Welcome And The Big Promise

SPEAKER_02 0:32

Welcome everybody. Hey there, Scott. And Scott, I've brought a guest for us today who has a solution to share. This is Jonah Lid. Welcome, Jonah. Hey Adam. Hi, Scott. Nice to meet you. Nice to meet you. Yeah, Jonah has a cool thing to share. He is a uh he has a cool background too. We're gonna get into his background too, but we're gonna start with his solution. But uh Jonah and I met on a boat and we started talking about physics and metaphysics and all the physics of the world. And uh it was a great conversation, and so we made we made friends after that. But he's working on a really uh interesting thing that is in the AI realm. I I know Scott Scott's not a lover of AI, but I think it's what's what's what's cool about what Joan is doing. I'll just say first off is that it's not your run-of-the-mill AI. It's not like a chat GPT rapper doing like HR for you or something stupid like that. It's it's really, really cool. So we're happy to, I'm happy to have invited him on because he has an unheard of uh but really potent solution to share. Oh, I'm delighted to be here. So thank you. Right. Do you want to tell us a little bit about Innovation Lens? That's the that's the name.

SPEAKER_01 1:47

Yeah. Well, why don't I give you a metaphor? And we'll start with the metaphor and then we'll see where that brings us. So Innovation Lens is like a map of what has already been done in science and what is likely to happen next. And if you think about a map of territory where you know people have partially explored, you're gonna have some areas that are covered with roads and trails and uh human buildings, things like that. And you're also gonna have empty spaces on the map. And the empty spaces call to our spirit, they call to our sense of adventure. And some people want to go into the empty spaces. And so innovation lens is like that kind of a map of the innovation landscape, the scientific research landscape, where you can see continents and islands of knowledge that has already been um has already been discovered. And also the empty spaces like the sea where people haven't gone yet and could go. And on top of that, we also have a predictive layer. So it's possible to see with a pretty good precision. So it's 70% precision, which means our predictions, seven out of ten of them, are end up being true. Um, where should you go next? So it's kind of like a um a predictive suggestion about where are we ready to go and explore next.

SPEAKER_03 3:07

Yeah. Now, where should you go next is a question that's not helpful for me because I'm not out there doing, but is this like researchers and scientists use that to figure out how to uh how to take the next steps on their project? Well, yeah.

SPEAKER_01 3:21

I mean, think about think about if you're I mean, the the first place that I started was PhD students. You know, I was a PhD student writing my thesis, and and this came out of some of the ideas that I had at that time. Um, you're gonna dedicate three to five years of your life to something. Um, you'd you'd want to choose a topic that's actually gonna be relevant, gonna be useful. And that doesn't often happen. So part of what I was trying to do was help um PhD students decide their topics with better precision. Uh, of course, that applies to early stage researchers, it applies to people who would like to get tenure. You know, you want to write an article on something that that gets big. And um, if there are patterns in the data, as I found, then perhaps you can, you know, you can choose this rather than that and get a better job later on.

Helping PhD Students Pick Topics

SPEAKER_03 4:06

So it kind of allows people to get ahead of the curve on things, even like as a as an idea to be in the right spot in the right place at the right time when it's when it pops.

SPEAKER_01 4:16

Yeah, so that's yeah, that's definitely part of it. Now, the the trouble I ran into was that researchers don't have any money.

SPEAKER_04 4:22

And so it's all that PhD student money, they really there's no they don't even have money for pizza.

SPEAKER_01 4:29

I mean, come on. Right. So I was trying to sell my predictions for a dollar a pop, and I still wasn't able to sell any.

SPEAKER_02 4:35

So you were just like the great Zoron sitting sitting at a at a at a carnival in Staten Island, like give me a dollar and I was telling you PhD.

SPEAKER_04 4:45

I could help you. You just have to have work with me. Just give me a dollar.

SPEAKER_02 4:49

All right, all right. You said yours is a map, but it's really more like a Ouija board. No, it's not like a Ouija board. It tells the future. Let's all put our hands on the innovation lens.

SPEAKER_03 5:01

So then if if that if that plan, the original plan wasn't uh wasn't taking off, what where did you what what was your next move?

Pivoting From Academia To VCs

SPEAKER_01 5:09

Yeah, so then we went to investors. Um investors have money and they have a real problem. You know, so you've got some amount of money that you want to deploy in the market, and you had better get it right, you know, at least some of the time. And so having an edge where you can say, you know, seven out of ten of our predictions actually turn out to be true is something that investors are interested in. So we've started actively working with VC firms. Um, we have a couple of different arrangements going right now. Um, but the best one was two years ago we worked with a company that was trying to deploy a future of computing fund. And they asked us for help deciding where to invest on that. And we gave them four suggestions, and three of them they really liked. And one of them was LLM generated code. So that's companies like Cursor, Codex, you know, these companies are lovable. Um, and everybody now knows that that was a great bet. That was, you know, a 200x bet over the last two years. So uh that really worked out well, and we have some others like that.

SPEAKER_04 6:13

Nice.

SPEAKER_02 6:13

Wow. Very cool. Sick. So that is a good, that was a good bet. And you what and didn't you? I think you told me about them. Didn't you also tell them not to invest in quantum computing?

SPEAKER_01 6:25

Yes, that's right. Uh now the reason for that was that they were looking to deploy 100 million, and we looked at what was possible and what was likely on the innovation frontier and said, look, with 100 million, you're just not going to be able to break into the technologies that are going to give you a return in five to 10 years. So why don't you invest in other things? And we, you know, we gave them besides the cursor LLM code generation, we gave them a couple of other other options too. What if you just started your own fund? Yeah, that would be a great idea. The problem is I'm not very good at fundraising, and so if I knew how to get the money, I could tell you where to put it on.

SPEAKER_02 7:03

I know I know people who not who know how to do fundraising. Maybe we should try that. Maybe we should try to put our own fund together.

SPEAKER_03 7:09

It's so often that the making of the thing and the selling of the thing are two different skill sets hold by two different people, you know.

SPEAKER_02 7:16

They really are, they really are.

SPEAKER_01 7:18

Yeah, yeah.

SPEAKER_02 7:18

I'll do some booking around about that. That'd be cool. Because it seems like that's I mean, if you can make a bet that you should put money into you know, lovable and and and cursor two years ago, that's amazing. I mean, that's a really good ability.

