SCALE DATA

“I think there's a lot of headlines about data center projects being planned or initiated. I don't think all of them are actually happening. So there's a lot of like phantom data centers that are being announced. Let's call it that way. I think data centers are definitely in a rush to get online fast. They do have massive supply chain constraints. Two years ago, it was maybe NVIDIA GPUs. Now, it's power. Now, it's actually not even hardware. It's actually labor.”

“We need to manufacture domestically. We need to source domestically. We need to make this using raw materials that don't come from China, that are not rare of minerals. And all of that is what we continue to iterate on. So every configuration that we design, we looked at the bill of materials or the BOM. And we asked ourselves, what can we eliminate from this? How can we make this simpler? And the same approach you see at SpaceX, right? Or at Tesla.”

“The main arguments would still stand for humanoids. One is that our world is sort of designed for humans. So one hypothesis is that if you design policies for like, they single out mobile managers, then once you solve a lot of tasks in that environment, then you see that it's limiting because many tasks in our world are like opening a bottle, or like opening a fridge and then taking something from it. So you have to keep the door open. Or even, I think some people say, well, you don't need wheels, but then what if you solve a lot of tasks on a wheeled platform and then there's a little curb on your floor or by a street side and then the robot is like stopped there. So I do think that ultimately, if you want to do a lot of tasks and be useful in environments where humans operate, you need to go to a human or as close to a human embodiment as possible.”

“Weimos says, and I think this is correct, that it's roughly eighty brass safer in terms of crashes are severe enough to turn down an airbag. Crashes severe enough to cause an injury, and also crashes involving vulnerable road users like pedestrians or bicyclists. So far it's been better than human drivers, and so far, I think the case for allowing them they continue. The experiment is very strong.”

“The main difference in their approach is how quickly they want to move. Anthony is very okay with risk. He gets one of these cars and he's driving it back, and he lives in Berkeley, works in Palauto. He's just using this car like the Bay Bridge every day, probably outside the bounds of what the team actually wanted, and he's not necessarily logging data. He's just enjoying his self driving car and taking it all over the place.”

“I don't think people contextualize how much energy that is. So if you talk about a gigawatt data center, that's almost a million US households of energy. So if a data center says, I'm building a five gigawatt data center, I'm basically building a city that's five million households. And they're building in a very tiny footprint, although not so tiny. The funny thing about data centers is, when they start talking about buildings in acres, you can start realizing how big these buildings are.”

“Like literally and maybe to kind of just put this a bit more concretely, you know, if you have your robot in some given initial condition and, you know, you try something with RT1, RT2, it doesn't work. Well, you're kind of out of luck. You can try the same thing over and over again. You can slightly maybe rewrite the language instruction, like instead of, you know, pick up the cocaine, you can write like maybe like lift the cocaine, but you don't really have the granularity you need to be like, actually, you are two centimeters, you know, too low. You missed the table because it's at a new height. It's kind of obscured by shadows. So you want to like be more gentle and approach more from the left. There's no really way to do that right now with the interfaces, the language interfaces that we train RT1 and RT2 on. But with RT trajectory, the idea is maybe if you have this kind of like line sketch of a course trajectory of how the robot should do the task, you could, under the same initial conditions, just change the prompt a little bit, do some prompt engineering and actually see qualitatively different behavior from the robot.”

“What's new about our approach and our hypothesis is making this in a modular fashion, making this in a factory, and then scaling the production to millions and ultimately billions of these, which ultimately allows us to go down the cost curve and generate electricity at our ultimate goal of one cent per kilowatt hour. And so that's why we started the company. That's kind of been the goal North Star from the beginning.”

“In the last moments of Alane Herzburg's life, the robot spent an indefensible five point six seconds trying and failing to guess the shape in the road there was a human body pushing a bike. Over those five point six seconds, the robot kept reclassifying our whishing an unknown object a vehicle a bicycle. During that time, spent wondering the car did not slow down. Soon after Elaine Hertzberg's death, Uber halted its testing program.”

“I would say that the open models are are pretty limited. So, you know, even us when when we try to optimize with our, like, our DGX Sparks, for example, and we try to put them on open models, you know, they're they're getting better, but they're not quite the best. Right? And so just know that when it comes to really strategic work, you're probably gonna be wanting to run on the Frontier models, which are going to cost you money.”

“Well, one of the aspects is, as you mentioned, rules are sort of subject to interpretation. And even if you have the same language, there are multiple ways to interpret it. So here's an example. So we said, well, don't do things that or don't interact with anything that's harmful. And I think there was something in the data set which like it's it's all a cigarette. And then it was like, well, I'm not going to pick up a cigarette because it's going to be harmful. Currently, I think our robots are more the problems don't come from the fact that they are too smart to work around the rules. It's just that I think they are too incapable of doing zero-sharp things in the real world.”

