Episode #506: How AI Turns Podcasts into Knowledge Engines

Speaker 1

Welcome to the Crazy Wisdom Podcast. This podcast is for you. If you have an insane drive to find the truth of things, it's not the good answers that we seek, but the good questions. I interview a range of different guests from many different fields, all with the intention to uncover the simple truths that are hidden in plain sight. Most people don't want to go there.

I go there, my guests go there, and you benefit. Please let me know if you enjoy these episodes and as always, subscribe on itunes, Spotify or wherever you listen to the podcasts. Welcome to the Crazy Wisdom Podcast. I've got Kevin Smith here and he is the co founder of Snipt. Welcome to the show.

Speaker 2

Hi Stuart. Thanks for having me.

Speaker 1

Yeah, we're having some great discussions about Protestants and Switzerland and all this different stuff, but we don't need to go there. Snipt seems really interesting. It seems like you guys have a good handle on the intersection between AI and podcasting. Can you what is your current framework for that?

Speaker 2

So in general, the way that we think about Snipt is it's the AI powered podcast player for anyone who listens to podcasts to follow their curiosity to learn to become more knowledgeable. So one of the actually the feature that we're most known for is also where our name comes from. It's called snipping. What that allows you to do is to save any insight that you hear in a podcast simply by tapping your headphones.

So our 8i then saves the moment that you that you just heard and summarizes the insight for you such that you can easily go back to it, share it with someone, or sync it to your notes app. But yeah, just on a more higher level we really. It's exactly what you were saying that we believe that we basically have two core beliefs at Snipt. One is that podcasts is one of the largest knowledge sources in the world and heavily underutilized in that regard.

And two is that AI is actually now changing the way that we can interact with this knowledge library and basically enhance our experience to get so much more out of it.

Speaker 1

Yeah, that's really cool. So I opened it up now and I'm going to start listening. I should have done this beforehand, but I'm going to start listening to podcasts and so can. I don't have the Bluetooth headphones that come with Apple. I have my own over ear ones. Can I. There's a way I can do it in the ui, right That I can clip something.

Speaker 2

You can even do it with your headphones. So it doesn't have to be AirPods. It works with any pair of headphones that have the ability to do a skip back action. So usually for over ear headphones, that's a triple click. Usually there's like one action button and you triple click it and that skips back. And that's basically where we've hooked in our Create snip button.

You can even use it in if you often drive, if you often listen to podcasts while driving, you can use the button on your steering wheel all. Now you can, you can of course also just open up the app and click the big button.

Speaker 1

Okay, great. So, and it feels like there's going to be a very quick or a voice agent appearing inside so that you could speak with it. Right? This is, this is where I was thinking, so let me give you my whole, my whole understanding of the nature of books and the nature of works. So I'm a knowledge management professional. I've, I've was the director of knowledge management at an up and coming AI company and I learned the huge problems of knowledge management inside of an organization.

It's so absurd, it's such an absurd discipline because it deals with knowledge and knowledge is not material. It's not something you can hold on to. You can't say like, here is the knowledge that I have. It's all, you know, it's like in the neurons and, but not just in the neurons, it's also in the world. All tangents that are unrelated to what I was just talking about, which is that, so one of the things we learned in knowledge management and ontologies, which is a strange other rabbit hole we can go on, is that each author has a work, so an author has a work.

And that that work is like a book. But it's not just a book. It's also the podcast that's related to that work. It's also the, the metadata. All, all of this stuff goes into a work, like, you know, magazine article, all these different things. So that's their work. And then the book is downstream of that. As soon as AI came around and chatbots came around, I realized, okay, now I can have a conversation with a book.

And that's really interesting because you're having a conversation with a book when you're reading it, but that conversation isn't explicit. It's dead. The book is not going to answer you. You can pretend it's going to answer you, which may get you in trouble in certain countries, but you know, like, that's like, it's just not going to answer you.

Now, with LLMs, you can have not only a conversation with that specific book, but you can have a conversation with the entire work. And you can even have a conversation with a simulated double. Simulated twin, a digital twin. And so all these things are coming. And so when I said that about like, oh, there's going to be audio here where you can just talk with the podcast, have you guys thought about that or are you thinking about it or what's your theory there?

Speaker 2

Yeah, so some of the things that you now mentioned is actually already possible today in the app and others are on our roadmap. So what you can already do today is actually with any podcast, you can chat with the episode, so you can ask your questions in text and you will get it back in text with any, with any podcast. The audio version is of course, something that's very interesting for us, given that a lot of our users love engaging with knowledge on the go.

