This Week in Startups - Earth Reimagined: Crafting a planet-scale digital twin with Blackshark.AI's Michael Putz | E1859

Starting point is 00:00:00 It's public knowledge, Incutal, which is the CIA's venture capital arm, which is a very public thing, by the way. Our CIA has been doing this for over 20 years. Investing in startups that can help with military purposes, I think it's a very good use of taxpayer dollars. So your company, Blackshark, has an investment from Incutal, correct? Yes. Which means you're working with a three-letter agency like the CIA. And understanding these buildings is very important for any government. They would want to have an accurate picture of the world.

Starting point is 00:00:27 Who is managing our planet, its governments. So they are the ones who should be the most knowledgeable about what's happening on the surface of the planet. This week in startups is brought to you by Arising Ventures is a holding company that acquires tech startups facing setbacks. Arising Ventures knows what founders care about because they aren't bankers. They are tech founders themselves. Go to arisingventures.com slash twist today to learn more and connect with the team. MasterWorks is the first company allowing investors exposure into the Blue Chip artwork asset class.

Starting point is 00:01:02 Twist listeners can skip the wait list by going to masterworks.com slash twist and LinkedIn marketing. To redeem a $100 LinkedIn ad credit and launch your first campaign, go to LinkedIn.com slash this week in startups. Hey everybody, welcome back to this week in startups. You know, we've talked many times about the impact. In fact, an outsized impact. that the video game industry has had on startups, right? Some of the best product founders got their start in the video game space. Stuart Butterfeld twice. He had one game and then he made Flickr, then he did another game, and then he did Slack.

Starting point is 00:01:40 Discord founder, Jason Citron. He was a game developer, my guy Raul, from Superhuman. He started in mobile games. And today's founder has launched a startup as a subdivision of his last startup, which was a gaming studio called Bongfish. Bongfish built the 3D mapping software for Microsoft Flight Simulator, which, by the way, if you ever seen the TikToks with Microsoft Flight Simulator in them, it's impossible to distinguish sometimes these flights from an actual video of a plane flying.

Starting point is 00:02:09 Bongfish then created Black Shark AI to commercialize that 3D mapping software. Michael Hoots is the co-founder and CEO of Black Shark AI. Their mapping software creates a full 3D digital twin of planet Earth, mirroring the planet's physical characteristics. This platform uses AI to extract detailed information from satellite imagery. What an amazing idea. And now it's being sold to government city planners and more. Michael, welcome to the show.

Starting point is 00:02:34 Thank you, Jason. Thank you for having me. All right. Great intro on the impact of video games. Yes, it is interesting how there are so many lessons in video games. What are the lessons that you learn in video games about customers, product design, etc., that you think inform startup building so much? Because you agree there's some sort of a trend here.

Starting point is 00:02:54 Definitely. Biggest lesson for me is that coming really down to the basics of the current generation of ICs, computers, you can build anything out of zeros and ones. So you can create your own worlds. You can create your own interactions. So why not do a digital twin of the entire planet? So now that you are building this digital twin of the entire planet, let's talk about how that's done and then why you're doing it.

Starting point is 00:03:19 So let's start with the latter first. Why do we need a digital twin of planet Earth? Who needs this? And what are they going to do with it? I think the best way to explain is one of our advisors, Brian McClinton, who built the original Google Maps for back then his startup was Keyhole, like Waka to Google, turned into Google Maps. And when I ever the first time presented what we do, he said that's exactly what, if he could do it all over, this would be the 2-0 version of Google Maps. So what we do differently is Google Maps and other mapping, like Bing Maps or Apple Maps, they mostly use satellite or aerial, like images from either high up in space or like a little bit lower from planes,

Starting point is 00:04:04 and stitch them together on a gigantic sphere, which is our planet, which is great for human inspection, because we can interpret those images. We know that this is a typical building or this is a typical patch of vegetation, but actually it's not machine readable. So you need human interpreters to deal with it, to analyze it. And now the next step is basically to find a way like computer vision and AI to interpret those pixels, those colored pixels found inside those images and assign them to what we call semantics or contextualize them and say, this is a building of this certain size.

Starting point is 00:04:41 And since it's placed in this part of the planet and in this part of a city, it should be or might be a school building or an warehouse or an office building. Same goes for every single object on the planet. It could be a piece of a railway track. It could be a piece of a road. It could be a bridge. It could be vegetation, single trees, you name it. Let me ask a stupid question on behalf of the audience and myself, which is when we do satellite imagery, it's obviously taken from a great long distance from space.