“We will get a lot of business from data centers, but not just on the installation. Data centers need roofs too. Data centers need waterproofing, very much so. Data centers often need lumber related products. So so data centers are big consumers of building products. Now TopBuild itself has single digit percentage exposure to data centers, but it's very fast growing.”

“The three elements are the optical table, so these lenses, and then you have the heat battery and the PC, or the power conversion unit, all in this kind of container. And the idea is that we capture energy from the sun throughout the day. So think of this as like a solar panel, but that has a built-in battery, and that essentially you can scale linearly to any project size. If a datacenter customer needs 100 megawatts or a gigawatt, you just stack these next to each other.”

“Now we call it agent mining because we record all these decision traces, these contexts, what the users are entering into the system. And then you can either use it to say like, hey, wait a minute, this is actually an anomaly. The folks in, I don't know, in UK from our company or the folks in Australia shouldn't do this because the standard operating procedure is this. Or you say like, oh, that's actually a very good improvement. And then you can elevate this to be the new standard operating procedure, maybe not just for Australia, but maybe for the rest of the world.”

“TopBuild is the largest installer and distributor of insulation. And as you were just saying, everybody needs insulation. Every house needs insulation in the walls. Every office building needs insulation everywhere. It's a needed product, and it's not going anywhere.”

“The last thing about everyone coming out of Tesla, you're obsessed with vertical integration, you want to build everything in-house. We haven't done that at Exowatt to that extent yet. We've started working with contract manufacturers because we know we need to scale fast quickly, so they have that infrastructure. But Elon really taught us how to vertically integrate everything and build it yourself and build it better, faster, cheaper, and don't take no for an answer.”

“I had experiences of losing people in my life to traffic accidents, and I felt we lost over the million people in the world to traffic accidents. Wouldn't it be amazing if Dabok invented something that would save a million lives a year.”

“A lot of people like to say, oh, it's Hermes versus OpenClaw. It's OpenClaw versus Hermes. No. You wanna have both because there's a a chain of accountability that wasn't there previously. And now if you have this on your agent fleet, things are gonna run a lot more smoothly.”

“If you're simply, you know, trying to predict the next video frame, that's not so difficult. But what you actually want to do is understand the consequences, likely consequences of actions minutes into the future. And to do that, you actually need much more of an abstracted semantic model of the world.”

“When I was actually working with Nvidia on the Synthetic Data Foundation Model Training Project, we were actually generating a lot of these synthetic data and showing that these synthetic data are actually as useful as real-world data when it comes to multimodal pre-training. But then, there's a lot of dollars being paid out to external vendors or other folks to manually curate these types of data.”

“SAP and these large customers, right? They always have a problem of scale. Okay, what do you know with 100 documents? Well, it becomes a little harder. A thousand documents becomes a deeper engineering challenge. And now if you go into Julia or Sarah, you're maybe an SAP US employee, right? Of course, if you ask a question, of course, for travel policy, for example, of course you expect a very different answer than me as a German employee would get. So you now need to connect this actually with your master data.”

“All of the evidence from neuroscience and psychology is that most of what comes into people's eyes is never processed. You're doing fairly fine-grained processing of exactly what you're focusing on, but as soon as it's away from that, you've sort of only processing top-down this very abstracted semantic description of the world around you. Human beings are working with semantic abstractions.”

“Perhaps recently, you know, you know, for example, with this work, Pivot, maybe the answer is that actually there is some very good amount of physical intelligence already contained in these like internet trained models by themselves without any robot data pre-training or fine tuning. Again, I don't, I also don't think that like internet data alone, just watching, you know, Reddit threads and Wikipedia is enough to solve contact rich robotics. But I do think that we've so far just been like seeing the tip of the iceberg for the knowledge that is already contained in these, you know, large VLMs.”

“SAP is the market leader in enterprise, software applications and platforms. It has 400,000 enterprise customers. Usually, I just running their finance, HR, and supply chain, manufacturing, execution, logistics, warehouse management, and then of course everything on the customer side, sales services, commerce, procurement, you name it. End-to-end, like SAP, we always say we have the broadest portfolio in terms of end-to-end running the business end-to-end. This is where SAP started with, giving real-time insight. Usually, I really describe this as it's not just software in itself, it's kind of the operating system of a company essentially.”

“This essentially focuses the light coming in from the sun onto the battery material. It gets very hot and then stores that energy. So, essentially, our battery is a heat battery, so what that means is we're heating up rocks, right? And then storing that energy in formal heat, which is very cheap as compared to doing lithium ion or any other type of electrochemical battery, and you don't have to get lithium or magnesium or cobalt or any of these other fancy chemicals.”

