Julie Fredrickson, Managing Partner of Chaotic Capital, rejoins Infinite Loops to talk about Montana’s Right to Compute Act, why verified human data may become one of the most valuable assets online, what nuclear power and the encryption wars teach about regulating AI, why chatbots keep recommending the same sunscreen, and what 2,089 days of daily public writing taught her about being wrong. It is her second Infinite Loops conversation, three years after Episode 181.
We’ve shared some highlights below, together with links & a full transcript. If you like what you hear/read, please leave a comment or drop us a review on your provider of choice.
Subscribe to The OSVerse to receive your FREE copy of The Infinite Loops Canon: 100 Timeless Books (That You Probably Haven’t Read) 👇👇
Links
Show Notes
Julie Fredrickson & Jim O’Shaughnessy
Julie Fredrickson: Managing Partner of Chaotic Capital, an early-stage venture fund, and a former founder in e-commerce and beauty. Chaotic Capital
Julie on X: Julie’s preferred way to be reached. Follow Julie on X
Julie’s daily writing: More than 2,000 consecutive days of public writing, started January 1, 2021. Read Julie’s blog
Jim O’Shaughnessy: Host of Infinite Loops, founder and CEO of O’Shaughnessy Ventures, and author of the New York Times bestseller What Works on Wall Street. Follow Jim on X
Full transcripts: Transcripts and highlights from every Infinite Loops conversation. OSV newsletter
Julie’s Work & Projects
Montana’s Right to Compute Act: The first state law protecting people’s right to own and use computers and AI tools, which Julie advocated for. Read the Right to Compute Act
Valar Atomics: The small modular reactor company founded by Isaiah Taylor, where Julie invested. Valar Atomics
Julie’s first Infinite Loops episode: Episode 181, “Circumstances change. Humanity doesn’t. Plan accordingly.” Watch Episode 181
Books & Ideas Mentioned
The Beginning of Infinity by David Deutsch: Jim’s reference for why no one in 1900 could have predicted the internet. The Beginning of Infinity
Carlota Perez: The economist whose work on technology booms and build-outs shapes Julie’s view that AI is still in its invention phase. Carlota Perez
Rainbows End by Vernor Vinge: The novel Julie cites for a future where people earn a living doing small tasks only humans can do. Rainbows End
Snow Crash by Neal Stephenson: Julie’s reference for how ideas spread like a virus through programmers. Snow Crash
Fall; or, Dodge in Hell by Neal Stephenson: The novel Julie describes, in which Americans mourn victims of a nuclear attack that never happened. Fall; or, Dodge in Hell
The WEIRDest People in the World by Joseph Henrich: Jim’s reference on how long cultures take to adapt to new technology. The WEIRDest People in the World
Nineteen Eighty-Four by George Orwell: Jim’s starting point for cognitive security and invented history. Nineteen Eighty-Four
From Googlebot to GPTBot: Cloudflare’s report on the rise in AI crawler traffic that Julie cites. Cloudflare report
The Vesuvius Challenge: The project reading the Herculaneum scrolls, subject of a documentary from Infinite Films. Vesuvius Challenge
Agrippa’s trilemma: Jim’s reference for why chasing a claim to its source often ends in sand. Agrippa’s trilemma
Highlights
The daily writing practice
Julie Fredrickson
Yeah. I really recommend a daily writing practice for everyone, if it’s at all interesting. Just start. Don’t apply too many rules. When I started, I thought I would be happy if I got a couple of paragraphs down, and I was aiming for 30 days. And now I’m on year five. I began it January 1 of ‘21, as a sort of mid-pandemic: what’s happening? What am I thinking? The world’s changing. And I tend to view writing as a thinking process. If it’s in my head, okay, not so bad, but I need to get it on the proverbial paper to actually understand where my understanding gaps are. And I’m so glad that I did it, not only because everything is tagged and there’s lots of useful metadata, and what my areas of interest are, how they’ve changed, but it’s also a wonderful way to discover where you were wrong.And we have a tendency to think of problems as... the stuff we got wrong doesn’t matter, it’s only the stuff we got right. And I think that’s very much what I’ve taken from it: that I am wrong so often that it’s almost embarrassing. And I don’t ever want to let go of that feeling, because now I’m no longer embarrassed by poor writing or shorter writing or having a bad writing day. Part of the reason I do it is I want to discover the areas in which either my thinking is flawed or I haven’t really thought something through. And now that I’ve got such a large body of work, I can toss it into an LLM. And honestly, coming up with a duplicate of me would be pretty simple just based on my writing now, because there’s just so much of it. But it’s very helpful.
Jim O’Shaughnessy
I think the thing that you said that I absolutely applaud, as one who has kept journals since I was 19 years old: the ability to identify where you were wrong is so important, because our memories overwrite, right, with what you’re believing.
Ask a Chatbot for Sunscreen
Julie Fredrickson
So I’m a huge skincare junkie. I love cosmetics and makeup.Jim O’Shaughnessy
You do have lovely skin.Julie Fredrickson
In a former life, that was... I came out of e-commerce and beauty and fashion, so I still really love it. But when you search for skincare advice, you’ll get served up lots of information about what’s popular on Reddit, because Reddit is a really rich source of user-generated data on caring for your skin, different problems. And there’s a brand that is owned by L’Oréal called CeraVe that is very well represented in Reddit’s makeup and skincare forums, and that if someone can dethrone... CeraVe is currently the most popular drugstore brand, at least in America. I think it’s reaching kind of different peaks other places. But I encourage people to try this. Ask ChatGPT or Claude or Grok to suggest the best SPF sunscreen or the best retinol moisturiser. And the chances are pretty good it’s going to be CeraVe, because they have simply dominated on user forums.And this actually gets to another problem. Well, that was organically generated in some places, but L’Oréal is quite a large cosmetics conglomerate, so they were able to do a lot of seeding of products with influencers who then wrote things on Reddit. And so Cloudflare has this pay-per-crawl service. They’re sort of security between you... when you search for a site, they are kind of the first line of defence, and they know if it’s a bot versus a human coming in, generally, and they can send a 404 to the bot and say, “No, you have to pay for this information.”
So datasets like Reddit that are really rich in user information about things like skincare, things that are a little bit more niche than general information, are actually now quite valuable. And the more valuable your specific professional information might be that you’re putting to people, the better off your sites and your businesses probably will end up being in terms of getting served by the large language models, because they’ve done a great job of, “Oh yeah, my absolute favourite sunscreen is CeraVe.” Now, there’s better sunscreens depending on what your needs are, but they have figured out a way to be the largest conversation on all of these forums. But what happens when those forums become synthetic, and there’s armies of bots creating the content? What happens?
Can anyone dethrone datasets that already existed with verifiably human content? Somebody launches something amazing... Korean brands are really hot right now, because this is the first time we’ve approved a new sunscreen in America in decades. So maybe that sunscreen and those new ingredients will dethrone them. So that’s just a whole mess right there.
Edited Transcript
Jim O’Shaughnessy
Well, hello, everyone. It’s Jim O’Shaughnessy with yet another Infinite Loops. I am delighted to welcome back Julie Fredrickson of Chaotic Capital. Julie, it’s wonderful to have you back. I was thinking it’s just such a shame that nothing has happened since the last time we talked.
Julie Fredrickson
It’s been a very quiet, calm couple of years. Nothing. No major world events, no investments going anywhere, no new technology. Yeah, totally normal couple years.
Jim O’Shaughnessy
Same shit, different day. But okay. With everything, with the tsunami that has happened over the last two years, what is keeping you awake, what is really annoying you, and also what is delighting you?
