All right, all right. Welcome, everybody, to Silicon Zombies Brainwaves. That’s right. And super grateful to have Lutz here, Lutz Wenger.
We’re going to be discussing a little bit about the future of artificial intelligence and all this crazy policy that’s hitting us. Lutz, it was such a treat to have you on. Quickly told, I’m German, so my accent, you have to suffer through that now. Sounds so cool.
Yeah, yeah, well, I don’t like a little bit of authority, right? No, so I’m German. That’s cool. Yeah, well, look at you.
Faculty at Cornell, I’m a visiting senior lecturer there. I worked at LinkedIn, Google, Snapchat before, and I have now my own company in this space. We can learn a little bit more there, too. Maybe.
A photojournalist in your past as well? Oh, yeah, yeah. Well, I was a Fulbright scholar in the U.S. I’m a physics guy.
I actually studied. I studied quantum physics. That’s right. And then I got a scholarship from Fulbright to come to the U.S.
And, well, University of Chapel Hill, North Carolina. Yeah, Tar Heels go. Excellent, excellent. But they are actually quantum physics.
The physics wasn’t that good. So I actually decided, well, to take a year off and study photography. Yeah. Like fun stories for a later time.
I went across the country as an illegal passenger of trucks and did a photo story on trucks. Wow. I was a Davidson guy. You know, this is the we’re going to totally go down a different weird direction.
But one of my favorite books is On the Road by Jack Kerouac. That’s awesome. And when you when you mentioned that to me, it reminded me of like him just totally checking out and taking on like weird jobs and meeting unique people and writing about it. Just being in the now being in the moment.
Speaking of which, Jack Kerouac has written his like on his typewriter while drinking at the Black Cat. It’s just like a. Couple of blocks down. Oh, funny.
Yeah. So a lot of a lot of rich history in Los Gatos, California. Well, it’s been a really pretty strange week or two. It’s been a strange couple of months.
Maybe set the table for us. What did you cover when you were on the A.I. in the U with Jahan and the rest of the folks? You guys became best friends.
Well, I mean, there was such an amazing discussion and shout out to the EU. We had a very lively discussion. And for the ones who haven’t listened to it. So I’m pretty sure you can put a link into the show notes to actually get us there.
And a lively discussion about how to balance innovation and regulation. Right. And if you think about regulating, you always should like think about that. A.I.
as a tool that has three groups that can help it can help the citizens. And we have seen a lot of discussion about how do we protect citizens from harm? It they are. Um.
There’s the industry, the companies and they have IP and you need to protect those and help them thrive. And we have obviously nation states. And over the last few weeks, we have seen a huge like back and forth about all those three constituents and what this means. Yeah.
And it seems to be like the news cycle seem to be speeding up. Elon just tried to buy OpenAI for a hundred billion. I don’t know. Was he just trying to like throw it slowly?
Yeah. Like so close and so close and so close and so close and so close and so structure is so that Alam can’t buy them. It’s pretty simple, doesn’t work. That’s cool.
However, OpenAI needs a lot of money and we can go into this. In order for them to be successful, they need more access to customers, more access to data, they need to build out services, they need to make sure that their models get applied. In order to do this, they need funding and they just raised fundraising and investors normally want their money back. So essentially, they become a business.
And if they’re a business, that means they are not a non-profit anymore. So OpenAI is changing from being a non-profit to become a for-profit and that’s nothing unusual that has happened before. It is kind of unusual. Just push that because I feel like there’s a bunch of different kinds of innovation that a startup can introduce.
Legal framework introduction or innovation just sounds messy to me. Just focus on making the products better. Don’t try and hide the cheese by creating a structure and where you’re going to change it later and it gets sticky. I think there was a mistake and this is like how some Altman had envisioned it probably.
But it is not unseen that you go from a non-profit to profit and you split those up and you make this counterintuitive decision and so on and so forth so that you don’t be so so But it is not unseen that you go from a nonprofit to a profit and you split those up and you make this clean. That’s what Sam Altman tries to do by, and now the question is in that entity, what’s the value for the nonprofit side versus the profit side? By Elon Musk coming and put a price tag on the nonprofit side, he makes the fundraising deal for Sam Altman actually not very good looking. That’s all what it is.
He’s pooping the party. Because I know he’s trying to raise, Sam’s trying to raise 40 billion right now, and he’s doing it at a $300 billion valuation. So you think Elon’s trying to throw a- A wrench in it. Totally.
Totally. So it is- It’s a power move. It’s a power move in order to make it- Make it harder for him. Now, why would Elon do this?
Well, probably there is some pettiness involved in terms of his approach. However, we also have to see that, I mean, Elon has launched the chatbot called Grok. He has launched XAI, and you’re looking like exactly, most people look at like, what are those? Yes, they haven’t- I know what both of those are.
