Timothy Busbice, founder of Chrome, argues that ChatGPT and other large language models lack real-world context because they're trained on static data and can't perceive physical environments. He's building AI inspired by fruit fly nervous systems instead, training it on live sensory input to control drones for wildfire detection and suppression.
Chapters
All right, welcome to the show. This is my special guest, Brooklyn Ollie. He's an Australian Shepherd, 3 years old. And my other guest, Timothy Bustos, the founder of Chrome and the father of biologic intelligence. Sponsored today by LaCroix. No, I'm not, just kidding. Not really. Um, what keeps me going. All right, buddy, come on, get down. You're gonna do something. Okay, lay down, bud. So yeah, so let's see here. I do want to get an update. I think, you know, since the The last time we spoke publicly, we've had a little something called ChatGPT and large language models take the world by storm. So I think it's probably worthwhile to talk about the difference between deep learning, LLMs, and biologic intelligence, which AI is beginning to exist on a continuum. Throw machine learning in there as well. And regression lines all the way on the left end. So yeah, maybe we just start there with a quick overview from your perspective. You've been in this space for a long, long time. What do you think about that? Start there and then—
Yeah, that's fine. Yeah.
Okay, cool.
So you wanna start with ChatGPT?
Yeah, sure. I mean, I think maybe one of the things is how, why couldn't you put an LLM in one of your robots and have the same performance that Chrome has in one of your robots? What's different? What's the same? What's different? You know, people aren't experts, high level.
Yeah. Yeah. LLM. Well, you know, there are people working on a robotic version of that, but it reminds me a lot of the IBM Watson. It's really semantic databases. We call them now vector databases, but it's just, just pretty much the same thing. But vector databases are a little bit more high performance. And so like I use ChatGPT regularly, okay? And, and, and many— there's many forms out there for development That I also plug into. I've used Bard and Gemini and all the different ones. For programming, it's still pretty shy on expertise. It'll give you— it'll write a program for you, but whether it works or not is a real gamble.
Mm-hmm.
And usually the type of programming I need it for, it doesn't work, or it Yeah, just sometimes it creates things that don't exist, things like that.
So it's God. It's a God that's creating things that don't exist. No, I'm just kidding.
Well, I mean, things like, you know, they'll put things in the program that, you know, oh, you should import, you should import this library. Well, that library can't be found.
Yeah, you made it up.
Yeah, yeah, they made it up. Yeah, so it's gotten better. It's gotten better from a year ago. Anyway, uh, but it's still, uh, I would say a year ago it is a real 50/50 chance that anything they gave you would work. Now it may be closer to 60% chance that it'll work, but still very far away from being a programmer or developer. And I know people are working on it, um, and working on— there's a product that just came out which I have not seen or used called Devin. It's supposed to be a programmer.
Yeah, I've seen that. I've seen that video.
Yeah, so, you know, I, I, I, I think it's got a long ways to go. And of course, you hear all the hype and garbage that's out there, uh, about OpenAI and the things they're supposed to be doing, and they talk about version 5 and how it's gonna, you know, revolutionize Everything. Well, of course you heard that from going from 3 to 4, etc. So, and then they talk about things like, uh, hallucinations. Well, hallucination is really just the application going amok because it doesn't know what to do or how to do it. It— the bottom line is biologic intelligence, what we're developing, has context to what is real. What, what is the What it senses, it has context to understand. ChatGPT and LLMs and the like have no context of understanding.
Until you give it to them.
No, I mean, well, you can tell it, you know, draw me a picture of a cup, but it has no clue what a cup is. It knows what it knows, where it can grab libraries to say this is a cup. But it doesn't, it doesn't know. It never used a cup, never sensed a cup. All it knows is that it's been trained to say this is a cup.
Mm-hmm.
And that's the huge difference between general intelligence and AI. And, and from the biggest perspective is that AI, even today, even with ChatGPT 4, 5, 6, 7, whatever, if it doesn't have a context of what it's talking about, It'll never— it never will progress to a point where it, it really fully understands and could be general. I, I don't see it. I mean, I just really don't see it. I see the failures all the time. Uh, I, I listen to Gary Marcus a lot. I don't know if you're familiar with Gary or not. Uh, I've met Gary and, um, he's actually chimed in on some of my Twitter, uh, feeds, but Um, you know, he's very critical, and he— and I have to agree with just about everything he ever says. It's just, just, it, it doesn't have context. It's never going to be able to generalize what, what, how to operate in a world in a general sense.
Yeah, I mean, they, they use terms and phrases like context windows, so I think most people who probably are listening and have studied it will go, well, that's wrong. They have context windows and the context windows are getting larger and we use larger chips and larger data centers. And I think the difference of what you're saying is it's training data, right? Like we gotta give you 1,000 cups, 10,000 cups, and we've labeled them as such using humans that then you, it's cup. And so here's the pixel orientation from showing it.
Yeah.
10,000 cups. But then what you're saying about biologic intelligence is there is no training data. You just turn it on and using its senses, it, I don't know if understand is the right word, but it senses its environment. And then maybe years later you might say, well, that thing's a cup and then it would map a cup to whatever. But Either way, it doesn't matter. I mean, if you're on Mars or in a well or in a mine and you've never seen a rock before, do you care? Or a pipe? Or do you just want to like get out of the way of the steam or falling rocks? Or do you need 10,000 images and videos of falling rocks before you save your own life? Like, I think that's maybe where the difference is.
But that example there, you know, you have, You could create a robot today that would realize that this object is coming towards it, and it doesn't— it may not understand what that object is, but it's smart. It would have enough intelligence to know, I don't want that object hitting me, so I'll move myself out of the way. But again, it doesn't know if that's a piece of foam rubber coming at it or, you know, a 1-ton rock.
Mm-hmm.
And so, It's going to move no matter what, but it has no context again as to what actually is coming towards it. Is that a piece of fabric, you know, like a parachute or something that, uh, you know, was ejected, uh, upon entry? Or, you know, what, what, what exactly is that material coming at me? All it knows is something's coming at me and I don't want to be around anything that's coming at me. So it'll jump out of the way or move out of the way, but We, we would know better, you know, humans would know better, we— or animals would know better, is that, you know, oh, this is just a plastic bag, you know, getting blown across the landscape, right, versus, you know, a 5-ton rock coming rolling towards us ready to smash us into the ground, right?
