Dileep George - Co-Founder of Vicarious AI

(Joel Beasley at 00:00:00) Hello, my friends. Today we are talking to Dileep George, the co-founder at Vicarious AI. And we discussed their long-term goal of building artificial general intelligence, learning from failures to come back even stronger, and the relationship between feedforward and feedback information. All of this right here, right now on the Modern CTO podcast. Here we go.

(Joel Beasley at 00:00:23) This is the Modern CTO podcast.

(Dileep George at 00:00:34) Hello? Hey, Joel.

(Joel Beasley at 00:00:36) Hey, how are you, my friend?

(Dileep George at 00:00:37) Good. How are you?

(Joel Beasley at 00:00:39) Amazing. Let me just make sure to get my audio to come through my AirPods. All right, that should be good. So we're recording right now, and we can edit anything. So we record the whole time, and we just hang out and just talk. I was so excited for a couple of things. The first thing is these days in my life when I get to record these episodes are the best days of my life. I absolutely love them, and especially when I get to talk to people like you. So I did a ton of research on you.

(Joel Beasley at 00:01:12) I was super excited to get to speak today because you're a nerd like me, and we think a lot about different but similar things. And I was reading your thesis about, you know, how the brain might work, and I was reading about your company and its values, which I love that you post its values on the website. And then one of my favorite graphics on the entire website was you had the vision of the company, and you compared it against other companies. You had Microsoft, and you said at the beginning they started as a BASIC interpreter, then they built the operating system, and now there's a computer on every desk. And then for you, you know, you had industrial applications, AI for robots, and general-purpose robots. And that really helped paint a picture for me because I saw the robots and the arms, and then I was like, but then I also read your content. I was like, how do these all connect? And then that brought it together for me.

(Dileep George at 00:02:03) Oh, wow. Thank you. Yeah, I am really excited to talk to you too because, you know, I am a nerd at heart, and I like going deep into technology and science and building companies with a long-term mission. That's really exciting.

(Joel Beasley at 00:02:22) So what is the official long-term mission of Vicarious?

(Dileep George at 00:02:26) The official long-term mission is to build artificial general intelligence, and that's machines that are intelligent like us, which means building rich models of the world and being able to use those models of the world in completely novel situations that we haven't encountered before, and machines that have common sense and can know what's the right way to behave in new situations. So that is all what makes general intelligence. And building machines that have general intelligence is the long-term goal of the company.

(Joel Beasley at 00:03:03) And I've thought a lot about this.

(Dileep George at 00:03:05) Okay.

(Joel Beasley at 00:03:06) A lot. So about three or four years ago before I started the podcast, I had two paths in front of me. And the first path was going and building an AI startup, which specifically focused on storing memories so that you would be able to interact. It was for a social concept, like you would socially interact with maybe Alexa, and you could develop relationships, you know, companionship-type concepts. And I was doing the papers and understanding and researching it just because I thought it was the future. However, I said I couldn't come up with a good commercial application for it. Right? And I didn't think the world was ready for this type of commercial application. And so I said, okay, well, if I put my money into this and I'm off by a year or two, then I'm out. I'm out of the game. I'm not out of the game, but if I put money into the podcast and build relationships and get to know people, that will benefit me no matter what long-term. So I analyzed those two paths, and I picked the podcast one.

(Dileep George at 00:04:14) And it worked. It was good. That's great. I think it's a good decision because some of those applications, it's really the timing. And it has to be the right sweet spot in terms of how good the technology is and how well it works together in terms of how real-time it is. So a lot of things have to align, and the market needs to align at the right time. And probably this is not there even now with even all the advances in AI. So I think you made the right decision.

(Joel Beasley at 00:04:48) I was like, this is way farther off. Like, we're going to need you guys to have—it'll be a problem you're solving when you have robots in every house and people are starting to want to interact with them on a deeper level. But it was fun nonetheless to map it all out and attempt to solve it and come up with everything. But you found a way because, you know, you have this vision forward, and you found a way to make it commercially applicable and relevant today.

(Dileep George at 00:05:16) Yeah.

(Joel Beasley at 00:05:16) So what are you doing?

(Dileep George at 00:05:18) So initially when we started, we were focused purely on research, and it was just a bunch of friends who shared this mission of building artificial intelligence as a long-term goal, building a company around that idea. And for the first few years, it was a very small team, just six people, and focused on just doing research. And we always knew that we would want to build applications in the trajectory of the company, and we wanted to find applications that align with the long-term mission of building general intelligence. And it should be okay—as you build more and more powerful algorithms, it should benefit your applications. And robotics is, in many ways, the right application because, you know, as you build better understanding of the world around you, as your algorithms start understanding the world around you, we can put those algorithms on the robots to make them smarter, and they can be applied in more diverse settings. So algorithms getting smarter and more applications opening up align very well in robotics. And another good thing about robotics is that it has all the right ingredients. It has a body. It has—you need to interact with the real world. It is not just manipulating text, it is not just language. You need to understand the world and be able to interact with it. And that interaction with the world is a core requirement for an agent to have intelligence. And so that's also one reason why we picked robotics as the problem to work on.

(Joel Beasley at 00:07:07) I love it. And so you have these—today it's like these robotic arms, and you're doing these—it's called picking. Can you explain that?