Julie Fredrickson
Well, I’m definitely delighted by the direction that artificial intelligence has taken in my own life. But I think the unexpected thing that I wasn’t completely prepared for, in terms of shifting, was just how much it would affect the real world, like the infrastructure and brick and mortar. And so I spent a lot more time thinking about unpopular topics like data centres and energy production than I expected, which is both exciting and annoying at the same time. I also got a lot more involved in politics. I care a lot about something called the right to compute. It started out as a little bit of a hobby project, and now we’re talking about what rights we have around using our computers of any sort, whether it’s your phone or your laptop or large agglomerations of data. It turns out we haven’t really figured out how we treat compute as a subject.
Jim O’Shaughnessy
Yeah. And do you think that there’s a First Amendment loophole way into the right to compute?
Julie Fredrickson
It’s an interesting question, right? Because I think that in this era, how could you be said to be exercising your speech rights without using the mathematics that underpin all of compute? Because it’s really just algebra. We’re just adding up lots and lots of matrices. And I tend to think of math as a subsection of speech, not the other way around. So I think that the First Amendment does apply in Montana, which is where I’m based, although I’m not there right now. Otherwise I’d have beautiful photos of mountains and not a background like this. But we passed a law that was signed into law by Governor Gianforte called the Right to Compute, which is a bit of an amalgamation of the First, Second and Fourth Amendments, because compute is a little bit speech.
But it’s also a little bit property. There’s some cybersecurity and defence implications. So in Montana, we took it as a little bit of a strict scrutiny perspective, looking back at things like the browser wars of the 90s and the encryption wars in the 80s: how do we incorporate a new technology that rests on things that we actually already have a constitutional right to interact with? So that was the perspective I took.
Jim O’Shaughnessy
Yeah, no, I like that perspective. Elaborate a little bit, because I hadn’t thought about it through that lens. I was around for all of the fights about browsers, for all of the fights about sites, cybersecurity. But this is just yet another extension of that.
Julie Fredrickson
No, I would think so. And maybe I’ll find out that other people really disagree with me, but, and I’m not a constitutional scholar in any capacity, I tend to think that rights are innate, and that we should have to come up with, they say, strict scrutiny as a standard, which is you have to make a case for why something should be banned, as opposed to why we would have the opportunity. There’s no reason, there’s no compelling reason to ban something. And we may already have laws. Fraud and violence are already proscribed. We already have a judicial body for that.
But it does seem strange that we should be taking something that is fundamentally speech and mathematical and deciding that it needs a whole new body of laws, that maybe we’d want it to build and have a better understanding of what areas of life it’s going to impact before we decide to write a whole bunch of new laws around how it can be used, who can use it. Obviously, regulatory capture was a big part of the concerns around the browser wars in the 90s. In terms of the encryption wars before that, America opened itself up to some troubles. It was more hackable because we put back doors into things, and that turned out to be a real problem, and we would never have enabled the internet as we know it today without strong encryption. And we almost banned it. And now can you imagine not being able to shop online, or being able to do e-commerce transactions, or take meetings safely? So we don’t always know how things will come together. And that’s part of the reason I think we should lean back on the systems we already have before deciding, “Oh no, we need a whole new body of regulations and moratoriums.” And I don’t know, it’s The Hitchhiker’s Guide to the Galaxy: “Don’t panic.”
Jim O’Shaughnessy
It reminds me of the Warren line: “It is a mistake to understand new things too quickly, for often we have misunderstood them and we end up killing them in their cradle.” Your reference to the cybersecurity or the encryption is the perfect example. I remember all of the passion on each side of that, but we didn’t know at the time. And that’s the problem. As I watch these wars going on online between the effective altruists and the e/acc types, it’s just like, guys, if you’re in favour of AI, that does not make you opposed to AI safety. You can’t say, if you are in favour of AI safety, then you must follow our manifesto. And I think that either/or is not an argument in good faith.
Julie Fredrickson
Yeah. And I’m actually a little perplexed as to how we came into that frame in the first place, because to suggest that wanting to bring forward benefits quickly is inherently more dangerous isn’t necessarily true. And we have seen... actually, I’m going to brag a little bit here, but in terms of production on energy over the last two years, we have moved on nuclear relatively swiftly, in ways that we have not done for decades. There’s now four new reactors that have been approved for use in the span of two years. And previously it would have taken a decade to get a new reactor to criticality, online. And this is just to say that just because something is faster, or we’re accelerating into something, doesn’t necessarily mean that it won’t be safer. Sometimes you want to lean into the curve in order to make it out the other side. So I tend to think of them as not being either/or. Safe and fast actually is probably more likely than safe and slow.
Jim O’Shaughnessy
Yeah. And I think one of the other problems is we don’t know yet what problems are going to present themselves. If you’re imaginative, you can come up with all sorts of scenarios that get you to Terminator or get you to paperclips, but I don’t think that necessarily is the way it’s going to go. And so I find myself often in the very awkward position of being what I call a rational optimist. Of course there are going to be problems. We’re not denying that, but we think that we’ll come up with solutions to those problems as they present themselves. I mean, am I just naive here?
Julie Fredrickson
Well, and I’ve definitely asked myself that as well, because I have kind of strongly gone in favour of the... I think compute is as much a right as speech or property rights. And people will say, “Well, but there’s all these new ways that we could commit crimes.” Humans are very ingenious. They come up with new ways to commit crime all of the time. But we don’t necessarily need to have new punishments for old crimes. Theft is still theft, murder is still murder. And the idea that we can’t come up with solutions as problems arise just sort of flaunts a whole body of history in which the only way we actually solve problems is by experiencing them and mitigating them, because we couldn’t always imagine what those problems were.
I don’t think it’s naive. I think it’s actually just very practical. If we imagine a problem that’s too esoteric, we’ll fix the wrong problems, and then we’ll be blindsided by something as it happens, because we took too long to actually implement something.
Jim O’Shaughnessy
Yeah. And I’m guided by the way David Deutsch looks at it. He has that great quote in The Beginning of Infinity, where he says: what were thinkers and scientists thinking and saying about the internet in 1900? Nothing. Nothing. Because it wasn’t around. And the whole idea that we can prophesise what is going... That is one of my things that really does trigger me, where somebody puts a completely speculative prophecy up. I mean, just think of all... And media loves it. They love doom. They love to scare the shit out of people.
Julie Fredrickson
If it bleeds, it leads.
Jim O’Shaughnessy
Exactly. And all those terms are all pre-internet, and newspapers did the same thing. Go all the way back to the Federalist Papers during the American Revolutionary War. Newspapers were incredibly biased, and they were propaganda sheets, essentially.
Julie Fredrickson
Yeah. Like, yellow journalism is from a war a hundred years ago.
The problems are almost comically human in nature. Any time I watch a new sci-fi conceit in which they’re saying, “Oh, we’ll have reproducing nanobots and we’ll be turned into grey goo for food for them,” okay, I mean, that’s going to require a lot of power that we don’t currently generate, a lot of compute that we don’t currently have made, a lot of chips. I love all of that science fiction, I really do. I’ve read so much of that kind of work, but I just don’t... And I think we’ve discussed this too: my attitude has always been that circumstances change but humanity doesn’t. Humans remain human.
Jim O’Shaughnessy
Yeah, circumstances change minute by minute. Human nature doesn’t budge millennia by millennia. And that was always what I would say about investing. Your edge is arbitraging human nature, because it literally does not change.