We've done some internal experiments. For various reasons, we thought that it wasn't there yet, but this is something that's on our roadmap and like to go in deeper. Um, and another one that you mentioned that's on our roadmap is what. What I find very interesting is not just having, not just being able to converse with a single episode, but actually the entire work of.

Of an author, as you now described it. Right. So some, you know, you take someone. Let's take some of them, someone very controversial like Elon Musk. You know, he's been going on podcasts for. For men.

Speaker 1

Right.

Speaker 2

So you could even communicate and research. Okay, what has he been saying about certain topics three years ago, five years ago compared to today? How have these thoughts and opinions changed in general? This actually does bring me to, again, my belief. I actually see in the beginning, I mentioned podcasts is one of the largest knowledge libraries, but another way of, let's say, formulating it is I see podcasts more and more becoming the knowledge capsule of humanity in a similar way as blog posts.

I think blog posts was sort of the. For a long time believe like this is where a lot of people will share knowledge in a very distributed manner. But if you look at the last couple of years, what's happening? I know very, very, very few people who actually still write a blog post. But I know a lot of people who just jump on a podcast and talk about what they'd been doing.

You know, a product designer has been designing products for 10 years. He's not going to take the time to sit down and write a really well written blog post, but he's going on podcasts and talking about, you know, hey, I've learned about this, I've learned about that. And you know, now, yeah, that makes it very interesting to, you know, create an interface where you can interact with this and search through it.

And then of course. And that, that's something that I always want to. Want to stress. Like in the end, it's again about listening to, to the original content because that's what, what people engage with, what people love doing. Right?

Speaker 1

Yeah. There's so many threads we can go down on this. The, the, the. No, everyone likes to post things written by AI. No one likes to read them. I like reading AI only for technical stuff. Anything where there's a bias, where there's a political bias. I love reading AI because although it has a political bias trained into, it's very easy to jailbreak around it and call it out on its, on its bullshit.

So the, it's like having a conversation

Speaker 2

with

Speaker 1

an episode of Elon Musk, but having the whole background. There's a bunch of technical problems I want to ask you about related to vectorization. Summarization. Are you guys doing a lot of. I imagine that you guys have to do a lot of this stuff because the like, I've done a lot of this stuff. I've done. I've now vibe engineered vectorized databases.

I've done a lot of transcript. Are we there yet where we can do streaming audio, multimodality? I imagine it's expensive if it is. And are you developing this or is your co founder developing it? Who's developing it? If you feel like sharing.

Speaker 2

Yeah, a lot of questions. Let me dissect that. So. Yes, of course. So in general, I mean, there's a lot of AI engineering going on in the background, but for all of our features, you know, like there are many AI features in the app that we now even haven't even spoken about, interestingly enough, just because you explicitly mentioned, I guess, vector databases and embeddings as of right now for the features that are live in the app.

So, you know, now we spoke about a couple of things that are on the roadmap, but the stuff that's live in the app, we're actually not doing as much embedding and vector search as you would imagine, mainly. So, for example, if you're now chatting with the podcast episode, you actually have the entire episode in the context. So we're not splitting it up, not chunking.

And the reason why we're doing that is A, because it's now possible, you know, like two, three years ago, that just wasn't possible yet. But B, the increase in quality is quite remarkable because the, you know, you really understand. The AI really understands the entire episode. What we spoke about. And not randomly the vector search, returning two chunks, but missing out for other chunks.

That was super important to understand this. Yeah, so that's the situation for us right now. But for some of the features that we now spoke about, like engaging with the entire. Everything that Elon Musk has ever said on all podcasts. Of course you run into these limits again with, with the LLM context windows and of course also, you know, costs and yeah, there will do much more in that direction.

Speaker 1

Yeah, it's massive. It's just like context windows, they're giant now, but they're not giant enough to get all of Elon Musk's entire work into them because it's, it really comes to. Makes me realize just how much content humans create. And like, like, like. And this is not new. You know, we were talking about the 1500s before we started recording.

And in the 1500. Well, let's go to Isaac Newton. Isaac Newton on. In his side quest of astrology. Actually, no, it was his main quest, but he. His main quest of not astrology, but alchemy. Isaac Newton had his main quest where he basically wrote 1 million notes and he never shared them with anyone. So. So that, but that was his main quest.

All those other things were side quests on this, trying to turn lead into gold. And so he created 1 million notes, like, and he's not publishing or anything like that. And this was before, you know, it got easier to write. He was writing that with like, probably like a quill or something like that. And, and you know, now we're. I can just type out a bunch of things on Twitter.