(Dileep George at 00:07:16) Yeah. So this is something a five-year-old child can easily do, right? Which is, if you tell the child to pick up objects and pack them in a box, that is extremely easy for a five-year-old child to do. But most of the industrial applications or most of the manufacturing is just that, which is basically picking up objects and putting them together in a particular way. And that's our first application of robotics. In a warehouse or in an assembly line, pick up objects which are presented in a cluttered bin and being able to manipulate them the right way and putting it in a box or doing light assembly. And this looks like a very easy task, you know, because this is not theorem proving or playing chess or playing Go. This is something a child can do. But it turns out this is actually a hard problem and requires quite a bit of intelligence because you need to understand—for one, you need to have good perception of the world. You need to be able to understand how objects behave when you poke around them, and you need to be able to manipulate them and orient them in the right way to precisely place it in the assembly. So it has many, many different challenges, and it has wide variety of applications because this is 90% of manufacturing.

(Joel Beasley at 00:08:49) This is awesome. So I want to do a thought experiment with you. I think it'll be fun.

(Dileep George at 00:08:55) Okay.

(Joel Beasley at 00:08:55) All right. So imagine that we are on Mars, and we are looking down at Earth, right? And we're watching time pass from the early 1900s to today.

(Dileep George at 00:09:11) Uh-huh.

(Joel Beasley at 00:09:12) And, you know, all the lights start to come on and all the infrastructure change. And then you and I are up there, and we're aliens, okay? And we ask ourselves the question, what are the humans building?

(Dileep George at 00:09:25) Yeah. Oh, okay. We are building roads, man. We are building roads. Interesting. That is a very—would they be able to find a pattern? That's—oh, wow. That's a nice...

(Joel Beasley at 00:09:49) Yeah. I've been thinking about it a lot, and it, you know, I get different thoughts all the time. And so I'm now asking people this to see, you know, what pops into their head randomly.

(Dileep George at 00:10:02) Yeah. Yeah. I'm not able to find the connection from that to intelligence. I mean, we could say, oh, humans are building general-purpose intelligence or as a society they are becoming more intelligent. But the recent evidence is all against that. And so...

(Joel Beasley at 00:10:20) That's what I've heard too. Yeah. I feel like we've been getting smarter, but then I hear a lot of people say the evidence points against that.

(Dileep George at 00:10:27) Yeah. So, yeah, not sure.

(Joel Beasley at 00:10:31) Yeah. I guess a couple ideas that I had was it kind of looks like silicon is taking over the world. It's like silicon showed up, right? And it's like, if you look at the humans objectively, they're staring down at their devices all day, and then they make money to spend their money to build more devices, to build more technology, and they just—it's like we're worshiping technology and doing everything we can. And then, you know, I look at the hospitals and stuff, and that's all just, you know—there's infrastructure and energy, but all of these systems are just to help us do this work of making more technology. That was one idea that I had.

(Dileep George at 00:11:14) Yeah. Yeah. It's been compressed, though. You know, that's totally the last 10 years. If you look at from the 1900s until now, the electronic revolution—it's just so compressed, and most of the effects of that are being felt only now, right? We are sometimes seeing the Black Mirror kind of situations where we are all just plugged into a social network and being manipulated, kind of things. But that's only recent. Although the buildup for the technology revolution has been happening for a long time, the network-level effects of technology, I think, has happened only probably in the last 10 years.

(Joel Beasley at 00:11:54) If you could, would you go into the computer?

(Dileep George at 00:11:58) I would. But yeah. What would it be for? Because going into the computer doesn't tell you any more than what you are seeing from outside. So...

(Joel Beasley at 00:12:07) It would be like going into a video game, I guess. We'll narrow it for...

(Dileep George at 00:12:11) It will be like going into a video game, but you will enjoy it for a while. But finally, it is just a freeway and a bunch of switches and traffic lights. So, you know, that's what it is.

(Joel Beasley at 00:12:28) Yeah. That's interesting. I often think about—I'd say last weekend, I rewatched the Matrix movies.

(Dileep George at 00:12:39) Uh-huh.

(Joel Beasley at 00:12:39) I hadn't seen them for probably, you know, since they came out. So 10 years.

(Dileep George at 00:12:43) Yeah.

(Joel Beasley at 00:12:43) And I was blown away by the thoughts because when you're younger, you don't necessarily have the capacity to wrap your mind around the story. It's just a cool action movie.

(Dileep George at 00:12:54) Right.

(Joel Beasley at 00:12:54) But then as an adult, I was like, I wonder what that was actually going for. And then I watched it, and I was really surprised about how deep these thoughts were across these movies.

(Dileep George at 00:13:07) Yeah. Yeah. Yeah. Actually, yeah, I had the same thing because when I watched Matrix the first time, I didn't even understand what the thing was about. So everybody said this is a cool movie, and this is when I wasn't watching that many English-language movies. I was in India. I was a kid, and so I didn't understand what the big deal was about. And probably one year back, I watched the movie again. And wow, it makes—you know, it's beautiful what they have thought about. And of course, the premise is wrong about how the story comes about. But still, if you just suspend your disbelief and take that as the—you know, just humans are being used to extract energy, et cetera, just as the premise. But the rest of the things that they built around that is just amazing and totally makes sense because, you know, all that we see as the world are just electrical sensations coming into our nerves, and you can completely emulate that. And you can make you feel that you are in the real world just by giving the right impulses on your nerves, and you won't know the difference. And that is definitely beautifully captured in that movie. And then do you really want to know the real world versus do you, you know, would you rather live in an imaginary world where everything is nice? You know?