Julie Fredrickson
Might be the only arbitrage that doesn’t change quickly. Everything else fades and can be moved. But humans are surprisingly stubborn.
Jim O’Shaughnessy
So what would you propose? If we made you the AI czar, right, what would this current debate look like if you could say, no, okay, now we’re going to have a long discussion, an honest-minded discussion? If you were the czar, tell me who you would invite, why, and what you would hope to accomplish and achieve.
Julie Fredrickson
Well, I do think that leaning on what we know, because history doesn’t repeat but it rhymes, would be very helpful. I think that bringing people in who have studied the encryption wars, the browser wars, because they’re with us, that history is not that far back, that would be an important construct to understand how we’ve integrated new technologies in the past, because we’ve got quite a bit of that. And that’s one reason that I just simply leaned on... I think the American Constitution is a wonderful experiment and has worked pretty effectively. We had a 250th birthday this year. Perhaps constitutional scholars would be more appropriate to bring to the table than every type of specialist.
But on the flip side of that, as much as we need to lean into the systems that we already have and understand and have worked well for us, there are types of risks we can predict. We understand that humans are prone to certain kinds of diseases. You want biosecurity. You would want to think about all of the security risks that we already deal with from cybersecurity. Just basic things, human error. So much of hacking is people testing the limits of your own human frailty, and it has absolutely nothing to do with the hardened walls of actual encryption. So I think being very precise about what problems we think are more likely, and then building the pipeline of specialists. But then also, I don’t know if we need a FINRA for artificial intelligence yet. How much business was conducted in America before we incorporated FINRA into our banking system? Quite a lot. It’s the same for the Securities and Exchange Commission. We did business for a long time without it.
Jim O’Shaughnessy
And I would push back a little there. That’s my old world. And literally most of Wall Street pre some of the regulations, and remember, this is coming from a very free-market person, a lot of it was literally rigged, right? Bucket shops were rigged. Pools would rig security prices. They would get together a group of insiders on Wall Street, they would run the price of a security up to ridiculous levels, and then that rise attracted the attention of the naive public. They would all try to jump into that particular group of securities, and guess who was selling those securities to them? Those very same pools. So, well, I’m very sympathetic to: yes, of course, much of our country was built pre all of the various regulations.
I’m not opposed to just sensible regulation. But I also think that you’re making the case that we already have a lot of sensible regulations that could be applied to AI. What do you think about the idea, though, that this really is a step change in technology? We have our own on-prem hardware, software. We’re developing a series of programs that we think will be commercially successful. And being able to do business the way we are now doing it is literally very different than when I started my first company in 1987.
Julie Fredrickson
I do think that it is significant. I am, however, not sure that we’ve reached the significant portion of the developmental process. I’m a big fan of an economist, Carlota Perez, who writes about boom and bust cycles. I’m not actually sure that we’re in the deployment phase yet. The idea of the boom and bust cycles that she talks about is that there’s sort of two parts of build-outs. One is the innovation and the invention, and that step is then followed by a build-out. And I think we may actually still be in the invention stage and not the build-out of the success phase. I think there’s still actually quite a bit of innovation that we haven’t really figured out.
Not to say that large language models are not incredibly innovative, but I’m not convinced it’s the last step in how we’ll actually put together successful businesses, and what we’ll consider success might change rapidly as different areas of the industry either commoditise or become much more driven towards service and specialty. And what the bar of commodity versus specialty is, I think, is unclear to me still. And this is kind of speaking as someone who, I’m not invested in the labs in any capacity, but I care an awful lot about making sure that we have energy and chips to actually be able to run those things. So I do think they’re important enough that we obviously need more infrastructure, but I don’t think that they’re the end game.
Jim O’Shaughnessy
Yeah, I actually agree. I think that we are in the developmental stage now, but even with the tools as they exist today, you can develop some really interesting and pretty cool new ways of doing things that simply were impossible before. I do agree that you’ve got to be committed to pivoting. You’ve got to be committed to being really fast at updating your priors, and you need that kind of flexibility. So now put your investor hat on. What are you looking for now in terms of the entire... and it doesn’t have to be just the AI stack. You mentioned energy and data centres. What else are you contemplating in terms of investment?
Julie Fredrickson
The biggest problem that I’m starting to see is actually that the inputs are not as openly available as we might hope. And this is kind of a function of: the way humans browse the internet is not the way machines are browsing the internet, and the ecosystems, and frankly the economies, of the open web are changed simply by the scale. It’s actually, I think, kind of astonishing to me just how significant the scale changes. And I’m trying to pull up some notes I had on it, because what we’re starting to see in terms of Cloudflare is actually probably a great example. There’s an 18% rise in AI and search crawler traffic, they said in their blog recently, from May ‘24 to May ‘25, and then immediately after that it rose by 305%.
So if you compare the Cloudflare AI search crawler traffic in that period, ChatGPT traffic rose by 2,825%. I’ll send the links along, but the machine-readable web is an order of magnitude, orders of magnitude, larger than the human-readable internet. And I think the web itself just wasn’t built for agents and the level of speed that they operate on. And that’s probably going to put some real pressures on... National borders are part of it, where not everything is machine readable depending on what region you’re in. I’m actually in Prague right now, and the European Union has a fairly different approach to how they handle data and speech to how America does. There’s no First Amendment where we’re at.
And all of that problem of what datasets are machine readable versus human readable, I think, will probably really factor into what kind of products win, because the novelty of information is actually a huge component of how people build edge and actually build out businesses that can be sustained. And I’m really curious to see what those unique datasets will be, but also, how do we preserve access to the information commons? Because if you’re not actually going to the pages that were serving you information, but you’re just getting it from a chatbot, that’s a very different transaction of information than what we had previously.
Jim O’Shaughnessy
Yeah. And I saw today that Amazon is shutting down the ability of Meta’s Muse to shop at Amazon. And it reminded me of, and I think I got my thinking wrong here actually, but it reminded me of when VHS came out, and the movie industry spent years and millions of dollars trying to kill VHS, without understanding that it would be the massive source of revenue for them that it became. Yeah. But what do you think about that? I know people who are using Muse. In fact, I had a conversation last night with two early users, and they were like, “It’s amazing, it’s magic,” in terms of the shopping. But my initial insight, seeing Amazon shut it down, was like, “Well, that’s not going to go well for you, Amazon.”
Julie Fredrickson
Yeah, actually it’s such a great example too, because it opens up a whole new stream of business. So I’m a huge skincare junkie. I love cosmetics and makeup.
Jim O’Shaughnessy
You do have lovely skin.
Julie Fredrickson
In a former life, that was... I came out of e-commerce and beauty and fashion, so I still really love it. But when you search for skincare advice, you’ll get served up lots of information about what’s popular on Reddit, because Reddit is a really rich source of user-generated data on caring for your skin, different problems. And there’s a brand that is owned by L’Oréal called CeraVe that is very well represented in Reddit’s makeup and skincare forums, and that if someone can dethrone... CeraVe is currently the most popular drugstore brand, at least in America. I think it’s reaching kind of different peaks other places. But I encourage people to try this. Ask ChatGPT or Claude or Grok to suggest the best SPF sunscreen or the best retinol moisturiser. And the chances are pretty good it’s going to be CeraVe, because they have simply dominated on user forums.
And this actually gets to another problem. Well, that was organically generated in some places, but L’Oréal is quite a large cosmetics conglomerate, so they were able to do a lot of seeding of products with influencers who then wrote things on Reddit. And so Cloudflare has this pay-per-crawl service. They’re sort of security between you... when you search for a site, they are kind of the first line of defence, and they know if it’s a bot versus a human coming in, generally, and they can send a 404 to the bot and say, “No, you have to pay for this information.”
