This Week in Startups - Figure’s AI Robot, AI News, Cognition’s Devin, and the biggest AI bet yet! | E1915

Episode Date: March 19, 2024

This Week in Startups is brought to you by…LinkedIn Ads. To redeem a $100 LinkedIn ad credit and launch your first campaign, go to http://www.linkedin.com/thisweekinstartupsGusto is easy online payroll, benefits, and HR built for modern small businesses. Get three months free when you run your first payroll at Gusto.com/twist.*Gusto pricing shown in ad is based on pricing prior to March 2025Attio - A radically new CRM for the next era of companies. Head to attio.com/twist to get 15% off for your first year.*Todays show:Sunny joins Jason to dive into AI news, including the CTO from Open AI talking about their training data for Sora (5:11), the FIgure AI robot (12:26), Devin, the first AI software engineer (38:17), and more!*Timestamps:(0:00) Sunny joins Jason for a fun AI day!(2:08) Elon and open source Grok vs “Closed” AI.(5:11) The CTO from Open AI talking about their training data for Sora.(9:28) Exploring the fascinating AI robot from Figure.(10:57) LinkedIn Ads - Get a $100 LinkedIn ad credit at http://www.linkedin.com/thisweekinstartups(12:26) Comparing the FIgure AI robot and what we have seen from Boston Dynamics and why this is impressive.(19:47) Gusto - Get three months free when you run your first payroll at http://www.gusto.com/twist(21:10) Thinking about how humans learn and applying that to AI.(28:28) The idea that robots may one day be 1:1 with humans.(30:50) Attio - Head to http://www.attio.com/twist to get 15% off for your first year.(32:01) New bet alert! And it’s a major one. Sunny and JCal bet on how soon we will all have access to AI robots in our homes!(38:17) Introducing Devin, the first AI software engineer!(40:24) The “maestro” concept when working with AI.(47:21) Hot news: Apple in talks to let Google Gemini power iPhone AI Features.*Subscribe to This Week in Startups on Apple: https://rb.gy/v19fcp*LINKS:Cognition’s Devin: https://www.cognition-labs.com/introducing-devinFigure AI Robot: https://www.figure.ai/*Follow Sunny:X: https://twitter.com/sundeepLinkedIn: https://www.linkedin.com/in/sundeepm*Follow Jason:X: https://twitter.com/JasonLinkedIn: https://www.linkedin.com/in/jasoncalacanis*Thank you to our partners:(10:57) LinkedIn Ads - Get a $100 LinkedIn ad credit at http://www.linkedin.com/thisweekinstartups(19:47) Gusto - Get three months free when you run your first payroll at http://gusto.com/twist*Gusto pricing shown in ad is based on pricing prior to March 2025(30:50) Attio - Head to attio.com/twist to get 15% off for your first year.*Great 2023 interviews: Steve Huffman, Brian Chesky, Aaron Levie, Sophia Amoruso, Reid Hoffman, Frank Slootman, Billy McFarland*Check out Jason’s suite of newsletters: https://substack.com/@calacanis*Follow TWiST:Substack: https://twistartups.substack.comTwitter: https://twitter.com/TWiStartupsYouTube: https://www.youtube.com/thisweekinInstagram: https://www.instagram.com/thisweekinstartupsTikTok: https://www.tiktok.com/@thisweekinstartups*Subscribe to the Founder University Podcast: https://www.founder.university/podcast

Okay, January 1st, 27, a humanoid robot will be available for purchase. Yes. And delivery. Yes. You can have it in your home. You can buy it and have it in your home. Buy January 1st, 2020. 27.

Okay. And you're asking me. No, the bet is. I take the over or the under is my. Well, you take over under. You're setting the line. Okay. And whoever's right, the other person buys the robot for that person.

Wow. This is the largest bet in the history of This Weekend Startups. There's 20 grand on the line here. What are we doing here on this podcast? I don't know. I mean, this is getting out of control. This is a big major bet. It's a big bet.

This Week in Startups is brought to you by LinkedIn Ads. To redeem a $100 LinkedIn ad credit and launch your first campaign, go to LinkedIn.com slash this week in startups. Gusto is easy online payroll, benefits, and HR. built for modern small businesses. Get three months free when you run your first payroll at gusto.com slash twist. And Adio, a radically new CRM for the next era of companies. Head to adio.com slash twist to get 15% off for your first year. All right, everybody, welcome back to this week in startups.

It's Twist. It's Madra Mondays. But we're moving from AI Monday since we record on Monday. So we're going to just start publishing AI Tuesday. It's going to be AI Tuesdays going forward with me. I like Modra Mondays. I know. We record on Madra Mondays.

Okay. And that is my partner in crime, Mr. Sunny Sendip Madra. You can follow him on Twitter at Sundip. Definitive intelligence has been merged into Grok. And so I'm now going to refer to you as what? Co-founder of Grok or chief technology officer? What is your title going to be? I am the general manager leading the Grok Cloud.

and our commercial business. Got it. GM of GROC cloud, GROQ, as opposed to GROK, hey, which leads to, I think, maybe a good place to start. I saw that Elon, you know, has been in this lawsuit with closed AI, formerly known as Open AI. He said he dropped the lawsuit.

It's got a sense of humor, obviously, if they just changed their name to closed AI, but they open source GROC. How big of a deal is it that he open source GROC? And what has the reaction been to GROK? which is Elon's. Yeah. Open sourcing.