(Joel Beasley at 00:14:31) Yeah. I thought it was interesting, the human part of it, how they originally tried to make a world where everything was perfect, but then nobody liked it.

(Dileep George at 00:14:41) Yeah. Yeah.

(Joel Beasley at 00:14:42) So then they made a world where you work. And I'm like, well, that's an interesting concept. If everything is always the same, you know, we are naturally in an environment where chaos runs through the environment, right? And so how weird would it feel to be in a state of perfection where nothing ever changes? I think we would almost hate it to some degree.

(Dileep George at 00:15:04) Yeah. I think it will be like being in Singapore.

(Joel Beasley at 00:15:10) That's hilarious. I've seen one video about Singapore, and it was this video blogger who had talked about why he moved to Singapore. And he did make it seem perfect.

(Dileep George at 00:15:23) Yeah. It is very ordered. Nice. No trash on the street and very nice traffic patterns. Beautiful airport. Yep. So it's great.

(Joel Beasley at 00:15:36) Well, it's newer too, right?

(Dileep George at 00:15:38) Yeah. Yeah. It is newer too. Yeah. And it's well run. It's run like a company. It's a country being run like a company.

(Joel Beasley at 00:15:45) Really?

(Dileep George at 00:15:46) Yeah. Because it's a small country, and it's almost like a CEO at the top, I think. That's the way it is run.

(Joel Beasley at 00:15:54) That's interesting. I'll look more into that.

(Dileep George at 00:15:56) Yeah.

(Joel Beasley at 00:15:57) I haven't heard that analogy before or that relationship before. That's pretty cool. Do you know how big it is?

(Dileep George at 00:16:04) I don't know the population size, actually. No. I don't, on top of my head.

(Joel Beasley at 00:16:09) Come on. You're supposed to be prepared for this.

(Dileep George at 00:16:15) But I'm pretty sure my 10-year-old knows. He is just studying the population of different countries, and he just comes up with these numbers sometimes. And I always think that he's making it up, but then when I go and look it up, he's actually right.

(Joel Beasley at 00:16:36) Our producer just messaged me. He said 5.6 million.

(Dileep George at 00:16:40) 5.6 million. Okay. See, that's a small—that's a whole country. That's small.

(Joel Beasley at 00:16:46) Yeah. Yeah. So your son's 10, you said?

(Joel Beasley at 00:16:49) And he's interested in population sizes. What else is he interested in?

(Dilip George at 00:16:52) He's interested in companies. He is interested in SpaceX. He's a big fan of Elon Musk. Even when he was four years old, we had this car, electric car, the toy car that we got. He made me print out the Tesla logo and put it on the car.

(Dilip George at 00:17:11) And so, yeah, he's in those space, some amount of AI, but a lot about companies and now quite a bit into video games and trying to learn coding, et cetera.

(Joel Beasley at 00:17:27) Oh, that's pretty cool.

(Dilip George at 00:17:28) Yeah.

(Joel Beasley at 00:17:28) That's exciting. The world is so much more available today to all the technology. It's so cheap. You can Amazon engineering kits to you. You could do so much. I'm so excited for this. I'm excited to become old, right? Because if I've seen what happened when I knew how much technology was available when I was a kid, and it was not much at all. It was very rare. The place we'd go would be like RadioShack, you know? That's what my dad would take me to. And yeah, but now it's just everywhere, and it's cool.

(Dilip George at 00:18:09) Right. Right. Right. Yeah. Yeah. It is amazing. Compared to living in India before the Internet period, I grew up in a small village. And we had this small library in the — we call the Panchayat Library. This is the local library. I used to just go there and I just read up all the books there, so there was nothing left to read in the whole village. And then I had to go get books from some teachers who I knew had books, and I would borrow books from them. And so it's so different now even in that small place because now because of the Internet, information is available everywhere, right? So when I was growing up, I had no idea — okay, what is Stanford like or what does studying in America look like or what is a startup company? All that information is available now. Kids in my place, the same place, now twenty years later, they know all these things because it's available on their fingertips, which is amazing. Different time to grow up in.

(Joel Beasley at 00:19:16) Yeah. I'm looking forward to Elon Musk finishing his Neuralink because then that'll change the world entirely. You just have the information right on tap.

(Dilip George at 00:19:24) That's right. Yeah. I'm looking forward to that too.

(Joel Beasley at 00:19:28) Have you seen the pictures of the Neuralink machine, the giant alien looking device that they'll use to actually do the implantation of the device?

(Dilip George at 00:19:41) No. I haven't seen the Neuralink one, but I have seen similar ones. You know, it's basically a surgical placement machine, right? Yeah. Yeah. So I haven't seen the Neuralink one. Yeah. But those are cool.

(Joel Beasley at 00:19:54) Would you get the Neuralink? How long would it have to be out for you to get it as a consumer?

(Dilip George at 00:20:02) Yeah. I don't think I will get it anytime soon. I don't want to mess up my brain directly like that, you know? So—

(Joel Beasley at 00:20:15) I think the word's enhanced.

(Dilip George at 00:20:18) Yeah. Yeah.

(Joel Beasley at 00:20:19) For me, I would have to wait and I would wait until a person — I would need to see the people I know that get it and have it for a while and have incredible benefits. Yeah. Yeah. I would need to see it like that, or it would have to be nonsurgical. I'd be more interested in it if it was nonsurgical. Right?