So datasets like Reddit that are really rich in user information about things like skincare, things that are a little bit more niche than general information, are actually now quite valuable. And the more valuable your specific professional information might be that you’re putting to people, the better off your sites and your businesses probably will end up being in terms of getting served by the large language models, because they’ve done a great job of, “Oh yeah, my absolute favourite sunscreen is CeraVe.” Now, there’s better sunscreens depending on what your needs are, but they have figured out a way to be the largest conversation on all of these forums. But what happens when those forums become synthetic, and there’s armies of bots creating the content? What happens?
Can anyone dethrone datasets that already existed with verifiably human content? Somebody launches something amazing... Korean brands are really hot right now, because this is the first time we’ve approved a new sunscreen in America in decades. So maybe that sunscreen and those new ingredients will dethrone them. So that’s just a whole mess right there. It might be better shopping to allow the bots on, but what are the problems?
Jim O’Shaughnessy
Yeah. As I was listening to you, I was thinking, wouldn’t it be ironic if one of the biggest new job categories was you being a human, opening an account at Reddit and actually typing yourself, not having the AI type it? But I could see, if you were a large-scale company like L’Oréal or any of their competitors, why not hire 10,000 people?
Julie Fredrickson
An army of young people.
Jim O’Shaughnessy
An army of young people. Give them a script, give variations, let them... or even better, don’t give them a script.
Julie Fredrickson
Just give them the product. Let it be.
Jim O’Shaughnessy
Yeah. “Here’s the product. Say really great things about it and we’ll pay you.”
Julie Fredrickson
So I genuinely have one. There’s a fantastic mathematician who recently passed, Vernor Vinge, who writes science fiction novels, actually well known for his work on the singularity.
Jim O’Shaughnessy
Yeah, I know Vinge’s work.
Julie Fredrickson
So in Rainbows End, which has an interesting plot about deconstructing a library and the value of a dataset, a millennial grandparent is trying to make the transition with their grandchildren. And the grandchild’s profession was somewhere between an influencer job and a gig worker job, in which they were creating little bits of content, or little bits of things that the machines couldn’t touch, but only humans could touch. And I think that actually might be closer to what happens: that there’s going to be so many different things that are human mediated, that being very literate and very capable and interacting with real people, those skills probably do become more important, and we may actually care a lot more. There’s the joke, “You should have studied STEM and not English.” Well, I think there’s a good chance that people who are better storytellers actually do quite well in a world in which unique data, and human unique data, is the most important asset on the internet.
Jim O’Shaughnessy
Revenge of the humanities.
Julie Fredrickson
And I think it’s becoming a rather popular theory that actually being human is going to matter quite a lot, because synthetic machine attention is not the same thing as a human having seen it.
Jim O’Shaughnessy
Yeah. The other thing I wonder about is datasets. Again, given my old world, the dataset was the most important thing. When I was still at OSAM, before we sold the company, we did a three-year data cleanse, because people had no idea how dirty the data is. Literally. And I would give the example: pull up the best-known company, pull up at the time Apple or Amazon, and just look at the P/E ratio. Okay, this is what you get from Bloomberg. And now let’s go over to Reuters and pull up the P/E ratio, and it’s a different number. And any financial data stream you find, different numbers. So we elected to spend a lot of time and treasure cleaning the dataset.
But then when you talk about Reddit... I’m not on Reddit, so I’m naive about the quality of the posts and everything on Reddit, but it’s kind of the same thing, where you have a self-selected sample bias, and the type of people who spend a lot of time doing reviews, et cetera, you could probably map over to their Big Five profiles, et cetera. And that is something that I could see you building. You could build an algorithm that looked a lot like a human.
Julie Fredrickson
Yeah. Well, and how fast do we do that? How much is it going to cost to duplicate what, in marketing land, we would call a psychographic: the idea of the person’s life, and what their house looked like, what their family looked like, what car they drove, what brands they had affinity for? We’re already actually quite good at that kind of work in commercial advertising, and I imagine we’re only going to get better at that. But then, and this is where I think it’s going to be very hard, to decide what actually reaches a human at that point. And maybe we do have to have some verifications to suggest that it’s a real person at the end of it. And that presumably is going to involve any number of different kinds of checks. And I don’t know how much people follow things like know-your-customer rules and KYC-type problems, but proving your identity is going to become a much bigger problem over the next few years, I think, because of it. And given the cybersecurity implications, do you really want to be uploading your passport to a commercial website?
Jim O’Shaughnessy
And what do you think about zero point proofs and things like that, where you can do it right on your device itself?
Julie Fredrickson
I think that’s probably our only real option. The zero knowledge is going to... And I would encourage people to actually learn about how they work and why it’s important, because the idea of trusted intermediaries is just... as far as I’m concerned now. And being able to have trustless systems. And this is probably where crypto goes, right? We spent so much time thinking about what are functionally financial products, when in the end... not to say that the financial rails and stablecoins and things of that nature aren’t huge and very important, but actually being able to do a zero-knowledge proof and say, “This is the entity you’re interacting with,” and be confident that you’re getting the person. I think we’re very close to needing that for almost everything we do.
Jim O’Shaughnessy
I agree. And I also agree with the assessment about... and it illustrates the importance of being able to pivot, right? Crypto, blockchain, et cetera: that piece of it might be the most important thing to come out of it.
Julie Fredrickson
Yeah. Which, it’s so funny that we have gone through... It’s the same cycle of: things start looking like toys. You’re like, “Oh, this is so fun and so silly. Oh, those goofy pictures.” And then you realise that there are dead serious problems of making sure that who you are is who you are to the computer, and is readable to machines. And this is where I was going with the question of what do the datasets look like, and who is integrating information, and at what scale? Because the further we get from source material into digestions and summaries, the more likely it is that the abstraction doesn’t reflect what the material at the source says. And that’s just going to be its own other problem.
Because I do think we’re experiencing fragmentation at a geopolitical level. The internet in Europe looks very different than the internet in America. Russia and China have totally different sets of systems. And there’s even stateless data now, where we think of the fall of the Soviet Union as being so far past, but there are Soviet patents that are not technically owned by anyone. And have we done anything to bring them online, make sure they’re machine readable? What do we do for all of those types of cultural and heritage institution types of information? The Herculaneum scrolls are an amazing example of people coming together and saying, “Wow, now we can read this data from Pompeii and Herculaneum.” But what other datasets exist that are like that? And how do we make sure that everyone has open access to them? Because we see patent data as being an open-access problem. But what if the state that issued the patent is no longer there? The Soviet Union’s not there. It’s a set of Baltic nations and Russia and Balkan nations.
Jim O’Shaughnessy
Our Infinite Films studio did a documentary on the scrolls, which I am very excited by, because to me at least, it is incredibly exciting to be able to finally make legible, through a particle accelerator, and then translatable, through AI, these ancient texts. And I’m also really interested in all of the papers, works, et cetera, that have not been digitised. For example, there is all of the works of William James. We actually have a project going right now where we’re at the libraries up in Boston, trying to digitise all of the works that are on paper that have never been digitised.
Julie Fredrickson
Yeah. And most information is in that process now, where UNESCO has been sounding the alarm for years that we simply have not digitised most of our historical artefacts, because they’re individual countries’. And how do we make sure that everybody has the credit and the lineage and information, while also still recognising that it belongs to the people of these nations, or the owners of the copyright? And perhaps there’s more value in them being broadly reachable. To use the VHS example, it might turn out to be whole new businesses that we haven’t even really seen.