What I would say... It's a huge deal. It continues to show the constant battle between open source and closed source. And the biggest deal with their model is the size of it. So it's larger than any other model that's been open sourced. And so it really gives a glimpse to the community of what someone who has at scale resources is, doing and opening that up because what we see from a lot of companies is the open source models, but sometimes even their smaller ones. This was, you know, they're sort of the primary one that

they were creating. So I think from that perspective, it was really interesting. It also shows a different problem where because the size and scale of it, it's not as, I'd say, usable by the community. Because in order to use a model of that size, you definitely need like a really large cluster. So I think what it's going to do, it's going to show folks that it's going to give people the ideas of what's happening with the larger model. Now, they didn't open the data sets that came with it. So in terms of the spectrum of openness, they made the weights available, but they didn't make the data that was available. And that's, you know, for obvious reasons. Yeah, because it's their proprietary data and they can't do that. So other models, like Facebook's model, do they release the

data it's trained on or they just say, hey, we used open crawl or something. show you exactly what it was, they used to train it on. Yeah. Now, when an open source project does that, in today's climate, where we have a big debate, people like Freedberg are like, yeah, you can train on whatever data you want. It's on the open web.

You know, it's a ridiculous position he has that, you know, nobody has any rights to their content. When it comes to training data, I think it's absurd. But, you know, this is how a lot of Googlers think, you know, he's a former Googler and a lot of tech people think Because if it's on the open web, I can do it every one. Free use, right? Yeah.

If it's on the open web, it's fair game, which is ridiculous. Like, a lot of things are available in the open web that you just can't take and go leverage. But anyway, putting that aside is now the reason why people are not going to share the data is they don't want to open themselves up to, hey, somebody posted to Facebook, you know, somebody's copyrighted material. Then it got ingested into the model. and now you have, let's just take the biggest IP of all time, Disney, Marvel, Star Wars, somebody put up, you know, in their G drive or on YouTube and it hasn't been taken down yet, an Avengers film. And now you've trained on the Avengers data, which you don't have the right to train. Yeah, I think it's like a big smoking gun in some ways. You know, I didn't get a chance to listen to the All In podcast, but I saw it in the show notes. Did you guys talk about the little mishap? Yeah. What was your take when the CTO of Open AI talked about the training data for SORA? What data was used to train SORA? We used publicly available data and licensed data.

So videos on YouTube? I'm actually not sure about that. What did you think of her reply there? Do you think she's lying? She doesn't want to set up herself for a lawsuit because she's a CTO. Shouldn't she know the data he was trained on? Well, that's where I was going to go. If you're maybe talking to someone, even a low-level engineer that's not working on the project, they may or may not know they're working on the code for it.

But I think someone in the position that she's in will have a full understanding of all technical aspects of what's happening and then data aspects of what's happening. And that's why I think when the question came up, she had some very specific answers. and as the interviewer probed, it started to get much more uncomfortable. Yeah. Yeah. And I think it followed what you were just saying, which is, well, if I can go,

and I think, you know, I don't use it this way, but I think you can get to Instagram via the web now. Like someone can send you a link and you don't have to open it in Instagram. Yeah. Yeah. And same with YouTube, right?

And so I think what the argument that was being made the way I at least read it was, well, if you can just get to it on the internet without a login or anything else, then it may be inside. It's fair game. was what she was saying. It's on the open web. By the way, a lot of New York Times stories are on the open web.

They don't put everything behind a paywall. That is not the, just because something's on the open web, doesn't mean you have the right to exploit it for commercial benefit. Let's give two letter grades here. What do you grade the CTO's response at Open AI, since we like to give letter grades? Well, look, a response in that interview with the Wall Street Journal gets. The response was not great.

It would probably be, you know, I'd say like, a C. You know, I think in that home. I give it an F. Oh, my God. It's a big fail. You have to be media trained on that. What are they doing at opening?

They should say we don't talk about the training data. We license data and we use data that's openly available, but we don't talk about what data we use. That's a perfectly good response. That's what I'm saying. But we shouldn't blame her for that. We can maybe blame the organization. You know, I think it's like she's doing great work. Yeah.

So, yeah. I don't want to single her out, but yeah. The organization failed her and she failed the organization. They need to be prep for this stuff. You know, the comms organization gets a bad grade. Oh, my God. They get fired. Yeah.

If there is, even if there is a comms group. Yeah. I mean, is there a comms group at Open Air? Maybe they just don't need. It sound to me like an organization that doesn't have a comms group and doesn't speak. Yeah. Offily about these things. Like, you can't wing it when you're open AI and in a lawsuit with the New York Times.

Yeah. Like, they're in a lawsuit with the New York Times. It's the, this is the content. I would say this is. is the content lawsuit of the century, of the past hundred years. You can't blame the CTO, though. But that's why I blame the comms team. No, I blame her too.

Yeah, she's got to think better on her feed and be thoughtful. Anyway, I give it an F you, give it a C. What do you give the impact of Elon and Twitter and Grock and all this stuff of open sourcing X open sourcing, I think impact from, you know, what it means to the community and like advancing open source, thinking around a, AI, A, you know. Yeah, it's an A. A plus, we can't give it that because like we'd want to see.

There's no data. Right. We'll just give it an A&A. Exactly. So, you know, there's a room to be better. Yeah. And I think this goes back to maybe what the core mission of open AI was supposed to be advance AI for all of humanity.

And so by doing something like that, it does, it does advance things for everyone. Let's get into other news. We have the first two new stories out of the way. We'd love to do demos here on AI Tuesdays. Go to this week in startups.com slash AI to get to our playlist of all the AI demos we're doing. We're well over 100 now, folks. Yeah, so this week we have actually two, one that, you know, we just can't demo unless we

maybe get the robot. We can go see the folks that are doing it. The first one is figure AI. And what I'm going to just show a couple of things here. You know, obviously I don't have access to this robot, but I'd love to go and maybe we can go, they're here in the Bay Area so we can go and see them. And also these folks have created this robot very much like, you know, the Tesla. What's the Tesla?

Optimus. Optimus. Yeah. And what they did, that was really interesting. And I'm just going to show another video here really quickly. This is a real world humanoid robot. It is done with two legs, two arms, a head. Which, why it needs a head makes no sense.