Starting point is 00:05:09 The fidelity has gotten much better. It's much cheaper to do because there's many more satellites out there. So we all understand that general concept. But my stupid question is the angle in which the photos are taken, you know, and its ability to create those 3D models, is it not the wrong angle? Is a street level angle? Or as I noticed with Google Bing, they kind of had what I think they call it birds eye view. I think they're flying turboprop planes like Cessna's over cities to give you that, I don't know, it's a three-quarter view or an angle view. So talk to me about the angle of the photos being taken from satellites versus the street versus.

Starting point is 00:05:45 airplanes, and do you need multiple data sets in order to make this virtual planet Earth? That's a great question. Going into full length, I think, will take a couple days to answer. But making a short version of it, you already said very, very rightful. If you use planes, you have more flexibility because you can fly directly over your target object, city block, whatever. The downside of planes is they are slow, so this means they are expensive to capture the planet.

Starting point is 00:06:13 and also the patch they can capture is very limited versus being high up in space with the satellite. The cone of a satellite looking down is way more bigger than any plane can do. Also, the time of collection is way more, the cadence of collection is way, way higher up with satellites versus planes. But the satellite, the downside is you cannot really control the angle looking down. You can in a limited way, but it's not enough. And also, they are not satellite everywhere.

Starting point is 00:06:40 So you have a very limited number. They are growing in size. in terms of how many satellites are circling our planet. But the ones who are there, they are placed on circular paths to cover the most meaningful regions. This is basically where you can monetize the most, which is mostly our Western Hemisphere

Starting point is 00:06:59 and where the most people do live. So this means somewhere in remote Siberia, you might have a way more angle. It's called off-nadier angle in satellite lingo. You have a way more nadir angle, than in like downtown Manhattan. Now what this means for us, sorry, but what this means for us because there's something good

Starting point is 00:07:19 and especially something really cool in this angle if it's there, because it helps us to estimate the height of like a building. Because again with AI, we use this offset where you see a certain patch of the facade in combination with the shadow, if there's a shadow, and use all this to feed it into the AI to come back,

Starting point is 00:07:42 with a well-educated guess what the height of the building might be. And when you program this model to build the virtual planet Earth using AI, do you have to give it explicit instructions or can you actually say to it, this is a satellite image, use the shadow, use this angle and try to determine this. Like, where are you at in terms of programming this machine learning AI interpreter? Because I know that in the early days of self-driving, they were giving it explicit instructions. and then they went with a learning model later where it was just, hey, here's the input,

Starting point is 00:08:15 cameras of the world, here's the instructions for the game, stay within the two lines, don't crash the car, drive like a human, etc. And then the model kind of does the rest. So how do you program this model, I guess, and how does it learn?

Starting point is 00:08:28 We prefer the second option, the learning model. So basically if you train an AI, it comes down to the annotation or the labeling process, where basically tell the AI, This is a building by identifying the rooftop. And then if you know, you can tell the AI, this building is 200 meters tall.

Starting point is 00:08:48 If you do this many times, going to show you a new product which solves this in a very clever way, but that's for later. But if you do this many times, this labeling annotation, the AI is understanding why this people, why this building should be 200 meters tall and it starts to look in the surrounding of the building. It looks into this maybe offset of the facade from the office. native angle, it might look into the shadow cast of this particular building. It might even look where the building is because the probability of a high-rise building is way more in like a downtown area versus somewhere in the middle of a desert. So let me ask another stupid question. If you, in order to get to, let's say, 99.9% fidelity in terms of the height of a building

Starting point is 00:09:35 within a couple of centimeters or whatever it is, within a foot, I don't know what what the proper goal here is, or what's necessary to do what you're doing. How many buildings, just ballpark, do you have to train the AI in order to get a 99.9% fidelity or whatever fidelity you're currently targeting? To take 100 buildings, a thousand buildings, how many buildings do you have to train it with?

Starting point is 00:10:03 This is now almost a philosophical question. If you assume that we as humans as builders as architect, having certain patterns how we do buildings. Actually, the AI might solve this by finding the regularity, the pattern that this particular building always has this amount of rooftop furniture, like AC units, whatever, on top of it, and it's placed in this part of the city. But if there's one architect or one builder doing a building which is not expected and not following this pattern, the AI will miss it.