My quizzical look. My quizzical look, Lutz, was because I’m thinking maybe he was trying to offer 100 billion for ChatGPT because maybe Grok isn’t going as expected in the training rounds, or is that a possibility? So Grok is a chatbot, and we can talk about the uselessness or the usefulness of chatbots. We have a technology and we’re trying to replace ourselves, and then we are scared about replacing ourselves and all of this is …
Yes, sorry, my language, and we can go into this. But so, probably, let’s park the chatbot. Yeah. I’m going to park the chatbot thing because we can go there.
Sure, sure. But the whole point is Grok hasn’t gone anywhere close to what the potential or the hope was. But more concerning, XAI, that’s essentially the OpenAI clone, and you haven’t heard about stuff. You have heard about Mistral, you have heard about Antropik, you have heard about DeepSeek, obviously.
But XAI didn’t go far. Yeah. So one of the big model efforts for Musk didn’t play out. So you could argue, in my opinion it’s wrong, but you could argue that he’s doing it to catch up, as much as he kind of called the ethical concerning card of, oh my God, this is all too dangerous, we should do a pause, in brackets, as some people said, in order to catch up for him.
Interesting. But all of this very calmed down since he got the direct line to the president. By the way, you have a red telephone here, just like on the desk. Is it getting us to Trump?
That goes straight to the man himself. Yeah, perfect. Perfect. Let’s call him in later on.
That’s true. President Trump, if you’re listening, you want to dial in, or anybody else, it’s 408-357-0330. Thanks for turning that up. That’s good stuff.
We’ll leave the line open for you. Thanks. We’ll leave the line open for you. So, Lutz, let’s zoom out a little bit.
You’re pretty bullish on AI innovation for a European. Do you kind of think yourself as more of a European, or with American sensibilities, or how do you view yourself as a futurist? I’m so European, right? I have a US passport, but I wasn’t raised in Berlin, so I’m definitely European.
Yeah, sure. I’m a bit of a European. I wouldn’t say that Europe is not bullish. I see many people in Europe who are bullish.
Europe hasn’t produced the amount of business applications, however- Or funding. Or funding, but they have produced a huge amount of papers and knowledge. So Europe is actually pretty strong in the creation of knowledge. So don’t, and this is where we should give a shout out to Gerard from our last discussion, do not discount Europe.
Now, at the moment, obviously, and I made this joke last weekend, I was at a conference and I said, okay, let’s take the top 10 companies that work with AI in the US, take the enterprise of those. And now let’s take the top 10 European companies that work with AI. Yeah. Okay.
tall do you think are the european ones god maybe like a little figurine i would have to guess yes four inch four inch guys and like for the listeners this is so much here right yeah you see this is a little bit bigger than the microphone now if i stand up that’s dwarfing it so if you think about funding and possibilities to build up the u.s is way better sure well you know just just yesterday excuse me just this morning i saw vance who was at the i think the paris accord yes he was there yeah and i’d love i’d love your take on this he said that you know roughly 700 billion is going to be spent on ai invested in artificial intelligence this year and the u.s will likely get half of it so i mean from your perspective what kinds of what kinds of shifts need to change in order for for europe to to get back in the race with china in the u.s so um uh ozala from the line announced 200 billion uh uh president uh macron uh announced 112 billion or so and i scratched my head thinking like hey hold on france had a problem to bring their budget through um um parliament and where does he suddenly come in like somewhere yeah um but note that money in itself doesn’t solve the problem right and if you look at the stargate announcement uh in the white house for sam altman musk wasn’t there some altman uh together with softbank announced that the 500 billion and if like all like listeners like this is like numbers right like let’s count everything which you’re also left on the line and macron put together this is 300 billion and then we have just one announcement of a private company like a private consortium in the u.s it’s 500 billion so again an unequal setup but it’s not only about money it is how do you spend the money what is the ambition level to set with the money and i i think we are a little bit mistaken to kind of think oh we need to put this all in research or we need to put this all into models because what we actually have seen with deep sea because like models are like fast to build there’s an open source community so um 500 billion invest today might be 50 billion worth tomorrow so very simple um a neural network is actually not something very complicated so what i didn’t say in my intro i said i’m faculty but i i just recently launched an online certificate for econel it’s open for everybody it’s meant for people like you and me you do not need to code because i essentially replicate myself i built a virtual looks and looks ai that codes for you that helps you to support you along and has all the discussions with you like it’s you do have to listen to 100 hours of real it but then you get a nice certificate these are my courses and yeah yeah i’m like here we go it’s called like it said like if you look look for my name let’s finger at uh key cornell you will find the course um designing and building ai solutions but in there i discuss what are neural networks sounds completely complicated but it ain’t a neural network is a bunch of linear function in an activation variation function that sounds complicated make it even simpler you know what a logistic regression is you have heard this in in your high school we stuck many logistic regression on top of each other something like logistic regression and that we call a neural network so instead of having one logistic regression to be so You have an interest. You say, that’s a cat. And that book says, okay. Is that like a logic trace or a reason trace to show how you feel?
No, there is no logic. So you show, and by the way, reasoning. Let’s try. Let’s try and like, I’ll put you now on the spot.