So then the software then that's in the robots, like Right. I mean, this is a critical thing. It's like people are using LLMs to power robotics and obviously we've been using our proprietary biologic intelligence artificial Connectome software to do it. So I mean, basically without all the how the sausage is made, what type of, like if grandma or 5-year-old kid was sitting there, what type of differences do you think they would experience? Watching both ones go side by side.
Like both ones of what? I lost you there.
Like, like one robot powered by LLMs and another robot powered by biologic intelligence.
Oh yeah, because the one powered by biologic intelligence would be, would adapt to its environment like we would, where LLMs, it would, it would have to pause and figure out what, what the environment was. In other words, we can cycle through our environments without any forethought. It's a no-brainer for us, so to speak, because we, you know, we understand the environments that we, we move into, even ones we'd never been into before.
Mm-hmm.
My— I always say, you know, I, I could take a 3-year-old child that's never seen an office chair in its life, uh, bring that child into my, my office, and the child would have no issue, no problem understanding that my office chair was a chair.
Mm-hmm.
Probably climb into it and, you know, do whatever. So whereas, uh, a typical AI— and I include deep learning and LLMs and everything in that same context— they— it would have to be trained on, oh, this is an office chair. Versus this is a sofa chair, or this is a dining room table chair.
Yeah.
It would describe it as something. And there's, there's plenty of classic examples out there where they've shown pictures of things to, um, classic AI anyway, and it still identified what it was, only because it's never been trained, or, or what it's been trained on of resembles what it's seeing in the image, but it really doesn't. It's really not anything to do with the image.
Yeah, I guess it's like real world versus digital world.
Yeah.
And one—
well, one of the biggest—
custom built for digital and one's custom made for the real world. And it's just 2 different use cases.
Yeah. And it's interesting because I've been approached a number of times about people talking about virtual worlds and training in virtual worlds. And Uh, you know, I, I've given some deep thought and found, you know, basically I go back and tell them, well, the issue you have is that your virtual human or whatever your, your, your object is that you're moving through the virtual world has very limited senses. 99.9% of it is visual.
Mm-hmm.
And even if you apply a lot of physics engines to it, it still will never experience that virtual world like we experience the real world.
Mm-hmm.
So it'll never understand contextually what, what's in the real world, because all it's got is some vision of being applied to some virtual reality.
Yeah, math. I mean, in essence, it's— you would have to accurately create all of physics in the universe in mathematical models.
Well, also the millions of sensors we have.
Yeah, right, for sure. So it seems to me like that's the essence of connectome, right? I mean, it is, right? nervous system. It's the branching throughout an entire brain and body and all the quadrillions of sensory inputs, all the cortical processing and the muscle output that is feeding back on itself in real time to create learning, manipulation, automation, right? Autonomy.
Um, yeah.
Manipulation.
I mean, we know, we know not to run into an object because we know it hurts.
Yeah, and, and how different objects hurt differently in the physics. We have a feel of them. And then I feel also like some of the other thing is, I mean, I just read an article in Quanta yesterday where it's like, oh, dark energy is decreasing, or dark matter is decreasing. And they thought it was a constant in the physics equation, which accounts for something like 95% of our explainable universe. We just say, oh, well, here's what's left over to make the equation balance. We're going to call it dark matter and dark energy. And now they're finding that it's not a constant. And so it's like, even if we could, even if we could accurately capture everything in the universe in an equation, program that and put it into a computer, even with the help of advanced AI, it's, we'd still be missing something that hundreds of millions or even billions of years of evolution is missing. So the closest we can get is actually take the thing that's already there from nature and then put that into robotics. It's as close as we're gonna get. But then here's my thing, right? Like, say we're able to do a full human connectome. What is it like 100 trillion neural connections?
Yeah.
Yeah. How, how would we— the technology may not exist— how would we recreate like a humanoid, like in terms of us? I mean, would you physically have to like put in the full nervous system and connection and like actually tie up every single sensory neuron to that piece in the connectome inside a robotic system?
What were our hardware Well, you got to understand, your peripheral nervous system is a connectome in and of itself. So—
What do you mean peripheral nervous system?
Well, like, you know, your hands and your skin and, and your muscles. And, um, because that's feedback. Your muscles are feedback into your brain as far as, uh, you know, it's a sensory input. In fact, it's interesting because, um, One of the updates I wanted to talk about today is my exploration of Drosophila. And one of the things I recently, you know, discovered, or I mean, it's been out there forever, but I just discovered it, is that it has like a spinal cord. It has what's called the ventral nerve cord. And in that—
And this is a, what animal is this?
The colopus? Drosophila. The, the little, um, no, the little, uh, fruit fly.
Okay.
So those, those neurons within the ventral nerve cord talk or connect to like leg muscles, for instance. It's not a direct connection from the— not necessarily a direct connection from the central nervous system, but central nervous system to the nerve core and nerve cord to, to the muscles in the leg. And the leg also has sensory input. So it feeds that information back through the cord to the central nervous system. So just like us, I mean, we, we have, uh, we have neurons that come and allow us to move and, you know, do things and whatever in our peripheral, but we also had feedback, whether it's an object cold, hot, the pressure that we're, what we're pushing or whatever, back into our nervous system to give us that feedback. So even though the little Drosophila has that to a very primitive degree. Yeah.
Interesting. So fruit fly would map to drones.
Yeah. And that's, and that's, I've just purchased a drone. I'm waiting for it to come. Should be in a couple weeks. And, uh, what I'm going to be doing is applying the Drosophila nervous system to the drone. See what happens.
Yeah, fly away with your lunch. Yeah, come back here, get back here. Uh, yeah, take over Amazon's delivery for it. Um, yeah, it's actually not a bad idea. But the question— so I saw, obviously, like we've been talking in one of your videos, is like the treating forest fires. Yeah, which Is interesting from an ESG perspective and carbon capture perspective, but also just like oxygen and land and the whole ecosystem. But how would that work? Like, how would that be better than, I don't know, the existing technology of drone?