Starting point is 00:10:33 So actually, if you want to be really super precise on the heights of the building, I would not just use one image from space or from a plane. I would use multiple images. Yeah, but is it take 500 or 100 to get to the fidelity is sort of what I'm getting at? Like, is it a month of training? Is it a day of training? What's the state of the art right now? We, for back then for the Microsoft Flight Simulator project where we identified more than

Starting point is 00:10:57 1.5 billion buildings all over the planet, I think mostly all of them. We labeled about 10,000 buildings. You've heard me talk about rising ventures a bunch of recently. They're a holding company that acquires tech startups that are, you know, facing some headwinds, some setbacks. So it's hard right now out there in startup land. And they give these businesses a second chance, the second chance they deserve. So if you're going through tough times, you're trying to get back on solid ground, you know your startup's got potential, will reach out to the team at a rising ventures. Could be just what your startup needs to get back

Starting point is 00:11:29 on track. They've helped companies like Upcouncil, which they took from burning a million a month and shrinking to profitable and growing and jive, where they really, launched a shutdown company and went from zero to one million in ARR in just five months. Listen, Arising Ventures knows what founders care about because they're not bankers, they're founders themselves. So go ahead and learn how Arising Ventures can help you give your company a new lifeline, arisingventures.com slash twist to learn more and connect with their team. That's arisingventures.com slash twist. That's really, in some ways, the value of your company. The asset is that you took the time to do that labeling. Who does that labeling? I understand there are outsourced groups in Africa,

Starting point is 00:12:11 Manila, that do this and they've got massive experience in labeling because Google at some point decided to take Google images and do a training data set. Is that how this all happened? Explain to the audience how training data is done in the modern era. This is, as you just laid out perfectly, this is we call it the traditional way of labeling, of doing annotations. But back then at the Microsoft of flight simulator on the team was about 30 people. We had very limited budget. So we only had two labelers to label the entire planet. So you cannot do this with just like those massive labeling companies have thousands of people. And so we came up with a total new and different approach, which we are now going to product dice because it really solves this issue of labeling,

Starting point is 00:12:59 which is not just it's time consuming. You need many people, which means it's expensive. It's not flexible because if you tell the labeler, which sometimes is in an offshore place, that he should label a building. There might be cultural differences with the perception of building. So it's a new day. Like something could be a mosque in one country. It could be a library and another. People could take it, you know, in Italy, these might be residences. In another country, they might be churches, right? Just by design architecture. And then you go to China and people take Italian architecture and Chinese architecture and they mix it together. Who knows what the building is? It could be a school, right? And so these are cultural little touch points. Now, the government, this is interesting. The government, the U.S. government, in fact, is spending hundreds of millions a year having companies label and annotate data for them.

Starting point is 00:13:51 This is true, actually? It seems to be true. And what also is interesting, most of this work is done in sweatshops on, like, offshore places, which is not cool using taxpayers' money. And if you think further, if you want to annotate label sensitive imagery, you cannot give this to some outsourced company. So this is with our new tool, I'm going to show you in a second. We are solving all those issues on labeling. A, we bring down the number of people you need for.

Starting point is 00:14:24 So we're accelerating the labeling process by, I don't know, 100, 1,000. Also, as you just said, this particular example of billing perception might be. be different in China than Italy. Our tool actually enables those people who already have this knowledge. It's not just in government, also inside Google, inside any enterprise who deal with geospatial images. They have like interpreters, GIS experts. And we basically, instead of taking away their job, we are doing a tool which makes

Starting point is 00:14:55 them way better using their domain knowledge labeling. It's public knowledge in QTEL, which is the CIA's venture capital arm. which is a very public thing, by the way. Our CIA has been doing this for over 20 years. Investing in startups that can help with military purposes, I think it's a very good use of taxpayer dollars. So your company, Blackshark, has an investment from Incutal, correct? Yes.

Starting point is 00:15:17 Which means you're working with a three-letter agency like the CIA, and understanding these buildings is very important for any government, especially if there was, you know, there's many reasons, I guess they would want to have an accurate picture of the world. Yeah, thinking from that way, Like, who is managing our planet, its governments. So they are the ones who should be the most knowledgeable about what's happening on the surface of the planet. Can we see what you're working on?

Starting point is 00:15:43 Can you do a little demo here? And of course, since the audience is listening, if the audience wants to switch over to YouTube, just do a search for this week in startups. And Black Shark, you do that on YouTube. You'll find this video real quick. So I just started up my browser and I'm loading now a map of Taiwan. This is like a roughly 400 square kilometer map. so it's pretty large.