Like, what is the purpose of the head except for design? I mean, you could put all that in the chest, I guess. So I'm going to show you why. So in this video, it's just like a very short clip. You're going to see that this gentleman is speaking. And you standing nearby with your hand. on the table. Great. Can I have something to eat? Sure thing. They asked the robot to describe what was on the table. On the table were some plates, some glasses, a dish rack, but one thing you could

possibly eat an apple. And they did a conversation, so it did voice recognition of him asking, I need some, I want something to eat. The robot assessed what's on the table. And I guess assume the only thing edible on the table is the apple. Am I correct? Correct. So we've described it for people who are not watching. If you want to watch, just type in this week in startups on YouTube. Navigating the B2B maze can feel really tough, huh? If you're trying to hit the mark with all those top-tier executives, you want them to pay attention to your enterprise product. But where can you find all those big fish, the whales? The ones who call the shots make the buying decisions for corporations, for startups, and everybody in between. Well, here's where LinkedIn ads is going to solve that

problem for you, and I've used this. It is one of my secret weapons. LinkedIn means business. Business equals LinkedIn in people's minds. When you're on LinkedIn, you're in the business mindset, so you're going to really be thinking about business products and services. You're open to those opportunities. And LinkedIn recently passed a billion users. A 180 million of those billion are senior executives, 18%. But hey, we all know about the 1%. 10 million C-suite executives. That's your CFO, CTO, CIO, these are the people who are always looking for a new product or service to make their organization run better, but they are on LinkedIn. That's why LinkedIn's ad platform delivers two to five times greater return on investment compared to other social media platforms. So easy to understand why this is,

because this is where all the business people are and they're in that business mindset. Super easy, call to action. Make your B2B marketing everything it can be and get a $100 credit on your next campaign. Go to LinkedIn.com slash this week in startup. to claim your credit. That's LinkedIn.com slash this week in startups, no spaces, no dashes, terms and conditions apply because they're giving you a hundee. Why is this impressive? It seems very basic, yet it is kind of impressive. So why should we be super impressed here? Let's go in the following order, right? In terms of like creating humanoid like interaction, I think this is where the head comes in, right? You know, of course, you could have done the same thing and you could have had a

robot without a head there. Yeah. And it would have been 10 times freak here. So one is just creating sort of, you know, comfort, I think, from a human perspective. Making the humans comfortable is why it has a head. Yeah. But two, to hear something. And, you know, Elon says this when it comes to FSD. You know, the FSD with no LIDR arguments or no ultrasonic's arguments is he goes, look, we're a human and, you know, millions of years of our biology have given us two eyes

that give us stereoscopic vision. And that's what we can use for depth perception and everything else. And so the argument is if you mimic. the human form and you give it, you know, two cameras, it has sort of all the features that humans have. And as you enhance its reasoning capabilities, it can do it the same way that millions of years of our biology landed us having a head with two eyes like this and ears at the side. So these are all evolutionary traits that there's millions of years of biology behind, but we can just copy it immediately and say, well, most likely we put the speaker right

here, we put the microphones right here, we put the vision things right here. Biology has already figured out that's the best format. The same way Elon says, you can drive a car with your eyes, so a computer should be able to do it just with its eyes. Yes, it's just a matter of the software, what the gray matter is in our skulls being able to catch up, and that's the language model, obviously. So here we are starting to see the beginning of a language model being put into a robot. Now, we have, obviously, the Boston Dynamics robots doing backflicts, getting picked, doing all kinds of very interesting what you would call verticalized software, right? Like very vertical software solutions. Here's how to walk.

Here's how to pick stuff up. What this is doing is very different. Explain to the audience why this is so different than the Boston Dynamics, incredibly powerful, you know, demos that we've seen in the last decade. And my guess is like, this is building on the learnings from what Boston Dynamics. were doing in all those crazy videos. And what they were figuring out was movement. And they were figuring out movement and stability and agility. What now we're bringing into it is what I'd call reasoning.

And in that particular video, you saw the human walk up to a counter. And you saw the robot have like a, again, the sensors that we have. And basically to speak, to listen and to see. And the humans say, can you give me something to eat? And the robot be able to reason over what's in front of it. and basically decide, you know, this is the thing you should eat. And you can do this. Yeah.

And an action of handing to the person. You can do the same thing yourself, right? You could take a picture of a table and there can be five things there and say, you know, tell me the thing I should eat. Exactly. And it most likely would tell you the right answer. Yeah, chat cheap people would know that. Yeah.

It would. And so it's emerging. So I'd say Boston Dynamics was much more on the like physics, movement, agility side. And you need that. Yeah. Exactly. And now. you're bringing the capabilities of these large language models into place to do speech to text

and then reasoning and then text back to speech and then tell the human, here you go. And then combine that with all the agility stuff to hand the apple back to the human. How does it know to hand the apple to him? Because he says, what should I eat? Yeah. And it hands it to him. So they've programmed the, they've additionally programmed that language model to take an action. And that I think is the most fascinating part. There's no action when you take out your chat chip EBT4, take a picture of your counter and say, what can I eat? And it says, oh, you can eat the apple. You can't eat the spoon and the fork and the plate, but you could eat the apple. But it said, it took the action of handing it to them. So they have built software into this to know to give the apple to the human,

that the human wants the apple. There's some other code here, right? There is. And so that's a really great question, actually, super insightful, Jake. These LLMs now, they have. They have. And have something called function calling capabilities. And so when you're seeing a lot of the more advanced stuff coming out of builders, what they're doing is they're using the LLM to perform reasoning. And then basically in that reasoning task, they basically allow it to have access to a selection of tools. And the LLM decides,

so all of this is usually happening in the background. So if we go through the whole workflow there, and I don't know this for sure, but like I'd say this is accurate, probably within 95%. It's initially speech to text, right? You got to take what the human says, you've got to turn that. We've all seen that happen. That's what happening for a long time. Then you've got to parse that the human said, what can I eat here? That then triggered a function and say, well, let me take a look at what's in front of me, take a picture of it. And then I can ask

the LLM the same question you or I would. Is there anything here I should eat? And then when it says, yes, it figures out there's something there. Then there's probably some function calling that happens says, well, based on what else was asked, what should you do next? And if one of its, one of its functions is hand that thing, and they could have hundreds or thousands of these. Yeah. And so what's really interesting here is that's most likely what is done. It says, one of its function calls, if someone asks you for something and you have it, we'll give it to that person. Right. Now, if that had been, if I had said, what on here is the most explosive item? And it's it, Well, this nitroglycerin is.