Well, one of the things we, we have right now, and like forest fires or large areas that we can survey in the mountains, even locally here, Uh, in the Greater Los Angeles area, it's— they have towers that people man during the fire season, and what they do every day is just sit out there and look for smoke.
It's like real. You laugh, but I laugh, but it's real.
Yeah, it's real. Yeah, yeah. And, and I told— I think I told you I'm working with a gentleman from, uh, Portugal, uh, a scientist, and, uh, it just happened, uh, that his father is like Part of the— there, there will be like an equivalent to the United States Forestry Department, and they have the same thing. They have— and Portugal and Spain suffer greatly from wildfires, just like we hear in California. So they also have towers, and they just sit out there and look for smoke. So even— Uh, for one thing, you could obviously replace that person with a smoke detector of some sort. You know, constantly span, uh, scan the horizon and see if anything pops up. Then you could send, you know, some fleet of drones to that area to say, okay, if there's something there, if there's a fire going on, put it out.
Mm-hmm.
So that it'd be— so what the Drosophila gives us is gives us fully autonomous navigation and gives us the ability to use sensors that could give us the ability to target, say, a flame or smoke and just spray it with water or whatever chemicals we want to—
Phos-Tek.
Or whatever we want to use. So if you had, say, like a fleet of drones, You could immediately put out a fire waiting for, instead of waiting for the fire to get it to, you know, 500 acres of burn and then have to bring in the heavy equipment like planes and helicopters and everything else.
Yeah, it's interesting. It's sort of like drones and waiting and then there's a couple different sensory neurons and then they just go tackle the issue immediately. And then there's some control software around the outside about like, return and some safety mechanisms. And then you just need to fill up that tank of water unless there's something nearby where they can get. Yeah, I guess that's the other thing is you could actually detect it to use natural water sources that are as close as possible to the spot to do like optimization challenges. But essentially, I mean, our hypothesis is that some of that is inherently programmed into the fruit fly. Which is part of the reason is because that's a survival mechanism. So they'd be inherently attracted without training data, without tons of programming. So ultimately the value of this whole thing is just speed of development and yeah, it's like speed of development because you don't have to do all the edge cases. You don't have to do all the training. You just have to sort of Tell them where to go. And, and not even really, right? Because I mean, if you had a, a smoke sensor, yeah, or monitors set up, right? And you could see infrared. I mean, think about, yeah, Arnold Schwarzenegger and Predator. Yeah, Predators. I mean, even Predators have infrared. Yeah, just give the So that's interesting.
They could, they could make a Terminator to terminate fire. Yeah.
Yeah. And just have lots of safety protocols. Yeah. We don't want to go to—
Well, that's, that's the thing too, using Drosophila as an example, whether that fire is in a forest, in the desert, in some grassland, it doesn't matter. The navigation's autonomous. Yeah.
It's interesting in neighborhoods, you don't think about it, but There, I mean, here's a perfect example. You know Jeff, right? My buddy I grew up with.
Yeah.
So I was up at his place the last couple Thanksgivings and in his development, there's one house that burned down and it's been a big HOA nightmare for a long time. We won't get into about like trying to rebuild it, but I mean, it happens.
It happens.
I mean, you're right there in suburban America and poof. And those fires can, I mean, our houses in suburban America are pretty close together.
Close together.
So a little bit of wind, I mean, as a neighborhood watch goes, I mean, I'll tell you an HOA I lived in for a couple years and they weren't even expensive HOA fees, but I think they spent like $5 million a year out of a $6 million budget. You know what they spent the $5 million on? Landscaping, flowers. And it's like, okay, that's great. But given a fire, I mean, there was a fire station in the community, so that's sort of helpful. But yeah, that's interesting. I mean, you could see the use case there on almost insurance. It's like just an insurance policy. It costs you, I don't know what times the number of people in the community, $5 a person. It's like, look, it could save your family in the middle of the night. When you're sleeping.
Yeah.
Like, what's $5 a month, you know?
Yeah. Well, and interesting, I don't know if you, you're aware, in California it's almost— it's getting fire insurance is almost impossible now. I live in Arjola, and Arjola has just been— we've had to have special assessments because they, they just can't get— the fire insurance is so outrageous. Mine almost doubled this year.
What, like, what's the level, like dollars per month?
Well, I paid like $900. I paid $600, I think, the year before. I paid $900 last year. And this year I just got the bills. Like, it's like $1,500.
For the whole year?
Yeah. Geez.
So it doubled in 2 years.
Yeah. Yeah.
So basically, if we could come in and say, hey, that fire insurance thing you have, which doesn't stop the fire from destroying your life. No, like, you could pay us $400 instead of $1,600 and we'll make sure your stuff's safe. There's a lot of people in California.
Yeah, no, and, and yeah, I hear the night—
it's a lot of stuff.
It's getting worse and worse, you know. Uh, there's a lot of companies won't renew, like, uh, fire insurance now. State Farm is probably the worst. They're not renewing anybody's. There's— I think they said they're gonna not renew like 700,000 Or $70,000, $72,000 homes, they're not going to renew this year.
Come on, get out of here. This is, this is interesting. I think that's State Farm calling about their innovation department saying you're about to disrupt their business model and they need to talk to you. Because I mean, think about that, right? I mean, flood insurance is the other big one that nobody's giving out anymore. And fire insurance— flood, I don't— we'd have to talk about that. But the fire insurance route is interesting. It's like You're going to pay all this money times how many million? There's probably, what, 30 million people in California, maybe more. I don't even know.
It was like almost 40, almost like 38 million, I think, now.
And how many homes? And it's like insurance comes later and then they won't always pay your claim. It's like, route— like cancel your insurance, all of California for fire, send your money to Prohm.
Yeah.