Jim O’Shaughnessy
Yeah. And on that front, and we’re talking about datasets and everything, kind of back to the cosmetics illustration you gave: I myself might prefer a dataset where it takes every sunscreen, since that’s what we were talking about, sends them to a lab, has the lab results come back: “These are the most efficacious. These have these awful ingredients, you probably don’t want to buy these. These are the top three, based on the underlying ingredients and the efficacy that multiple tests or studies have shown.” I would trust that. If I was choosing a sunscreen, I would go to that rather than what a bunch of Reddit users think about it.
Julie Fredrickson
Yeah. Well, and people are building websites like that using user-generated datasets like Reddit. And I think that actually gets to be a little bit scary, is that we have to think about, okay, well, who owns the academic publications? Who owns the user-generated information? I love going PubMed spelunking, because you just never know quite what you’re going to pull out of the scientific corpus. And being able to make those distinctions of, all right, where did the data come from? What’s its national origin? Who owns it? Who paid for the studies? Because all of the context of the information is probably quite salient to making those choices. And it sounds a little silly, but this is totally how I think about cosmetics: what’s my Pareto-optimal routine? And if you want to spend... get that 80%, and it’s good enough. But maybe you’re a completist and you want to have the 100% most perfect skincare, because you don’t want skin cancer and you want to age gracefully, where all of that context might actually be really important to going from Pareto optimal to complete. And my guess is that there’s probably markets like that everywhere, and cosmetics is just an easy one to think of.
Jim O’Shaughnessy
Yeah, I would agree. But I also agree with the follow-the-money. You bring up the various studies and the spelunking you do at the various academic sites, and I was remembering, in the 80s, when homelessness all of a sudden became a huge issue. And I can’t recall the author, but somebody wanted to find where the papers that said there were... I can’t remember the number, but it was a big number. It was like 30 million people homeless in the US. And this is all back in the 80s, pre-internet, et cetera. So this guy was literally obsessed by, “Okay, where did that number come from?” Because he found it cited in all of the academic papers written about it. And then he would say, “Okay, well, this cites this paper, I’ll go to that paper.” Then he would go to that paper. Yeah, exactly. And it would cite... And it turned out, and this might be apocryphal, so let me insert that, because I never did a deep dive on whether this was actually proven...
Julie Fredrickson
But apocryphal stories prove the point regardless.
Jim O’Shaughnessy
Yeah. What this blog that I read long ago concluded was, he finally got the number from either a high school or college student that made it up, that pulled it out of the air. And the reason I loved that story was because of the Agrippa dilemma. All of these systems, when you keep trying to find bedrock, it’s just sand and mud and sand and mud. And I think that actually using AI to do that... because obviously that guy was a little obsessed to be willing to go find every one of those papers, but that’s easy peasy for AI, no?
Julie Fredrickson
Well, and it isn’t, because not all datasets are machine readable. And this is why I mentioned stateless documents, the Soviet patent system. There’s no one library that has all of this information. And so actually being able to send a computer might not be an option, depending on what you’re trying to access and where. And I definitely think this is going to be... in my head, one of the entrepreneurs I’ve been working with, we’ve been calling it the Open Information Project: how do we actually have protocol-level accessibility of saying, “This is where the information came from, this is the provenance,” and all of the categories? Just in the way that Wikipedia will show all of the different aspects of how something was comprised, even if you don’t necessarily agree with how it was written, at least you’re given the sourcing and the origins. I think that sourcing and origin portion of the internet not only doesn’t exist yet, but it’s probably going to have to be at a supranational protocol level, because not everyone is going to be obsessed.
And I’m sure if I were a nation state running an intelligence agency, I’d be thinking of all of the possible ways of dropping information that becomes apocryphal and then becomes truth. There was a great Neal Stephenson book that had a premise where there was a set of Americans who really believed that a nuclear explosion had happened, when in reality it had not. And they celebrated a memorial for the victims of a disaster that never occurred. And I wonder how often that will happen as we go forward, where you’ll be so certain that something happened, but your ability to actually track down the source of it might be more limited than you think.
Jim O’Shaughnessy
Yeah. And that is a very deep rabbit hole, because literally, if you study history, society has been astroturfed since Jericho put up its walls, right? I mean, the idea that people who want to run society have not been inventing all sorts of things... I think of ancient Babylonia, and they wanted to rule Mesopotamia, and so they were like, “We’ve got a puny god. We’ve got to do a complete rewrite of our god.” And they did. And so it went from a puny, kind of regional god to the god of all gods. The god that contains all gods, who created the skies and the earth and everything.
Julie Fredrickson
And it worked. Yeah. And also, a great Neal Stephenson sci-fi book on that one, Snow Crash, where Babylonian... and they inserted that meme set into computer programmers. And I sometimes jokingly wonder how much of the effective altruism and rationalism and effective accelerationism is actually just Stephenson’s Babylonian memetic cult in our own interpretation, where it was like we could kind of imagine the ways that linear, programmatic, deterministic thinkers would operate. And it’s the perfect meme to have massive distribution. And now here we are thinking, do I actually want to live in a Bentham utilitarian, everything-has-a-utility-maximising approach? I don’t think I want that. I’m not a utilitarian.
Jim O’Shaughnessy
Nor am I. He had real style, though, in having his body preserved, and I believe it’s still on display at some college in London. Jeremy Bentham. And when you go back and you look at what he came up with, you’ve got to do it with context. I, like you, am not an adherent of utilitarianism. But when you go back and look at his Panopticon, the reason he recommended a Panopticon was that at the time prison conditions in the UK were horrific, and his idea in that context made a lot of sense, right? So it doesn’t mean that when you switch context to the modern world... well, then the Panopticon doesn’t make a lot of sense at all. It becomes the all-seeing eye of Sauron, and a couple of hands control it, and we all live in a totalitarian nightmare.
But I’m also fascinated by the idea, because of what we’re talking about right now, of invented history. I’m sure you’re a fan of 1984. The Party’s last and most enduring command was: ignore the evidence of your eyes, ears, et cetera, and believe what we tell you. And I’ve been thinking a lot about cognitive security. Again, this is coming from someone who is incredibly pro-AI. I think it’s the technology I’ve been dreaming about since I was a young man, and yet it will be able to do persuasion at scale, right?
Julie Fredrickson
I think it already is.
Jim O’Shaughnessy
Yeah, I agree. And propaganda 101, it’s real simple: repetition, repetition. And people then suddenly take that into their ideas that can’t be questioned. And how do we prepare for the onslaught? Again, because we’re back to humans, aren’t we? If I’m an evil genius and I want to control the world, I’m going to use these tools to their maximum effect. I’m going to have the psychographic profiles of my target audience. I’m going to try as best I can to have them begging me to protect me, to keep me safe, and “I will give up any degree of freedom to feel safe.” And then, well, you think, maybe there’ll be the white hats who are like, “No, this is what you’ve got to do.” But I don’t know that the average human right now is ready for what’s coming on the cognitive front.
Julie Fredrickson
Yeah. And it’s actually quite scary, I think, because I guess I consider myself a creature of the internet, and so I’m inoculated against all sorts of... There’s been a lot of techniques over the years, and the deep dark recesses of the internet have some pretty gnarly things. Stuff that percolates up from 4chan up into Reddit and then over to Tumblr and then up to Instagram, and then suddenly it’s on Facebook and everyone has access to it.
Jim O’Shaughnessy
Well, if it’s on Facebook, then okay.