“The time is clearly over where you design software, where the dump software that requires the intelligence to sit in front of the computer. If you look at classical software, what did you do? You decide to use an interface. This is over. It's now, we call this Generative UI. The UIs get dynamically generated. If you have analytical questions, for example, or if you want to do your deep research, not just the deep research you find on perplexity or the usual chatbots, but deeply rooted, let's say tariffs are being introduced or new taxes or the straight-o-formals, what does this mean for my supply chain?”

“When we complete the merger, we'll be number one in insulation, we'll be the second biggest in roofing, will be number one in waterproofing, and will hold number one or number two positions in certain geographies within lumber and building materials. So we'll have a huge addressable market, several $100,000,000,000.”

“Large language models are not made for this, right? In a way, how they generate just one token after another essentially in a sequence to sequence modeling, I mean, they're language models, right? And they do this phenomenally well. But if you still want to do these predictors where you have to go back to these classical machine learning approaches... What we said all the time is, okay, look, we have all this data stored in these tables, right? Thousands of tables, right? Where all this information is stored. Can we not apply the same idea that large language models or multimodal models did for the unstructured world, actually for the structured in order to start predicting things?”

“I think it's one of the most exciting takeaways for me, at least, was the fact that the line, the boundary between what are perception problems, what are open world object recognition, and what is robot control. This line starts to blur, right? We do not have a pipeline system where you first take care of perception and you solve that and then you solve control after. We're literally just treating both of these problems as a single VQA kind of instantiation.”

“I think it's fair to say that, you know, vision understanding sort of stalled out; you got to object recognition and then progress just wasn't being made. There's really an interesting research question as to why that is, and at heart, the ideas behind Moonlake are an attempt to answer that, believing that there can be a really rich connection between a more symbolic layer of abstracted understanding of visual domains, which aren't in the mainstream vision models, which are still trying to operate on the surface level of pixels.”

“Open claw is an autonomous agent, and a lot of people talk about it. And open claw is great for being the execution. Okay. You want it to be your arms. You want Hermes to be your brain. And the cool thing is you could have each of them hold each other in check.”

“I 100% agree that we are a few breakthroughs away from general purpose robotics, you know, that it's the dream that we are working so hard for. I think, again, if you want something commercially viable, something that will maybe make money or help some people in the world, I think a lot of those ingredients are already ready to have a larger impact than maybe even just a few short months or years ago. But for the true full vision of embodied, you know, AGI, I do think there is still fundamentally a few open research challenges left.”

“The reality is that although the visuals do look fantastic, those visuals actually aren't accompanied by an understanding of the 3D world, understanding how objects can move, what the consequences of different actions are, and that's what's really needed for spatial intelligence. So, I mean, a term we sometimes use is that you need action conditioned world models, that you only actually have a world model if you can predict, given some action is taken, what is going to change in the world because of it.”

“I think for me, the understanding was like people used to think that all the robots are so different. All of their data is like so different. And every lab has or like they invest in like a couple of embodiments. It was just I think, post RTX, the idea was that people moved in the direction of thinking that all robotics, all robots are kind of similar. It's like, it's only as different as like English and Chinese or something. And the concepts are similar. It's just the manner of expression that's different.”

“Tony Tether, who had been a door to door salesman in his use definitely has that flare in that way of thinking, says, let's have a contest. Let's see who can put all of these ingredients that we've developed together into a proper self driving car. His original idea is we'll drive him down the Las Vegas Strip that's almost immediately next because it's insane.”

“For me, it was always very clear. I mean, for the most part, SAP software is seat-based, licensed today, with a few exceptions, like Concur or Fieldglass, for example, or the business network. But very clearly, with AI, it was very clear for us that step by step, it will go towards this consumptive world, first consumptive, and then maybe in the next step, once we have more verifiability in the system, then also towards maybe an outcome-based license model to, for example, what Sierra is doing and so on and so forth.”

“The old economy is still here. It's not generative AI, but it's arguably maybe it's even more important. You couldn't have the new economy without cooling and heating. The actual business of making things and moving things, heating and cooling, is very much enmeshed with the new economy, which is all about generative AI.”

“My Open Claw and Hermes agents don't do any of that. They run every morning before I wake up. They get better every day, and they cost less than one month of a junior hire. Here's why I would take them over the vast majority of marketers I've ever worked with.”

“But once you have Obsidian added in, it makes it easier to manage everything. So, look, it obviously can't be all sunshine and rainbows. Right? So there are some things there are some limitations here.”

“Now I do wanna emphasize this again. Hermes learns from every session. So imagine this. After 20 to 30 tasks in any one domain, it's going to be measurably better than it was before. OpenClaw doesn't do that.”