So I create a huge amount of content on Twitter. I do a huge amount of podcasts. It's funny, I actually never. I still can't get over the hump of doing a medium form a book. I want to do those things and I don't want AI to write those things. I actually do enjoy writing, but. So we just create so much content and we can't fit it all into the LLMs.

And the LLMs are helping us to create even more content. The LLMs are even more verbose. Like, it's so fun. Do you know why they're so verbose? Have you figured out like, what about the. The whole thing? Because it feels like at a cost saving perspective you'd want them to be less verbose. But maybe it's that human principle of just like if you.

It's it, it. You know, I would have, I would have made this shorter, but I didn't have the time. That famous quip basically like, do you.

Speaker 2

Do you know, I think at the end of the day the, the LLMs. The optimization function behind the LLMs today is not saving costs. So at least with the output, so, you know, you can really see that they are optimized for consuming the output by reading. And that is something that you really notice once you start playing around with generating AI generated audio, which of course, as you can imagine we've done, you really notice like, hey, this is not how someone speaks, this is how someone writes.

Speaker 1

Yeah, exactly that. Huge problem. Yeah.

Speaker 2

As if we were having this conversation and actually wrote down word for word what I was going to say and then I'm going to read it out. It sounds completely weird and it actually, and that's the more interesting part for me, it's not just that it sounds weird and that it sounds unnatural, but it is much more difficult to follow.

Speaker 1

Yes. There's no linear. Wait isn't. Because it's not linear. Is the conversation nonlinear? Is what we're doing right now not

Speaker 2

so? There are a couple of things that I've noticed with AI generated audio. So one thing is that sometimes it's too dense. There's like one sentence where if you read it, it's the worst because you can take 10 seconds to read it again and read it again. With audio it just continues. And especially if you have the AI voice not understanding that, oh, this was the most important sentence.

Maybe you should say very sl slowly and give the other person a break. Then it's very difficult to follow. And the other part is like sometimes you notice that it's very difficult to get some of the standard LLMs to not think in terms of a header or in terms of a bullet point list, at least in the output. It's fine to think like that. But then I won't just say a title.

Speaker 1

How did that happen? How did that get in the training data? How did these. How did these LLMs all. Because it's all con. It's all, it's all trained on written content. So it's all trained on listicles and bullet pointed, all of those things. Do you think that's. That's right. And you can disagree with me if you're.

Speaker 2

I would say, as far as I know it's. I would rather say it's fine tuned at the very end for primarily that use case. You know like the, the biggest way that people are interacting with it is you know, in chatgpt, you know Claude, whatever your favorite is, it is the text base is first even though, you know more and more of the providers are now doing voice.

But the biggest foundation and also in the, in the training data is still, it's still text.

Speaker 1

Yeah, it's super interesting. So I'm getting the sense that you're, you're coding a lot of this stuff. You're, you're, you're building this or are you, are you?

Speaker 2

Yeah, yeah, yeah. So the team, well I mean the team with four people, we're all coded. My, yeah my background is in AI so I like my story is originally I studied mathematics and economics with all geared towards becoming a quant in.

Speaker 1

Oh, interesting.

Speaker 2

So all of you know doing like the mathematical models for derivatives pricing and quantitative trading, all of that stuff. And the interesting thing that happened for me was that I always liked the mathematics of it, but I just never, I, I was just never so passionate about the field that not a single time in my free time I actually read like an academic paper or something about it and everything changed when one of my best friends came to me.

So we, we were like at the very end of our studies and he says like, hey, I, I, I just took this, like I started taking this course. It's a bit weird but I think you will absolutely love it. It's called machine learning, Krista. So like what, like machine learning? What, what the heck is that? So he sent me over the PDF slides and within a weekend I went through it and I just knew like what the heck, like this, what year was this? Exactly.

So that was I think 2013.

Speaker 1

Okay, wow. Yeah, 14 maybe.

Speaker 2

14. Yeah, I think it was 2014. Yeah. And I just knew like oh my God. Like this is all of the parts that I loved about this mathematical modeling but without the finance world.

Speaker 1

Interesting because the finance world uses the machine learning. I mean there's a whole bunch of machine learning algorithms. I think if my understanding is correct, it definitely sounds like you have more understanding here. There's a lot of financial models using machine learning. Some of those were more successful. One quant told me about Microsoft and some random library which is in my notes somewhere, which is a whole other topic.

