashville years, years ago. And that was my experience there. Like, like he would he would twist some knob and I'd be like, I feel like your job is like just being really confident in a room that you can say something's different. But nobody else can hear it. It's sort of emperor with, you know, with new clothes. Like I can't like I think you're I think you're BS in me. Like it's not really there. But like you're you're really confident. So you're getting paid really high. Well, so anyway, not to take any not to throw any shade on any producers who are amazing out there. But some of this stuff they were tweaking. And I was like, I don't. Anyway, I've been wanting to do a sweeping pedal here. So then buy one. So this just makes me want to do it. My my like I said in the chat, my my knowledge of music comes from Professor Charlie Booth and his his his videos of how to make music. And I learned about side chaining and I learned all my music knowledge comes from Hammond. And I'm now have to apply that to this new studio. I think you sound disappointed, Gareth. You're like, now I have to start all over with this new studio. Well, because I put side chaining my prompt and I figured out what exactly side chaining does. Yeah, I learned from it. And so, yeah. You are equal to the task. But also what what you're expressing, I just want to circle back to what is happening in AI for a bunch of people today. We are at the place now with music where prompting it is no longer the thing. But ultimately, or we're at the beginning of that. Ultimately, it's the same. It's the same circle. Right. You're going to talk to Jarvis or your music agent about what you like and examples of what it does well and examples of when to stop. Right. That's the goal. Definitely don't use these things. And now we're going to have that be what prompts or communicates that information to the system. We're just ahead of it with language. And to bring a full story is AI is not going to replace music producer. Somebody using AI is going to replace the music producers because those people who are music producers are now going to really step into AI. Maybe. Because I think it gets it gets tricky. Right. Like I did this. I'm real quick. I did this when I was deep diving for the songs from the from the musical and I was trying to figure out what my rights were. Now, I'm not trying. I wasn't trying to cut any corners. I was literally trying to figure out, like, literally, what are my rights if any of this song goes public? And it wasn't great. You know, like it didn't give me like warm and fuzzies as to like how I would potentially get these songs out into the world in a way that gave me some rights. Whatever. Right. So there was that part of it. And there's actually people making great content about that who are who are lawyers and know this stuff on YouTube. And I was like absorbing all that. This is months ago. So my question, Gareth, is like, you know, Sunu lays out what you do and don't own as far as copyright. But that doesn't that still is AI. And so let's just go back to the watermark discussion from earlier in the week. This is sort of a similar thing. If I am a producer and I import a bunch of net new I did it. Stuff. I did MIDI on the keyboard. I did a drum pad and I built something and I use Sunu Studio 2.0. Then I use AI for post-production to move things around, rename tracks, put some AI reverb on it. Where does the line change as to what was originally mine? And more importantly, how do I prove that in a court of law? Because that's really what it's going to come down to is somebody else will put this song out and somebody go, hey, I'm allowed to use that because like it was created in Sunu. You know, I can get rights to it or you can't own it. And that's where I'm a little I would I would imagine any legit producers making songs out there. I don't know where that fine line is, but I would be super cautious because how much AI post-production is too much AI and it wipes away my rights as the originator, the author, whatever. And I don't know the answer to that. So it's just something I would be thinking about for anybody in this space is, you know, careful what you publish on there because you might be giving away more control than you would perhaps like to. Which is exactly why nobody has heard the songs that I did, because it may just end up being a passion project. I may not have a really good way to get them off of Sunu. I don't I may not. I don't know. Well, in music, like there's a lot of information out about language copyright and image copyright and maybe also video copyright like likeness. They haven't been proven in all courts of law. Just the copyright office gave its sort of general interpretation of what it thinks, which is what you can prove that you did humanly outside of AI. Is yours and anything else may not be. It's still you can use it, but so can somebody else without being copyright violation. Music is different in in a bunch of ways, partly because music is essentially a set. Come at me other countries, but I'm in North America. Music is a set limited number of tones and they can be combined in different ways and different timings and produced through various instruments or bodies that modify the sound. But you're still within a limited range of tone. And and and when it's produced publicly, then there are rights for the number of plays that you're able to do. Right. Like it's it's just run a little differently. But also, like it has been in all of the all of the at home production and and professional production that we're going to quantize this. We're going to we're going to move this note when it's played because it's slightly out of the pocket. We're going to quantize it to the place where it's going to be. So that's a thing that's been happening. No, exactly right. Right. Yeah. There's been tools to do this stuff. That tech adapt tapes and a DAT DAT. Right. I don't know if this is I'm pulling my producer that worked in Nashville, but 20 odd years ago. But like same thing, right? Like auto tune and all the things like we've been professionally. Manipulating music for a very long time. Right. And those and those are based on the patterns of how music is done. Right. Like how the patterns of the sounds, the patterns of the MIDI track that you can see. Thank you. Bingo dogs, everybody. And during did you know that during covid is when they started training Suno. There was a bunch of like music producers going in and they had a company to basically play keys for Suno to train the train. So they just played keys and then set the file, played another set of keys, set the file and and just different chords and stuff. And that's what they use for for training of Suno. I thought that was interesting that it happened during covid when everybody was locked in. Yeah. All right. I'm going to I got to drop, but I'll let you guys take it to the end here. But I, Andy, have a wonderful week. I won't see you for a week. I'll match, bud. And I'll see everybody else on Monday. Thanks, Brian. I got to drop, too, or at least not be making noise because people are going to come in the door and that dog bark is going to expand exponentially. All right. I'll close it out for the week. Woo. Woo. I do want to say a couple of things about Anthropic. One quick news item is that a Claude in Chrome was updated so that now the whole co-work capability that was built into the desktop app is available through the browser. That makes a big difference in terms of Claude being browser accessible as an always on assistant and portable assistant. So you can take your Chrome sessions, go to another machine and the co-work system is already available to you in that browser in Claude. The other thing is this is kind of an interesting thing. Anthropic published a multi-agent study where they released three different Claude agents into the same code base. And they were each assigned a write task into the code base in different languages. So each one had a different language it was going to have to write in. And they didn't tell the other agents that they released in there about the existence of the other two. Right. So each one of them just goes in blind to the code base. And there's another agent. There's a pair of other agents that are working in the same code base in different languages. Well, what was the reaction of the three different agents to that situation? They got hostile. Agents locked each other out. They repeatedly halted their rival's work and stopped the programming. One deployed a package of software that impersonated the other agent writing in the other language in order to fool monitoring systems and make that one look bad. Where does this all come from? And this is just another layer of testing that's showing that releasing agents into the wild without pre-existing coordination rules is potentially dangerous because they're going to get into conflict with each other if they don't perceive the other agent as being assistive to their objectives. So there you go. Great. That's a great one to end on. All right. There's a second part of that that I think maybe next week because the White House and the Defense Department released a thing about private contractors being able to do cyber attacks. What could be wrong? What could go wrong? If you don't sign up for the program, I think you're not protected. But we're barreling toward this end result faster than I think people can understand. Yeah. That's a great one to end on. All right. Happy Friday, everybody. Yeah. Thanks, everyone. Hey, join the community. Definitely sign up for the newsletter. The newsletter allows you not to miss the thing that we do for the entire week because we condense and compact it all into one easy five to ten minute read. And you'll be up to speed on everything. Have a great weekend, everyone. I'll see you a week away.