“Sebastian, I think you should build a self diving car that can drive anywhere in the world. And my immediate reaction was, no, taking the technology we build for this empty desert and put it in the middle of Market Street in San Francisco is going to kill somebody. And Larry would come back the next day with the same idea, and I would give them the same answer, and both of us got increasingly more frustrated. God damn it, it can't be done, and eventually came and said, look, Sebastian, OK, care, I get it. You can't do it. I want to explain to Erk Schmidt the CEO at the time and Sergey Britt my cofounder, why it can't be done?”

“I saw that all the teams treated this like a hardware problem. They looked at this and say, we have to build a bigger wheels and bigger chassis and so on. And I looked at this and said, about wait a minute. The challenge really is to build a self driving car. They can drive for the desert. I can get a rental car. They can do it just fine, provided as a person insight and the challenges we need to take the person out of the driver's seat and replace it by computer. That is not a problem with bigger tires. That's actually be a software problem.”

“I think my one-sentence explanation is that with the era of internet scale foundation models, things that used to work maybe 20, 30 percent of the time are now working 60 to 70 percent of the time. And in robotics, right, as a very complicated, dynamic, engineered system with many pieces, in the past, if every small component of your entire system only worked 30 percent of the time, it would take many, many iterations to get a whole performance system working at scale. But now when every single part of the entire stack just works that much better, from the research iteration process to the engineering scaling process to the data collection engines, I think you can really just see the pace increase when you just have many more successes and a much higher hit rate when you're going about and scaling up your research.”

“Most people don't think about data integrity. They don't think about keeping their CRM clean. They don't think about any of this stuff. Okay. And so you're gonna have to keep that clean. The other thing too, is if you don't have good skill dot MD files, meaning that you don't have good processes within your company and you don't pass them over to, like, your Obsidian to have that central memory continue to get better over time in terms of how you make decisions, well, then this is not gonna be that helpful because garbage in garbage out.”

“It's very intuitive. So if you like, try to learn new, new sports, like do you go surfing or skiing? I feel like during the day, like when you started, it's really hard. But I found that like once you, once you like, if you go surfing for like two days or skiing for two days, like initially it's like really hard. And then you go, you sleep overnight and then you come back. And then you're immediately much better. And I like that in some way, the learning to learn faster paper has sort of mapped it into like, as Ted said, the day cycles and the night cycles, where the day cycle is sort of like in context learning, where you collect more examples, but then it's in context. And then the night cycle is like where you go retrain or find you change the weights of the model.”

“The hypothesis is that, of course, once the hardware matures in the quantum space, there are certain problems that you can address that are hard to address today. What we are focusing on is the optimization domains, obviously, and then if you go into things like logistics, traveling salesman problems, knapsack problems, like all these kind of usual hard problems in computer science, these are interesting problems where we believe that could be interesting for the future, for maybe a different kind of computing paradigm to solve for.”

“The AI race is so intense. It's now time to power, right? I need power, I don't care how you make it. If you talk about a gigawatt data center, that's almost a million US households of energy. The grid is just really not built for that. What that means is the data centers now have to bring their own power online because it's also important that when you plug in a data center, it consumes a lot of power from the grid.”

“And I would even emphasize that to expect such a result where the generalist outperforms specialists on the very niche domains that, you know, the specialists have kind of been overfit to, this was actually quite shocking to me. You know, like, I think there's been so many examples over the past years where people have tried to scale single task methods to multitask methods. And you definitely get a lot, you know, maybe you learn faster, you learn a more robust policy that's less brittle to small perturbations. But oftentimes, you have to give up raw performance, right? Generally, in a lot of cases, the only way to max out your performance on this one narrow regime that you care about is to train a specialist and overfit to that domain. And so it was really exciting here to kind of see positive transfer, where the generalist outperforms even this presumably very tuned baseline from the individual labs on their setups themselves.”

“Four point eight million Americans drive for a living. It's one of the most common jobs we have, and these workers do not plan to surrender to the California tech companies. They're doing this because they stand to make an unfathomable amount of money if they eliminate driving jobs for working class of people. These drivers are represented by unions backed by politicians and in cities across America blue cities. They're organizing. So far they're winning.”

“What hasn't changed is what customers are seeking for, which is outcomes, right? Outcomes and return on their investment in order to get the things done, right? And of course, now AI is an amazing technology that again helps to get more things done in the enterprise, right? And then that is actually what SAP is standing for, right? And so what we are really doing is in given, of course, also the breadth of the portfolio and the customers is, of course, to help customers to achieve more by deeply embedding AI, AI agents, and of course, transforming now the user interface.”

“The acquisition of Beacon following the acquisition of Kodiak, which is followed now by the acquisition of TopBuild, takes us from eleven months ago where we had no building products revenue, let alone EBITDA, to the second largest publicly traded building products distributor in North America with more than $18,000,000,000 in combined company revenue and more than $2,000,000,000 of combined adjusted EBITDA. It's a big deal in the industry and a big deal in the market as a whole.”