But then there the real utility of machine learning algorithms started to come with the Facebook social media feed of how do we actually deliver a piece of content to this person. Then after that then it was the Transformers. I wonder. Yeah. So then the transformer model attention is all you need. But there are financial modeling from machine learning.

Right.

Speaker 2

So at the time, so I mean it's very important to like separate what they're doing today and what was happening now.

Speaker 1

All of those gone now basically 10 years ago.

Speaker 2

So at the time it was mainly. So there were a lot of mathematics involved. Right. And a lot of optimization. That was basically what we were doing in learning. But it was actually not classical. Machine learning had very, very little to do with. Was actually more that you had these, I would call them economic theories or like there were a couple of theories that were sort of accepted and I think one of them even won the Nobel Prize or like something related to that.

And based on these models you created mathematical functions that described how the stock market would behave or for certain financial product would behave. And after that most of your work was actually more in fine tuning the parameters because these models had like two, three, four parameters. And now you got your data and then you, you created statistical estimators to find what the right parameter is for this.

But that is like very far away from machine learning where basically the. And that was really what also freed me up like or like was felt so freeing to me is it's actually, it does not start with a, with a very, very restrictive prior as you would say a very restrictive theory. It says oh no, it could be anything. And then it just goes off and learns the entire pattern just purely based on the data.

Speaker 1

Oh cool.

Speaker 2

But because you need a lot of data which, that's the, that's the ending thing in the finance industry, of course again you need to, need to check a bit like what, what part of the finance industry you're talking about. But if you, if you're interested in something like, you know, how the stock market changes like on a day per day basis, that, that is not a lot of data.

You know, you get for per, per Stock you get one data point per day. That's 365 data points in a year. Like that's nothing. Right? Prepare that to, you know, it's very difficult to learn to train an entire non biased machine learning model just based on these couple of data points. Right. So yeah, that's also why it's difficult to sometimes apply these models.

Speaker 1

So interesting.

Speaker 2

At least back then with the model, with the models from back then.

Speaker 1

Okay. There's one question of like how does it work differently now whether you're paying attention to that still. But the kind of thing that jumped out to me, the nerd sniping. Sniping. You should use the nerd sniping for your marketing and snipped. So you said early machine learning models for finance were based on economic theories. That seems like so interesting that, that there are a bunch of economic theories that were being tested in real time but with not enough data.

Is that accurate? Is that an accurate representation of what you were saying with machine learning models in 2013?

Speaker 2

So depends on how technical you now want to get.

Speaker 1

Yeah, I would love to go more technical.

Speaker 2

I'm.

Speaker 1

I'm not sure how much, how much of my listeners keep on going. I just did this. I published a crazy bitcoin episode two weeks ago about, about really technical details. I have no idea whether anybody. How many listeners I lost. But yes, please, let's go as technical as possible if you're interested.

Speaker 2

Yeah, maybe, maybe to just add one comment there. So I wouldn't call like the thing that most of the people in the finance industry like the. Most of the quants as they were called were doing. You wouldn't call it machine learning. It was classical statistical modeling. Okay, so that's from the terms. Yeah, yeah. Like around that time it then did start to pop up a bit more.

So basically after I fell in love with it, that was actually my first try. Like, or where, where I tried to apply it because I was already, even though I was studying, I was already working at like the largest bank here in, here in Switzerland. And just everywhere I tried to use machine learning. And luckily at the time my boss was actually quite open to that.

Speaker 1

Wow. Called.

Speaker 2

I realized that, you know, I felt a bit like, what's the classical saying? You know, I was a hammer. So everything, every problem was a nail. And at some point I realized, look, if I just want to do this every day, all day long, then I need something else. So I then actually took the decision and said, okay, I'm going to leave. I quit my job and I joined a small tech startup here in Zurich where I could build up the AI team from scratch and can do this 24 7.

So that was the way to.

Speaker 1

Then you then entered machine learning and not just at a university level, but in a company.

Speaker 2

Yeah, so it was a small early tech startup here in Zurich and I got the opportunity to build up the AI team from scratch. So there was no one else there before, before me. So it was a completely green open field. It was still tendencia. It was still related to the, to the banking industry, but we were. So we were building software that we were Selling to banks.

But as you can imagine it was a small startup, so a completely different culture. And in particular what we started doing were there were two things. One was understanding transactional data. So whenever someone buys a, buy something in a supermarket to understand what that transaction means, who's the merchant, what merchant is that? Is it a supermarket? Like we, you know, we would categorize it and were there any recurring transactions identifying certain patterns.