Yeah. Now, does it decide to hand me the nitrolycerin? Does it decide to hand me the grenade? Well, this is guardrails, right? Now we're back to guardrails. We're back to guardrails, right? And, you know, Isamov's rules around robots and not hurting humans, right? Right.

This is back to prime directive. What is the prime directive of these robots? So now, science fiction, right? In the movie I robot, there are directives. In the knockoff robocop, there are directives, right? and you can't hurt humans. So are we to believe that this company has in fact given prime directive to these robots? Because now it's doing interactions in the world.

It must have don't hurt a human in there. So if I was building one of these, and it might not be as simple as that, but imagine the whole thing is driven by a prompt. Yeah, there you go. The number one item would be do not hurt a human no matter whatever you do. And then I'd have guard rails outside of even. of the LLM. This is really fascinating because where this is intersecting again with Tesla is you've seen this in the evolution of FSD 1 through 11 and then 12. Yes.

So the big jump from 11 to 12, which is not a jump because it's a completely different way. Exactly. And so it's odd that it kind of continue the numbers, but they just had to do it, right? Software versioning. But what they did was they went from creating code that was, you know, full of all kinds of branching, you know, if this, then, that, right? Yeah. And, hey, there's a stop sign to do this. There's a speed limit to this. There's a car runs across the road.

A child runs across a road. Do this. Yeah. To basically an approach that's built off vision and learning by watching the video over and over again. Listen, as a founder, there are things I love doing, like building products or meeting with partners, hanging out with my team and dreaming up new ideas. And then there are chores that I,

don't want to do. I don't want to do HR. I don't want to do payroll. I don't want to deal with all that. So I use Gusto. Gusto is the best for payroll, for HR services, and for running a small business, it makes everything so much easier. Even a mid-sized business, man, I get a lot of portfolio companies that are pretty sizable using Gusto because it is designed for you, the small business owner. And payroll is something you definitely do not want to mess up. You've got to get it right. And Gusto is going to make it perfect for you by calculating paychecks perfectly. Also, payroll taxes. You got to get your taxes right. You can't make mistakes there. And you want to set up open enrollment. You want to be good to your people. Gusto handles onboarding, health insurance, 401k,

time tracking, commuter benefits, off the letters. And they even give you access to HR experts. So Gusto takes all of this off your hand and lets you focus on important stuff. Your product and your customers, it's super easy to set up and get started. And if you're moving from another provider, Gusto will transfer all your data for you. Here's your call to action. Because you're a twist listener and you're part of the family, you're going to get three months free. Incredibly general. is totally unnecessary. Thank you so much to our friends at gusto.com slash twist. You must go to Gusto. Again, gusto.com slash TWIST to get free months free. Thank you, Gusto team. And so, you know, I was just watching the GTC keynote earlier. That's the big Nvidia conference.

And what someone was saying, what Jensen, sorry, someone was saying in that conference, which I'd never thought about before, about like, how do humans learn? And we think about textbook, and reading on the internet. It's mimicking. But one thing he said is like, you know, a big way we learn is just watching TV. So imagine you give that robot, you just let it watch TV for, you know, an amount of time or consume things. And so the ability to figure out even what those functions are and then how those functions operate can also become an AI task in and of itself. Wow.

So just to pause there for a second, insert the clip here of the fifth element. When Mila Jovovich, the character, wants to learn, she just like touches a TV or something and all of a sudden, br-drhs, the whole entire history of humanity, including the Holocaust and the moon landing, just every similar thing. She learns from a television. As well, in the movie The Matrix, it predicts I know Kung Fu. He just watches every Kung Fu movie ever. He watches every martial arts fight scene. And now that's in his programming. So now we start thinking about science fiction. It literally predicted this that you could just watch the history of television, all recorded, and then learn everything. Learn. Wow. It's just extraordinary. And then all this work that Waymo did, all this work that Tesla did on full self-driving,

Waymo is done, doing conditional statements, 100% of that code is just going to be thrown in the garbage and never used again. It was all learning up until this point. Now that we have these models, we're just figuring out, hey, what has, everybody else done at this intersection? And what would a human do in this intersection? Yeah. When a bicycle cuts across it, you'd stop. Or how you learned to drive or I learn to drive the same way, right? You know, we looked at some rules. We watched TV for a long time. We knew that you shouldn't do X, Y, Z. This is crazy. Yeah. Hopefully it doesn't watch the French

connection. For people who are under the age of 40. Great seat. Great scene. I mean, it's the greatest car chasing ever. Just type in French connection. car scene. Okay. So this robot is super interesting. Like, it is going to take time, but because it's a human factor, what's very interesting about picking the human factor is evolution made humans to the dominant species on the planet. Right. So whatever it is, you know, you might look at tigers and say, that's more efficient or sharks in the water are more efficient. For some reason, our frail little bodies have done pretty well with a giant brain. Okay, fine. Now you start to think about, well, we also built the modern world. So we built the world, the door behind me, windows, chairs, office spaces, cities.