Julie Fredrickson
Yeah, that must be real, right? And I think younger people may actually be better equipped. And this will be a problem where millennials and Gen X and boomers, we will all have trouble realising the degree to which we’re swimming in different memetic propaganda streams. And Gen Alpha will have been simply born into it, and they’ll be like, “Oh yeah, no,” but they’ll bat away all of the problems, because they’ll understand it in a way that we just don’t. But I do think that one of the best ways to manage cognitive security is simply to question everything that comes at you. And this is Infinite Loops, right? If I don’t question everything, I will continue, and I will never get out of the set of noble lies that I don’t even see. I’m a fish and I don’t know that the water’s nice.
Jim O’Shaughnessy
Yeah. And illustrating that, and getting people to actually do it, that’s the problem that I see. And when you’re in a liberal democracy, you have a lot more attack vectors available to bad actors than you do in a more closed society. Over the weekend, I reposted a thread I did on, for lack of a better word, cognitive security, and the ability of mimetics to really... Mimetics are very powerful. And anyway, then I thought, “You know what? I should give this thread that I wrote, last year or two years, three years ago, to our AI and tell it, ‘Hey, tell me where I’m wrong.’”
And one of the things that it came back with is, “You’re only looking at one side of the liberal democracy, the attack vector space, which is larger, but it also has a much larger corrective space.” True, true, but the corrective space requires that there aren’t all sorts of regulations telling you, “No, no, you can’t say that.” And so it is a conundrum. It doesn’t make me less optimistic. It makes me just more aware that we are really going to be facing a ton of challenges, I would say, that we really haven’t, in a compressed amount of time. I mean, the Industrial Revolution was one thing, but it took like 100 years. People had the opportunity... The WEIRDest People in the World, the book, talks about cultural lag. It takes humans and society a time to adopt.
And I like your thought about Gen Alpha. All my grandkids are Gen Alpha, and I was with my grandson, who’s oldest, last night, and he was correcting me on a bunch of stuff: “Oh no, Papa, it doesn’t work that way. It works this way.” So I think you’re right about that. But I also think, do we get to a certain stage in life...? Let’s take TV news. I am a marginal boomer. I like to argue that... I was born in 1960, and the traditional definition of boomers goes up to 1964. But the demographer Jonathan Pontell posits that the people born between ‘56 and ‘64 virtually share nothing in common with the boomers from 1946 through 1950, whatever. And so, boomers... I’m the youngest, so all of my older siblings still watch TV news all the time, and they believe it, because they were brought up on Uncle Walter Cronkite: “And that’s the way it is.”
And I made a decision in 2010: I’m not going to watch any TV news anymore, just because you could see that it’s patently false, especially when it touches on something you know a lot about. I got interviewed a lot back then, and I would look at what came out of the grinder, and I just got to the point where, if this is what happens to something that I know I said, and then it was done this way, I am opening myself to just huge bias if I continue to watch TV news. So literally, I don’t. I mean, I see clips, anything that’s put up on Twitter. But one of the effects was, like a month later, it was like this brain fog lifted. It was really quite exhilarating. But you’re not going to get people who are in the habit of keeping their favourite propaganda channel... And that’s the other thing. I’m not saying this in a partisan way at all, because MSNBC is for the left and the blue, and Fox is for the red and the right, and they’re both propaganda machines. And again, I even try to say, if you look back at the Revolution, that’s the way it was back then too. And so it’s not like it’s anything really new. But trying to navigate, especially for the different generations, I think it’s going to be a Herculean task. I don’t know if you’re probably too young, but have you seen the movie Network?
Julie Fredrickson
Oh, yes, I have. My mother loves it, so she’s got good taste.
Jim O’Shaughnessy
So Howard Beale, the increasingly insane newscaster: “Turn off your TVs right now. Turn them off. Turn them off.” And I’m like, yeah.
Julie Fredrickson
But what are we tuning into after? And this is where context and information provenance... And how do you even become a critical thinker, where we had things like civics lessons and media literacy, and those turned out to just be different variants of, different angles of, propaganda? And I was recently in a hotel that had Iranian state TV, Russia Today, Russian TV, all of the European channels, and Fox and CNN. And it was just such a strange thing to go from Iran, Russia, America, and realise everybody is always consuming some different perspective. And maybe that is ultimately the new context of the Panopticon: that we actually might be outside of it, looking in.
Jim O’Shaughnessy
Yeah. It fascinates me because, as you alluded to, you can’t analyse the water in which you swim, right? And there are a million, probably many million, strings that attach us to the culture in which we exist. And so I just think it’s fascinating. I think we will come up with techniques that make it easier for the average person reading or watching or whatever. But I think it’s a huge issue that a lot of bright people like yourself should be thinking about, because I think the next 10 years are going to be quite tumultuous.
Julie Fredrickson
The great weirdening has accelerated. I guess maybe the accelerationists are simply pointing out the facts on the ground: that things are going faster, and it doesn’t appear there... I have not seen a handbrake for it. I mean, you can’t stand athwart history and yell stop, right?
Jim O’Shaughnessy
So we must, as we always do, muddle through. Carry on, carry on. Stay calm. Don’t panic, as good old Doug Adams would have it. Let’s move on to what is exciting you right now on the investment front.
Julie Fredrickson
Well, gosh, I have had spectacularly good luck over the last several years.
Jim O’Shaughnessy
And I’m going to not let you be modest, and say you’ve been very talented over the last several years.
Julie Fredrickson
Well, it certainly helps to be early. It always seems a little bit silly to say that I’m a fan of the movie Margin Call. There’s only three ways to make a buck, and it’s just a heck of a lot easier to be first. And my risk tolerance for the highly unusual has, I think, stood in good stead. Of the companies that made it, it was the wackiest ones. In particular, I mentioned nuclear earlier, where simply everyone that I talked to said, “There’s just no way nuclear can go any faster. Small modular reactors are absolutely not going to happen. And if they are going to happen, it’s certainly not going to be youngsters. It’s got to be an old, serious team.” And I was able to work with a young founder from Idaho, Isaiah Taylor, who founded Valar. And it was three years to get to the point in which he was able to... this is the summer he raised a billion-dollar round from Sequoia. And I was there when it was... he hadn’t even started it yet. He was on his prior startup.
And to see something happen in the physical world that fast was just an astonishing process, because it was like 24 months of everybody thinking that we were absolutely crazy, nothing was going to get done. And then a couple of regulatory hurdles were removed. The government became modestly more... instead of being in the way, actually our national scientific labs are quite helpful, and it actually turns out we have amazing scientists working for us as Americans. And seeing how quickly all of these young nuclear companies, not just Valar, which is where I invested, but we had multiple reactors that went critical and were able to start producing energy by the 250th birthday, after it was like a year in which regulations changed to... We turned things on. And I had not previously experienced that speed in anything other than an internet or information or financial company before. And to see what’s functionally... I shouldn’t call it a factory, because they’re not really factories yet. But seeing a reactor get built from scratch in a year was just, wow. Actually, the world is so much more malleable than we give it credit for. And I’m not necessarily sure that it will always remain that way, but because we could do so much more of the work, the paperwork and the research and coordination and all of the work that goes into building a business, it is now expedited so significantly by artificial intelligence tools that the actual human work goes faster. It’s Steve Jobs and computers, like riding a bike. And it’s a lot faster to bike.
I grew up in a world in which things only happened fast on the internet or in finance, but you certainly didn’t build hard infrastructure quickly. And to know that you can do that, and that so much of the slowdown was a function of rules and regulations and didn’t necessarily make anything safer or better, might have been the most important lesson that I got over the last three years: that someone young and motivated, with great tools, who really wants to get something done, can find the people who want to support them.