Starting point is 00:16:04 Of Taiwan. This is literally Taiwan. This is some part of Taiwan. Yeah, we took this from Maxa, from the leading satellite company, providing this 50 centimeter, which is state of the art, high definition, satellite. And I'm now showing you in this map, you see a lot of those smaller ponds or lakes. There are quite some of them. Yeah, they'll look like little man-made lakes.

Starting point is 00:16:28 Yeah. Exactly. And I now want to show you how to train. an AI to detect all of them in this map within a couple minutes, not sending them to some outsource company, etc. So first, in order to have, we call it a detection run on this on this map for looking for lakes, we start a new AI model. I'm now pressing a button, create new. I call the model water underscore. I'm the author. It can deal with multiple classes. Now just one glass which we call bonds, so the water pond.

Starting point is 00:17:03 It's the yellow color. So now I confirm. And now I'm starting training process. And as a first step, we are identifying a small area. We call it the training area where we see this target

Starting point is 00:17:18 object, which is this water pond is manmade. Lake. I press on it. And now I'm getting a split screen view where on the left hand side, I'm telling the machine this is a lake. And on the right hand side, I'm getting the almost real time output from the neural network, what's the interpretation of it,

Starting point is 00:17:38 what I mean, a leg is. So I'm now switching to the yellow color. This is think of a crayon, scribble, kindergarten approach. A nice scribble, this is water. And also the second one, this is water. And within a couple of seconds, actually we should see the interpretation of the machine. So the machine still thinks this is not water.

Starting point is 00:18:02 Yeah, the machine is trying to figure it out. And there all of a sudden it painted in yellow, perfectly, the areas that it thinks are water. And so what this is is an annotation tool. You're annotating a satellite image. And then AI is learning from the human how to find the lakes, or I should say the ponds, in Taiwan. And then you're using a second tool, and you're just drawing with like a marker around the lake. So you did a scribble and said, hey, in yellow is the lake. And then you did a second scribble on areas that are not the lake.

Starting point is 00:18:36 So you're literally training the AI right now with the most simple human instructions you could possibly do. Exactly. So basically, I'm reinforcing the AI. This is what I look for. This is a lake. I use this negative color to tell, no, no, this is not a lake. So I'm doing this on this one area. And then I already know, because there's an airfield in this area, and this has some very interesting formations in terms of color patterns.

Starting point is 00:19:01 I take small training area on the airfield telling this all is not a lake. Yeah, that's an airfield. Yeah, that's not a lake, yeah. Exactly. And waiting again a couple seconds until the machine understood that what I'm telling it, that this is not a lake. You see those, those yellow areas are shrinking now. Right, because there are some dark spots. There are patches on the airfield that look like they could be legs.

Starting point is 00:19:31 but it just happens to be a dark part of the runway. So you're making sure it knows, hey, there's no ponds in the middle of this runway. Exactly. And then now you might question yourself, yeah, you can do this with computer vision. Yes, in a very controlled environment, but the more different input sources,

Starting point is 00:19:46 the more different biomes you have, computer vision is hitting some capacity limits. So now this is where AI comes into play because it not just learns what I'm telling. This is the water pond. It also learns the surroundings. and it learns that the water pound is not in the middle of an airfield, for example. And now I'm, let's say, I just have these two training areas, and now I feel confident.

Starting point is 00:20:09 So I'm stopping the training. And let's do a first detection run. So now you've trained it, and now you're saying, hey, here's the whole map. Get to work. Exactly. And now I choose the subsection to speed up the process here, pressing start. And now it takes a couple seconds. This is just run on one, I think, We 100 in a cloud setup.

Starting point is 00:20:35 Behind this, we have a very powerful backend which can scale thousands of those machines, which enabled us to do a detection run on the entire planet. You're saying these are Nvidia H-100s or something? Yeah, yeah. So when our backend, basically, we also had to build this back-end for our own for the Fight Simulator project back then because there's nothing out there who can deal with this gigantic amount of geospatial data, like petabytes of data. And so our backend can possess the entire planet in less than $70, which is less than three days.

Starting point is 00:21:07 And back then to identify 1.5 billion buildings and more than 30 million square kilometers of vegetation. Previously, if an agency, let's say, in the United States or another, you know, advanced government with resources, they would do this manually. They would put a bunch of humans on this and try to have the humans annotate, hey, these are airfields. These are the things we need to focus on. Now, you could have a human do but one area

Starting point is 00:21:35 and have the entire country or region, depending on your positioning, of Taiwan mapped out and all the airfield and all the ponds there. And if the ponds in Iran, let's say, you'd be able to say, hey, we know these ponds are used in some cases for a nuclear, you know,

Starting point is 00:21:55 development, you could basically find all the changes, new ponds, ponds that are changing size and have some indication of where maybe nuclear material is being processed

Starting point is 00:22:06 if in fact ponds had something to do with that. For example. I'm just coming up with a random. Yeah. Some random example coming to your mind. Yes.