And we will, we will protect, we will protect your home, your family, your belongings. And, you know, maybe there's some insurance involved as well because it's— that's a better underwriting scenario. It's not just I'm waiting for something to happen and then I may or may not pay out a claim. It's like, no, I actually have a mitigation. Oh, this is interesting. Like, the insurance companies would not like us, but—
Well, power companies would though, because the power companies are the ones causing the wildfires by their lines getting snapped. And wait, they say more. They're putting out— they're putting out millions of dollars in claims right now because, because their power lines, um, the power lines, the trees fall on the power lines during heavy winds and they spark fires. Paradise Fire was caused by, uh, a line going down. And it, you know, pretty much wiped out that whole community.
Yeah, above-ground power lines rather than underground.
Right, right. And there's still plenty of them around our city where I live. They, they've got some programs putting them all underground. They did a major one a few years ago, uh, that ends right down the, down the bottom of the hill from where I live. Uh, but they, you know, and then it goes above ground to the, to, to, because I'm right on the Ventura-Los Angeles County line, so it goes, so it goes over to Los Angeles County above ground.
Jeez.
But those guys are going, I guess.
Yeah, that's— oh man, I used to live in LA. I don't, I don't want to see that happen. But yeah, I mean, talk to Elon. He's got some tunnels going in. Maybe he can drill the city solid, you know, like—
But they're even— I remember we were walking because we go out and exercise all the time, walk and Um, somebody was marking the trees and we stopped and asked them. They said they were marking some trees because they were near power lines to be taken out because they were concerned that these particular trees might snap and cause some, some major fire.
Yeah. Hmm. I mean, this might be interesting. Keep me updated on how this thing goes. Um, yeah, obviously you've got a ton of connections and insurance and they've, they've got buckets of cash. And they're more worried about the payout risk profile and the big one. So if we can not be a threat but a solution to them as well and also help people along the way, you know, I think this is actually a really good use case that can help people today, you know, help businesses today.
Yeah. Well, my ultimate plan with with the quadcopter and the, um, Drosophila nervous system is to actually, uh, connect a little water tank to the bottom of the drone with a little pump and have it go out and actually extinguish a fire.
Yeah.
You know, start a little— I'll go up to my property up north and my 40 acres, create a little fire, have it go out there and see if it'll put it out.
Yeah, Tim, make, make sure it's surrounded by rocks and open clearing.
Well, it's all green, you know, it's all green right now. It's beautiful right now. It's all beautiful green. But yeah, in about 2 more months from now, it'll all be brown.
Well, only you can prevent forest fires, you know.
Yeah, exactly. Yeah.
Um, hmm, interesting. I mean, that's the big problem is the weight of the water.
Yeah.
And available source is getting— but there's other fire mitigation. I think there's like, I am not an expert, but there's like dust particles or something that can help snuff out fire that isn't as heavy as water and has a bigger spread and more of a dampening effect. I'd have to do some research on it, but—
Yeah, I was wondering. Okay. So, The smallest helicopter that carries water for fire dousing is— holds about 500 gallons. I think— I, I, I, I know we could easily get up to 50 gallons with current tech, with current human drivable quadcopters.
Uh-huh.
Um, but I think if you were to take the same kind of design, I think you could probably get 100 gallons. per drone. So, so I always envision this kind of fleet of drones, right? Not just a drone, but a—
Right.
You know, and maybe, maybe in a shift kind of situation where you have, say, 5 drones attacking the fire, 5 drones getting water, and 5 drones coming back. So you kind of rotate, uh, the firefighting. But yeah, the ultimate goal would This concept to me is to put the fire out before it spreads. I think once it spreads to a major fire, I mean, then you gotta have, you know, more conventional firefighting.
I'm, I'm Googling fire dampening chemicals right now. I have a—
Well, there's, there's Phos-Tek. Yeah, there's Phos-Tek. And I, I think if I remember right, Phos-Tek may be even heavier than water. It may not be, I'm not quite sure, but it's not, it's still heavy. It's not, well, I wonder because it doesn't really, it's more of a, prevents the fire from spreading more than putting out the fire from what I understand. So what they do is they spread it around the periphery. You never see them dump Phos-Tek directly on a fire. It's usually around the peripheral of the fire.
Yeah.
You know that red, you're seeing 'em on the mountains here in California, the red The red, you know, line that's going around where a fire was, that's where they dumped a whole bunch of Boztek. And they, they, uh, it's supposed to prevent the fire from, you know, moving any further. Sort of like, you know, putting a line in the sand to say, okay, you can't, you can't fire, you can't go past here. And it works. I mean, that's why they use it. Right. But they, but they still use water to dump directly on the fire.
I wonder, I have some friends at BASF and some chemical manufacturers. It, you know, if this thing goes down the path, I mean, they're geniuses, chemical engineers. I feel like—
Sure. Yeah.
A requirement of needs to be lighter than water and can spread further and dampen fire faster. Like, I'm sure they can come up with—
It's another— Yeah, smother the fire, right? Yeah. Well, yeah, the other thing is you want to smother the fire without damaging the environment, right?
The environment, exactly.
Yeah, that's why they use Phostek, because it has like a fertilizer or something built into it so that it, you know, once the fire is gone and the next season comes, it'll continue to grow vegetation, which I don't know is a good thing or a bad thing. Because, because, you know, it's like we had, we had a record number or amount of rain this year in California, which is wonderful. Everything's green right now. But, you know, the more water we get, of course, we're not in a drought, which is wonderful. We're not being rationed water right now. But, um, at the same time, the more water, the more vegetation, and the more vegetation, the more fire fuel. So it's kind of a vicious circle.
Yeah.
You know, when it comes to all this stuff. But as far as like water sources, I, uh, we had a fire here a couple years ago. I mean, probably, I don't know, 50 yards from my— 100 yards from my house.
Mm-hmm.