So basically making this data set of transactions finally valuable to banks and understandable. And the other part was we built more classical machine learning models to identify customers that were likely to be interested in a new financial product. So basically helping the salespeople identify leads, if you will.

Speaker 1

Yeah, okay, that's really interesting today. And that was early machine learning still within finance. But you actually got to understand how these things can actually work for products. The various products that were in Switzerland and probably in the US probably in London, probably in New York and that's probably what a lot of defined a lot of the activity within the finance from 2013 to 2025.

There must be a, a revolution going on right now.

Speaker 2

And.

Speaker 1

Well actually, yeah. Are LLMs useful for finance or are they all doing their other weird esoteric.

Speaker 2

Okay, I mean I'm completely outside of that industry now, so I'm really the wrong person to ask but like 100%. I mean now LLMs are also a completely new generation. Right. Where you can come up with these very different ways of interacting with them. Right. You don't need to now train something from scratch to have anything useful. You can just have a reasoning model, basically make decisions for you or you know, gather data together, reason, reason about it or be very good in dissecting real time news in a fast way and making then the decision based on that.

So, so that's what a lot of things.

Speaker 1

Yeah, so that's what the transformer opened up was this ability to not have to train a model from scratch every single time. Is that right?

Speaker 2

I wouldn't say it like that. What I would say is that. So the transformer architecture in a way is, is, it's been much more successful than the other architectures. But in and of itself it's also just, you know, a machine learning model that learns from data and you need to. The, the big change was that, and I would actually say that you were able to pre train these models and that the community learned that you could put so much knowledge into these models and if you just put more and more data into it, they learn more and more and more.

And all of these emerging capabilities emerge that it could actually start reasoning. You could actually start chatting with it. Of course, I'm simplifying this. There was a lot of engineering work that went into that, but now it's almost like talking to a person. Right. The same way how. You know, we just hopped on this call and I didn't start with.

We just start with zero. Right. There's already a lot of knowledge that you and I have, so we can just start talking on a much higher level.

Speaker 1

Brilliant. And so that relates to the bitter lesson that. The bitter lesson was that we just throw more compute at the pre training.

Speaker 2

The bitter lesson, the way that I understand it is actually that the methods that rely predominantly or exclusively on compute, those will in the long run, sooner or later outperform anything that is reliant on human design.

Speaker 1

Interesting.

Speaker 2

Wow. So if you will, the transformer, I mean this is something that. I don't know how much you've been following it, but the inventor of reinforcement learning who has coined this term, he was recently on. Yeah.

Speaker 1

Dwarkesh Patel's podcast.

Speaker 2

Podcast and corrected some of his. Some of this theory or statements. The transformer revolution, the LLMs, yes, they were much more bitter lesson pilled compared to the more classical AI community where everything was designed where. So there was a time before all of this, let's say machine learning phase, where the predominant theory was that we as humans, we just need to record all of our knowledge and design systems that are smart.

So basically you could think of it as one really, really big if then else chain. So if Stuart says this, then say that this is only the system and that gets you somewhere in the very beginning. That actually gets you further than a machine learning model if you only have very little data to train on and very little computer. But over time we got more and more data available and more and more compute and then it actually these methods that just learned on their own, that basically leveraged compute, they outperformed systems that were purely designed by humans as.

And now the big question is, or like my understanding of what Richard Sutton's point was when he was on this podcast and the Dwarkesh podcast is that currently this phase we are still reliant on human input because we are training all of these models with data that is created by humans. You know, there are. There are big comp. We trained it with the entire Internet and now a lot of.

There are these companies like scale AI that are. That are millions in revenue to create new data sets, very specialized data sets for various specific tasks or Very specific areas. It's up to a point where these companies they only work with people who have a degree or like a PhD in certain areas and they're creating more and more training data but that will run out.

Or you could argue that we've already almost done out. So he argues this is again not 100% lesson built.

Speaker 1

Yeah, I like it. So this is, this is where my knowledge management experience comes from because I worked for a company. I was the director of knowledge management at an AI company doing RLHF reinforcement learning through human feedback for OpenAI and for all the other big tech companies. Have you heard of Surge AI?

Speaker 2

Yes. Correct me if I'm wrong. They are the company that got very famous once Scale AI got so exactly.

Speaker 1

Are not acquired by semi acquired. What is it? Yeah we won't go into that tangent but Serge AI. Yes exactly like these. These guys are. He was a. The founder was an early Twitter and Facebook employee. He's only done two podcasts. I keep on asking him on LinkedIn if he wants to do a podcast but he really interesting because I'd never heard of them.