SPEAKER_01 7:31

December 2023, man.

SPEAKER_02 7:33

So yeah, I was I was thinking that yeah, early. Yeah. Yeah, wow, that's cool. And even if you only put a third or quarter of your capital in that, I mean, the the the the benefit, you know, because it's that was the the Pareto distribution right there, you know, it would have blown it up. Yeah. Pareto distribution. Now you're going over my head. Oh, come on, Adam. You know what? I know that's what that one's 8020. That sounds like a parental distribution, like a parent. Here is your allowance the parental distribution.

SPEAKER_01 8:02

Is that is that Scott was the comedian.

SPEAKER_02 8:04

I well, you're the expert today. You I'm like proposing the solution in Scott.

SPEAKER_03 8:09

Yeah, you're you've led Adam off the chain doesn't have to be.

SPEAKER_02 8:11

I'm off the chat. I'm I'm I'm I'm loose. The juice is loose, you know? All right.

SPEAKER_03 8:17

So Jonah, let me let me find so now Scott's being the serious one.

SPEAKER_02 8:20

Okay.

Winning Bets And Avoiding Quantum

SPEAKER_03 8:21

Yeah, we're switching roles. It's great. It's not crazy. But I as a person who doesn't understand AI or use it all that much, how could could you just walk me through like how does this work? I mean, I I'm imagining it's not just like type into a computer, what's the next smart thing? And it goes, well, here it is. So like that's my basic understanding. How can you how can you help me flesh out what's what's actually going on with this thing?

How Text Becomes Math Vectors

SPEAKER_01 8:45

So all right, I'm gonna try to be a little bit brief here, but uh the technical details are interesting. So the big breakthrough, um, now it was foreshadowed in plenty of other work decades earlier. So I'm not, I don't want to say that this was the only thing on the on the scoreboard. But in 2013, a Google team led by Mikhailov um published an article called Word to Vec. And this article showed uh a way that you could take the output layer of a transformer, not a transformer, output layer of a neural network, and um use that as a representation of text. So, what does that mean in layman's terms? You could make a vector, um, which is basically saying like a position in space that is associated with every word in a language. And the the famous example that they use in their paper is that if you take the vector for king and you subtract the vector for man, you add the vector for woman, you get the vector for queen. So it's like there you can do an algebra in this abstract space where the meaning of words is converted into basically physical locations. So based on that idea, um you can take all kinds of different text and turn it into vectors and then do algebraic and geometric reasoning on the distribution or the relations between the spaces in different different words. You can take a phrase and represent it as the sum of all the vectors of the words that it contains, and then you can do also more complicated things than that. So essentially, what I was doing during my doctorate was using this technique to um automatically read very large corpora. I was working on ancient history and archaeology, and um the ability to read gigantic databases like the entire Bible and all of the Babylonian religious texts and compare them by doing natural language processing was uh the field of work that I was that I was in. Okay. What I realized was that this space of um vectors that represent texts has mathematical characteristics that can be studied in their own right. And now the the interesting thing about that is you're taking AI, so you're taking the um the large neural network representation of everything that humans have written and using that to represent some subset. So, you know, whatever texts you're working on, I was working on uh comparing the psalms with some of the Babylonian religious texts.

SPEAKER_03 11:14

Okay.

Goldilocks Zones For Discovery

SPEAKER_01 11:14

And um then you can use the mathematical characteristics of those two groupings, those two clusters, to draw certain kinds of conclusions. And those conclusions are inner referential. So this is something I really cared about and at that time in particular, uh for motives that we might get into later with with Adam. But um, I wanted to study text in a way that wasn't making reference to outside information, but was all self self-contained. Um, and I I wanted to figure out how to you maybe you can't say very much, but I wanted to say things that are very, very certain about the relationship between texts uh without making any appeal to authority, without making any appeal to metaphysics outside of these text representations inside of the neural network. So to answer your question directly, Innovation Lens uses this representation of scientific literature now. So not not the not the Bible, but scientific articles. And by by looking at the relationships between the places, the locations where these articles are are you say it embedded in um abstract space using the text vectorization, um we can see patterns that have predictive value. And now that's the part that's patent pending. So that's the part I don't want to really say exactly how it works. But we can say that there's a Goldilocks distance where um there are fields that are so sparse that it's very unlikely that if you do new research in that field, you're going to hit gold. And there are other fields where the field has been dug up so much that it's very unlikely you're gonna find any gold because it's already been extracted. But there is a Goldilocks distance where you are most likely to find some treasure if you go dig there. Now, of course, you're gonna have to be good at digging, you're gonna have to be good at writing, you're gonna have to be smart. So it's not like the output of Innovation Lens guarantees success. But what it does do is it says the research landscape is ready for a discovery in this area. Go look there, that's your best bet.

SPEAKER_03 13:21

So, like like uh like an oil oil driller might be like, we don't know exactly where is the best spot to drill, but if someone can give us based on you know, based on other clues, most likely here, it gives us a better chance at striking what we're looking for. That that makes total yes. Is that do I have that? Okay, yeah.

SPEAKER_02 13:38

Well, it's like when an old man holds those like a like a stick with a device-shaped stick, and he starts walking around and then all of a sudden it tips down like that's oil.

SPEAKER_01 13:50

You know, my parents literally did that on our land. We built our house and they literally had a well maker come and find where to dig the hole. Yeah.

SPEAKER_02 13:58

And it worked first try.

SPEAKER_01 14:00

It worked, it worked absolutely. Now, I don't know. I probably there's a water table you're gonna hit it one way or another.

SPEAKER_03 14:07

So, Jonah, it kind of sounds like, and forgive me if I'm not understanding this right, but it kind of sounds like innovation lens, or it's like the the program is what if there was a person who could read every single piece of science literature and have perfect memory, and then all of a sudden go, oh, I think these things might be connected or drawing connections that a normal, like an actual human could never do because of our limits of our abilities. Exactly.