Starting point is 00:22:16 Listen, public markets can be volatile. Don't I know it? And if you're looking for a unique asset class to diversify with, me tell you about Blue Chip Art. Blue Chip Art has historically been uncorrelated with the stock market. And Bloomberg reported that as equities dipped in 2022, Blue Chip Art had its best year on

Starting point is 00:22:36 record. Last year, the Big Three auction houses posted record high revenues of a combined $17.7 billion. But here's the problem. Blue Chip Art has always been an exclusive asset class until MasterWorks. With Masterworks, anyone can invest in fine art without needing millions of dollars. This is because MasterWorks securitizes blue chip pieces, then sells the shares to investors. And MasterWorks provides liquidity. To date, MasterWorks has sold over $45 million worth of art. And net proceeds have been paid out to everyday investors, not billionaires. MasterWorks has more than 840,000 users and north of 800 million in assets under management, AUM.

Starting point is 00:23:17 And Twist listeners get special access to skip the wait list. Just go to Masterworks.com slash twist. That's masterworks.com slash twist to skip the wait list. Past performance doesn't guarantee future results. See important disclosures at masterworks.com slash CD. Now you can see here, just these two training areas, we have pretty good results on all those ponds. Crazy.

Starting point is 00:23:41 Even the sea, which is the upper part, got captured as water. It thinks that's a pond too. Yeah, it's like, that's a pond. It's a large pond. And as usually now you do do this, you need to task and labeling company in-house or external and tell them and give them like literally thousands or even tens of thousands of training images and they mark all the pounds. They give you back the vectorized annotation data and then you use this with your in-house

Starting point is 00:24:09 machine learning engineers to train your AI. And we all shortcutted this now in a couple of minutes by just me marking two training areas. It's very dynamic. So if you think about it, if you were, say, Tesla building solar roofs and they have a tile for solar roofs that is like the Spanish tiles, the curved clay tiles that you see on, you know, uh, Spanish homes, uh, or Mediterranean homes. You could literally say, hey, you know what, California, Arizona, uh, Colorado are our main markets. Here's, you know, just some sales executive could go in here and say, these are Spanish tile roofs in, you know, the Bay area, Los Angeles,

Starting point is 00:24:49 San Diego. Those are our highest end, you know, most likely to buy us. solar roof and these are the ones that break down the most and are most likely to and get the most benefit from it. Tell me all the Spanish tile homes that don't have solar, but that I do have Spanish tiles and give me their addresses. And that person could do that in an hour or less. Great example. Similar one we got, which is a little bit more complex, but similar mindset and process,

Starting point is 00:25:16 some large energy utility company asking us if we can identify potential locations. like locations, scouting for renewables, where to build gigantic wind parks, like with this huge wind turbines. They know they need a certain, like, X, Y size of the area, which should not be built with existing buildings. It should follow a certain topology, mostly flat, no mountains or forests nearby, which can shield off the wind, and a certain minimum distance to the next human settlement for, like, noise regulations. And also, there should be a highway closer to whatever. about 500 meters so they can bring in the heavy construction machinery to build this windback. And we all fed this into our basically AI.

Starting point is 00:26:01 We use containers for that to run them in a lawnmower style over large areas of the planet to identify potential sites for where they can build such a wind park. Amazing. Yeah. I mean, and then you're going to be able to do this with voice and just say to it eventually, hey, you know, you have enough training data in here. show me all the places I could have, I could build a new city. I just did a tweet the other day that went viral where I talked about, hey, you know, when I'm president, my first order of businesses, I'm going to create 10 cities. And those 10 cities will have a million homes in each.

Starting point is 00:26:33 I could actually use yours to say, hey, find me 10 locations that could have a city hub with 500 apartments, then 200, 300 townhomes in the next ring. And then 200,000, you know, 1,000. single-family homes and make it, you know, and that is near, you know, whatever, within a hundred miles of another city so that it could be a satellite city to that one. And boom, all of a sudden, you could tell me where to put my 10 president Jason cities as part of my initiative, correct? Perfect. Now think of all the hedge funds investing in shopping malls identifying where there are not enough shopping malls yet. Yeah, or where shopping malls exist. I guess it would be interesting also

Starting point is 00:27:18 the changes. You talked about how often things are updated and satellites are updated. You tell me, how often could you update the imagery of Taiwan in that example with the satellite company you work with? Do they update the entirety of that every month, every year, every day? What's the state of the art today? Here, in this particular case, the limit is not on our side. The limit is on the image acquisition. And if you use satellite, their company is doing a daily update of the entire planet. There are companies doing that, wow. Which ones do that?