Yeah, please. Explain to the listeners, if they have never seen a cat, how does a cat look like? You have seen a cat, right? I’ve seen many cats.
Are there, can I make an assimilation to something similar, like perhaps a smaller tiger? No, you can’t. You have to describe it so that I actually can see it and understand it. I don’t know what a smaller tiger is.
So, okay. Well, I suppose it’s kind of like the famous story about, you know, you’ve got the five blind guys in a room with an elephant, and they’re all experiencing differently. Yeah. From the trunk to the tip.
Yeah. If you say four legs, I say it’s a dog. If you say, like, a furry, then I kind of said it’s a fox. I mean, I will, yeah, like.
Wow, you said that. I did. But the point is, the paradox is, we know more than we can actually explain. You fail here to explain what a cat is.
Yeah. The only way for me to learn what a cat is, is you show me cat. Yeah. Right?
I experience a cat. So, we build neural networks in a way that explains what is a cat. So, the network now learns it. And now we can, because we have all those stacked logistic regression, this is it.
This is totally simple. Each of those neurons is a mathematical equation. And the mathematical equation that we learned in high school. Okay.
So, now we can look at it and see how that network learns what is a cat. And it learns first, shapes. Colors. Structure.
And the further you go down, the deeper you go into the network, the more complicated things it learns. That’s it. I’m fascinated by abstracting these ideas and maybe learning about how the mycelium and mushrooms grow. Or the underground roots that are connected in a forest.
And then maybe trying to reproduce that in a lab somehow. I think this intersection is just a pretty… It’s a pretty fascinating space. We can talk about mushrooms.
And taking mushrooms is the same thing than retraining a network. And we can do this. Is this like neurogenesis? It’s kind of like…
If you… So, we train a network. And we train it by showing things. Now, if we show them…
And showing things could be really showing them a cat, the network a cat, or like training it with training data. And if we show them the wrong training data, right? Then… The network would learn the wrong things.
That’s what we then call bias. It’ll tell us that that is not the thing and we can reward it for being correct. No, the not like… No, AI is not that intelligent.
It just will tell us what we trained it for. You’re saying that there’s no reinforcement learning with AI? Well, that is one way to train it over time. But if we reinforce it to the wrong thing, it’s, you know, it’s like we are reinforcing the wrong behavior.
The network, like us humans, by the way, learn the wrong thing. So then, if we want to correct that behavior, then we have to selectively correct the neural network. And that is nothing else than taking mushrooms, right? So you selectively…
I’m not a, like, a doctor. And I haven’t done that. This is financial advice for anybody. This is not medical advice and you should not take mushrooms.
But… Dr. Lutz, I’m right. But…
I’m right. Go ahead, doctor. No, but I’m like, the total idea is to release parts of your neural network and retrain it. And this is how we train quickly.
Now, meaning if I can get a neural network trained on shapes and forms and images and colors, the only thing I haven’t trained it on is my cat and my dog. So all I want to do is I peel back two layers of my network and retrain it. It goes pretty quickly. Without a lot of data.
That’s what we call fine-tuning or retraining. Now, if you do that, the value is actually in the first few layers is common sense or common knowledge. And if you think about common knowledge, how do you protect common knowledge? If I show you yet another cat, you will say, it’s a cat.
Duh. And how much can I charge you for this? Nothing. Right.
Right. So, okay, it’s a cat. Thank you so much, Nick. So now, actually, let’s go back to today’s world.
All the people who built now large language models have managed to create the knowledge of how we humans talk. Okay. And now I give you the network. And now it talks like a human.
It’s like, yeah, I have heard humans talking before, especially you. So it’s not a big deal anymore. Yes. It’s strange to me that we seem to be keeping to move the goalpost.
On what general intelligence says, you know, Gates had told Sam a while back, they said, well, when your model can take the bar exam and pass it over 80% or something like that, then you will have achieved a sufficient bar to build the law. And now, I mean, that was years ago. Okay. Let’s put here a clear notion now.
Yeah. about general intelligence or AGI, artificial general intelligence, is slightly overblown by business thoughts. And we can go into this. But I do and many other in the research field do not yet see AGI being very close.
We have a saying in research where we’re saying, AI is everything that has not been built. Because think about it. Tonight you go to Netflix and you switch it on. And then you click on your little Nick icon.
And suddenly Netflix shows you all the movies you really want to watch. Something embarrassing. Yeah. Some of them might be that your wife wouldn’t have known that you want to watch it.
Do you think that’s general intelligence? No. This is just collaborative filtering. I explained to you people like me have watched this.
So what’s the big deal? It was a big deal as it first came out. Now we’re saying we understand how this works. So does that mean that AI will never really be achievable?
So to say? Well. Because that’s a broad. It’s kind of a vague and broad definition.
It’s a vague and broad definition. And Sam Altman, actually, who uses AGI for marketing sway, actually has moved the goalpost for his AGI. Anyagri as well and like one of the early like month ago so he made an interview and he said, well, AGI will be here soon and it will be way less than we kind of expected. Meaning what he kind of I think he alluded to is look I now know how we can make models to challenge themselves and going into an iterative loop and and reason.