And I was just so impressed simply by watching American scientists who work for our national laboratories kicking in and saying, “Well, heck yeah, we do want cheaper power, and nuclear is great and clean, and we have so much demand. Let’s turn on however many reactors can do it. We’re going to commit the resources of American scientists to actually helping companies build more reactors.” Which for me was just, oh, wow. We can build websites fast is one thing. We can build nukes fast is a whole other ball game.
Jim O’Shaughnessy
Yeah. And it goes back again to our earlier chat: if the society and the zeitgeist in which you grow up is one where nuclear power... When I was a kid, it was just axiomatically true that nuclear power was incredibly dangerous, and there was no evidence that would sustain that conclusion. There was evidence that was like, we’ve got to be really careful about our safety protocols and whatnot. But Chernobyl and Three Mile Island, and more importantly the fictional accounts of Three Mile Island, turned society vehemently against nuclear power. And how many decades of lost power resulted from that? And I kind of think we’re doing the same thing with data centres right now.
Julie Fredrickson
I am concerned that’s what’s happening as well. And it is only an accident of circumstance, I think, that we were able to pass the Right to Compute law in Montana, and that I had initially put it forward as something for cryptography: that we want to protect things like privacy and autonomy. And that had been the original thought: well, we should have a right to compute in the same way that we have a right to speak your mind. And there are so many other approaches that one could take to any technology. But if we don’t start from a place of thinking, how can we develop actual safety, we’ll never get the chance to do something, because we’ll never build the thing to find out the problem.
Jim O’Shaughnessy
Yeah, the precautionary principle has literally sent various societies to ruin. Like, literally to ruin.
Julie Fredrickson
Well, and I think the precautionary principle in nuclear really did set us back. We could have much cheaper power, and we wouldn’t have had the full-on effects of... how is carbon input into the system affecting our weather? Is global climate change something that we could have avoided entirely if we had simply gone on this different route? Would we have different problems? Maybe it would have gone in a somewhat different direction. Apple TV just continues to produce great science fiction, and their For All Mankind just had a totally different set of problems because they invested in nuclear.
And I just so sincerely hope that we don’t get that problem with artificial intelligence: that out of fear of all of the possibilities, we have a precautionary principle that stops us from something that could really do genuine good.
Jim O’Shaughnessy
Yeah. Again, history gives us mixed results, and I think it’s great to illustrate what happened. The Chinese invented gunpowder, and they didn’t make use of it the way the West did, and suddenly you had this sea change in the power dynamics of the world. And I think that thinking that you’re going to be able to prejudge which technology, which innovation, where it’s going to lead, I think is delusional, honestly. But at the same time, I very much believe that we should, as we began our discussion, have honest-minded discussions about the potential of what might happen. Like, 80% of it could be wrong. And obviously those discussions should be ongoing. It shouldn’t be like we’re going to do a big seminar, and then we’re going to publish the results, and then we’re going to be done with it. No. It should be an ongoing process that brings together a diversity of minds and points of view.
Julie Fredrickson
Well, because when they speak about alignment, the next question everyone asks is: aligned to what? And then we have to answer the question of, well, all of us. And if there’s only so many models, and we’ve got such a strong regulatory stance that you’ve got five choices, which is better than one choice, but it’s probably not necessarily... what if in that five you would prefer to have your own model and not a hosted model? And how do we make sure that people have the tools that they want for their own job, and that you have a right to be able to build and own the fruits of your work? And this is where I think open source just remains the clearest indicator: that the internet and most of our, all of our computers are built on the body of work of open source. Most of the world is powered with Linux.
Jim O’Shaughnessy
Yep. And Microsoft did everything they could to kill Linux, right?
Julie Fredrickson
Yeah, they really did.
Jim O’Shaughnessy
So people who are looking at the labs now, the frontier labs now, it’s the same playbook. And if you look at it from their point of view, well, of course, if I’m going to be that type of person, I’m going to go for regulatory capture all the time, because it makes things a hell of a lot easier for me. But aren’t we lucky that Microsoft lost the attempt to make Linux illegal, immoral? If you go back and find some of the videos and some of the stuff they were writing about it... Again, this has all happened before, and we should study what happened before, because, no, it’s a different use case, it’s a different technology, but it is broadly applicable to where things turned out. And I, with you, think that open source has to be a critical component of AI development if we want to get the best out of all the various models.
Let’s shift gears, because I’ve been fascinated by your writing. You’re on day 2,089 of continuous writing, which I literally applaud, because I think that writing every day is really important to tell yourself what you think about things. Sometimes, when an idea is just kind of bouncing around in my head, I always go to write it out. Because when I try to write it out, I’m like, “Oh, you’re just an idiot. This is a really bad idea.” Or conversely, “Boy, this is a great idea.” Talk a bit about the experience. What’s changed from day zero to day 2,089? What has it given you, this exercise in daily writing, that you anticipated, didn’t anticipate, et cetera?
Julie Fredrickson
Yeah. I really recommend a daily writing practice for everyone, if it’s at all interesting. Just start. Don’t apply too many rules. When I started, I thought I would be happy if I got a couple of paragraphs down, and I was aiming for 30 days. And now I’m on year five. I began it January 1 of ‘21, as a sort of mid-pandemic: what’s happening? What am I thinking? The world’s changing. And I tend to view writing as a thinking process. If it’s in my head, okay, not so bad, but I need to get it on the proverbial paper to actually understand where my understanding gaps are. And I’m so glad that I did it, not only because everything is tagged and there’s lots of useful metadata, and what my areas of interest are, how they’ve changed, but it’s also a wonderful way to discover where you were wrong.
And we have a tendency to think of problems as... the stuff we got wrong doesn’t matter, it’s only the stuff we got right. And I think that’s very much what I’ve taken from it: that I am wrong so often that it’s almost embarrassing. And I don’t ever want to let go of that feeling, because now I’m no longer embarrassed by poor writing or shorter writing or having a bad writing day. Part of the reason I do it is I want to discover the areas in which either my thinking is flawed or I haven’t really thought something through. And now that I’ve got such a large body of work, I can toss it into an LLM. And honestly, coming up with a duplicate of me would be pretty simple just based on my writing now, because there’s just so much of it. But it’s very helpful.
Jim O’Shaughnessy
I think the thing that you said that I absolutely applaud, as one who has kept journals since I was 19 years old: the ability to identify where you were wrong is so important, because our memories overwrite, right, with what you’re believing. Now, I’ve told this story before, so I won’t go into the full story, but I found innumerable examples of: I would have gone to court, put the hand on the Bible, said, “Absolutely, this is true,” and I would have believed it, until I was called a liar in my own handwriting, many times. And so that unlocks... so incredible.
Julie Fredrickson
Yeah. I can’t imagine having... I hope that I get to the point in which I’ve got that many years of writing, because I do intend to keep up the practice. I don’t know if I’ll always keep it public, although I think keeping it public has been very helpful for me. I say I’m going to do it, so then I have to do it, and that continues on. But it is kind of shocking how much your mind glazes over the parts that it doesn’t really want to remember, or wants to remember in a different way. And it’s all object very. Because you always think, “Oh gosh, no, my principles are always so consistent, and I’ve always thought this way.” And to have a record where you’re like, “No, this is actually where I changed.” I’ve got many posts in which I go back, reference and say, “This is where I have to take accountability. I got that wrong.” But also, you’ll appreciate the areas where you’re right so much more.