(Dilip George at 00:20:44) Yeah. And also, what are the long-term effects? That too, you know. So let's say they had a great ten years, and then it just goes down, you know, after that.

(Joel Beasley at 00:20:55) Oh, that's true. That is true. You've got that going on. Or yeah. There's all sorts — my mind is just iterating through every single possible issue right now, and it's just too many to even list. But they have a very smart approach. They're solving very real problems, people with disabilities, people with horrible diseases, and they're starting small and there's money there. You can solve these problems and get paid. Then you have — and then it's basically you've got this engine that you're growing that's got all this collective knowledge and all of these great people. And then you can finance your own internal R&D projects and constantly re-improve the product. And yeah, it's very, very cool.

(Dilip George at 00:21:39) Yeah. They are very disciplined, and they do have short-term goals that are very, very practical and can make money mostly in the research setting, but also probably in therapeutics. So it's — I think it's very cool. Although the final outcome is science fiction-like, the route to that science fiction-like outcome is nowhere like science fiction. You know, incremental steps onto that destination.

(Joel Beasley at 00:22:09) Yeah. The amount of time it takes is important because it gets us all adjusted to it, and it just kinda happens. One of the cool conversations I was having with Sri, who's CTO of PayPal—

(Dilip George at 00:22:24) Uh-huh.

(Joel Beasley at 00:22:24) He was talking about ambient payments. So when you get out of a Lyft or an Uber, when do you actually pay?

(Dilip George at 00:22:31) Right.

(Joel Beasley at 00:22:31) The payments are happening in an ambient way, and it's like our technology, I feel, progresses like that too. It just kinda happens.

(Dilip George at 00:22:37) Right. Right. Yeah.

(Joel Beasley at 00:22:39) So are you excited? Do you get to be mad scientists over there and run a bunch of futuristic projects, or is it just focused on the current clients and the arms?

(Dilip George at 00:22:50) Well, it varies. I think part of the challenge in running a company like YK is that you have to balance the long-term vision and long-term futuristic projects with the short-term needs or the customer needs. And you need to have a very clear judgment on what's the purpose of each project that you're doing. You know, some of those projects, you are very clear, right, that this is not something I'm shipping to the customer in the next month or two months or even a year. You know that the outcome of this project is going to be a paper, and it is cool. It is — the real applications is three years out, but it is something that you need to do now as research on the path to general intelligence. So some of our projects are like that, and you need to be very clear about that. You don't want to take your research projects, real research projects, and say, oh, we are going to force it into the product right now. That never works. You really need to be clear about the engineering side really needs to be driven by — okay, these are the customer problems that need to be solved, and this is the approach that will solve those problems and with the right constraints. You know, because when you're thinking about a robot that needs to work in a warehouse 24 hours a day with high accuracy and a very low failure rate, there are some very stringent requirements for the kind of algorithms that we can put in. And also, you know, the speed of operations, there are stringent requirements on the cycle time, how fast should these algorithms work. So that is not — you don't want to always push your most sophisticated algorithms into that production all the time. So you really need to run the engineering part of your organization based on real customer needs and real customer constraints. And then the research side, yeah, you really need to be ambitious there. And there, you do want to keep the problems from the real world in mind, but not necessarily all the constraints. For example, the speed of processing. That constraint, you don't necessarily want to keep on your research side because, you know, that's just a temporary constraint based on the processing power available now on a particular GPU. So there, you can take those risks. You can still be informed by the needs from the customer side, the problems from the customer side, but you don't want all the constraints from the customer side. So you have to keep that in mind while doing the research.

(Joel Beasley at 00:25:46) And how did you come up with the name?

(Dilip George at 00:25:48) Oh, Vicarious. Well, yeah, we thought about lots of names. And we thought about, you know, what would be one word that signifies the kind of intelligence that we are trying to build. And if you think about, you know, the highest form of intelligence is being able to be in someone else's shoes, being able to see the world from somebody else's viewpoint. Basically, being able to model how you think about the world in addition to my own model about the world. And so that is experiencing things vicariously. And so that's the highest form of intelligence. And there are multiple levels to this. One, basically saying that just being able to think about something without actually doing it is also called vicarious evaluation. You know, it's called vicarious trial and error in science literature. So there are different levels of depths to this word, so that's why we liked it. And so that's why we chose that as the name for the company.

(Joel Beasley at 00:26:51) I love that. So this highest form of intelligence, is there some ranking scale, or is that just subjective?

(Dilip George at 00:27:02) No. It's not, I don't think it is subjective. I do think you can put them on a scale. So just think about the evolutionary history of intelligence, right? So we, you know, the evolution has run for millions of years, and we have had very simple creatures who can just crawl around in the world and react to the environmental stimuli. And they have — they evolved some responses to environmental stimuli which were adequate to make them survive in a niche. Most of the creatures in the world are like that. Just, you know, adaptive responses that make them survive in their own ecological niche. So those creatures, yeah, they have some very rudimentary model of the world, but nothing like a model of the world that humans have. So even that rudimentary model of the world is enough to make them survive in — if it is just survival, you can just do with some rudimentary stimulus-response mappings learned through evolution. But once you get to animals like mammals, you know, they have a completely different architecture. It's not the same kind of circuits that powered reptiles that is powering mammals. You know, we have something called the neocortex, which is a completely different architecture compared to the old brain. And that new architecture allows these animals to build better models of the world. So you can say that horse's model of the world is quantitatively, you know, richer than a mosquito's model of the world. So you can measure that power of modeling the world. And now once you start increasing that power of modeling the world, you start, you know, modeling things around you, how other objects behave, then you start modeling how other intelligent agents behave. That becomes part of your model of the world. And when you start understanding how other intelligent agents behave, that's when you can start being really vicarious.