SPEAKER_01 14:33

No, no, you've you've hit it exactly. I mean, you want you guys wanted to start with the solution and end with the problem, but that is exactly the problem that I was I was frustrated by. I was working on this doctorate, working with a research team, and we would argue every week for many hours about what we were reading and thinking, and you know what I realized after a couple of years was that we kept arguing about the same things without ever really addressing the upstream reason that we were disagreeing on them. And so I started to build out graph representations of our arguments to try to arrive at what was the node, what was the real thing that we were disagreeing on? And then I ran into the problem that you just mentioned, which is there's a vast amount of information that we couldn't ever take into account because it's just too much. And yet it's there, so we ought to, you know. So uh I built my programs because of that. I wanted to see if I could make an intellectual prosthetic that would allow me to take into account more information than I can natively.

The Real Problem Information Overload

SPEAKER_03 15:26

That's that's awesome. And I mean, I get a rap for being an anti-AI guy, which is partly true, but this is the space I want AI to be using, you know, like I want AI to be doing supercomputing on things that human brains can't do, you know, like figuring out research and science and things that are helpful. I I my bristle up more when it when people are like, we should use it to make paintings and like movies and like music and stuff. And I'm like, no, no, no, humans already do that very well.

SPEAKER_02 15:52

So you're you've you mentioned a few times like debates you were having and you were talking about the Bible. You I know that you used to be a Catholic priest before you was a mathematician. Or before you became a got a PhD in math, but you were probably a mathematician a little bit before then.

SPEAKER_03 16:10

Is that why you were comparing religious texts earlier?

SPEAKER_01 16:14

Well, let's say it was a bridge. It was a bridge between different worlds because uh yeah, I was a Catholic priest from 2020 until 2019. Um and in 2020 I started my doctorate, and part of what I was doing was um trying to find a way to survive. You know, it was COVID and I had to find money, and I, you know, I'd been cut off from any sort of support from the church. So that was your first thought was to become a grad student. Well, it wasn't my first thought, but it was the only thing that was concretely possible.

SPEAKER_02 16:46

I'm going to commit myself to eating ramen for the next four years.

SPEAKER_03 16:50

You're like, if I'm in isolation that's wealth, unable to go meet other people, I might as well better.

SPEAKER_01 16:56

Yeah. Well, I I'm grateful for it. It was it was a hard time for everybody, but uh I'm I'm grateful I had a roof over my head. But you switched from Catholic priest to PhD in mathematics.

SPEAKER_02 17:06

That's a pretty isn't that a no, it wasn't okay.

From Priesthood To Research

SPEAKER_01 17:08

It wasn't a PhD in mathematics. This is why I call it a bridge. It was a PhD in ancient history. Um and I was doing digital humanities, so you know, I was applying a lot of math, but I previously studied physics. That's what my bachelor's degree was in physics and well, astrophysics at McGill. And uh so I had a math background from there. Um I wasn't I didn't get into the PhD thinking I was going to be as computational as I ended up being, but I have to say that um I've been programming since I was a kid. I was seven when I started programming on my TI 99 foray. Um and I've loved it, I've loved computers basically all my life. Um and yeah, so I I was I was trying to figure out a way to to survive. That was that was certainly part of the story, but I was also trying to use what I had been working on for 20 years in the Catholic Church in a new space, you know. So I had I had left the church for basically two reasons. One was um I wanted to get married and have children, and the other was I um I don't believe in authority, and I don't believe that authority is a road, uh you know, a trustworthy road to truth. And um so that was one of the main drivers behind my research was how can I discover truth in any domain, but whatever. The the sandbox I was working in was ancient history and archaeology. But if you're working on humanistic texts, it's applicable to many, many places. So, how can you find truth in a humanistic domain um without reference to authority? That was really an important question I was asking myself. And so I went to math as a way to represent things and manipulate them and compare them without reference to any external reality.

SPEAKER_02 19:02

So you built a mathematical Ouija board. Exactly. That was it. No, I like your map. I like your like, you know, heart of darkness analogy, like look, the interior of Siberia, the interior or the vastness of the ocean. We don't know what's there. The here be here there be dragons kind of analogy is quite cool. Yeah.

SPEAKER_03 19:23

It also sounds like in truth searching, especially with historical text, you're relying so much on the writings of a person, you know, and and that comes with it all the biases or different things that that person might put in their writing. So being able to kind of look at all of them and smoosh them together and compare them might I I'm assuming it might mitigate or you're trying to mitigate some of that personal uh input on the historical truth or yeah.

SPEAKER_01 19:48

No, that's that was that was a big part of it. Now, applied to history, it's actually very hard to do what I try to do. Um, and I don't think I succeed in that domain, or at least not very well. You know, my my doctorate is more about the limits of my approach than than its successes. But in other domains like quantitative science over the last 30 years, where we have a very large body of published work that is generally speaking truth tell truth-telling, you know, that's an important distinction, I guess, because Innovation Lens really works on text that tries to tell the truth. It tries to map the empty spaces. Uh, whereas historians do to I think a lesser degree for a bunch of reasons. One is the story itself is much more um it's much more difficult to synthesize. It's much more fun to tell a story that's riveting and interesting than it is in physics. You know, in physics, people really don't care about the riveting story unless it's also mathematically formalizable and true. Um in history, you can definitely tell a riveting story about the way that the Egyptians lived or the Babylonians or and and get a long way with that, you know? You can argue just for the sake of argument. That's a big part of it too. It's sort of a um Some people make their whole careers doing that. Yeah, I mean it's a sport, right? It's it's a it's sort of a that that is part of what's going on. So um, but for me, the thing that I really cared about was trying to cut through all of the debate and arrive at the structures that are um invariant, and that's that's what I've tried to do.

SPEAKER_03 21:27

Does does innovation let so I'm I'm in my head, I've got the visual of the continents, you know, kind of and the spaces in between. Does does your product does it like uh create new continents that are in between the two? Or is it like do you does it fill in spaces?

Testing Predictions Against History

SPEAKER_01 21:46

Yeah, yeah, yeah. That's a great, great question. Right. Because the problem with continents is that we generally don't see them emerging from the waters, you know. Um, but that does happen in science. That if we if you look at what what we had studied 100 years ago and what we know now, you can see that there was a very low resolution image of the universe that gradually becomes higher and higher resolution in certain areas, and then other areas remain very much unexplored. So, yeah, innovation lens as a map shows you the time. dependent development of science as well as the current state. So you know you can you can query it and look back twenty years ago, what did we know? And what do we know now and what's the difference?