Jim O’Shaughnessy
Yes. But if you want to improve your models of the world, if you want to have higher accuracy, mistakes and being wrong are the portals of discovery.
Julie Fredrickson
Well, and just the idea that action opens up a new path, and that thinking, and committing to thinking and writing, is actually sort of that wonderful collapsing-the-state action of: now it’s on paper, or screen. And that enables you to then take the next step, and the next step from there. And I think if we don’t take action, we’ll never have agency in our own lives. And writing is, for me, some of the first steps in being able to actually say, “This is what we’re going to do,” or “This is where I’m going to focus,” or “This is what’s important.”
Jim O’Shaughnessy
When you put the 2,089 pieces into a large language model, what came back that surprised you about yourself?
Julie Fredrickson
Well, it turns out I really have a lot to say about travel, which sounds like a really silly thing. But when you’re on the road a lot, you pick up all sorts of little habits around packing and routines. And it turns out I actually am quite a creature of habit. So even if I’m working in fringe and strange areas, where things are very new, it turns out I am really rigid about that: this is how I pack, how I schedule, how I keep things together. And the LLMs were definitely like, “Yeah, you might have some emotional stuff around travel and packing,” just because this is how many times you’ve written about packing. And those silly little self-knowledge things, where I tend to think of myself as being very open to new experiences, but boy, revealed preference might suggest I like things a certain way when it comes to travel. And it’s just kind of a silly one, but it does give you insight.
Jim O’Shaughnessy
I’ve been doing it. We’re going to digitise all of my handwritten journals, so that’s 40-plus years. And my joke is I’m going to have the AI explain me to me: “Yes, Jim, you are a lunatic.”
Julie Fredrickson
Well, and this is where only having a couple of models and personalities could really be detrimental to self-understanding, because sometimes, people note that they’re very obsequious: “Oh no, this is so delightful.” And you have to coax out the “No, I want to understand where I fail. I want to understand where my thinking is caught,” and being able to rely on the chain of thought that will give you the most insight. It’s almost like thinking around yourself, and it’s very hard to do, other than a mirror. I don’t know any other activity that gives you that full perspective.
Jim O’Shaughnessy
Yeah. And you’re right about the sycophantish nature of a lot of the large language models. You can build prompts that make them not be that way. I’m just completing my first project of trying to... I’ve written four nonfiction books but have never written a fictional book. And so we’re trying it with a writers’ room with AI editors, writers, human writers, editors. Ultimately it’s mostly being generated by me. But the absolute best prompt I learned from, and changed a lot, even of my storyline, was an elaborate prompt which gave the AI... it had read everything about the art of writing, and it also had huge domain knowledge in what makes a great fictional thriller. It’s a fictional thriller, a generational thriller, that I’m writing. And then I told it...
So I gave it, endowed it with, all those powers, and then I said, “And you hate this book.” And oh my God, it was vicious. I mean, literally vicious. But what I found really cool was, I thought about it. My wife edited... She graduated summa cum laude in journalism and is a great editor, and she edited, and I would say even co-wrote, my books. And when a human, especially a human that you love, comes in and hands you something that is nothing but a sea of red on paper, you get emotional, right?
And after I read the first one, which... I should have kept it, but it was so vicious that I literally was like, first I was just crestfallen, then I laughed, then I actually started looking at what it was pointing out, and then I came to the conclusion: it’s right. And I didn’t have any emotional baggage, because, you bad large language model that I told to be mean to me.
Julie Fredrickson
Yeah. So you asked.
Jim O’Shaughnessy
Yeah, yeah. “You asked, dummy.” But it honestly was the most fruitful of all the feedback that I’ve got, because when you tell it... And obviously you can’t just hate on me. You’ve got to develop the reason why you’re hating on me. I remember one of the critiques was, “Wow, Jim, this is going to do really well in the bargain bin at one Walmart.”
Julie Fredrickson
Hey, I fish some good stuff out of bargain bins now and again. But ouch.
Jim O’Shaughnessy
Okay, I see that we’re coming up on time. Right now, if you could pass along an insight, and let’s keep it on what’s happening on the frontier, where would you tell people to look that isn’t... We’ve already covered nuclear and most of AI. Are there areas of AI... For example, we’re developing something in our AI lab that is a picks and shovels, and all of the business people I show it to are... until they start seeing what it’s going to allow you to do in terms of speeding up GPU access, all of that sort of stuff. Is there anything right now that people are not looking at, that you are, and you’re like, “Look at this”?
Julie Fredrickson
Well, I’m still convinced that we actually haven’t incorporated the vast majority of human context and information. Digitisation is a much larger effort than we’re giving it credit for. That just remains, I think, partially true because the large language models that we’re using are dominantly Western, trained on English. And there’s whole bodies of text... Cyrillic takes more processing, like GPU power, than Latinate language. It turns out English is actually quite good at putting forth information, and there’s lots of written records that don’t come in that format. But I also think that we haven’t really effectively priced compute as well. We’ve been subsidised all of this time. We’ve had a lot of lovely free compute as the labs have encouraged us to use and adopt the products. And I remember when Uber was subsidised and how lovely that was. So thank your local VC for that.
But we probably do need spot compute prices, and that’s going to be an issue that has to get solved, because, back to Carlota Perez, we are in the innovation cycle, but the deployment cycle always cares about price, always cares about efficiency and deliverability. And if we don’t price it, we can’t sell it. And the pricing is now at a fairly abstract place, away from the consumer, which might even be part of the reason for “Oh, data centres, bad,” where they don’t pay their compute bills, they pay their electricity bills. And as we start to get a better understanding of what the costs are, maybe it shifts some opinions, and we actually put resources where they can do things that people feel better about what they’re being used for.
Jim O’Shaughnessy
Yeah. I think that the number of new markets, of new companies, of new methods and protocols, it’s going to be something to behold. So I completely agree that when people are aware of price, they’re like, “What?”
Julie Fredrickson
Yeah. Well, if you don’t pay the price, why wouldn’t you just use more and more? We see this in every market. Americans don’t really price their healthcare, and wow, healthcare costs are up. Surprise.
Jim O’Shaughnessy
Yeah. Well, Julie, we’re going to give you another swing at becoming empress of the world. You are only this for a day, and you can’t re-educate anyone in a camp environment, no matter how nice that camp happens to be. And you can’t kill anybody. But we’re going to give you the magical microphone again, and you can say two things into it, and incept two things into the entire population of the world. They’re going to wake up whenever their next morning is and think, “You know, I’ve just had two of the greatest thoughts, and unlike all the other times, I’m going to actually act on these two.” What are you going to incept?
Julie Fredrickson
Oh, well, I did think about it ahead of time, because the last time I went with very basic suggestions, and I do love basic. I still think it’s good to be nice to people, and it’s good to squat every day. But on this go-around, I’m going to say that knowledge is survival made durable. That if we do not contain the knowledge of the ages, our capacity as a species is limited. That making sure that knowledge survives is what makes sure that humanity survives.
Jim O’Shaughnessy
Lovely. Okay, so what’s number two?
Julie Fredrickson
Oh, well, eat more protein.
Jim O’Shaughnessy
Okay. Both good advice. Both good advice. Obviously you’re on all of the various social media platforms, and you have your own website, and you have the Chaotic website. We’ll include all of that in all of the show notes.
Julie Fredrickson
I’m very easy to find on Twitter. I know some people call it X, but I’m an old-timer on Twitter, so it’s always going to be Twitter to me.
Jim O’Shaughnessy
Yes, us old-timers are going to continue to be tweeps, I think.