We built these for our form factor, cars, etc. So by building, even if the robot would be better as a little tank with a, you know, four wheels or, you know, tall or small or whatever, by building it to be six foot tall or five, ten or whatever the average human is, five, eight, probably. Well, it now can operate in the real world, a factory, a kitchen,

a car, you know, walking through a city or a town. So it's kind of brilliant that we're making them to look like us. Yeah. And I think about, you know,

one of your great investments, you know, I saw yesterday, I was in the airport, the CafeX. CafeX, yes, at SFO. And think about the evolution

for CafeX, right? Could it be that you can, and I know you guys have done a lot of work with the robot there using the standard machines, how does it fundamentally change if you could use the figure robots and think about sort of all the software that would just fundamentally go away, where it's you walk up and say, I'd love a double espresso, you know. It is obviously the future of it. Now, I think it would be slower than a purpose built. we're now talking about narrow AI, verticalized applications versus wide. What Optimus and this figure, it's called figure of this company. Figure, yeah, figure AI. What they're doing is they're going for that, like 10 years from now, this thing,

you could general purpose robot, just like, you know, chat GPT is general purpose or Gemini's general purpose. That means it's not going to do coffee very, it's going to be slow and klugey. It's not going to move wicked fast in a contained space with walls or, around it that customers can't get in between it. So you're going to still use vertical AI to win at chess or to win at making coffee. But if you were doing a restaurant, man, this robot would be pretty great to work at a holiday in.

Like literally, you know when there's a holiday in and they don't have overnight food? Like this robot would be pretty dope, do that. But the ramen machine would be better at making just perfect ramen, like a verticalized ramen. But it also could be like an expansion for CafeX. Like, they get it right, and you sell them into Starbucks, right? They want to keep their storefront form factor, and you maybe want to talk to a human. Yeah, drop one of these in.

Because, you know, it just increases their throughput. So, I mean, it's a brave new world right now, and this is moving at a crazy pace. I mean, looking at this, I think it's like a B. I mean, it's not super impressive to me. They've glued together the LLM and the robot. Congratulations. I'm not blown away yet. but I am impressed that somebody has finally put a connection together.

But it's pretty basic. I mean, if it put a bunch of vegetables and fruit on the table and said, make me a salad and then it asked me a couple questions, like, would you like onions or not? Maybe that's a challenge. Maybe that's a challenge to the team. That could be a cool thing. Yeah, here's my challenge. We can get one.

I give this a C plus. I'm giving this a C plus. I'm not impressed. with the robot because the robot's not as good as Boston Dynamics and I'm not impressed with the AI implementation here because it's base. Basic.

You get a C plus for plugging the two things together. I want to challenge the team. Put a bunch of stuff on the table and in the refrigerator, give it a knife and say, make me a salad. You make me a salad to my specification. I'll be impressed.

Wow. That's my A plus. So there's your, that's it. That's your test. You're a tough customer. That's my turn. test. What do you give it? I give it an A because I know how hard it is to put the robots together. I know how hard it is to make all these things work because I do it day to day.

But there's a lot of, you know, I'd love to try one. So you know what, if they can reach out or come on your pod or something. You want to see if the apples any good when he hands to the apple. No, I want to, you know, more experiments, right? That's just like a two minute video. So, but I see a lot of potential there. You know, I see a lot of areas. where there can be some improvement, like, you know, the voice reaction was a little bit slow. Maybe that could go faster. But their overall, I really like where this goes for. And I do believe, I think Elon has said something, is like there's going to be something like 10 billion robots on the planet within the next 30 years of these humanoid robots.

Yeah, one for one. Everybody will have their own robot. It'll be like C3P. Yeah. Why wouldn't you have a robot? I mean, if it costs. I mean, the idea that everybody would own a car was farcical, like for a long period of time. I'm like, why would anybody need that?

And now everybody has a car. I think there's more cars in the United States than people. I don't know if that's that's true or not. But so, yeah, why wouldn't there be? I mean, if they can get this thing, I think they can get these down to 10, 20 grand. And then it's like buying a used car. What would you rather have? Let me ask it this way.

Would you rather have a robot that could, you know, do anything a human can do? Or would you rather own a Prius? And you have to take public transportation and have your own. robot or you can have a Prius and not which would you rather do. I take the, I take the robot. 100% everybody takes the robot. Everybody takes the bus and the robot.

You could have the robot making money for you. You'd be like, hey, you know what, go find me, you know, go, go pick, go to fish strawberries, go, go forage from mushrooms in the, you know, in Portland and come back to the, shovel the snow at my, in Tahoe. In Tahoe, go shovel the snow.

Right. I mean, it's 200 bucks an hour. has been waiting for this, hasn't it? Yeah. Robotics has been waiting for this. This is going to be bigger than AI itself. I think when you think about it, like...

It's like a related though, right? You know, I know, but AI without the robot means like the problem set is like, you know, 10% of the problems in the world. Yeah. AI plus robot equals 90% of the problems in the world. Like, I think it is almost everything. What couldn't the robot do? like maybe fly or go into water or maybe there's some problems they can't solve.

But, you know, this feels like, whoa. This is like, whoa. I'm kind of in the same spot as you. I think seeing multiple companies do this, you know, between Boston, you know, Boston Dynamics, the Tesla with Optimus, and then these folks, I feel like we're closer than ever. And I'd be willing to say definitely within, within, three years, people are going to have these at home. Wait, wait, how many years?

Within three years. There's a bet. Hold on a second. Yeah. Startups and small businesses, listen up. You want a CRM that neatly organizes all your customer data so that you can avoid missed opportunities and you can deliver a personalized service. Rid CRMs can adapt to your fast growing needs and that's where audio comes in.

A TTIO delivers the goods. It's a custom CRM that's flexible and deeply intuitive. ATO is built for the most. modern company headed into the next era of businesses. It connects your data sources, adjust easily to your specific setup, and suits any business approach, whether it's self-serve or sales-driven. Adio automatically enriches all your contacts. Think about that. You might be missing a first name, a last name, an email, an address, all of that stuff. It's going to sync your emails and calendars. It's going to enrich those contacts, and it's going to give you powerful reports.