And we were out there and we're watching the helicopters come, and I understand they went to the local golf course and sucked up the water at the lake there. But, but even around here, what they did is, you know, They, what they did is they figured out there's certain areas where, like after the Woolsey Fire, which kind of wiped out all of Malibu and, and that came right down to where I live as well. They, they, um, created these tanks, these kind of portable tanks that they can put out there in the wilderness and fill with water so the helicopters can come pull the water and then go back to the fire. So there's all kinds of things you can do with that. Um, the other side of all this is what kind of got me started on this was when we had the Woosley Fire. Uh, it came down a mountainside right outside of, uh, our community here, which it's only maybe probably 2/10 of a mile from my house, or 1/10 of a mile from my house, and it burnt down the side of a mountain. And there's another side Went up because, you know, kind of in a valley. So it burned down to half more of the valley, which there was a fire road there. So it stopped there and the fire— the firefighters were able to contain the fire at that point. And of course they have to move on because they can't stick around. They got, you know, this fire was huge and they had to go fight it wherever they had to fight it. So about, this was like on a Friday morning. The fire department stopped the fire about 11 o'clock in the morning somewhere in that neighborhood, and they took off. I think they left one truck here for a little while, but by mid-afternoon they're all gone. A message went out on next-door neighbor telling us that there's a guy who was trying to put a fire out in a tree, and the wife was asking the people. You know, anybody around, can you come help? I went to grab my shovel and go help, and I saw another neighbor coming back and he says, oh, it's already taken care of, we, we took care of it. Uh, but then we realized the whole side of the mountain was all still burning embers.
Hmm.
And it was windy. That's what causes our fires here, the winds. And, and it hadn't lit up the other side of the canyon yet. But we knew unattended it probably would. So we— I spent Friday night running up and down this mountain putting out embers.
Any glow?
It was great when it got dark because you could see it pretty easily. You could see the embers glowing, so you knew where to run to, and you throw water or dirt on it. But even then, on Saturday, uh, we had a couple little hot spots flare up. that we put out right away. Even on Sunday, there was a tree, the whole inside of it lit up.
What?
And it was all around the brush. Yeah, all the brush was around it ready to be ignited, but just the center of the tree. And we're lucky enough, we— one of the neighbors was out walking their dog or something and saw it and, you know, told everybody, hey, we got to put this fire out. So we all ran over there and put that out. But it was so bizarre that just the middle of the tree was on fire. And all the brush around it was, was fine. So those little embers, they flare, they take off in the wind. Wherever they land, they can cause a lot of damage and a lot of harm. So even— There's, there's a whole, I think, a whole drone or robotic, uh, you know, use case to do what we call mop-up. Fire mop-ups because after the crew professionals leave, there's still all these embers around and all this other stuff that somebody doesn't take care of, it's gonna just burn something more. They'll come back but after everything else is burnt down.
Well, I think it's time for a poll. Check this out. Can you see on your screen?
Sure can. Yeah, all those things. Oh, you can say like mine exploration, like deep sea mine exploration or deep sea mining is another good one.
So I don't know if this picks up because I think only you can vote. So it's a sample size of 1. But if the recording doesn't pick this up, I said best use case of biologic intelligence in parentheses, Brome. And the options are firefighting and fire mop-ups. Option 2, self-driving. Option 3, mine exploration. Option 4, other dangerous human need. And we have 1 vote that's come in so far.
That was me.
Tim, our chief scientist here. Technology officer has chosen firefighting and fire mop-ups because it's a clear and present danger. And this is okay. I'm gonna end the poll because I can't vote, I guess. So I guess I don't count. Need to register next time. So let's share results. Okay, there you go. You can see your own results. Yeah, there you go. 100%. You know, you're you're predicting the future one day at a time. No, but that's super interesting because the whole, like, I mean, you can't stand out there 24/7, 365. And now it makes sense why you've got—
I spend hours. Yeah.
Yeah. And I mean, talk about a waste of time and money and the, but the risk is so high if you don't that it's like you have to. So your life, what, like your life stops? And what if it's during a workday, during the week, in a neighborhood?
And I'm saying this, like, this happened Friday morning, came down to our neighborhood, right? It lasted through Sunday. We were putting out fires on Sunday.
Yeah, there goes your weekend.
Yeah, one way or the other. Yeah, everything was alert. You know, everybody was watching and alert. And there was probably about 3 or 4 of us neighbors that were really tentative to it all, uh, but there was other neighbors that would, you know, come in and just, you know, walk around the mountain or whatever to try to— I mean, I'm talking about steep mountain too. You carry 5 gallons of water up and down these mountains. It was a workout.
Yeah, the 36th chamber is not too far ahead of you.
So we had shovels too, but I mean, shovels, I don't think do as good a job as, uh, as just dousing it with water. Water would get rid of the problem real fast.
It's slower too. Interesting. Okay. All right.
So I mean, that's what we're doing right now with Drosophila. Um, I've got a, I've got a little project plan figured out. You know, I gotta build it. I got a lot of different testing to do, a lot of different Uh, things that, uh, I need to apply, you know. And like any, any project, you want to test one thing at a time. You know, I'm not going to build the whole thing and say, okay, does it work? You know, you got to test the Drosophila nervous system before I start applying things like, uh, water tanks and pumps and things like that to it. But, um, interesting. Yes, but that's, that's, um, That's my goal probably over the next month or two, to get our, uh, proof of concept for the Drosophila nervous system.
Yeah, interesting. I like it. I think it's, uh, I think it's smart. It's, it's a clear use case. Um, yeah, I wonder who else is out there doing the same thing. Seems crazy that there aren't like 50 companies Trying to tackle that and save people's homes and big market.
Yeah, I haven't seen, I haven't seen anything out there. Um, I know in that, in that Twitter, um, exchange that we kind of had, there's several people kind of chimed in, uh, thought it was a great idea. Obviously, you know, like I said, even Gary, Gary Marcus said there were some guy he knew at some university or something that was thinking along the same lines, but I've never seen anything. I mean, I've seen firefighting drones. There's some cool things out there. There's, uh, one set of drones, but there's a difference between wildfires and, uh, structural fires, right? So the ones I've seen so far were drones, which are pretty awesome. The drones could pick up the hose and fly up to a building, you know. So it was a multi-story building and the fire was, you know, 5 stories up. Uh, the drone basically would pick up the hose and pull the hose up there and direct the water onto the fire. But I think it was— I think it was all, you know, manually done.
Sure.