And let’s call this now AGI, which is not necessarily the idea of AGI. AGI really means that it can set a plan independently in all of those potential scenarios, which our world has, our world is complex, and then act on this. See, I’m more curious, and I think society at large is as well, Lutz, is when is it going to have a meaningful impact on our day-to-day life? I mean, like, come on, you can go, if you are on my course, Cornell, you can talk to me all day long.
After this interview, you don’t want to do this anymore. But like, the question is, how much meaningfulness do you get? No, I just repeat what you could have written or read, or like, I write a lot. Yeah, like, if you go to my Forbes page, okay, now you get to summarize.
A Forbes contributor as well. So let’s talk about this a little bit. If distillation is the process of a process of a process of a process of a process of a process of these foundational models, providing information to these agents, how come the agents are seen as the experts? Wouldn’t it be kind of something?
Or what’s the relationship between these two? Actually, before we go down to agents, let’s stop on distillation, right? Distillation is a task we do. And, you know, if you think about how universities or schools evolve, there was a time where we had to memorize the code.
And we had to memorize things. Like a good friend of mine just recently spoke in one of my classes, and he said, I was a better doctor than my colleague if I could memorize seven courses for a disease, while he or she could only memorize five. So there was memorization. Then the internet came, and data was freely available, and it was more trying to summarize the information.
And now we have a tool that does summarization. And it’s a tool that does summarization. so yeah but I realized that it’s not about reasoning it’s not about summarization it’s about empathy yeah like people might have sent we could have told you but the key moment here in one of my research which we did in my startup is we built something to help people to get away from diabetes and one of the people we talked to was she she said like look I’m pre-diabetic I still eat my doughnut in the morning and now leave me alone right so the idea is me explaining on AI explaining this isn’t necessarily helpful so what should a doctor do empathy what should a human do with empathy so we need to focus more on this and this is good for us this will change a lot of our health care and it will change a lot of our human interaction so I give you another example where summarization in itself doesn’t work so as I created the course there is in like I go through many different industries and in every industry I talk about bias and risks and I obviously use chat GPT and I used my own clone to create the course but the answers were so lame they were so plain vanilla they were middle of the road and why are the middle of the road because and I give you an example life is like a box of chocolates like you are such a brainwashing child right white chocolate well because you know we got Forrest Gump in the mix Forrest Gump yes like the blockbuster movie for Americans like ask the same question somewhere in a country that didn’t watch so you’re saying it’s all contextual I mean yes I like look at it like why should life be a box full of chocolates well why can’t it be a box full of surprises yeah or like unexpected ideas or whatever it is or luck or happiness this is a contextual average so if you have a summarization model you get average but we humans never really wanted average so okay well I want to get into that a little bit further but real quickly I think it’s interesting that six months ago the prevailing wisdom lutes was that these wrappers yes that people were creating outside of the these foundational models had like little to no actual value and that’s why I think it’s so interesting that the! so Let’s look at the innovations.
Just real quickly, since we’re at the top, or half hour, you’re listening to Pirate Cat Radio. That’s 92.9 FM in Los Gatos. That’s KPCR. And then KMRT 101.9 FM in Santa Cruz.
Brainwaves on Pirate Cat Radio. So getting back to it, if the idea is that the utility exists in the wrappers, is this kind of where you’re playing now in the e-commerce space? Help us understand. So let’s define wrapper.
At the moment, we have this. Yeah, right. So wrapper could be that. But wrapper in the sense for AI models is you can do amazing things with OpenAI or ChatGPT, and it looks good.
And a wrapper is you just put a nice interface. And now saying, look, now I have a page. Like I put a wrapper around an LLM and called this wrapper. This is now a copy from Lutz.
I informed the LLM about my knowledge. I put a wrapper around it and now my students can interact. Did you use 11 Labs? I actually, I used 11 Labs for my, in order to make this course fun, I actually cloned Kenneth Cookier.
He is one of the economists, like from The Economist in London. He is an amazing guy. Yeah. He has a very good voice.
So I asked him, may I clone you? And he allowed me to clone you. So I cloned his voice with 11 Labs. So good thing you didn’t ask Scarlett Johansson.
Well, I didn’t know anybody, but I started like, what was the number again? Like there’s a red phone. Scarlett, four-way, 555. No.
At the moment it’s occupied with Donald Trump. But once he’s off the phone, Scarlett, please call me for my next gig. No, but okay. So a wrapper is essentially, how do you put the model?
Together. Now let’s, everybody gets so excited about generative AI, but honestly, most of our world is working already on AI. If a generative AI means I create something. If I have a model that can identify a cat, then I turn the model around and say, dear model, imagine a cat.
And this works the same thing with our brain, dear listeners. Let’s do a quick test here. You all know what I mean? You all know what a cat is.
Now close your eyes and imagine a cat. And yes, do you see it? That’s it. Your brain just created a cat and that cat you saw is probably not real.