It's also going to let you quickly build Zapier style automations. If this, then, that type automations. The next generation deserves more than a one-size-fits-all CRM. Join industry leaders like OpenAI. Replicate and 11 labs and scale your startup to new heights with Adio. Head to adio.com slash twist and you'll get 15% off your first year. That's ATTIO.com.com slash twist. Let's make this a proper bet. We got a new bet alert. This week in startups.com slash bets is where all the bets on the show are being put. We've got to go back to the, now that we have AI indexing the whole site, we need to go, back and find Alexis O'Hannian's original bet.

But look at this. This was this company, Rude AI that got bought. Yeah, yeah, yeah. Unfortunately, it was like with a SPACT company and I think it went to zero. Okay. Yeah, it's kind of sucks. Look at this robot. Yes.

Zip, zip, zipping. This is three years ago picking tomatoes. And, you know, the hand is just perfectly designed for this. And it's using computer vision. And this is before LLMs. We're in the mix. This is a verticalized, you know, application. But look at how precisely it can go pick,

cherry tomatoes. Or we can do raspberries and strawberries. Here it's doing bell peppers as well. And so this idea that you would actually pick vegetables is going to be over soon. Now, this thing only worked in vertical farms. So you see this as an indoor, it's not out in the field. And it's not the dimensions of a human because you need a different type of hand. But if you had the optimist or this future one, you could take.

take the, it could take its own hand off. Of course. Yeah. And put the regular hand on, right. So what's your bet? Frame a bet here. Come on. I want to get a little active.

Well, I just want to comment on that thing. So what's going to happen is the same thing that's happened with LLAPs, right? Where it's become a superpower. And what people are going to do is the robot's going to be the blank canvas. And then people are going to program them and prompt them. There will be a lot of different ways that they'll operate with them. And so the amount of energy it took.

to create that robot that you just showed us was immense, an amount of engineering and testing. And then you had this kind of had to adapt it to a certain form factor for it to work. But going back to where we started this conversation, if you have a robot in a form factor of a human, which means a human should be able to walk there and go and pick tomatoes, cherry tomatoes. And you give it the slight difference in the hand like a tool. I think we're going to see that very, very quickly. So coming back to the bet, I think what. what we'll see is, yeah, within, okay, here's the bet. This is a, it's going to be their biggest

one yet. Okay. Within, within, within three years. So at the end, so 2027, and end of 24, 25, 26, beginning of 2027, you will be able to get one of these for yourself. Okay, January 1st, 2027. Seven, yep. A humanoid robot will be available for purchase. Yes. And delivery. Yes. You can have it in your home. You can buy it and have it in your home, buy January 1st, 2027. Okay. And you're asking me. No, the bet is.

I take the over or the under is my. Well, yeah, you take over under. You're setting the line. Okay. And whoever's right, the other person buys the robot for that person. Up to a cap of $10,000 for the robot. Let's say like $25. I think they're going to be.

Wow. This is a big major bet. It's a big bet. Yeah, it's a big bet. But you're setting the line. So I pick. Yes. So do we have to put the dollar amount of the.

robot in there? Like, that's just a cap. Like if they're, if they're, if they're, you can buy them, but I would think if you went and you offered a hundred thousand. It would be because I'm, yeah, you're saying January 1st, 2027 in America. Yeah. Around the price of a Prius, around the price of a Prius for the price of a Prius. Okay. So now we got something to pin it off of for the price of the entry model, the cheapest Prius. Yes. Um, so we can have two bets here. The over under on the price. And, and the date. So we could do those. So I'll go first. I'll pick the date. Okay. You pick the over under on the price of the item, which is a Prius, which I think right now at entry level Prius,

somebody can look it up, but I think that's probably like 40K. Yeah, I was going to say 35, 40,000. It seems right. Yeah, 45. I don't know. Anyway, we know we're kind of where it is. All right. So I'm going to take the under on January 1st, 2027 for 10K. And then you take 10, of the bet for over under on the price. Entry level Prius, the cheapest Prius new you can buy. Yeah. You think it's going to be over or under that price. Under, under.

Ooh, I think you made a bad bet. So I have the over. Okay. I think it's going to be like 50 grand for these 75 grand. No way. No way. Really? No way.

Okay, it's the entry price of the robot versus the entry price of the Prius, which I think is having to look at these things. I think these things are the equivalent of a Prius. Yeah. And I think it's got to be before January 1st, 2027. Gotta be. It's got to be before that. An Applevision Pro, which has all the brain and sensors is only $2,000.

This is the largest bet in the history of this weekend service. There's 20 grand on the line here. 10K and 10K. I mean, it could cancel out or somebody could sweep 20 dimes. Yeah. Oh, my, what are we doing here on this podcast? I don't know. I mean, this is getting out of control.

Our poker game is making its way into this. It's making it's way into this. I mean, are we doing this podcast in order to place bets? It feels like it. And that's okay. We're making it interesting for us. We have a bet that now is going to go through the rest of 24, 24, 25, and 26. So it's a two and a half year bet, folks.

It's almost a three year of bet here. But I like it. We'll be sitting here with more gray hair. Yeah. And yeah. Yeah. So I give this thing a C plus as done here. I want to see you guys make a salad.

I think if my salad challenge. I think they should come on. the show and maybe or we can go on site and see what it can do. Let's go on site and see him make a salad. That's like a real life because we haven't done one of those yet with someone. Let's go go on site. Yeah, I can get somebody like to do like a really good like get a, we get a red camera or one of those like really good HD cameras.

We'll get makeup. We'll get some old man makeup going. Yeah. Okay. I like it. All right. Let's do another demo and then we'll wrap here. We got a lot done today.

Well, the next one I want to talk about as well is not a demo and then I have an actual demo. All right. So we're just doing all top, all topics today. I know I have five queued up, but these are big ones. Okay. So you remember last year, you were starting to get excited about auto GPs. Yes.

This was, I think baby GBT's and auto GPTs, yes. Exactly. So this company, which was founded by a couple of amazing engineers, the CEO is this guy, Scott Wu. Interesting background. This guy was a competitive coder. And so he had won competitions doing competitive coding, incredible. And what this team did was they basically took the idea of auto GPTs and they really took it to the next level. And there's a few examples here.