I I thought I did all the research to try to find all the other guys found scale and. And he called scale a body shop and a body shop meaning they just go and find people. And I know that to be true knowing many people who worked at Scale AI and just all the ways that both humans game the system and the AI's game the System. That's the most interesting part is that the AI is also gaming the system.

So Sergey, I've. I believe that they figured out a way. It's hard to know it's all proprietary but they're doing it seems like they're actually doing good work in a. In a world that seems very ponzied. So yes, I agree that the. The data is the limiting factor and not only data, the high quality reinforcement learning. This gets into RL environments which are really interesting.

If you have anything to share on your learnings and RL environments I'd love to do that. One of our mutual connection, Christian Ulstrup is basically he's the guy I go to to try to figure out what's next and he talks about RL environments, prediction markets, that prediction markets might actually have a really important avenue for RL environments.

Yeah. If you want to talk about any of that we can go into it. Otherwise there's a whole bunch of more questions on the platform as well.

Speaker 2

No, I think rl. So reinforcement learning is an area where I haven't done any practical work, but I completely agree that it seems more and more obvious that it will be a very, very important and large part of the solution of the next generation of AI.

Speaker 1

Interesting. Yes, because we got the robots coming. The robots coming, and they need their data as well. That's a whole other tangent. Let's. I want to more find out more about Snipt.

Speaker 2

So

Speaker 1

Snipt. Okay, so you take. I'm going to start using it as I record more podcasts. One of the challenges I have with using Snipped rather than Spotify to listen to podcasts is that Spotify also has audiobooks, and audiobooks are really important. Maybe that's a maybe, if you guys aren't thinking about that already, to start thinking about audiobooks and how to include them.

They have this really sneaky one, which I'm pretty sure they have machine learning models which are trying to make me pay more money to them, which is that. So I, I have a certain amount of audio book hours that I can listen to on my premium subscription. And then every once in a while they say, oh, you've run out. But they don't tell me how many hours I've actually had.

And I assume that if I just click on say no, that they'll never actually charge me and I'll just train their model to, to just to, to target me as a person who's just not, like, not going to pay them more money. And so there's some interesting business model that they have with those guys, which is interesting, but.

Speaker 2

Well, yeah.

Speaker 1

So all the stuff we've talked about now, what are the podcasts that you're listening on? These subjects that you already know are. Is on Snipd and what like either on the technical stuff we've been talking about or any of the other interests that you're listening to. What are the best podcasts you're listening to right now?

Speaker 2

Yeah. So first of all, important to note is we have, we have all of the podcasts that are also available on, you know, like the Apple podcast app or Spotify, or at least 99.9% of them are also available on Snipped. So with respect to podcasts that I love listening to. So I actually listen to a very broad range of podcasts I'm subscribed to.

I don't even know, probably 150 podcasts.

Speaker 1

Can I reframe my question? I would like to. I don't want to know of any of the podcasts that are really well known, like the Dwarka Patel or all the other ones. What are your sneakiest technical people who are completely off the radar.

Speaker 2

Yeah, let me open up. Snip. Seriously, to have a look. So one of the things that always comes to mind, he's getting bigger and bigger. I think recently he actually also hit like the top charts in Spotify, but still way too unknown is the so called founders podcast by David Senra. Every week what he does is he reads an autobiography of an entrepreneur, you know, one of the, of one of the all time greats, and then basically dissects it.

And he then, you know, makes these connections between all of these different founders, which traits were similar, how they were different. And it's an incredibly motivating podcast to listen to if you're, if you're a founder yourself and you're going to learn a lot. The best way to describe it is it's like church, but for entrepreneurs.

Speaker 1

It reminds me we're going to do. Yeah, go for it.

Speaker 2

Go for it. Yeah, maybe just to explain what I mean by that. So I often coined this term. It was actually David Senway himself, the host. If you think about what church is like, every week you go to mass and it's always about the same underlying principles, right? Like the principles do not change. It's all based on what's in Pebble. Every week there's a new story.

It's, you know, the, if you have a good priest, he comes up with a new story and tells you the same principles packed in a new story. And basically that's what you realize when you read or listen to all of these autobiographies of the greatest founders, entrepreneurs of the last 100 years. In a way, they are all so similar. They have the same principles underneath, even though they have a different personality, different background.

But it all boils down always to the same couple of principles, like work hard, be dedicated, be completely passionate, you know, have a strong opinion, but also adapt to data. But it's easy to now just rattle these off, these principles to you and you're like, oh yeah, okay, now I know them. No, you don't know them. But if you hear them over and over again in like very, in different stories all the time, it just helps you to understand what it actually means to have a strong opinion and follow that passionately.