It might be if you are a cat owner, you might be having something very similar at home, but it is not real what you just imagined. So that is generative AI. Now the fact, and I would claim that most of our business actually needs more a model to identify something, than to imagine something. So if the query we think is X and X historically has a parameter of all of these different elements, then there’s a likelihood that it is the same?
Yes. So I give you both directions. And then let’s do this. Like e-commerce.
Yes, I can do it e-commerce. So normal AI is you get an ad or you get an email and that is targeted. We use AI for this. We’re trying to say, okay.
Okay. What type of shopper is that person? What like with all the traces you left Nick on the internet? What do you really want to buy?
And now I make a prediction and that’s traditional AI. Generative AI is that I’m trying not only to make a prediction of that you want to buy this, but I actually put the thing you want to buy in the picture or the frame or the setting, which is most likely. Now, these are the models. Now, down to agents.
What are agents doing? Agents taking one model at a time, sometimes traditional models, sometimes generative models and do steps as much as an employee in your company would do steps. And that is the agent idea. Now, so for e-commerce, like what I do is I realize that people target the ads, they personalize emails, all in order to reach customers.
What is not personalized is the interface. So my company are to decide, we are actually building this interface in order to make personalized interface for every web shop, for every company. Now, how does this work? Well, you come on and I have now agents.
I have an agent that helps you search. I have an agent that helps, advises you on what to buy. This is creating extra workload for the agents. Yeah.
Yeah. So what do you do? Do you just send the email to the end user or is it just kind of having? No, that is the important part.
What do you think when you talk to your Google home and you set a timer and Google said, and the timer starts now? What do you think? I think that I sent a signal, it received it and then executed a command. Wow, come on.
No, I think I like, he’s just stole three seconds of my life with that and I don’t get back. Like what the heck? Like just, just know. Just do it.
I see. Right. And then take mud off the racetrack. So, um, Reid Hoffman, um, started the chat bot company and we wanted to come back to chat bots and Elon Musk started the chat bot company and, um, they have different personalities and they are very chatty.
And initially you kind of think, Oh wow, this is a fun person to hang out with. After 10 minutes, he was like, dude, can you, just be to the point? Don’t check me up down. Like if I ask, you know, um, what’s the latest policy in Europe on AI and like, man, that’s such a good question.
Let me tell you about it. You have your spot on. No, Tell me. Right.
This is human stuff. I want to be emotional. I want to have empathy. I want to talk to you.
I don’t let the chat bot needs to deliver. It needs to read what the audience wants. Just like we do. So very important.
So very important for my company. And we did this with a lot of user testing became, it has to be in the flow. So perplexity recently, um, came out and said, we will help people to shock. So if you go online and you type something in, then, um, it tries to become the new Google, right?
It tries to put you instead of 20, 20 blue links, it kind of like places products in the stream and perplexity hopes to make money with it. And, um, we are working for it. And we are working for the shops and I realized that’s not the right thing to do. So if you go to our shops, you don’t necessarily realize that there is an AI agent in the background guiding you.
If you search for something, my search is just better. I mean, I, I normally deliver 11% more revenue to my customers only from search, but on top of it, I give those small little nuggets. If you look, for example, you look for a helmet. Like, let’s say you’re a motor, like a, like a biker, you’re a motorbike.
And you look for a helmet for somebody who has glasses. Then I show you the helmets that are for people who have glasses, but I have explained to you as well. What are the features you need to look out for in a helmet? So I give you this small little advice because I know that’s what you want.
I don’t say, oh, poor you, you’re getting old. You’re wearing glasses, right? Well, but still I give you a helmet. No, this is to change.
I just give you. And you’re able to, that seems like a pretty clever approach. And you’re able to obviously measure like a versus B ROI analysis. You know, the, the gentleman that I mentioned before the show, who ran center eyes, they’ve got a, I think a similar philosophy where it’s dynamically showing you products that you’re more likely to, to convert on.
So like, for example, they’ve got Nike as a customer. And I, it’s like, I don’t know. You know, it’s, it’s impressive, right? It’s kind of a similar philosophy, but how can you know what the customer wants before you?
What’s is that like the search for it? They go on, they go on and they type in something and search. You’re assuming like there’s some kind of trail of cookies, right? No, not like I can use those, but more importantly, once I have them, this is, this is very similar to how we humans would sell to somebody.
You go into a shop and you say, I like, you know, like, let’s say RRI. say REI, you say I need a sleeping bag and then I’m looking at you and kind of trying to figure out what kind of backpacker is this guy. Maybe you see me and think he’s homeless. Well, so I would suggest something but then see a reaction.
So that’s what we do. We use traditional AI to actually understand targeting. We use generative AI to communicate small nuggets. You’re not talking biometrics, are you?
No, Generally, no, not at all. Sleeping bags, I show you a couple of sleeping bags but I will say something, oh wow, it depends, Nick, where you’re going and what temperature zone is it. And then you can say, where are you going? You say, I’m going to Bolivia.