I'm not going to play the videos because I think everybody who has seen it at this point. If you haven't seen this yet. This is Devin. Yeah. But what they did, which I thought was and it's really, really insightful. Everybody else was using auto GPs in a very constrained box, like either in a term. or something like that. And it could kind of figure things out,

but it was really hard for it to think holistically. What they did was, they kind of flipped it around, in my opinion. They gave Devin an IDE that was built, so it was an integrated development environment. So it has access to code, it has access to a terminal. This would be like a replet.

Great example. They basically gave it a custom replet that their AI can drive. And so by giving that, the auto GPT, in their case, Devin, had access to much more much different set of tools. It's much more powerful.

And so because of that, and then they obviously took a model and they fine-tuned it, did all the good stuff, and they have great scoring on that model. They really broke through what I believe was the barrier in the, you know,

you sit down somewhere and you're like, hey, I want to make a site that basically does news and this is how it should work and all those kind of things. And so they, in their IDE, have like a chat bot. They have a terminal.

They have a web browser. They have all. all the pieces that you need to go off and do this. And so I really think that they've done something incredible here for that particular revolution of like the auto-GPT style of applications. Yeah, it's super impressive. And this is the future. I think, you know, we have verticalized apps. Okay, I can beat, you know, big blue chess.

I can beat Casparov. Great. Really programmed well. Okay. What's next? Co-pilot. Okay, really interesting. I'm working.

the co-pilots finishing my sentence, giving me an outline of my blog post, or working with me on my code. And then we're going to agents, right? We called them baby GPTs when we started this journey. But basically, it's a role. It is a worker. And so the way we should think about this is we don't use the term slave anymore. I also got corrected, by the way, the language police. I call it something a master bedroom.

You can't call it a master bedroom anymore because it's hard to get to slavery. That's what it. Oh, wow. Okay. I hadn't thought about that. So I think they call it primary. So now they're calling it primary rooms. Okay, I get it. And so, but essentially you're creating a slave, right? And that's what they used to call all these things in programming code, right? You have this slave. So this is like a role. I'm going to just call it a role, a job function. And if this thing can operate through a role, what I described on the last all in on the one, before was what I believe will be what I'll call maestro. The maestro is coming, the conductor. Oh, great name. Great name. Yeah. So the maestro is coming.

And what the maestro is going to do is I'm running my one person company. And I just say, okay, I need a developer. Oh, I need a little designer over here. Okay, I need a copywriter over here. And then what is my job every day? Well, I'm sitting there with these roles, right? With these agents. And I've got these virtual.

employees. And I'm just pushing them along. Hey, show me a new design for that app. Hey, I want to add to this app, Twitter login and Google login. Right now we go login through phone. But I want to add the Google login code. Boom, it does it. Okay. Hey, copy team. Let's write a blog post about this and do a tweet. Okay, boom, we got that queued up. Okay, now I want everybody to build the launch plan. Give me the plan that we're going to launch this new, you know, login with Google. And I want press release. Okay, boom. And it's like, whoa. Now we start thinking about what would modern day entrepreneurship be, it's going to be being the maestro. Meestro. You're just going to be a maestro with a bunch of virtual assistance. And then you might bring somebody in. And so there's going to be

a new class of company I predict where it will come in and it will be like, hey, you're using this AI to make your marketing plan. So we're going to have a human review it with them, right? a marketing maestro. Now you got a marketing maestro at the company. What do they do? Okay, they got the copywriter AI. They have the ad buying AI. They've got the logo and the brand building AI, you know, the tone of voice AI, the social media.

I mean, this is going to get really interesting. And I think this is where we get to a 10 person company, you know, making $100 million. Each employee makes $10 million. And then you have 10 person company, 100 million in revenue, equals a unicorn, right? 10 times revenue. And what would the margin on that company? If you pay each of those people a million dollars a year,

because why not? They're my Shadows. Yeah. And you spend 10 million, let's say you spend 10 million on other expenses, servers, whatever, make a 20 million on operations. Got a 70% margin.

Now you've got a $70 million profit company with 10 employees, 70 million times a 20 times EBTA, $1.4 billion dollar company. That's the future, bro. I think this is the future. I completely agree with you. You know, definitely it's an A plus, right?

Yeah. It's A plus. The only reason I'm not going to give an A plus is I don't have access yet. Oh, you go A, I go A plus. I go A plus. I'm going to give it to them on vision and be in the first out of the gate. And look, like, they're awesome. And what you're saying is super fascinating.

There you go. This is their scoring of their, on the software engineering bench. What I would say is when you just said what you were saying, I think about all the windows you kind of have open. And so what you're basically taking out to a next level is you'd have kind of like your Google Docs. And you can sort of do this today because if you enable in your Google Workspaces or your personal Google Docs, they've put in some of the co-pilot features now. And then you can do it within like your tweet deck equivalent as well. And this is a really fascinating idea that take all the startup functions and you basically allocate a co-pilot.

pilot to it. And then initially, one person is driving in. And then over time, you maybe layer it's some other people. I really like that. MISRO is the company. I want to incubate. So if somebody out there wants to do MISRO, really interesting times, should we do one more demo? You know, we had a lot of catch up news that we had to get to. So. Yeah. Well, you know, since we're on news, I'm just going to keep it on that topic. Okay. Keep going. Because, you know, this is one that you and I were talking about earlier. I mean, we can do news today. And I'll, I'll do another episode with you this week, man, I'm so deep on the AI. We can just do a demo episode this week.

Yeah, we'll do a demo episode because there was a lot of news to catch up on. A lot of news to catch up on. Yeah, but this was interesting because you and I have a couple of bets going with Apple and then there's bought off the press's news as well. So what's interesting is this is a

research paper that was released by Apple folks that talked about their work on a multimodal LLM pre-training and what they had done there. And so it's an excellent paper. It talks about, you know, relatively small sizes.