SPEAKER_02 22:29

Yeah, so so in order to test whether you to test whether your predictions are true, you can just look back and say, what would have we predicted? Yes. And then you can verify by then playing the playing the tape forward to the present or or whatever a few years and saying oh in in 2015 we would have predicted you know this, this, and this, you know, drones or something and this and that. And then you w play the tape forward and say, hey, three quarters of those or or or I think you said you know 70% of it not a 10 of those were right were right on the money. Right.

SPEAKER_01 23:07

That's right. Yeah that's it's publicly verifiable. You can go on it with a free account and you can see some of our predictions and and look at what happened since then and and see it. I've done some videos on that too that are on our YouTube channel where you can just see yeah it's it's a pretty easy overlay actually you just show the predictions from 2018 to 2021 and show the articles from 2022 to the present.

SPEAKER_02 23:26

And it's like I once I once wrote a short story about this. Like if there were people who could predict the future and then it was about like agents of the government trying to track them down to hunt them down to put them in prison because they were like screwing up all the stock market and everything by predicting the future correctly. Okay. So just keep your head just keep your head on a swivel because there might be like a secret you know agency out to get you if you start predicting the future too well. Seems like it would be hard to catch someone who could predict the future like I'd be at their house and be like, gotcha they're like actually I'm out of here I'd saw this no they can only predict the future like like Jonah's thing can predict the future like in these general sort of predict like what's going to happen tomorrow to me.

SPEAKER_01 24:08

They can't do it's you know this is this is the premise behind uh foundation right yes it's it is like psycho history psycho history yeah it's oh my god you know psychohistory jonah yeah wait catch me up what is foundation nope Scott's on the outside this is an inside thing between you and me now Jonah I'm just gonna let me in let me in I Jonah you want to tell him what psycho history you knew the word right away I was gonna call it mathematical history like a doofus so you know no okay so I'm not very good at it because Asimov like um I once heard it his his books described as um more typed than written um which I I I think is kind of accurate.

Limits In Humanities And Corpus Access

SPEAKER_02 24:51

So I don't like reading him very much the but the idea behind psychohistory as far as I remember it's been a while but um is that there is an order to history and you can sort of you you could perhaps influence the order in which things happen and make outcomes happen over a very very large scale you know like centuries millennia scale um billions and millions of people right yeah yeah yeah it was like it was like that you could like you could like so I think the big thing was like this one guy who invented psychohistory was like I predict in the in the Asimov foundation series which there's an Apple TV show about and you know it's like Elon Musk loves foundation the foundation series like a sci-fi series about interplanetary uh civilization like an intergalactic civilization and there's this guy in that who says uh at the peak of their intergalactic hyper technological civilization there's this guy who invents psychohistory who says he can mathematically calculate the future in these these big grand swaths you know and he predicts that the civilization will collapse and everyone's like that's crazy how could this collapse you know we have this interplanetary intergalactic you know hyper technological civilization how could it possibly collapse but he says no it really will and people start to believe him and he says we're gonna create this arc of knowledge to preserve so when the collapse happens there's this arc to preserve all the knowledge and technology of the civilization and so people start to come and flock to this guy and help him do his arc uh and you know and I think it I haven't read the whole series but I think he's right it ends up the civilization kind of collapses under its own sort of corruption and and so it's sort of like a Roman Empire or whatever. But it's like um but it yeah so that's that's kind of what what Joan is doing in a way he's he's found a way to he's found a way to not let it be dialed back a little bit.

SPEAKER_01 26:43

Yeah well let's go a little further joan is a horse doing Joan is betting on the races these days let's just put it that way he's at the track let's dial it back a little bit what I'm doing is I'm I'm saying that the the scientific community is ready to make discoveries over the next 48 months or 24 months sorry next two years right um in XYZ domains. Right that's a little bit less than saying everything that found you know if civilization's gonna collapse or not.

SPEAKER_03 27:16

No but you're you're saying there are smart bets to be made and there are ways to make them smartly and not just being like uh this thing or uh that thing you know like my friend or looking forward instead of looking back as to indicators of what's gonna be big.

SPEAKER_01 27:31

Yeah and and I can I can also compare that we have a published article on how how we did the validation um it's on our website innovationlens.org you can get the article article from archive but the the basic metrics I mean there are two that are obvious one is random guessing so you just do whatever the hell you want and see if it's good um and that obviously doesn't perform very well and then another one is follow your PhD advisor's advice. And so you look at the tradition of existing research and you go and find a gap and you write an article about that. And so you you find your location in the space of scientific knowledge based on the existing tradition. And that obviously performs well enough. You know that's the way that PhD advisors usually work. But our algorithm outperforms that by a lot you know it's by 200% in the physics or computer science domain by 300% in biomedicine. So what we've found is a pattern that allows you to do a lot better than what you would do if you just follow the the ordinary way that scientists choose their next topic of research.

SPEAKER_03 28:34

Yeah and and I also as a non-scientist non-researcher think the idea of spending three to five years on a subject you would want that extra help you would be like pick something that's going to be relevant and interesting all the way through and I'm not like like all right I guess I got to complete this thing now.

Why Many VCs Still Guess

SPEAKER_01 28:49

But that's yeah well I mean that's that's the crisis that most PhD students go through because you you hit a wall where it's like this is really I've exhausted what I care about here and I still have two years to work on it, you know, and and uh you've got it now that's part that's also important because you do have to carry the thing out all the way and you have to learn how to write articles and do the whole thing. But what I'm offering is an optimization where you can do a little bit better at the beginning in your choice of of topic and it also is helpful in a bunch of other ways. You can do a better literature review um you can see the startups that are working in that space you can find the researchers who are working there.

SPEAKER_02 29:24

So there's there are a lot of advantages that you can get from our model and I have a question as a PhD researcher myself but in the humanities so you said it's not very good at like finding truth or making predictions in like history or the humanities but can it help find popular research topics in the humanities for like a PhD student?