I’m saying, oh, Bolivia, if you go to the beach in Bolivia, you need a lightweight sleeping bag. And if you go to the mountains, it’s actually pretty serious business. And I give you those opportunities to scan, like to be able to scan. So what we do is we do a lot of research.
So what we actually see, and yes, we do testing and ID testing. First of all, at the end, we see more revenue because we get people what they want faster. But we see more engagement. Our engagement rates go up through the roof because people are like, oh, wow, this is actually helpful content.
Let me engage. And people are faster because they actually don’t see thousand results. They can narrow it down faster. And returns are less because they show up.
I see. So we shop with more confidence. I see. So if you’re, Lutz, if you’re, let’s say you’re building your castle.
First of all, what’s the name of the company? It’s R2 Decide. It comes from R square, R2, and then decide for the decision. Definitely go check that out.
Lutz is building his castle, R2 Decide. And maybe one of those moats is a unique data set, right? What kind of unique data are you capturing and can then perhaps monetize or continue to build the model? Any city So that’s what we do.
Now, how do we build up a moat? Well, if somebody buys, we probably gave the right information, right? This is we learn, you learn about a cat by seeing a cat. Hey, mama, look, this is a cat.
You learn not to test the stove when it’s hot. And I exactly. And I learn I like you talk about sleeping bags. I show you sleeping bags and I have a discussion about where you’re going.
And I think that’s something that sells better than me telling, giving you sleeping bags and saying I have five different brands, which is the huge difference. Because when you go normally on a website, you have only filters. And this is like for me, this in my journey, this was amazing. In 19.
99, Amazon 1999 and Amazon 2025 looks exactly the same. It’s a little bit more colors and bigger pictures, because we have more bandwidth. But the way we shop hasn’t changed. It’s filters and search.
So maybe that’s classic innovators dilemma similar to Google struggling with. So we’re going to change that. But we’re going to change this to generative AI and agents. So our listeners.
Love good stories. Also, you’re a world class storyteller. So thank you for the energy. What was the aha moment that you had when you realized that you needed to implement this?
This is a personal frustration of yours. Tell us about that. Tell us about the breakthrough. Hey, that’s cool.
The aha moment. Yes. I probably should have now the story that from a child on, I really hated web pages. This is my I’m going to revolutionize the world.
I have the НА moment is actually different. The moment the moment is look I worked for Google Health. I then went to a company to predict preventable health diseases. And I had now a prediction to prevent diseases.
I and I build up a system that we had doctors actually delivering better care. All this worked if the trend for the humans. Because the humans. Darn humans.
Yeah, darn humans. And that’s the reason why I’m so much like cautioning regulation, because we shouldn’t regulate the AI. We should regulate more humans, which we call laws. So I realized that humans need convincing.
So my aim was I built this nurse in the pocket, as I said, right? Similar as in my course, I built Lutz in a pocket. I built a nurse in the pocket. And I realized it’s probably not what people necessarily need.
So now I had this beautiful tool. I was like, where do people need information? Well, official. They need the right information.
And they don’t need, and this is the difference to why I don’t believe perplexity is going to do a big hit, because they supply too much information. If I want to buy something, I don’t want to read a whole page. I want to have the bits of pieces of information. You did.
That’s an interesting take. However, I’m going to push back on something you mentioned earlier, which is research for its own sake. Okay, let me back up a little bit. Ilya Soskovor just raised a bunch of money.
Yeah. And his whole focus is safe superintelligence. I think the company is even called SSI. So if they’re able to raise.
Yeah. I mean, I think that’s a good thing. Yeah. I think they’re able to raise a hellacious amount of money just based on him as a technologist and say, you know what, we’re not going to get distracted with business models.
I think that there’s maybe something to that because there can be more of like a foundational leap in the innovation. You’re not necessarily worried about the specific application. You’re just going to build something that’s fundamentally stronger, better, faster, smarter. And then kind of assume that the genesis is going to be a little bit less.
And then you’re going to assume that the chips will fall into play from business model. Is this kind of a weird way to build a company? If you’re not Ilya Soskovor? No.
I mean, we had this before. You remember the tulips in the Netherlands? Like everybody invested in tulips in the Netherlands. It was a huge thing until they realized it’s tulips.
Yeah. So I’m losing my audience here. No, We got a special guest. Brian, come on in.
I thought Donald Trump was coming. Like you said, like, hey. Hey. Good seeing you.
Hey, man. Any time after the party? Yeah. No, it’s not a weird way of building it.
Yeah. It is what he says is I built a foundational model and when everybody uses foundational models, I hopefully created enough moat. Now, the whole disaster, which we saw was DeepMind was that having a foundational model in itself is not necessary. Okay.
Okay. necessarily the protection for the market. Yeah. Right.
So I don’t think that we can build super intelligence and protect it. And I think we don’t change business dynamics overall. Or put it differently. If you worry, if you are today an engineer, where should I put this in?
Three or four actually. Or three. Put it in three. I put it in three.
I’m connected here. So. Can you hear us? Yes.