(Joel Beasley at 00:29:14) You are brilliant. Yes. Computational or the power of modeling the world as the scale of intelligence. And I guess, you know, my limitations are really I'm just trying to think as a human. So I can — I refer to it as a superpower. I talk about it a lot, actually—

(Dilip George at 00:29:34) Uh-huh.

(Joel Beasley at 00:29:34) About how I can understand how a character in a movie would think. Right? I can make — I can decision like a specific character that we're all — let's say, you know, Neo from the Matrix. We can decision like him—

(Dilip George at 00:29:47) Yeah.

(Joel Beasley at 00:29:47) After you've consumed enough of his decision making skill. And we're actually pretty efficient at it because we can think like people after watching a movie. We could think like one of the characters.

(Dilip George at 00:29:57) Yes.

(Joel Beasley at 00:29:57) It's really, really interesting. However, a limitation is me thinking in parallel. I can't of two different things at the same time. However, a computing system could do this.

(Dilip George at 00:30:11) Correct.

(Joel Beasley at 00:30:12) That would be the evolution, right? It's like what comes after — if you look at the single-celled organisms to the reptiles, and you're looking at their computational power as a scale for intelligence, and then you get to us, and we've got this vicarious concept. But then you apply multidimensional array to that and make decision making off of simultaneous thoughts. That is the next evolutionary level, I would believe. Right?

(Dilip George at 00:30:39) That is true. There will be many cool things we can do with — you know, this is, of course, science fiction development, but that would be one way in which you can try to make it more general than human intelligence because now you can have multiple conversations in your mind in parallel, multiple personalities, and then try to somehow, you know, converge onto something more coherent than being able to run one personality at a time. So those would be very interesting because some of them might not be algorithmic limitations. They might be purely because our evolutionary hardware limited us from being able to run, you know, multiple parallel personalities. So, yeah, those would be very interesting kinds of experiments.

(Joel Beasley at 00:31:26) Yeah. It would be, especially if you start looking at neural patterns or whatever the common term — I'm not in this space, so I don't know. Look at the brain patterns of, or structures of data. I was reading a lot about your research paper, and you've actually — the hierarchical concepts of how they store memories in the brain and how the brain works.

(Dilip George at 00:31:48) Yeah.

(Joel Beasley at 00:31:48) And if you look at a schizophrenic person, they seem to be able to have this ability to run different things. I don't know if it's simultaneously because it seems like they switch, but they still would have — they would — I bet you if you just look under the hood, there's something interesting happening there that give them the multiple personalities that are separate and unknown of each other.

(Dilip George at 00:32:11) Yeah. So one thing I think — and this one we can understand with even our current models of the brain — is the balance between feedforward information and the feedback information. So, you know, when you look into the brain, you see that there are a lot more feedback connections. So connections going in the other direction of the sensory system, you know, so from the eye to your inferotemporal cortex, that's a feedforward direction that's away from the sensors. You're taking it multiple processing steps. But most of the connections run in the other direction, which is — and those are the connections which are making the predictions about, you know, about the world based on the model it has learned. And it is this imbalance between the feedback connection — so if you can run the model in a way such that your sensors are not giving you much information. It's your feedback which is being fed into yourself. So you're basically running your own model of the world back to you, and the sensors do not pass through.

(Dileep George at 00:33:16) So you can get into a model like that, and that happens because of the imbalance between the feedback information and the feedforward information. And I think we can probably even induce that using drugs or by changing the mode of the neural firing. So some of these neural phenomena, which schizophrenia is one of them, I think we can explain those things partially using the current model. Some of the models that we have built can explain those things.

(Joel Beasley at 00:33:45) That is fascinating. So tell me if I pulled this in right. So you have—I had to make notes while you're talking—you have the feedforward information and the feedback information. And the way you were describing it, it sounded like you said that the decisioning is happening where there's the most connections, which in my mind reminded me of processing data at the edge versus sending it back to a central computer system. Are these decisionings happening at the edge? Is that what we're saying?

(Dileep George at 00:34:14) No. It's more, so you have an internal model of the world. You know, you have learned the model of the world, and you are all the time using this model to make predictions about the world. And your own perception is a mix of your own internal model of the world interacting with the sensors coming in. So you are never seeing the world as it purely is. You are always filtering the world through your internal model. And your final perception is the result of filtering your sensors through your internal model of the world. So this feedback connection is used for filtering or reinterpreting your sensory data based on what you've already learned, based on your model that you've already built. And now this mechanism, most of the time, works just right, but you can change the balance of these things. You can basically make the sensors go away, or you can make it feel like your own predictions are the ones which is being sent. So you can think that your own imagination is the real world, just by changing the balance of those feedback circuits versus the feedforward circuit.

(Joel Beasley at 00:35:25) Oh, you hotwire your brain. You can actually think your imagination's real. I think I dated a few girls like that. No, just kidding.

(Dileep George at 00:35:33) No. Just kidding.