SPEAKER_01 29:47

Well I don't know the answer to that question because I don't have access to a large enough corpus of humanities work. So one of the great things about physicists is they got really tired of um you know the pay to publish thing and they invented archive at Cornell University in the 90s. And so basically all physicists around the world publish their articles for free immediately as soon as they're ready for for eyes on them, you know, as free prints. And then that expanded computer scientists started doing the same thing and some other scientists but essentially computer science and physics are the two that have great coverage and those are the those are the the two domains that InnovationLens started with but in but in theory if you had access do you think in theory would it might help with that? Yes I think I think it would help. I would be I'm very very curious to know that and I I would love to find a partner who would like to deploy this on humanities research or history or economics or you know one of the other fields that we don't cover currently.

SPEAKER_02 30:43

Couldn't be economics because again it has to be based on trying to find truth.

SPEAKER_01 30:49

Yeah well actually one of the cool things dig one of the cool things about the algorithm though is that if if the articles are um how can I say this there there is a certain amount of ability to to overcome bullshit and it's not like if you pile on with a hundred articles on one topic that doesn't really sway the algorithm so um interesting that's that would be fads it's sort of it won't be completely swayed by fads or by no actually fads fads get get show they they don't show up very much.

SPEAKER_02 31:20

It's it's not about yeah I would love to see that I mean especially because I'm like a philosophy ethics researcher guy it'd be really cool to see like what your thing thought the hot new ethical topics would be in like two years. I could I could I could see although in some ways it's reactive to things like for example you know two years ago or five four years ago would it have predicted that AI ethics would have been a huge field you know only if it looked at the fact that AI would have emerged. You know what I mean if looking in just endogenously at the corpus of ethics papers you wouldn't have been able to predict that AI ethics would become this big thing probably I don't think well I think that's the point is that none of us are predicting machines the way that because we can't keep all the pieces in to repair them. Right.

SPEAKER_01 32:10

So maybe you would yeah maybe you would I think you should run that Adam go go to the platform and and put in some ethics research um but I thought you said you don't have the corpus. Well inside of archive inside of the computer science domain there are a lot of AI ethics papers and it would be interesting to see when did they start popping up and were we predicting them because we are we are predicting that domain it also could it also could be like a subtopic inside of AI ethics because you would have the corpus for that then because they're all there in Archive X.

Prediction Markets And Data Sources

SPEAKER_02 32:38

Yeah yeah yeah but it'd be interesting to know like I don't know are people going to be like you know legal normativists or something in five years you know or some you know some ethical you know topic is going to become super hip. I don't it's very it's quite interesting. But but I but I do like the hard science stuff really obviously because I want to enjoy the technology that comes from that right I want to enjoy the you know discoveries of drugs and gene therapies and whatever else gets invented. So the faster we can accelerate that the the better it is for for all of us. So I'm I'm I'm super uh stoked on this uh on this it and it seems so interesting I once I once taught I once stood up to a whole panel of VCs you know I live in San Francisco and I am in tech and I interact with investors and I have friends who are VCs and interact with people. Anyways I went to this conference and there was a panel of VCs and these were like big dog VCs. These were not like little dogs these were big dog VCs and they had you know the mic going around and I was like I'll just be an imp you know I share with you Jonah the same sort of distrust of authority. So I stood up and said all of you uh you know advance and talk about new technologies and you know improving everything with technology can each of you go down the line and say one thing that your company has adopted a technology you have adopted to improve your allocation of capital into startups and honestly they had nothing nothing they went down the line and they were like uh we use Google Docs like uh we use uh uh we've used some uh I mean you know their in their Wikipedia things have probably gotten better in the meantime but uh well this was like five four years ago five years ago yeah and they well I mean that's why I it was pretty interesting to when you said I'm actually making a technology that makes VCs because VCs just they just give money to like their friends or like what their staffers their staffers' friends you know they have these young staff members and the staffers go and like connect with startups that they know from like parties and then they introduce them to like their bosses at the VC and the money just goes to them. So it's not like it's not like they're doing some like you know they're not scientists.

SPEAKER_01 34:49

They're not using data they're not using they're just using like you know like their hunch the hunch they have well okay that's too broad a brush though man I mean that might be true about a bunch of people and that's and yeah I mean I I don't know those those uh young scouts either so I haven't been able to get a lot of calls with the VCs myself. Yeah you gotta go to more parties in San Francisco.

SPEAKER_02 35:08

Well I have a baby and uh Hawaii love parties oh my gosh babies love babies totally love San Francisco parties where people are all on ketamine they love most of the parties here are like 90% babies there's very few it's mostly babies actually yes yeah all right well I'll have to think about that but um no I'm of course being somewhat cynical but I it is true but I'm not you're not wrong either but it is a bit it is it is like that for a large part of VC you know I'm not gonna I'm not gonna claim to know everything about VCs but I'm just gonna give you one use case about data driven VCs that we're working with now.

SPEAKER_01 35:44

And they they've decided what they want to do is automate the whole pipeline. They've got a bunch of scouts um actually like a thousand of them and what they want to do is automate all the vetting and so the scouts throw their ideas into the machine the machine chooses them up and and vets them and then a very very short list goes to the LP and uh you know the the actual investors who are going to put their money in. So that's good. Here's the cool metric innovation lens right now is landing 60% pass rate of the machines. So that means 60% of our of our predictions are passing their hour long you know LLM that's going to vet the the uh startup ideas against all their different um agents so I think that's pretty good. Yeah that's really good.

SPEAKER_02 36:32

Definitely better than a better than a coin toss and better than a monkey on a c on a keyboard.

SPEAKER_01 36:36

So yeah well it's a lot better than the other scouts too so that's that's kind of oh you guys are the best scout.

SPEAKER_03 36:42

Oh so Jonah I'm hearing you talk about prediction and that's what your game is like that's the industry. But I also know like um this may not be related at all but prediction markets are a big thing right now here where everyone's trying to predict everything and betting on everything. How do you have how do you see a fast can we make some fast cash is what you're asking Scott well I guess I'm seeing like that that seems like a really inferior version of just humans trying to predict at each other versus what Jonah's interested in it sounds like which is more like getting to real truth and not the game part of it I guess.

Leaving The Church Without A Clean Break

SPEAKER_02 37:19

I don't know I I just want to see what I guess my question what your thoughts are on prediction markets or out of the was Jonah Jonah were you the guy who bet like$75,000 on like January 1st that Maduro would get out of Venice the head of Venezuela I don't moonlight as a security agent Noah um no that would be that would be a cool thing to get into Scott I um I guess there are a couple of things that um keep me away.