No. No. I put you in the one dome. How about now?
Now I put it in four. Okay. Can you? Oh, yeah.
Okay. Yeah. Hot mic. Yeah.
Hot mic. Nice. So. Business.
Yeah. Come on. Give me a high five. I’m in the child seat right here.
I feel like I’m in the bucket. This is good stuff. You were saying. Yes.
So. Yeah. Okay so consistent strides over many years. It’s because they’re focused on a core concept, not a technology, not a product.
AT&T is a perfect example of that. So maybe you’re saying Ilya is going to be focused on this core concept of safe superintelligence, and that’s why he’s going to be… Are you bullish or are you bearish on what Ilya is doing? I don’t know the details.
As I said, I personally don’t think that we are close to AGI, and if we’re not close to AGI, then we won’t be close to superintelligence, which would be the next level. Let’s actually define those. AGI means it is like us humans. They are so general that they can do everything a human can do.
Superintelligence means it’s so much smarter that we humans have the intelligence of an ant. So wonderful, Brian, in here. Brian, so you get to talk to all sorts of brilliant, founders and technologists, bright sparks. So what do you think?
Is AGI around the corner? Are we maybe one or two big leaps away? What are your thoughts? Well, this is so far out of my wheelhouse, but thanks for letting me crash your party.
I appreciate that. Yeah, for sure. You guys are in the thick of it, but I don’t know. I’m still a student of the game, so I really can’t make a call right off the top of my head and sound intelligent.
I appreciate that. But I have seen… I appreciate that. I mean, I have seen people who have a lot of good knowledge can go around and raise insane amount of funding because they make claims what they’re going to develop.
So their cost of capital is like zero. For them. Like, not for me, but for them, it becomes way easier to raise. Well, I mean, you’re a Fulbright scholar and ex-Google.
I mean, you obviously are very… I closed my… I closed my angel around. This is all working.
We need an applause button. Yes! Yes! But the main point here is whatever you built, you need, at some point in time, you need to return capital to your shareholders.
Yeah. So Amazon had a very big vision and said, in order to build this, we are going to lose a lot of money. And then we had the bubble where everybody says, oh, losing money. We can do this.
Let me get $1 of venture funding and sell it for 50 cents. Wow! And people take my dollar for 50 cents, right? It didn’t play quite out like this.
Do you think it’s kind of like that’s more of a Ponzi scheme? Like there needs to be somebody else to buy? You have to have fundamentals. You have to have people to give you something of value.
This is nothing to do with AI. That’s the fundamental law which we’re seeing in business. Yeah. Science versus the business model.
But what do you do if you’re Sam Altman right now? Sam Altman is doing everything according to the books, and he’s doing it well. And it has nothing to do with AI. So first, let me walk you through all the things he has done.
I mean, it’s stunning. We should learn from him. He is doing such a good job. Okay.
This is a hot take. You hear, like, first of all, he said AGI is so terrible. First of all, he says AGI is around the corner. And I, Sam, I know how to build it.
So pay me $200 a month for my system, because if AGI comes, you’re the first to get it. So please come to your customer. It’s a marketing take. Then secondly, he says, oh, by the way, this is so dangerous that I need government protection that only a few should really be allowed to offer.
Okay. So I’m going to do this. I’m going to do this. I’m going to offer the service.
Yeah. Again, market protection. So, like, please do not allow anybody to develop the same service and therefore protecting the market. Now, people started to build and he’s like, okay, well, then I actually go and make a clear statement that I have more money than anybody else.
So he goes to the White House and debt finance says Stargate. I have $500 billion. Do not try to compete with me. So he’s doing everything.
He’s doing everything. in a very good business fashion in order to protect the company he’s building up and i have a lot of respect for that well don’t you think that he may find himself in the wrong side of history as far as like the open source versus closed source like i get the policy or the privacy element but like everybody cannot oh half the world cannot be or will be beaten by like a small few that is playing inside a wall guard if you if you look at alex uh alex carp right with palantir palantir is what is it it’s a it’s a dashboarding application it’s something like tableau but only slicker uh so it’s it like initially well you’re discounting the utility of their models which are you know these are peter keel’s company well even though like he’s not messing around but it’s not the utility of the company like the utility of the model it is used case of the model what’s the difference the use case of the model is palantir works a lot for state entities police and other areas like yes and they have protected data which only palantir works on and they put all the models the mathematical models to work there so now we are coming down to it’s not the math it’s the application of the model which becomes important so much about the wrapper question early on you take a model model as a model what’s the application of the model which becomes important what you need to do is put it together in a useful fashion and uh palantir is doing this and oh may i don’t think sam is on the wrong side of history because go and ask any chat bot about how to build a giga factory he will not get a useful answer because it’s complicated now go to alan musk’s team and say guys you have done this help me to build a giga factory you will get a good answer can an llm summarize this knowledge keep that information in the company despite maybe people leaving that’s a good thing i’ll give you one more pointer here why is california such an innovative place because you can go from company to company we actually put barriers in so that you can protect like make um non-compete we don’t have allowing non-compete in uh in california and that has been linked kind of like the wild card you know like the wild west yes and that has been linked to innovation because you drop your pen here you go over there and you build the same and from like same idea up there yeah that has driven apparently like so so research says innovation now with llms you can actually still do this go across companies but still keep the knowledge within the company so alan musk might have somebody who building a so so research says innovation so i’ll give you an example so i’ll give you an example so for the next company, but Tesla will still be able to do it. I could see that. And when you mentioned innovation, to me that’s courage plus curiosity.