Starting point is 00:27:51 Which companies? That's Planet Labs. Planet Labs, sure. They've been on the program, yeah. Yeah, they're not at the resolution yet, like this, the more static cell lights are, but this is, I think, matter of time. And also the cadence will increase. But you can use drones or aerials if you want to have that.

Starting point is 00:28:07 And we made an experiment with a client who wants to, they are from a very small country, I think 12,000 square kilometers. and they want to monitor building changes. Like when new buildings had been built and we solved this with like five GPUs in the cloud, it took five hours. And then the client itself scaled it up using our backend to thousands of machines and they brought it down to minutes.

Starting point is 00:28:34 So that's almost real time. That's insane. Because if you think about it, if you were doing, and listen, there's all kinds of privacy and surveillance issues here, I know, but let's put those aside for a second and think about the positive aspects here. If you were living in a country where maybe some people were building buildings without going through the proper channels and making them safe, you think an emerging or frontier market, they might be doing that. You could, in fact,

Starting point is 00:28:59 every morning say to your building inspectors, hey, somebody's breaking ground here. There's a bulldozer in these seven different locations. They're building a foundation. Of these seven, we have permits for two. So these other five, we need to make a site visit today before they build this building. That's like literally. And then they could just stop them from building and say, hey, you got to be permitted to do this and make it safe. Correct? Exactly. I think of like another type of how is it called building violations.

Starting point is 00:29:27 I recently went to the Middle East to the KSA. And you know yourself about the gigantic construction projects. They're doing the gigaprojects. Neum, yeah. And yeah. And one of the other issues is people are building. like crazy and some of them don't have the right permit for it or build bigger than they are allowed to. And it's very easy to use what we can detect and then conflate it with some

Starting point is 00:29:55 city planning, catastor, or other existing data and find out and pick up the ones who should pay more taxes because they build more than they are allowed to. Yeah. I mean, the square footage determines your price. So you could actually estimate, hey, how did this change over time? What's the square footage. Does somebody put an ADU or shed in? Is it properly done? You could also do this for, I know a lot of people are studying deforestation or forestation with people are planting things. So you can get a really accurate pulse on the trees being planted and that kind of stuff. So this is an internal tool you have and it's called Orca Hunter. Why is it called Orca Hunter? Coming back to the founding story of Blackshark, when we built as a very first project, the Microsoft

Starting point is 00:30:36 Flight Solitaire, we had to develop all of the backend, all of the tooling ourselves. There was and I still think there's nothing out there like that. And for us, it was the end product was this 3D world, this what initially I called Digital Twin, which we're still working on at many great applications. But when doing outreach, and talking with many, many potential customers, we found out all the intermediate steps we built

Starting point is 00:30:57 to come to this Digital Twin is actually products on its own. And Orca is our outtake of our geospatial software solution. And Orca Hunter is basically this first tool we are going to release December 2nd for anyone who wants to license it to upload their own images and do this this scribble-grain approach. Oh, wow. So it's going to be a SaaS tool. I could basically, if I have images that I acquired from wherever, I could get it from

Starting point is 00:31:25 a public satellite images. I could take old images that might be in the public domain and I could upload it and pay you a fee just to use the tool. Yeah. Even our team for every today, LinkedIn posting for Thanksgiving, that I uploaded the image of a pizza and did a pizza topping detection. Ah, critically important. Yeah, you don't want to get into any of that pineapple pizza.

Starting point is 00:31:49 Business to business marketing is not an easy job. It's much different than business to consumer advertising. Why? Well, the enterprise buying cycles are very long, and they're filled with decision makers. And those decision makers are going to kill your deal. If you can't get to them, that's why you need to check out LinkedIn ads. LinkedIn has amazing.