SPEAKER_01 37:48

So one is the availability capital. It's the same problem that we said before for investing in traditional markets. The other thing is that from an information standpoint the idea that you have a special you know a super predictors the idea that you can figure out the patterns in the data better than others um certainly there are many many ways to to parse the data and try to make predictions the data source matters and what I'm working with is only scientific articles. And I think that that's that's kind of the specificity of what InnovationLens does. Other people can use other things I'm sure if I had Twitter data, you know, the way that the the fire hose used to be open source you could you know there are papers on this you can predict the markets based on Twitter sentiment very well you know so um definitely with different different data sources you can do lots of predictions.

SPEAKER_03 38:37

I was gonna say but I'm I'm on Twitter so if you start including Twitter you're gonna get all my dumb takes in there too I think it's it's keeping it to scientific research is probably not a uh not a bad idea there.

SPEAKER_01 38:48

Yeah well but I mean there's there's the fact that that the aggregate is useful you know the the average of what everybody is feeling about something does move and that average means something too. So i people can have the most random takes they think but as a statistical body you know as statistical mechanics the the whole cloud does move in one direction and that means something overall I have a question that is completely sideways here this is a different but uh you mentioned you were a priest and you're no longer a priest.

SPEAKER_03 39:22

I in my head I'm wondering how does one stop being a priest and and and my my imagined which is not true is that you march into like the Pope's office and you put your Bible in your collar down and you're like that's it we're done like a cop would do with a gun and a badge but what that's clearly not true.

SPEAKER_01 39:39

But like what what what is the how does it I think we're we've got a whole genre of movies we could start making I could do next I'm done bam bam yeah I need your bible in your collar now you're out you know Scott that would have been a better way that would have that would have been a better way I wish I wish that I'd had a moment like that where it was you know slapping down the badge and saying that's that. It's yeah it would have been more cathartic. The the unfortunate situation is that you know usually um if you believe in something like the Catholic Church for 20 years as I did um you cannot quickly make that kind of a decision. Like it's it was an extremely long and painful process for me. So there is there isn't a cathartic moment where it's like oh you know what everything that I believed and said and all the things that you my leaders have said to me over the years actually you know what it's all not true and you just hug each other and be gone. You know it can't happen that way it's it's much more painful. So um and it's in a sense an ongoing process because the the kinds of things inside me that led me to that life the searching for truth the searching to be a good man searching to contribute positively to the world um you know all of those things remain true about me. And I guess having bracketed computational and quantitative methods for those 20 years because I I really did focus on uh philosophy, on theology, I have a doctorate in theology as well. That that bracketing was something that also I've had to work through and I'm still working through. So going back to quantitative methods for me was was a part of the catharsis. It was like okay I've loved math since I was a little boy I've loved computers since I was a little boy. Let's start from where I started and try to rebuild life from here. Including all those questions about truth. You know so innovation lens is really kind of my my yeah my love child in the sense that it's it's all of those things in one you know it's an attempt to to support my family but it's definitely uh first of all an attempt to know the truth and communicate it and do so in a way that that is as free of bias as I can possibly do which is bias from authority as well as bias from fads and bias from bullshit and you know all the kinds of bias that is explicitly part of what I'm trying to do.

SPEAKER_03 42:05

Yeah it seems like an extremely noble endeavor and that's that's commendable very cool.

Philanthropy And Better Science Metrics

SPEAKER_02 42:10

Oh my god Scott just complimented an AI system. This is like bizarro land where am I making a convert here? Don't teach it to write poetry just whatever you do no but it's poetry is so bad. It's yeah yeah the poetry that it writes is pretty bad huh?

SPEAKER_01 42:28

It's not well but no but this let me just let me just say this this is actually I really believe and love this part. So what what do what is AI right? It's the aggregate of all the things that humans have written and then you can put in multimodal so you can put in images and other things too but essentially it's the aggregate of what exists and then that distribution has some sort of fancy characteristics where it's not just a big gray blob it's actually queryable in a way that makes wonderful text and wonderful programs and all kinds of things. But what do humans do humans invent languages right humans invented language yeah and that's something that you know we go outside of distribution. So my whole idea of you know going into the empty spaces on the map that's a a very deep thing about humans that what we do is we go into the unexplored territory and we map it.

SPEAKER_02 43:19

We discover what's there yeah um we we describe it we communicate it I was walking in Noe Valley where I live in San Francisco and I saw Dwarkesh who's one of the most famous podcasters in existence right now. And I was Dwarkesh if you're listening to us calling Jonah onto Dwarkesh that's what we should do. Because Dwarkesh talks about this a lot he talks about how LLMs can you know they kind of can spit out interesting stuff and they can kind of you know organize things and sort of say the predictable thing but they can't they're really not good at discovering new knowledge. And I was like Dwarkesh you should really talk to Jonah like he has an AI that discovers new or at least points in the direction of this new knowledge. That's right.

SPEAKER_01 44:06

It points at where you should go explore it doesn't tell you what's going to be there but it it tells you what what's likely to be important. Yeah.

SPEAKER_02 44:14

And it could be an agent in a whole slew of agents that Each one, you know, right now our agents are just all different little flavors of LLM, but there could be in the future totally different, you know, the agents could be based on completely different computational bases, much like the the parts of the brain. They're not just like all they're not just all, you know, um, you know, they're not just all just visual cordises all over the brain that gets sort of forced into doing something else. They're actually specially built networks for each part of the brain. Um, and it seems like that you're you're basically building one particular part of that.

SPEAKER_01 44:53

That's yeah, that's a good way to think about it. And that is all one of our offerings. We have an API and an MCP, so it's possible for you to take our predictions and use them as the basis for further queries. So an agent could go and take take our predictions and then go run the numbers, you know, do the market study, do whatever else you want to do.

SPEAKER_02 45:10

When you when you told me about this, my my thought was it's so hard to capture the value from this. I mean, unless you had millions and millions of dollars to start with, then you could just invest yourself or you could, you know, but but it's so hard being just like a normal person like us, like just normal non-millionaires, you know. But then you created something that is like that the value capture happens at a very at a very high level of capital. So, you know, it's it it's too bad if you have to like hopefully you can really sell yourself for or you, you know, this service for uh what it's worth, which is it sounds like it's tremendously valuable, but it's hard to capture that value. So well, yeah, you have to just thank you. Thank you.