And, you know, you can see here as an example, I'm going to zoom in where they're asking their thing is like, how much should I pay for all the beer on the table according to the price on the menu, right? And so, yeah. Right. And you can see they kind of compared it to different chats, right? Which is their set $12 and this other Emo, 37B said 59. and then Lava basically came up with a different number, right?

And so explain why. And it said here, look, there's two beers on a table. Each card costs six according to the table. So six times two is $12. This is a really, really, like, well done. So amazing. Yeah. This is where it gets really interesting.

It's not just like hand me the apple and the apple's the only thing. It's like it's really doing some logic here. Yeah. There's two beers. The price in the menu is $6. Therefore, I mean, it's, and it, the ability to show your work is, I think going to really help speed this up. So what we can learn here dovetailing with the other breaking news. The other breaking news is there was a Bloomberg story by a very credible journalist

over there at Bloomberg, and then perhaps the most credible journalists covering Apple, that Apple and Google are talking about a partnership where Apple uses Google's Gemini in the iPhone. This would be colossal. What do you think is going on here? Because, Apple not having AI means the end of Apple in my mind. They have to master AI. They can't give it to Google, can they? So I believe that this is slightly different, and it's related to something you guys talked about,

even on the All-InPod. This, I believe, is a business development deal versus a technology deal. Okay. Explain the difference. And this is a business development deal, because, one, the exchange and value of the search deal, is ridiculous. I think it's on the order of $15 or $20 billion a year now. And so for both of these organizations, there is a data need and a channel need, I'd say, for Google in terms of that's

why they pay so much for that default search and Safari. And on the flip side, that's a significant revenue stream for Apple today. So this is a business development deal to try to keep the status in place as much as possible, where Apple's like, man, I think our investors, would really probably be upset if we lost, because you got to think about that 20 billion or so that comes. Do you know what the number is? It's a big number, right? I think they might be up 30 billion for the search deal on this story in Bloomberg from Mark German, who is known for being one of the great people covering it. So you could be correct here in the framing is they're just going to put Gemini on the iPhone. There's just be a Gemini app preloaded and it lets you interact with your

phone in some way at a deeper level. So when you say, or Siri, I want to do this, maybe it would do, you know, might use Gemini to help. I don't know. But so that's what you think this is a carriage deal as opposed to replacing Siri. I think so. It's use your distribution mechanism that you're buying today to put yourself in place to then, because there's cascading effects here by having this deal. If Google and Apple work this deal out, Apple gets to keep that revenue.

stream, no startup can pay them that. Open AI can't pay them that. No one can pay them that amount of money. I would think there's exactly one company in the world that can pay them that amount of money, which is Google. And on the flip side, if Google can get that data into their ecosystem, then what that really does is it helps them continue to get a moat around the experience. And, you know, so what users are doing to build a better product.

So in a world of like, you know, all this regulatory stuff that challenges startups today, they probably couldn't go by perplexity right now, which is, you know, they'd get sort of challenged on all this. Yeah, sure. So here's what it says. The two companies are in active negotiations to let to let Apple licensed Gemini, Google set of generative AI models to power some new features coming to the iPhone software this year, said the people, blah, yada, yada, yada, Apple recently held discussions with Open AI and has considered using its model. according to people. If a deal between Apple and Google comes to fruition, it would build upon the two companies, search partnership as you talked about.

So this is kind of interesting because the way this is being framed, I think, is that this, they say it's for like a certain set of features. I wonder if this is just like image correction or, you know, search or something, or an extension of the search thing.

So when you search, it does something a little more intelligent, you know, in the search results, right? So I don't know, man. Or do you think it's white labeling it and to power series behavior? I bet on the latter. I think if you put anything else, it's going to power Siri.

I think if you put anything else in the market right now that doesn't have capabilities like we talk about on this show, I think you're going to put yourself behind. And so my guess is that holistic, and there's a couple of parts to doing this at scale, which I think people are underestimating. you can pay chat GPT or OpenAI $20 a month. You do. And it only lets you send 40 questions every couple of hours. Yeah.

Okay. You can go to Bard and use it as much as you want to. So Google has and is probably one of the only few companies that can operate the infrastructure at a scale that can also power these things. So it's a cloud play as well. Anyway, this is fascinating. It's a whole ecosystem play. You got to think, you know, with the Lena Con Biden administration, you know, and the EU and the UK is, because obviously UK is not in the EU anymore, the UK's regulatory departments, there's going to be a bunch of microscopes on this real quick because these two own 100% of the smartphone market. So this is very strange.

The search deal is already under scrutiny. This is now, well, AI plus 100% of the search of the mobile phone market. And remember, there was a rumor that Johnny I, formerly of Apple, who created the iPhone and Sam Altman amongst the many deals floating around that Sam was involved in, a consummate dealmaker, obviously, shout out Sam Waltman, that, hey, this was going to be, he was going to make an opening iPhone. So this is really fascinating. Maybe this is a defensive play for them to keep the duopoly.

Hmm. Yeah, competition makes for strange bedfellows. There's a lot of angles to it. A lot of angles. Well, we have our bet that there will be a model built in to the iPhone and the Android to smartphones, and you're going to win that one, I think, hands down. I think that's an easy bet for you to win. So congrats on that. All right, everybody, if you want to watch this show, because we had so many video references,

please, just type in this week in startups, go subscribe, hit the bell. You'll get the notification when a new episode comes up four days a week right now. AI Tuesdays, and we're going to do a demo show. So we'll book another show this week and we'll try to get the demo show out later on on this week. Follow Sundeep. X.com slash Sundeepe. X.com slash Jason for me.

If you want to be a Mench, go ahead and write a review on your favorite podcasting app, rate, subscribe, all that nonsense. You can follow us also on all the socials, TWI Startups. See you all next time. Bye-bye.