(Joel Beasley at 00:35:36) Oh, my high school relationships. No. But that's actually pretty interesting. I have a pretty strong imagination. Actually, a big shaping factor in my maturity was learning the limits and the abilities to control and discipline my imagination because it's active. And so I have to isolate it to tasks that require it. You know? Just because it wants to run all the time. Because if I let it run rampant, then I would end up doing nothing with my life but wandering around in an imaginative state about all the future things I could do.

(Dileep George at 00:36:17) Yeah. That's like, you know—I don't know whether you have read Calvin and Hobbes, the cartoon series. So in that one, you know, Calvin is this five-year-old kid, six-year-old kid who has vivid imagination. And whenever he's given a math problem, his mind goes into this vivid imagination. He converts it into being, you know, I am fighting with alien zombies, this thing, and and then the problem doesn't get solved. So, yeah. It's really, being able to balance how you use the imagination. And, of course, you know, even when you have the imagination, you don't get confused with what is imagination and what is reality. But you can with some neural pathologies. So you can get confused with, you know, oh, this my imagination is the real thing, and that is when things go wrong.

(Joel Beasley at 00:37:12) Do you need to grab your AirPod?

(Dileep George at 00:37:14) Yes. I will grab my AirPod.

(Joel Beasley at 00:37:16) We'll put a little bumper together for the episode. We'll say, you laugh so hard your AirPods will fall out. Right? That's how we'll promote the episode.

(Dileep George at 00:37:28) Perfect.

(Joel Beasley at 00:37:29) So we're geeky people. Have you gotten into the Roombas or the automated vacuums at all?

(Dileep George at 00:37:36) Yes. Actually, I did. So the funny story is that when we got married—this was in 2005 or something. I forgot the exact year now. My office mates gave me, okay, here is a check for you to buy furniture and whatever. You know? We were grad students, so that was a big check at that time. And what did I buy? We bought a Roomba.

(Joel Beasley at 00:38:01) It's smart. You gotta keep the place clean.

(Dileep George at 00:38:05) Yeah. Yeah. Yeah. And at that time, Roomba was—you know, I figured that I am cleaning Roomba more often than Roomba is cleaning the room. You know, it was like, it would get stuck on hair or, you know, it would suck in a piece of paper, things like that. And, you know, I have to clean it more often. So I ended up, I think, giving it away. And then recently, I haven't bought one because I have two boys, and they don't keep the floor clean at all for it. You know? I think my house will be the most adversarial space for something like Roomba to work in. So I mean, I haven't got one recently.

(Joel Beasley at 00:38:45) That's a good point to make because I have two kids too that are young, and there are just toys everywhere. And, you know, we try to clean up every night, but it's basically the moment you're done cleaning up, it's like their game to go break everything down.

(Dileep George at 00:38:59) Right. Exactly. Yeah. And yeah. I don't think we are very consistent in cleaning up every night, and sometimes we're like, it's fine.

(Joel Beasley at 00:39:10) We actually so we were gonna get the Roomba because, apparently, the new ones can empty themselves.

(Dileep George at 00:39:17) Uh-huh.

(Joel Beasley at 00:39:17) That's pretty cool. But instead, my wife went with this Dyson one that looks unbelievable. And I played with it a little bit, and it was fun to use, but it's like it looks like a cartoon gun or something. It goes when you use it, and it's honestly, it's one of the coolest things. I would never have told you that I got excited about a vacuum before until she used it or she set it up a week ago and used it, and I was like, this is actually pretty cool.

(Dileep George at 00:39:58) Oh, okay. My wife just asked me to buy one, so maybe I will look into the—and what I heard is that it's the pleasure, it's a feedback of seeing the dirt being sucked in. You know? If you can see it, that gives you the visceral feedback that you need. You know? That's probably the trick there.

(Joel Beasley at 00:40:17) Oh, I believe that. But the thing that got me going is the fact that it's battery powered. And so you just walk around with it, and there's no cord, and it lasts for more than enough time that you need. And then you just hang it on the stand, and then the stand's powered, so it just charges it. When you put it in the holder when you're done, it's being charged. And I was like, I don't know. I've been talking about it now. That's how excited I got.

(Dileep George at 00:40:43) It's amazing that vacuum cleaner is this category of things that has gotten so much attention from robotics, designers, mechanical engineers. I think it's just amazing the amount of innovation that has happened in the space. You know? I was watching one video from IDEO, and he was describing, okay, here, we designed a vacuum cleaner where, you know, you can keep vacuuming. And when you reach there, you can press a button, and the cord will unplug itself and come back to you.

(Joel Beasley at 00:41:17) You get a whiplash. It's gonna hurt. It does not sound good.

(Dileep George at 00:41:22) And then you can plug it in right there, then continue vacuuming, and then, you know, keep going. You know? So—

(Joel Beasley at 00:41:27) There's scratches all over the walls.

(Dileep George at 00:41:32) It's amazing the amount of things that people have tried to make this vacuuming so efficient.

(Joel Beasley at 00:41:39) Oh, I love it. Yeah. I'll send you, after the show, I'll talk to my wife because there's 10 different models to say which one did we get, and then I'll email it over to you so you see at least—

(Dileep George at 00:41:48) Perfect. That'll be great. Yep. Thanks.

(Joel Beasley at 00:41:51) Yeah. There I go. We should get that as a sponsor. So we should call them up. Yeah. Okay. I think it's Dyson. Yeah. Hey, Dyson. So when you got to go out and raise money and build up the company, did you get to actually meet Musk or Bezos?