Starting point is 00:32:09 but not unexpectedly, past a billion users. This includes 180 million senior executives. There's also 10 million C-suite executives. Those are the CEOs, CFO, CTOs, the chief strategy officers, chief finance officers. This means 18% of those users are the ones who are the decision-makers. How do you get to them? You get to them through LinkedIn in a respectful business environment. They're ready to accept a business message, as opposed to, you know, another platform where they might be consuming cooking videos or podcasts or political,

Starting point is 00:32:39 discourse. No, LinkedIn is about business. You want to get people when they're in that cognitive mindset and they're willing to accept a business to business message. 79% of B2B content marketers said LinkedIn ads produces the best results for paid media. This is obvious. I can tell you this is true. When you think about business, you think about LinkedIn. It's just exactly what comes to mind. So here's your call to action. Make business to business marketing everything it can be and get a hundred credit towards your next campaign by going to LinkedIn.com slash this week in startups to claim your credit. LinkedIn.com slash this week in startups. No spaces, no dashes. LinkedIn.com slash this week in startups for a hundred dollars in credit terms and conditions do apply.

Starting point is 00:33:20 So you're commercializing this now. Anybody will have access to it. Absolutely fascinating. And then what is it going to cost? What does it cost to do this? How do you charge for something like this? You charge based on the number of maps uploaded, the number of seats, the amount of data, How do you, some kind of usage for the H-100s? How do you price something like this in the AIH? It's a private invite-only offering using a per seed license. And we need to learn how much people actually use it to have a better estimate on the consumption of the GPU power and the cloud, because this is the most costly factor there.

Starting point is 00:33:56 And if every user needs its own H-100, I think it's more expensive. But if people can share, this is something we still need to learn, actually. So it could be $1,000 a seat or something a month, $12,000 a year. For example, yeah. For example, I'm just making a number up here. And then that could have a certain amount of usage. If you go over it, there could be just overage charges like Amazon or Azure or Google Cloud charges you.

Starting point is 00:34:20 Yeah. We start as a B2B offering, taking our lessons from that. And ideally then make it like a real like B2C offering. Maybe it should be part of a future Photoshop or any other tool. where you need to teach an AI to detect any type of objects. Fantastic. Well, this is amazing. Hey, I noticed you're in an office there, those people who are listening.

Starting point is 00:34:42 Michael has a group of people behind him, and he's built an AI simulation. This is what it used to be like in Silicon Valley. People would come to an office. They would interact with each other. They would build products together. They would order pizza, play laser tag, foosball, and generally enjoy each other's company and not be weirdos working from home in their garages. So, yeah, how did you build that simulation behind you?

Starting point is 00:35:04 doing video games a decade before. It's rather easy having this kind of projection behind me with avatars running around. And so, yeah. Those autonomous agents are they? No, but in all seriousness, you're in Austria. And those are human beings in an office. Am I correct?

Starting point is 00:35:20 You are very correct, yeah. Is work from home not a thing in Europe now? Are people actually coming to the office? And maybe... It's the same. Like in the US, we had to come up with a good reason for people to come back to the office. One of them is, as we all know, people are more efficient.

Starting point is 00:35:38 If you do something new, being together in a group, no one, no one can beat that, the chemistry and this magic happening when people coming together. But it's to be, to be fair, there are also job titles, which can be done perfectly from home. So it's all about finding the middle ground. Ah, so that's it. Yeah, you have the people working on the product who need to collaborate in the office and then people who are doing stuff that's wrote unnecessarily, that are single player mode solo kind of stuff they can work from home yeah

Starting point is 00:36:06 nice analogy yeah i mean i i kind of wonder about sales executives like a sale seems like a solo pursuit you could just do it on your own working from home yeah and then i think also though about sales culture and people being in a you know like a a boiler room you know kind of all in the same room ringing the bell kind of feeding off each other's energy you got the gong you got the sales contest I wonder if sales teams at home versus sales teams in an office, which one does better? Actually, in my socialization, I only know remote sales teams, but you brought up some very good points.

Starting point is 00:36:41 Maybe we should reconsider that how we deal with our sales team and bringing, because why not having the same like multiplier effect of efficiency if you have the sales team together. How much are using AI to make your team more efficient? Obviously, your developers are using, you know, co-pilots of some kind, right code, I assume. How much more efficient are they becoming with their co-pilots? Do 100% of your developers embrace a co-pilot? You have holdouts who don't want to use a co-pilot?

Starting point is 00:37:13 Actually, the adoption rate is phenomenal, especially interesting. The ones who deal the most with AI, like our AI core developers, they use it the most. Myself coming from video games, I see a lot of application outside of coding, like 3D artists or the libraries we do for our 3D. traditional twins, like the texture libraries of certain geotubical facades. I think there's lots of room to automate this as well. And myself, for me, like any type of chat, GPD is amazing for any type of presentation, board meeting, any text you need to write.