But then also adapt when, when new data arises and do what you're doing,

Speaker 1

which is to actually build a company. Because without actually building the company, there's

Speaker 2

no way to do it.

Speaker 1

And it's funny because the mba, there's always this back and forth mba. Oh, we gotta get the mba. Get the credential. Oh, no, we can, we can disrupt that. We could just do a podcast and call it an MBA and then back to the university, which actually that would be a really interesting conversation to talk about, given you're. You're in Zurich.

There's a, There's a university in Zurich, Right. That's the E F eth. Right. It always, yeah, it always, it always confuses me. Ethereum. And, and, and this university is like, are they talking, what are they talking about? What's the deal with that university?

Speaker 2

So ETH is. You can think of it as the Stanford of, I would say Europe or of the rest of the world outside

Speaker 1

of the U.S. interesting, interesting.

Speaker 2

So it's a highly technical university. It was the reason why I came to Zurich to study there. Um, you have, I think there are some very interesting statistics that I think it has the most spinoffs, like startup spinoffs of any university in the world.

Speaker 1

Cool.

Speaker 2

Like, it could be wrong. Maybe it's not number one, but then it's, it's up there and at the top. A large, large part of it is. Of course, these days, a lot of, lot of things is happening in AI, but I would say I think even more is happening in robotics. There are a lot of.

Speaker 1

Oh, cool.

Speaker 2

Biology, biochemistry. So, yeah, it's a very technical university, similar to, as you would know, Stanford.

Speaker 1

Okay. I would love if you have anybody to talk to in robotics, if you know of any friends who are. I'm gonna put a note to ask you later. We don't have to answer now, but yeah,

Speaker 2

it's just. Well, yeah, we could talk about that later. But yeah, the answer is yes. Because actually my, My wife.

Speaker 1

Oh, wow.

Speaker 2

For. For a. Robotics. Well, she used to work for. No, she does work for robotics startup. They, they do autonomous robots, but they were acquired a year ago, so now it's not a startup anymore.

Speaker 1

Wow. Okay.

Speaker 2

Yeah, it's a very interesting field and they work. Basically they get all of their. Or like 90, 80% of their talent are students that are coming out of eth. Yeah, it's, it's, it's quite nice.

Speaker 1

Okay. And do you like history? Do you like going into history?

Speaker 2

It's a very interesting question if I like it. So if you ask. So I enjoy it. If you ask whether I do it a lot, whether I read a lot of history or listen to a lot of history podcasts, then I think the answer is no. Except for maybe the founders podcast goes

Speaker 1

back into history of Founders. Yeah, because I'm, because I'm asking. Because I'm looking at the Wikipedia for ETH Zurich right now, and it's founded in 1854. And this is such an interesting time. Like it's the start of the, of the, it's not the, it's the end of the Enlightenment, it's the start of whatever came. I think it's the Romantic period.

I don't even know the term for what was happening in 1854. But essentially it's like when the west woke up and it was like we have the, the, we have railroads, we have media, we have science, we have technology, we have engineering and mathematics and just completely like went crazy. And this is my, one of my favorite periods of history. And it feels like Eth Zurich is similar to something we have in California, which is the university in Pasadena.

Pasadena, which Albert Einstein ended up at Caltech. So Caltech has a really interesting history because it took what happened in 1854, but it did it in California. And, and, and it was later, it was in the early 1900s. And that's where we get all the rocket companies. We had this crazy guy named Jack Parsons who was not universally educated at all.

He was actually this crazy occultist who ended up blowing himself up in a, in a rocket explosion after dabbling in crazy esoteric spiritual traditions. Really interesting story. There's a great book about it called Something Angels. So this guy started the rocket stuff. And, and, and it, it feels so important. This period from 1850s all the way until today, basically so important for understanding what's coming next, because we're basically all of the promises which originally were supposed to come in the 1960s are now coming now.

Like it's all the robots, the, all the Asimov's Laws, the rockets, you know, crazy stuff that none of the sci fi people predicted. So I guess for the last 10 minutes, if you do have to go at the hour, let me know. But for the last 10 minutes we can talk about that. Going into the future, where do you see we're going in the next five years, 10 years? What does this open up?

Speaker 2

That's a very big question. So let me start with the building blocks. So I think the three most important building blocks of today in the next 10 years is AI, robotics and energy. Basically you have AI for all of the, let's call it software and all of the intelligence, you have optics to apply all of that in the real world, and then underneath you have energy to actually power all of this.