SPEAKER_01 45:50

I mean, if there are any any VCs or or hedge funds listening, we could sell you a vector.

SPEAKER_03 45:56

Part of it is explaining what what it is and what's going on, and that that helps me uh for me understanding anyway, just having a conversation like this is is huge for that.

SPEAKER_01 46:07

Well, the other the other person, let me just say say one more thing on that. Because the other person that I'm talking to are philanthropists and um you know, family offices. So people who have an idea, they have capital and they have an idea about the world, um, about human flourishing, about how to improve the future. And those are great conversations. And you know, they're looking like also one of the first places that we're going to be able to monetize in a in a real way. Because um, these are people who see beyond just the prediction thing, but it's also about what of what of the possible futures would we most like to have happen? And you know, then align that with which are the most likely, and you know, there's a there's some leverage there. Interesting.

SPEAKER_02 46:49

So wait, try to like spell that out one for one more second. So I can understand the science, you kind of predict where the science is sort of headed and you kind of go there first. With philanthropy, it's like I'm a philanthropy, I want to reduce malaria or something. How does your thing, how does your peak product help with that?

NSF Grant Idea And Measuring ROI

SPEAKER_01 47:09

Well, so one thing is is giving you KPIs for the ideas that you already have. You know, if you if you want to dedicate a lot of funding to cancer research, we can help you decide which of the many things you could focus on are most likely to be valuable in the term of a grant, so the next two to four years. You know, uh that's that would be one way. And then another way that we're we're working with more kind of visionary uh philanthropy is if we're talking about the future of money or we're talking about the future of um you know capital itself. I know you like to talk about this, Adam, on online where you talk about basic income, universal basic income, and and different structures for voting. People who have that kind of interest are often also interested in how do we optimize science beyond metrics like um what we have, which is citation count or publication count, which everybody knows is a good heart law thing where you end up publishing bad articles and you know it's not very useful. So a tool like Innovation Lens in that domain is not directly maybe monetizable, but it's it's part of an ecosystem where it's like we want to optimize for the good of humanity, and this is a public service. So it innovation lens could become a thing like PubMed, where it's um a layer on top of existing research helping scientists do better. You know, that that's that would definitely be a very cool use for it.

SPEAKER_02 48:32

Yeah, I'm I noticed that in my in my research that they're really there really lack, there's a s a very large gap in trying to understand which journals are like high quality and which publications are high quality. There's just like there's this thing called like Ciago or some weird has a weird name, but SJRC or something. And and you look on there and then there's this H index, and there's like but all of them are I mean, you you just scratch how their their methodologies and you realize these are totally imperfect. Oh yeah, they're not very good. But that's all we have to go on. So people just try to get into the high H index journals and they try to get into the high Ciago uh metric journals, and it's like, why why? I mean, they just do it because that's what other people are using as a measurement, it's not because it actually is the most impact, it's just because that's the metric people are using. So it'd be cool if there was really like, no, this is methodologically, this these are the most impactful papers, and these is the most, these are the journals that are creating the highest, so you know, getting the highest impact papers are coming in these journals. Like that would be a real service to science. I mean, I don't I don't think it's a highly paid service, you shouldn't do it as your first well, yeah, I don't think I'm gonna get rich on that, but I have a I have a grant pending at the National Science Foundation to do exactly what you just said.

SPEAKER_01 49:53

So I said I said to them, look, with this tool, I can map the money that you have already spent on grants, the outputs from those grants, right? Articles, patents, and you know, because we also map the patent database uh besides the scientific articles. So I can show you, you know, what was the actual output of your of your very large investment, and then we can map that against my predictions and see if could I have helped you do better? That's an open question, but I think the answer is probably yes, yeah, probably yes. And I'd love to do that. Even if it's 20%, but you said sometimes it's 100%. That's billions of dollars.

SPEAKER_02 50:28

That's billions, yeah. Wow, that's a really cool shoot.

SPEAKER_01 50:34

Yeah, goddamn. It's gonna work, man. It's gonna work one way or another.

SPEAKER_02 50:38

You gotta find the right people, yeah. Yeah, very cool. And the philanthropy people, too, they might also just have connections out to this kind of thing, too. So that's interesting to have philanthropy be uh the V, you know, if you're talking to VCs and philanthropists, you probably are talking to people with enough power and enough, you know, connections to to if they say, hey, this would really work over here, they can make an intro too. So it's I think it sounds like you're barking up the right trees.

How To Reach Jonah

SPEAKER_03 51:01

Well, I was gonna say, since we know that Dwarkesh and these VCs are listening right now, how how would someone who's in interested in the innovation lens, how do they, what's the best way for them to get in touch with you or get get involved with this?

SPEAKER_01 51:14

All right. Well, you can go to innovationlens.org. You can email me at info at innovationlens.org. You can find me on Twitter, you can find me on LinkedIn, Jonah Lynch. Um we'll put this all in the show notes so people can see it right at the top.

SPEAKER_02 51:30

You can find me. Yeah. I realized when I saw Dwarkesh, I walked right by and I said to my wife, that's Dwarkesh, he's the most famous podcast.

unknown 51:39

Oh my god.

SPEAKER_02 51:39

You know, like I was totally like fanboy, like oh my god. And my wife was like, Why didn't you go up and like talk to him? And I was like, What do I have to say to Arkansas? Now he doesn't want to hear he's probably just swamped with fans all the time. Of course he wasn't, he was just walking down the street with his friend. But I was like, you know, after the fact, I was like, damn, I should have gone up and just said, Hey, your podcast's really awesome. I also have a podcast, it's called Solutions of the Multiverse. Like, if he could just listen to even a few, I think I think that'd be pretty cool. He might, he might think that's pretty cool. So next time, next time I see around because I think he's in my neighborhood. That's my suspicion. So uh next time I see him, I'll I'll I'll I'll I'll paparazzi him.

SPEAKER_01 52:18

All right, well, say hello from me.

SPEAKER_02 52:19

Yeah, I will. I will and I'll also reach out to some of my friends who run funds and be like, hey, could we start a fund around just like why don't we do our own fund and you and and we all collaborate? That could be fun.

SPEAKER_01 52:32

Well, maybe I should do a paper trading version and uh get some back testing and prove it.