(Dileep George at 00:42:09) Musk? Yes. Elon, I have met a couple of times. Bezos, I haven't met. But Scott has met Bezos at a conference and maybe once later too. Yeah. So, yeah, we met some of them.

(Joel Beasley at 00:42:23) Have you read their books, their life stories?

(Dileep George at 00:42:27) I haven't read the one about Elon. I just didn't want to get too disappointed with myself. So—

(Joel Beasley at 00:42:36) How so?

(Dileep George at 00:42:38) You know, what I heard is that it's kind of the Superman, the Iron Man kind of story. So I know.

(Joel Beasley at 00:42:48) There's still time, my friend. We'll get some marketing people, some PR—we'll put a PR team together for you. We'll get a superhero cast after you.

(Dileep George at 00:42:57) Yeah. Yeah. Yeah. But, of course, it's a very truly inspiring story.

(Joel Beasley at 00:43:02) I do that too, by the way. I put myself at a ridiculously high standard, and I'm constantly having to be patient with myself. But I found that I get very depressed if I'm not attempting to push myself beyond what I've done before, and I have a very low patience threshold. So I actually tracked it in a journal. Okay, I'm upset and I give up. How long does it take me to shake out of that? And it's sub a week. It's less than a week. I just bounce back and I'm like, okay, I gotta do something bigger. And I just—it's this fire inside of me and it's—I think it's a lot of little things. I don't think it's exactly one thing you can point to, but I think it's partly me choosing, partly things happening to me, and it's just kind of this experience I've been going through in life, and it's been such a big part of my life that I can only see it existing forever inside of me.

(Dileep George at 00:44:05) Yeah. And what I have noticed is that once you start taking up more challenging problems, you start treating the other, the earlier ones you thought of as challenging as, oh, that's not important. I've seen this before. It's fine. It will work out fine. You know? I don't need to stress out about that as I used to before. And I think that's one thing you learn with experience. You know, you can really absorb the punch from lots of problems and still recover. And some of them, you recover by even not doing anything. It's fine. You know, you don't have to stress out over every little thing. And I think oftentimes it is taking up a larger challenge that gives you those lessons. And when you don't have those challenges, when you have those bigger problems to work on, then you get distracted by the smaller challenges, the niggly little problems, and you spend a lot of energy on those little problems. Whereas when you have something even bigger to worry about, those other things just go away. You'll see that those don't matter anymore.

(Joel Beasley at 00:45:09) That is very true. How do you get bigger, better problems? And that's actually some of the ways I track my progress is what type of big problems am I solving now? Or I'll hear myself say a sentence and I'll say, whoa. That's—while that's a problem, it's amazing that I'm dealing with that problem. Because I'd given myself 15 years to get to this point, and I got here in three or four. And so I was—I'm just constantly amazed. I'm, and I'm also very grateful. And then that's an easier way to start the day. I go on a run, and I think about how grateful I am and how excited I am about the different things that are going to happen and how I'm pretty confident about what 10 years will look like for me. But I can't—I mean, 20 or 30 years? It's just exciting. It's exciting.

(Dileep George at 00:46:01) Yeah. Yeah. It is. It is. And that attitude is something important to latch on to and keep alive throughout. It's not easy. You know? It doesn't come naturally to me. You know? I have to kind of work on it. But I've developed some mnemonics to help me. You know? Whenever some bad thought triggers me, I have developed some tricks to kind of flip it. You know? One way in which, you know, we often get mad about the world is that, oh, this is so unfair. The world is so unfair. And why is this happening to me? And the trick I'm flipping is that the world is so unfair. It's so great, and that's why I'm able to do these things. So think of it—it is true. If the world wasn't unfair, I wouldn't be here doing these things. It is—I am the result of, I am exploiting the result of the world being unfair. So it is cool that way. So I've developed some of these tricks to kind of keep the gratitude up and focus on the positive things, which is very, very important.

(Joel Beasley at 00:47:08) Yeah. Do you have any other tricks?

(Dileep George at 00:47:11) I don't want to give away everything. Come on, Charles.

(Joel Beasley at 00:47:14) Well, I'll share one of mine. You wanna hear one of mine? And that's what we wanna do. We wanna give away all of them because 99% of the people won't do anything with it, and then there will be one person that'll come up to you in 10 years and say, oh, I heard this, and I used it, and I applied it, and it helped me. So one of the ones I use a lot, and when I first heard it, I didn't think much of it, but it just stuck. And I heard a Navy Seal speaking. And he said he has this concept for decision making, and it's called HALT, H-A-L-T. And he says, when he's hungry, angry, lazy, or tired, he won't make a decision. Because you make the worst decisions when you're hungry, angry, lazy, and tired. When you're in those states of mind, you make horrible decisions. And so he doesn't make important decisions, big things, when he's in one of those states. He gets himself into a good state and then makes the decisions.

(Dileep George at 00:48:12) Yeah. Yeah. Yeah. Often we, you know, basically say in the company, on some of these things, you know, let's sleep on it and absorb the context of the problem, and give it a day and get some rest before making a decision. So that definitely resonates with me.

(Joel Beasley at 00:48:29) Do you have one more tip to share?

(Dileep George at 00:48:33) Okay. Let me think.

(Joel Beasley at 00:48:35) Make it the best one. Make it the one you wanna share least. The one that's most impactful to you.