Brian, you must get to see all sorts of really cool companies that are innovating the space, awesome founders. Any founders you want to give a shout out to here? Well, I mean, we’re working with Alex Holt. You know, he’s familiar to Las Gadas.
Want to give a shout out to the hometown? Yeah, for sure, for sure. He’s innovating in a space that’s very much behind the curve in the hospitality industry, building some equipment to more efficiently run his restaurants. Oh, okay.
Which is traditionally kind of tricky margins. So it’s a good problem to solve. It is, with such small margins, it’s definitely going to be a fast growing company. But I feel like I cannonballed you guys’ show here.
I’m so glad you did. It’s always, anytime, anytime. Although it’s been a pretty fascinating conversation. You know, unfortunately, we got like four minutes left in this bad boy, but we should definitely do like a round two here.
We also got like the gray shirt memo. That’s exactly right. Yeah, that’s everybody’s gray shirt. Yeah.
So, you know, just to catch up a little bit. Lutz was on the show with GABA, the German American Business Association. And that was a really cool one. And it’s so crazy to me to think how much has changed, even in the past couple of months, that we had the Senate bill, not 10, but now it’s 1041.
Maybe it was, right? Well, essentially what has changed is, I think in very broad strokes, we are realizing more and more that application of AI is the thing which we need to focus on in terms of allowing companies to innovate, building applications and networks in Europe as well as here. Because I like, there was a kind of a wake up call, for whatever reason, like people in the industry weren’t so surprised about DeepSeq. And we could actually, like the surprising part, well, all the innovation they did, but they did it a year ago.
And they had to do it because the administration stopped them from getting access to high power chips. But that’s just like, well, there’s, you know, folks are saying that they might have gotten them from Singapore anyways. No, but like, if you look at all the design choices they did, all the design choices point towards, you know, And I think that’s a big part of the reason why we’re doing this. And I think that’s a big part of the reason why we’re doing this.
is that they did it. All the design choices point towards, they are working with lower power chips. And when people saying, yeah, they were cheating, they were using maybe OpenAI to train. So did Alpaca on Lama and all of this.
Yeah, they’re all using each other. And by the way, Sam Altman used us. And probably tomorrow. Probably you too.
Yeah, yeah, probably. Like you put this live and tomorrow, some, bots crawled to show and they learned what I said. I hope they do. And then the bot will talk about me.
So you got copied, man. Yeah. And maybe it’ll get back to AI at some point here. So one thing in the last couple of minutes that we’ve got, Lutz, I really want you to pull apart this concept of the value being at the application layer.
The modalities are changing. How we’re interacting with these, with these tools and technologies are changing. So what is, what does the future of that look like for us, Lutz? The future of it will look like that AI or this large language model.
Large language model is nothing than an interface. It’s, it’s now an easier way for us to interact with computers. It’s an easier way for a computer to talk to us. And that will have consequences.
It will have consequences in e-commerce. That’s the reason why I built this e-commerce company where you have way easier way to talk to us. It’s a way easier way to interact with the web shop. It will have consequences for healthcare.
Like think about the person who sits in the corner from a doctor who is a scribe. Now we don’t need the scribe anymore. A person like you just talk and tell and the computer writes to it. It will have impacts in how we navigate.
It will have impact in every single industry and every single step. So the future is now actually not the big ideas of AGI and CIGAR. Everything is changing. No, the future is we have a nice tool and let’s put this tool to work.
Yeah, beautiful. Beautifully put. Lutz, who can our zombie community connect you with? What is the title of the individual that ends up adopting your technology?
So if you have a web shop or if you know somebody who has a web shop, I would love to show you this. I would love to show them that we can make more money for them because we bring that web shop into the reality of the new AI. Your web shop will be dynamic, personalized, targeted on the fly. We give you better search.
We give you sales guidance. We give you an ability to make your customers happier than before. And even if you can convert an extra 2%, let alone like almost 15 and 11, you’re rocking and rolling. It’s beautiful stuff.
Well, we are at the top of the hour. So I can’t believe how quickly this time went by. You’re listening to Silicon Zombie’s Brainwaves on Pirate Cat Radio. That’s 92.9 FM in Los Gatos, 101.9 FM in Santa Cruz, KMRT.
And if you’re feeling generous, head to donate.kpcr.org and chip in for community radio. So awesome. Thank you. Great to see you, Brian.
Lutz, thank you so much. I appreciate it. All right. Cheers, everybody.
Bye.