SPEAKER_02 52:37

Honestly, if you could do public, if you could use these predictions for public trading, you know, then you could you could really, I mean, then it'd be really easy to raise money.

SPEAKER_01 52:46

You know, we could. We could all I'd have to do is pattern match our predictions to the publicly traded companies.

Trading Plans Open Source Debate Closing

SPEAKER_02 52:51

And um I think you should spend the rest of the day on that because that's the most direct access you could get to to money, you know. And also you can use leverage too. You can you could, you know, you can get three to one, four to one leverage, you know, if if you have the right traders involved in your team. And so if you really, if it's working, shoot, you know. I mean, obviously that it's right, it's risky, but you know, if you can predict 60% of the time the right answer, uh like what's gonna go up, you can become extraordinarily wealthy very quickly.

SPEAKER_01 53:22

Well, that would be nice. That would be nice.

SPEAKER_02 53:24

Yeah, yeah, yeah. You'd be like the next uh Berkshire Hathaway, right? The next Warren Buffett.

SPEAKER_01 53:29

Because remember, but you're bringing me back to to the first the first investor. You know, I have very few, but he he really helped me at the very beginning. Uh, when I told him the idea, he was like, Um, I said, Yeah, and it's a machine that'll help us know what the truth is. And he took a beat and he said, Who cares? Yeah. The cynicism of the capitalist. Yeah, he gave me 10k anyway, but uh, I think it was just to say, uh, go go play.

SPEAKER_02 54:00

That's awesome. Well, cool. I'm I'm super jazzed. I'm more jazzed than I was at the beginning.

SPEAKER_03 54:05

I am impressed that I have an AI of like a positive AI. This is you've done what I said earlier that the making things and the selling things are two different skill sets, but you've done the selling things on me where I I'm uh I'm on board. I think this is a fantastic idea, and I'm really happy that we had this conversation with you today.

SPEAKER_01 54:23

All right. Well, go to the website and subscribe, and you can give me 40 bucks. 40 bucks.

SPEAKER_02 54:30

I want to get on the ground floor as an investor, but I don't have enough money, so I can't do it.

SPEAKER_03 54:36

It's the chicken and the egg. We need the prediction to get us the money to invest in the prediction. Okay.

SPEAKER_02 54:42

That's right. That's right. Well, I'd go in on a I'd go in on something that that would just, yeah, where we just start trading.

SPEAKER_01 54:48

Well, let's do that, Adam. I'm gonna I'm gonna do a pattern match. I mean, Claude Code can write that for me tonight, and uh tomorrow we'll start the fund, okay?

SPEAKER_02 54:54

We'll start the fund. We'll put in we'll put in five grand and we'll see what happens, you know. New era. Yeah, well, we or we can test it against the past, right? We can say what if we what would we have done six months ago and then what happened? What would our returns have been? Yeah, so you can be pretty, pretty, pretty comfortable with it, I guess. I mean, I guess you're always putting your way putting money in the stock market, it's always kind of you gotta kiss it goodbye, you know, in case. I like the idea of not sharing it with anybody and just using it just because you basically are like like if you were walking down a beach and you found this like genie, and the genie was like, I'll tell you like 60 of the time what's gonna happen in the future, and yet you go you'll be right 60 to 7 of the times. I wouldn't tell anybody I had that, I would just use it for my own benefit.

SPEAKER_01 55:36

You know, okay, I gotta I gotta push back a little bit. You know, I know we're closing here, but my dream here is really to open source the whole thing and write an article about how it works and just give it away. I want to make a little bit of money first, but not that much. Yes, I'm on that.

SPEAKER_02 55:52

No, I've seen a lot of I've seen no, no, I'm I'm not against open source. Open source, if it's the right, but you have to open source the right things. I've seen things that are super valuable that get open sourced and then they never get applied because there isn't the way to get the funding to get the money to get people excited to go and do this to go and do and build it out. Like, but there are other things that it works really well to be open source, like Ruby on Rails, right? They open source Ruby on Rails, which was like a web framework, and it was still able to be applied everywhere because consulting companies used the open source framework to sell uh consulting services so they could still make money and they could they so it still became very you know, helped a lot of people. But other things like I know a kid who open sourced a like uh a health library for like impoverished, like a very uh affordable way to like do like eye health and like health stuff with cameras, and he open sourced the whole thing and then it never went anywhere because a bunch of poor doctors in the third world like didn't have any money to like do it. Whereas if he had started a company, kept it, you know, and he had applied it to wealthy markets and then subsidized its use in the third world, he could have helped million, you know, millions of people. So I think I think it depends on the on the product. And this product, you don't want to open source it. This is something where you want to be the gatekeeper who makes millions and millions of dollars, and everybody else has to come to you and kiss your ring. And no, wait, maybe I'm just thinking of the Pope.

SPEAKER_03 57:20

Yeah, it's like Jonah, I encourage you to not take the emperor of the world position as your goal.

SPEAKER_01 57:26

No, it's not my goal. It's not my goal.

SPEAKER_02 57:28

I would I think Joan would make a good emperor of the world, human flourishing. That's a very nice man. Yeah. Of all the of all the technocrats who could take over. See, this is the problem, Jonah. People like you, people like me, people like Scott, we want to like open source things, could just get enough for our families and open source. It's the evil, wet lizard men who want to own everything, right? So good people like us, we have to try to like, you know, we have to contend with like the the the Elon Musks and the you know the psycho, the psycho ol uh billionaires, you know.

SPEAKER_03 57:58

Well, it's yeah, it's the person who desperately wants to be king that you don't actually probably want to be in the spot.

SPEAKER_02 58:04

Exactly. You're like the pneuma pompilius. Pneuma Pompilius, the second emperor of Rome, who didn't want to be emperor, but they were like, that's the that's why we are making you emperor because you don't want to be emperor. So there you go, Jonah. You have to become Emperor God King.

SPEAKER_00 58:21

No thank you. It's been decided.

SPEAKER_02 58:23

It's been decided. All right, we should probably end. This is a long episode, but I think it's great. It's all episodes. It's very fascinating.

SPEAKER_03 58:31

Good good stuff. Thank you very much, Jonah, for joining us.

SPEAKER_01 58:33

Thank you guys. It's a real pleasure.