(Dileep George at 00:48:41) Oh, yeah. Yeah. Yeah. Yeah. Oh, I'm blanking so much on that now. You know, I have so many of these hacks, but nothing is coming to my mind right now. You know, one is, of course, writing a journal every day. No. That that is a habit. When I look back, I know I used to have that habit whenever things were challenging for me.

(Dilip George at 00:49:03) And then when things go easy, I kind of drop that habit. But I've always found that doing that is something that helps. And one of my wishes is to be able to stick to that habit, you know, just through and through. My mom, she has done it throughout her life. She has been really good and she has faced a lot more adversity compared to me and, you know, what she has done with her life.

(Dilip George at 00:49:32) And I think being centered by writing that journal every day is something that has helped her. And I think that is definitely a powerful moment.

(Joel Beasley at 00:49:41) So when you would do this journaling, this habit, would you do it in the morning or the evening, or how did you do it?

(Dilip George at 00:49:46) Yeah. In the evening. I'm not a morning person at all. So it's at 1AM, 2AM, that kind of thing. Yeah.

(Joel Beasley at 00:49:55) So you, at the end of your day, you're sort of sorting your thoughts. And one of the things that I have done, and I think you stated it beautifully, is I drop the habit when I'm not doing something that's too challenging. Because you start to seek these tools and implement these tools when you're doing the impossible because it's just so important. And I found that the thing that journaling helped me with is the focus, the persistence of the thought. Because your brain constantly wants to change these thoughts.

(Joel Beasley at 00:50:28) And so, like, building a business, it's very difficult. Right? It's not easy, because you have to be doing work, but making sure that work generates revenue at a rate to which you're cash flow positive. Right?

(Dilip George at 00:50:39) Yeah.

(Joel Beasley at 00:50:39) Yeah. And so, you're always focused on it. And I found that the journaling would just help set the tone. I would write at night and I would reread it in the morning.

(Joel Beasley at 00:50:48) I did that as a journaling thing. I also did, like, future authoring where I'd make a bullet point of, like, where, at the end of this week, I'm here or at the end of the month, I'm here, and I would actually write it out, you know, for twenty minutes at the start of my workday. Those were useful, but I think the most useful thing is what you said, which is when you realize you're not journaling, it's time to, because by the time you realize it, it's already happened for a couple weeks or a couple months.

(Joel Beasley at 00:51:19) So when you realize you're not journaling, you kind of have to reevaluate, like, what am I doing right now? Because you're not going to your full potential. And that's a tough thing, and I don't expect, like, I give myself seasons. So I started, this is something new that I started this year where I'm doing twelve weeks on, four weeks off.

(Joel Beasley at 00:51:42) So twelve weeks of all the best I can possibly do and then four weeks of maintenance.

(Dilip George at 00:51:48) Uh-huh.

(Joel Beasley at 00:51:48) And so I'm working on that because I tried doing, like, 110. And as you know, like, founders, we all reach it. We just burn out and my health is a problem. I have to allocate time to my health. Otherwise, I'm not gonna be here to continue the work. I gotta allocate time to my family and my relationships and all of that. I've gotta take a step away from the work to get that renewed energy again. It's all very important.

(Dilip George at 00:52:13) So one thing came back to me. This is something I always use. Treating constraints as a positive thing rather than a negative thing. You know? Saying that, oh, I don't have enough resources or this thing is blocking me and treating those constraints as something positive rather than negative.

(Dilip George at 00:52:31) Because, you know, even while growing up, when I was looking at, you know, I used to complain at that time that, oh, we don't have TV, you know, or we don't have this thing. But those are great not to have. And because that channeled my energy in completely different directions and made me focus in very different directions. And had I had much more resources than I was, you know, at that time, probably I wouldn't have been as, you know, I wouldn't have been able to use my energy creatively. So I consider constraints, and this is something you have to kind of consciously do.

(Dilip George at 00:53:11) But once you get the habit of doing it, you start seeing that, okay, I can use that constraint as a superpower. And so, yeah, that's a, I would say, good habit to have, and it's a good habit.

(Joel Beasley at 00:53:23) Yes. We have a phrase around the office. We say constraints breed creativity.

(Dilip George at 00:53:29) Yeah. Yeah. Exactly.

(Joel Beasley at 00:53:30) Yeah. Oh, this is so fantastic. Thank you so much for coming out.

(Dilip George at 00:53:36) This was so much fun. Very, very different from, you know, I thought you'd be just asking some boring robots questions or something. How do you make the, how do you solve the path planning problem? You know? How, you know, instead of RRT, what would you use or something like that? You know? But this was a lot more fun.

(Joel Beasley at 00:53:59) Thank you so much. I'll take that as a compliment.

(Dilip George at 00:54:03) Thanks, Joel. No, it was fun. Yeah.

(Joel Beasley at 00:54:05) It was unbelievable. You were the best. Thank you so much.

(Dilip George at 00:54:07) Thank you. Thank you. Makes me feel good. Thanks, Joel. And thanks everybody else for supporting him. Thank you. Thanks a lot.

(Joel Beasley at 00:54:15) Thank you. Have a great day.

(Dilip George at 00:54:17) You too. Bye bye.

(Joel Beasley at 00:54:20) Thank you so much for listening. And if you found this episode useful, please share it with a friend or colleague who you think would get value from it. And if you have topics that you would like to hear discussed on the podcast, either add me on LinkedIn or send me an email [email protected]. Every time I get an email or LinkedIn message, it absolutely makes my day and inspires me to keep going.