But it was cool that— the cool concept, in my opinion, because You know, for them to get the ladder extension and, uh, try to do it themselves. And of course the ladder can only go so far where the drone can go much further. So they could, you know, tackle a 10-story building or whatever to pull the hose up and, and, um, you know, get, get to the place where the fire is happening.
Yeah.
Uh, it, it, you know, of course most cities and stuff now require sprinklers, but Inside that, inside the, the buildings. Yeah, there's still a lot of buildings out there that have not been fit up with sprinklers, especially a lot of older ones. Yeah, so I think the home—
I mean, the consumer market in the home is way bigger. And you're not— I mean, most homes don't have like fire retardant on the outside and like fields around homes. Like, it just doesn't exist. Yeah, yeah.
What, what one of the big initiatives here to, quote, save on insurance, which is really a joke at this point, is called fire hardening. They, you know, if you can harden your house against fires, uh, the insurance company is supposed to give you a break. But when they're canceling your insurance altogether, who gives a shit?
I mean, what do you, what do you do, wrap it in like a cocoon? I don't know how you harden your house against fire, keep it constantly wet? Like—
Well, actually, I ran into a company here local, uh, then I thought about this concept a long time ago, is, uh, put these like heavy-duty sprinklers around the outside of your house, like, like in your yard kind of thing, that would pop up and just spray water everywhere around your house. Uh, there's actually a company here locally that does it, that, that does that. There's some They had some video on there where it shows like these powerful sprinkler heads popping up and just like sprinkling the house and yard and everything around it, uh, here in the mountains of— well, the Malibu Mountains here where, you know, that's where the Woolsey Fire was. And I bet— I'm sure those folk can't even think about insurance. They probably have to use the state program. Yeah, but makes sense. But Yeah. Yeah. But, uh, that, I mean, so there are, there are people thinking along the terms of how to keep your house safe from a fire, but it's not, it's not inexpensive.
Mm-hmm.
Helps if you have a pool, cuz you can use that as a water source. And of course you have to have the pumps and everything in place and around your yard or wherever you're gonna, your peripheral of your house. to make it safe. But with the fire we had, it was only like 50 or 100 yards away. I mean, um, I have neighbors on the end that, um, they were actually out of town when it broke out. And, uh, myself and my niece's boyfriend, who happened to be here at the time, ran over there and started dousing their house down with water because, you know, you could see embers coming.
Jesus.
You know, And, and actually, we were just, we just kind of got started and the fire department came along and go, don't worry about it, guys. We, we will, we're gonna do this. We got it covered.
So, wow.
We let, we let the fire department do their job. Yeah. Yeah. But we're gonna let our neighbor's house burn down, you know? Uh, yeah.
And they would get overwhelmed quickly. I mean, there's not, yeah, an infinite number of firefighters, you know?
Interesting. Yeah.
Well, we, we got about 10 minutes. Um, what else should we cover? What else is new and cooking in your world?
Well, the other side, the other, uh, thing I wanted to talk about, and I, I think it's becoming more of a subject matter these days, um, is SWAP, size, weight, and power. The one thing that, you know, I— so the central complex of the Drosophila is equivalent to our hippocampus. It's where we have— so when we navigate, we navigate with what's called place cells, grid cells, border cells, and head direction cells. And it's all part of this hippocampal-thalamus, uh, part of our brain that navigates. All these pieces of neurons fire off and allow us to navigate freely through our environment. Drosophila has a very similar— they call it the central complex. more advanced insects, uh, have the same kind of, uh, pre-hippocampal body in their, in their nervous system, uh, and it for navigation purposes, of course. So I, I was able to take the central complex where I think it was like 60-some— I forgot how many neurons it was, a lot of neurons. One of the largest, one of the largest complex parts of the Drosophila brain. And actually, I got a brand new Raspberry Pi 5, and I was able to run the central complex on it.
So that's ridiculous compared to LLMs. Like, yeah, like we have to build massive supercomputers using the most advanced cutting-edge NVIDIA chips in a data center. And you're like, nah, I just got an off-the-shelf Raspberry Pi. It's fine.
Yeah. And obviously to explore, I mean, to have an autonomous robot go out and explore the world, you can't, you can't, like, I think I might have shown you or sent you the picture I developed from DALL-E, you know, ChatGPT, you know, showing Mars. a Mars rover pulling, you know, all these semi-trailers behind it. Because I mean, that's what it would take. Yeah, yeah, that's what— and, you know, recently OpenAI and IBM— or not IBM, Microsoft, I think it was— you know, announced they're building this $100 billion supercomputer for AI. So here they're developing this AI that can't even tell an office chair from From a dining room chair, that it's going to be, you know, this mega data center. And even from an environmental point of view, a lot of people have noted that AI is just sucking up energy and causing tremendous amount of pollution and issues. Yeah, it's killing the environment.
Take that $100 billion and Just take $1 billion. Yeah, take $100 million. You know what, take a million. Just take a million, which is what, 1/10,000th or whatever, and allow us to prove you— well, we kind of already have, whatever.
Yeah.
You can't change the, you know, you can't change the world.
Everyone's really hype on LLMs and, you know, There's, I know they're dumping all the money they can into it. Yeah. And it really surprised me a couple days ago, Numena sent out, uh, uh, uh, I don't know, press release or something. I don't know what you would call it. They, they sent out a thing saying, if your AI can't run on a CPU, then give it up.
Yeah.
And that really surprised me and I was so happy to see it. Because it's like, you're absolutely right, man. If you can't run on low-power machines, if you need, you know, 10 zillion GPUs for your AI to tell me how to write a program that doesn't work.
Yeah, I think really, I mean, look, it's such an amazing breakthrough. Attention is all you need, right? The 2017 paper, it's such an amazing breakthrough. And the experience of changing one kind of pixel into a different kind of pixel and having it like actually make sense in words and visuals is really compelling. Yeah. No, it's great. But I feel like what actually happened is you just took every if-then-do-while loop statement possible across the entire internet that's ever existed. And then you put it behind a curtain in the Land of Oz and said, don't look behind the curtain.
Right.