Speaker 1

Cool.

Speaker 2

So anything future related, I Usually try to think about it in one of these three buckets. Does it belong to one of these three buckets? And if not then sort of like is it really that important or like will it remain important, you know, in 100 years time? Yeah. And yeah, I think that's, that's sort of the, the building blocks. What would this open up? Yeah, difficult to say.

I mean let's put it like that. I don't, I don't have a catchy hot take.

Speaker 1

No, that was pretty good. So that's a great framework, the AI robotics and energy. And let's dial in on the energy one because that's the most. A lot of people call AI the electrification of knowledge, which I agree, I think it is that. And then it's going to double back and meta change the robotics and energy because very soon I'm going to have people mule bring down robotics kits from the United States because I probably can't get them in Argentina, although I have not looked.

Maybe I can bring down robotics kits. We have this AI group here in Buenos Aires called AI Whisperers and we're teaching them how to vibe engineer right now. And they're already coming up with crazy things. Next step is to teach them how to vibe engineer robotics, which means I have to learn how to do it, which I'm excited about. But then the energy part is really interesting because Argentina also is related to the energy thing.

OpenAI just announced that they're doing Stargate Argentina and Stargate Argentina has all the risk will be on the Argentine partners to build an SMR small modular reactor in Patagonia and build a city around a small modular reactor. But United States has not developed a small modular reactor. I'd be curious if anybody's doing that in Switzerland, but Argentina has now promised to build a small modular reactor in one year.

Although it took China 11 years to build their small modular reactor. So I don't think that's going to happen. But, but there is an interest like there is a whole bunch more interest in building these, these, these tools for energy. What's your favorite energy mix up like solar, Wind? Yeah, yeah.

Speaker 2

I mean may just like. First of all, a couple of thoughts on what you just said. I do believe that in general the mainstream right now they're actually not aware how energy is more and more becoming the big bottleneck for the further development and deployment of AI. And one of the things that I've now seen very slowly starting to happen that I'm.

That I don't think is a good trend is that more and more people are starting to frame it as if it's AI or like the evil AI companies that are increasing your electricity bill and framing it as like, yeah, in five years time, it will take up so and so much electricity, such that you won't have electricity anymore to heat your house or something like that.

So that's like a, like, you know, like a very zero sum game kind of, kind of thinking. It's a bit like going back 50 years and then saying, well, just imagine a world where we have all of these smartphones and all of these laptops and electric cars. It will take up so much electricity, you won't be able to have electricity for the light bulb in your living room.

They know the solution is, obviously, we need to produce more electricity. And that's something that I think is, you know, that's very contrary to a lot of the cultural trends in the last 10 years. I would say, especially in parts where I have lived, you know, I live here in Europe. I would say Switzerland is a bit different than the eu, but, you know, our biggest neighbors where most of my childhood I actually grew up is Germany.

They very famously switched off all of their nuclear power plants 10, 15 years ago or whenever it was. Hasn't worked out that great for them. So one of the important things there, I think, at least in the west, will be as a society that we need to change the mindset from just seeing more energy as something evil. Obviously there are some energy sources that are not good, so maybe we should reduce the reliance on them.

But we, we should not just say, oh, no, energy is evil and let's reduce it by all costs. No, let's invest into new technologies. I think nuclear is one of the most promising ones where there's just been so little research going into that field or so little willingness, especially in countries like Germany, to go with new technologies. So that's one of the fears that I'm most optimistic about.

And beyond that, definitely solar. Solar seems like such an obvious one for at least many, many countries. Yeah. And that's at least has been one of the positive things in Germany that solar has been supported quite a lot.

Speaker 1

Well, thank you so much for coming on the show. And how can our listeners find out about snipd?

Speaker 2

Yeah, definitely. So one of the things happy to offer you and all of your audience is a. I can give you a promo link that you could add to the show notes with which all of your listeners can try out the premium version of SNIPD for one month for free. Obviously, we have a free version with which you can get almost all of the AI features that we spoke about, but if you want to give the premium version a shot, you can use that link.

So that would be the easiest way of finding us. Otherwise, just go online and search for snipt. It is spelled S N I P D and I'm sure it will come up.

Speaker 1

Thank you for listening and I hope you enjoyed this episode. As always, you can find me on Twitter. Also, don't forget to subscribe on Spotify or itunes for every weekly episode that I publish on Monday mornings. Hope you have a great day.