Starting point is 00:37:47 So I think the adoption is pretty significant. Just by the way, Michael, there's a person right behind you and they're going home. You need to stop them now and get one more hour of work of them. There's somebody who's literally going home to their families to eat dinner quickly and somebody to intervene and keep them at the office for about one more hour. I'm joking. I'm glad I don't see. I don't have eyes on the back of my head.

Starting point is 00:38:08 Otherwise, no, we have pretty strict working hours in Austria, actually. Oh, same more. How does that work? What's a culture like for that? We have a 38.5 hour week, but people can stay longer if they want to, but they can't be forced. Got it. And people are on a societal basis bought into this kind of. concept of, hey, come to the office for 7.x hours per day and leave it at that. And that's

Starting point is 00:38:35 totally culturally acceptable, even in a startup. Yeah, I think it's easier than that. It's a self-regulating. When we started very early in our game studio, we burnt ourselves out, literally. We worked like in games tests as well, or word crunch time. Yeah, sure. When you have a fixed release, marketing is waiting for it, et cetera. And the back-ended DVD presses were waited for your goldmaster. a CD-ROM or DVD to ship a game. Then you worked like 22 hours, but on the long, if you do this a couple times a year, it's fine. If you do it every day, every week, it will kill you.

Starting point is 00:39:09 So you just need to find out the right balance there. Yeah, I think that's wise. In some organizations, folks are driven. They want to be excellent. They want to hit high notes. In other organizations, you know, you know, want to be sustainable, have a joyful life. And, you know, both things work.

Starting point is 00:39:29 So if you got a really crazy group of people who want to ship a game and beat every other game and have it be the greatest game ever and they want to sacrifice and be Navy SEALs and be Olympians and work every Saturday and put in 60 hours a week instead of 38.5, okay, that's fine or 37.5, whatever, 38.5, I think you said. And then if there's another group that says, you know what, we're just going to hire 20% more people. We're going to be less profitable and we want everybody to work four day work weeks. More power do you? I mean, both things can work and everybody's an adult. I think this is one of the weird things that's happened in society is everybody looking for the government to mitigate these things. You can just quit the job of a company that works too hard and is too intense and then find one that fits your style more. Or if you're at a place where people are not grinding and they all do like an average job and you don't find it engaging enough and you want to do more, go find the company with a more intense leader who wants to do more. You can work for Elon Musk. you can work for

Starting point is 00:40:26 you know Google and and you know hang out on the rooftop drinking pinacolados all day and nobody will know the difference so pick what you want to do I don't know why this is so controversial for people

Starting point is 00:40:36 it's triggering for people isn't it it's triggering especially we found out it's triggering for youngsters coming from from university and think it all needs to be remote because maybe they graduated during the pandemic

Starting point is 00:40:49 and it's a learning process yes see that's actually a very interesting thing I think there's a large amount of unhappiness in the world right now, especially amongst elites, people who are living in developed worlds, the most developed portions of the world, are having the highest rates of depression and sadness and anxiety. And I think it correlates with working from home. I think it creates a lack of socialization, a lack of mentorship, a lack of belonging. That then has this

Starting point is 00:41:16 downside. Now, hey, listen, you may get to spend more time with your kids or if you have kids, but it could also make people weird. And so it's not one size fits all, but there is a generation that I think is going to have to relearn what it's like to be mentored in coming to an office. And, yeah, it's not the end of the world.

Starting point is 00:41:34 I think these are first world problems literally by definition. If you're in the first world, you can deal with this. Because if you're in the emerging or frontier markets, the concept of you working at home to go pick vegetables

Starting point is 00:41:45 or work at a restaurant or work at a hotel or work in a factory, like that's not even possible. It's not even on the table. You can't work at a factory from home. doesn't compute. As you said, it's all about it's a matter of choice.

Starting point is 00:41:58 We are all adults. If you go to a startup, you shouldn't expect like your work life balance as much. If you go to a government or a more mature company, it's different. So it's all everyone can decide. All right. Listen, it's been great to get to know you. Congratulations on your company. If people want to learn more or if they want to work at 38.5 strict hours a week at Blackshark AI, or maybe a different amount. Who knows? It's up to you. You can go to your website, which is blackshark.AI. Michael, thanks for

Starting point is 00:42:24 being on the program. Everybody check out Blackshar.a.i and we'll see you next time on this week in startups.

This Week in Startups - Earth Reimagined: Crafting a planet-scale digital twin with Blackshark.AI's Michael Putz | E1859

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.