And, you know, you have like, and that's really the issue is you just have this giant piece of mega machinery behind you in order to You know, change a sentence into a picture. And it's like, is that really helping people or is like fighting fires helping people? You know what I mean?
Yeah, that picture I created with like the rover with a bunch of semi-trailers behind it, that was, that was one out of maybe 6 or 7 pictures because, you know, and now the big thing is prompt engineering. Because if you don't tell the LLM exactly so it understands what you're asking for, not that you don't understand that it does, it understands, it gives you results that are not desirable. Until I can tell you, just trying to create a simple picture, like, you know, do a picture of a Mars rover with semi-trailers behind it.
Mm-hmm.
You wouldn't believe the crap that it generated.
Well, I did it. So I, in Q4 2023, I spent 3 months designing a men's fashion line using ChatGPT and I probably entered 10,000 prompts during that time to try to To try to fine-tune the variables. And what existed was it was an iteration friend, but I could never get, you can never get past a certain threshold, which is—
Right.
Hold this variable constant, change this variable.
Right.
Like it wouldn't get the spelling right when you wanted to add a word to the text. And so what it really allowed you to do was just quickly visualize a concept and get you from 0 to 1. It gets you from 0 to 1 much faster from the blank piece of paper problem. But then from 1 to 100 or 1 to 10, the human has to take over, the biologic intelligence has to take over.
Exactly. And again, that's the same with programming, you know, ask it to write you a program, actually ask it to give you a simple I mean, if you ask for very simple scripts like, you know, Python programming or C# or whatever programming that you want, it can usually generate real simple things.
Yeah.
You know, like if you say, you know, give me a Python program to add 2 numbers together, you know, it can do that for you without issue. But, you know, I mean, when you're really down deep in the weeds as far as like business programming and robotics programming and things like that. I mean, it just, it just doesn't have that. That's raw.
You know what I think this whole thing is and how it ends up? I think what we're talking about is a hammer, a screwdriver, and a wrench, more or less. It's, do they compete with each other? I mean, I guess in some ways, yes. They're, they're all different tools for different specific purposes in the context of building things in the real world, but each thing is good for something unique.
Right.
And I think deep learning is good for something unique. Large language models are good for something unique. Biologic intelligence is good for something unique as well. I don't know. I don't know where this is all going to end up, like, if they end up converging. I mean, because humans are flawed too, right? Biologic intelligence. In some ways, we aren't perfect, but there's perfection in the imperfection. I don't know. Who knows where it all ends up? Yeah.
I mean, there are—
I think real-world physical stuff, I think, is biologic. That's, yeah, it's always been my hypothesis. Like you get animal connectomes, animal intelligence that have built in the physical universe, it will work for applications in the physical universe. If you've created large language models in a digital universe to transform text into images and videos from one pixel to a different pixel, then it'll work way better in digital space, which is one of the things you were saying. It's like, Yeah. people want me to put biologic intelligence into a digital space and it's like, well, where's all your different sensory inputs? What's your motor output? It's like you're trying to almost shoehorn.
Yeah.
And I feel like in the same way, maybe LLMs and deep learning are trying to shoehorn that into like humanoid robotics or robotics in the real world. And so maybe that is really where the paradigm and the disciplines are just different in different use cases. What do you think?
Yeah, I agree. Couldn't tell you more.
I agree.
The thing is, with the hype that's going around, I mean, AGI is the subject matter, right? Artificial general intelligence. And somehow they think these LLMs or OpenAI is going to get us to AGI. You know, all they have to do is apply more computing power and scale it. Yeah, yeah, the scalability Is not there, and it's, it's, it's really silly. And the fact that no one's looking at this, this concept of swap that we've known about for, well, how many years now? Um, yeah, you know, and NASA obviously is—
Yeah, yeah, yeah, for sure.
You can't— you're not gonna, you're not gonna send, you know, 500 semi-trailers of computing power to Mars for a little rover to go around. In fact, there was even a Uh, there's been serious thought as putting humans in orbit around Mars, like in a space station, so that they can run the rovers.
Yeah.
Because right now, as we know, you know, it takes 6 minutes to, you know, back and forth between Mars and Earth. And it, you know, it's like, take, take a picture, send me the picture, which a picture is going to take more than 6 minutes. It's going to take a long time. Okay, looks safe. Go 20 more centimeters and then take another picture and send it to me. You know, it's, it's, you know, the rovers up there have only gone a few kilometers at best because it's just excruciating, uh, not to, to keep them from getting in harm's way. And, but it's all human-driven. So they even talked about, you know, putting humans in a space station in orbit so they could have real-time interaction with the rovers. But yeah, I think with our technology, you could put a rover out there and it would not— it would be fine. It would not, um, it wouldn't kill itself. Yeah, it wouldn't kill itself. It wouldn't get hurt. Um, if it did get hurt, we'd be able to somehow recuperate from that hurt. Yeah. Right now, if a wheel or something goes bad on a rover, it's like, oh, well, guess what, man? We're screwed.
It can't adapt. Yeah, it can't adapt. I mean, that's, yeah, that's interesting. I mean, there is no training data available for Mars, right? Or Venus. I mean, it doesn't exist. We have like 3 pictures.
Like, yeah, what are you gonna train? Right.
There's nothing to train.
Yeah. Like we've always said, you can't take a self-driving car here in the, in, in, on the world, on the Earth and put it on Mars and say, okay, now start driving around Mars.
Yeah. It's looking for the highway lines.
Yeah.
It's like, where's the, where's the stop sign?
Yeah.
Where's my, where's my Tesla plug-in to recharge the rover? Yeah, well, size, weight, and power. I mean, that's why they said it starts with a toy. It's like actually not a toy, like an RC car is actually, that is the test, the true test. Can you get an off-the-shelf robot with the world's most advanced artificial general intelligence software and run it on off-the-shelf parts. It's like a data center, it's commodity parts. And that's what makes it happen. I mean, if every data center in the world and the internet was powered by custom, each one was custom, each chip was custom, we would never have the internet. It just like wouldn't work. It's like, what? It's fucking crazy.
So, oh, and anyway, in the '60s, '70s, and '80s, that's what we had. And that, that's where people started pulling away from it industry-wise, because every computer system was its own, had its own operating system. You know, everything from an IBM 370, 360, all the way down to a DEC PDP-11 or whatever. I mean, everything had its own operating system, its own proprietary software. where you could not take any of the software and put it onto another system. It was not ubiquitous in any stretch of the imagination. And companies started getting fed up with that because they get locked into these really expensive contracts and they couldn't do anything with other than what they had. So that's where the internet allowed us the freedom that we have servers in the cloud. And those servers can talk to each other in a lot of, a lot of, a lot of ways. I mean, the system— I just got finished at my day job, you know, migrating like 13 systems to one. And, you know, I could do it because it was all cloud-based software.
Hmm.
I can get to it and pull the data and do what I had to do to massage the data to move it into the new system.
Yeah, universal standards.
Yeah, if it was in the '80s, man, that would have been— I don't know where you would start. Spreadsheets, maybe? You start as 1, 2, 3.
Yeah, you'd have to start over. Pen and paper.
Yeah, exactly. That's why you'd have to start the new system fresh. You would, you know, the likelihood of you moving data from one system to another was pretty slim.
You have a team of analysts or interns, like, you know, or maybe—
So with biologic intelligence, I mean, you know, the sensors, motor output, as long as you know what you need to feed it and what you need to do with the output, you can apply it to a lot of different Modalities. You don't need, you know, you don't need, um, like, like for instance with the Drosophila test proof of concept I'm working on, instead of olfactory is a big part of olfactory sensing is a big part of the fly, as you can imagine. So change that olfactory to smoke, smoked it, so it thinks the food is smoke. Or smoke is the food, right?
Go towards, go towards food, i.e., smoke or fire.
Yeah, yeah, yeah, yeah. So, so you can, you can manipulate the nervous system by adding it, but you could create any kind of sensor you wanted to, um, that you wanted to tell the drone to, to move forward to.
Yeah, pressure, density. Humidity.
Yeah.
Elevation, altitude, orientation. Yeah.
I mean, it's, it's unlimited as far as, you know, if you have a sensor, we could probably connect it to a biological nervous system, artificial nervous system, and, and use it in some way, fashion, or purpose.
And there's a ton of sensors that are off the shelf and cheap and good.
Big time. Yeah.
And Yeah, super high-powered magnets too. Like all this stuff has become commoditized at scale. So it's like just use the stuff that's already there instead of having to reinvent everything.
Exactly.
The thing you should reinvent is the software, which we've done. And then you can use all the other stuff that already exists. That's the whole point. Like stand on the shoulders of giants. We shouldn't be making new chips for new Like make new software that fits in with the old way.
Yeah.
Yeah.
Yeah.
Anyways, preaching to the converted. All right. Well, this has been rad, Tim. I gotta run to my next one. Let me, yeah, let me know what's going on with this use case.
Yeah. Oh yeah. I'll keep you informed. I'll direct— I mean, you'd be the first person to know.
Yeah. Maybe I start reaching out to some insurance folks and I could actually reach out to one of the chemical folks that I know and just see like thoughts, just get their perspective on this, see what they think.
Yeah, no, I would love it. Yeah, please do. Okay. If you have any connections with the power companies, yeah. I mean, if this works, which I don't see why it won't. Um, I mean, it, it could really, uh, improve our lives tremendously, let's put that way, and improve industry lives tremendously.
Yeah, yeah, for sure. That's why I do what I do. So hell yeah. Okay, cool, brother. Well, good seeing your face again.
You too. Yeah.
Love seeing all your little Your little furry friends behind you. Not so furry yet, but maybe one day. And, uh, yeah, just give my love to the family.
I don't know if you can see the workbench back there or not.
Oh yeah, that's the, uh, yeah, that's the mad scientist laboratory right there. I love it. Nobody has to know. Yeah, got the compound out in the, out in the woods with Solar, like satellite internet, autonomous robotics.
That's, that's our robotic proving grounds, the 40 acres up in southern Monterey County. Yeah, Starlink's up there and I got it all wired and I can look at, I just, just looking at the live feed from there about an hour ago or so.
I mean, if we figure out autonomous food creation, like call it a day. I just talked to a guy last week who was pretty rad, but he created a solar panel manufacturing business that grew in 15 years to something pretty insane in terms of amount of dollars per year. And like super duper cool. So, and then he automated the entire manufacturing line. So it's like—
That's fantastic.
You know, you can get like, there's the solar farm right there. You put some compute next to it, you know, you get and the robots can come recharge, get the water regeneration going and then figure out a way to like plant and till and mend and rewater. And then that gets figured out and that's a whole ecosystem right there.
And that's what you think about.
Yeah.
Agriculture is, I still, I mean, even though there's a lot of robotics being applied to it, there's still nothing that's general.
Mm-hmm. Yeah.
You know, I could just manage a field.
Yeah.
Yeah. I've made a good friend with a guy who owns a winery and, and in, you know, has, I don't know how many acres of grapes and, you know, labor shortages are always a huge problem. And they've got a lot of different things they have to do throughout the year. And, you know, everything from harvesting to pruning to getting the weed— getting rid of the weeds and all that kind of stuff. I mean, it's, it's so much ripeness there for, for robotics. But again, without that general navigation capability, it's— I mean, it's almost— it's very difficult to, to create robots that really can go out into field and do the job like a human can.
You know what I just realized? Life is, life is just a funny thing. So in this episode, this call, we started with fire and moved on to farming. And wouldn't it be funny if we're essentially rebuilding the thing that our own biologic intelligence, i.e., humans, uh, started with first fire, then tool making, then farming and agriculture.
Agrarian. Yeah.
Like what? Like, and we got there independently. Uh, this is, yeah, this is some wild stuff. All right. Well, fire it is.
Yeah. I know you gotta get going. Yeah.
Okay, cool, man.
All right, man.
All right. Well, enjoy the rest of your week and, you know, stay safe and yeah. We'll talk to you on, on the internet.
Sounds great. Thanks. All right, talk to you later.
Yeah, bye.
