Dave Nadig: Hey folks. Dave Nadig here, financial futurist at VettaFi, bringing you this conversation I had with Jim O’Shaughnessy, legendary investor, currently the executive chairman of Stability AI. You may know that from the stable diffusion and mid-journey generative image models, Jim’s been a real front-runner in the AI space for the last decade, honestly. Really fascinated by the ways that machine learning and how we think about data change the process of investing. In this conversation, we really dig into how AI is going to change the global economy, the balance between labor and capital, and how society is going to respond. I found it a fascinating conversation. I hope you enjoy it. Cheers.
Could you start maybe by talking about the great reset? It’s a framing that you first explained to me maybe two years ago, and now you’re sort of living it out loud, so can you give us the background of what you mean by the great reset?
Jim O’Shaughnessy: Sure. Well, first of all, correct. I call it the great reshuffle.
Dave Nadig: You’re using the World Economic Forum’s. I think far more nefarious version. It sounds scarier.
Jim O’Shaughnessy: It does. It sounds a lot scarier. I think that the great reshuffle is essentially been… It really got kicked off, honestly, the beginning, nascent part of it got kicked off during the explosion of the internet because it gave us the first universal variance amplifier in human history, and it gave us the ability to move our digital zip code far more easily than moving our physical zip code. It gave the internet alone, just on its own, gave people the platform in which they could build communities. They could build like-minded associations, et cetera, with people from all around the world.
As you know, the cultural leg takes a long time to catch up. I learned that the hard way by trying to form an internet investment advisory service called Net Folio in 1999, 2000. Boy, I thought I was really…
Dave Nadig: But at the end of the day, you ended up selling Canvas.
Jim O’Shaughnessy: Indeed.
Dave Nadig: So you got there eventually.
Jim O’Shaughnessy: We did.
Dave Nadig: You just had to take a couple of tilts at the windmill.
Jim O’Shaughnessy: And what’s interesting about that is it neatly demonstrates the cultural leg that is inherent in the way we humans adopt things. You’ve seen the classic adoption curve for technologies and whatnot. That still persists right now. Now it’s sped up because of a variety of factors that we no doubt we’ll get into. But the point is we started then, but not much of anything happened, right? We used it for shopping, we used it for social media, we used it for a variety of those things, but we really weren’t moving on to the next stage until you started coming online digital natives. So Zoomers…
Dave Nadig: Which coincides with the pandemic that became really important at the same time.
Jim O’Shaughnessy: Indeed.
Dave Nadig: So has the pandemic really been the accelerator of the great reshuffle?
Jim O’Shaughnessy: Absolutely. As a matter of fact, we think that certain of the trends, for example, I have a home office now. We don’t have a central office. We have a clubhouse for O’Shaughnessy Ventures, but it has bedrooms so people can stay there. It has great workspace so that we can collaborate in person, but one of the things that accelerated was the work from home movement and we all see the efforts being made by the companies, trying to get people to go back to the office and it’s not working out so well and it’s specifically not working out well at all with younger workers.
Because they came of age and they have the skills being digital natives to be actually… If you’re a knowledge worker, let’s be very clear, but if you’re a knowledge worker, you have the skills that you are going to be a lot more efficient, a lot more fulfilled really if you don’t have to face it to our commute each way, and you can then spend that time to take a walk. You and I are big advocates of touching grass, and so you see the campaign, it’s just not working. But yeah, the pandemic accelerated the work from home movement significantly, I would say by as much as seven years.
Dave Nadig: Yeah, that makes sense.
Jim O’Shaughnessy: It accelerated the use of these remote style tools that were there, but nobody was really using because they didn’t really have to. And then when we all started doing Zooms for everything… I talked to a banker friend of mine who said their revenues, at a very large investment bank, and she said their revenues went up like 40% because what happened was they loved the fact that they could have all of these meetings. They didn’t have to get on a plane and go to Tokyo, or have to fly out to California from New York. They could just do the meeting. And so what happened was a lot of the senior bankers got back in the game.
Dave Nadig: But this implies that the great reshuffle is going to have some winners and losers in it, right? That’s an example of a winning scenario where what was a constraint actually becomes a huge opportunity and a huge accelerant on productivity. We’ve certainly read the nightmare stories on the other end. Companies that have tried to work remote, really struggled in the pandemic and have brought folks back and reportedly to great success, but they’re different kinds of businesses. So in that kind of reshuffle, who are the winners and losers and what should investors do about that?
Jim O’Shaughnessy: So I think we have to be careful designating this winner, this loser, because there’s going to be various aspects that work very, very well for certain types of companies where they’re detrimental to other types of companies. But really another part of the great reshuffle is all the old playbooks, all the old models are collapsing, things like when we adopt this more online posture, what becomes important? Authenticity becomes very important. Not trying to bullshit people, corporate speak, PR speak.
People can smell that a mile away. And so all of the old tactics of trying to speak in the passive voice, it was decided by who? Who decided that? So any kind of hint of inauthentic or kind of weasel words are being judged much more harshly today than they would’ve been say 10 years ago. But to get specifically to winners and losers, well, so winners are going to be anyone operating in the real physical world because you’re not going to be able to automate a lot. If you have a plumbing business. Congratulations. If you are a licensed massage therapist…
Dave Nadig: You going to have a dynasty.
Jim O’Shaughnessy: If you’re a licensed massage therapist or a physical therapist or any of those jobs, the world is your oyster. Talking to a PT recently, literally they can pick a job anywhere in the country and improve their compensation, improve their hours, improve everything. So anyone operating in the physical world in terms of those types of occupations, your job is not only not going to be threatened. You’re going to probably have a great decade. Losers are going to be people who are lower level people doing repetitive work.
Dave Nadig: Particularly repetitive knowledge type work, right?
Jim O’Shaughnessy: Yes. Specifically like junior bankers, junior associates.
Dave Nadig: Beginning actuaries.
Jim O’Shaughnessy: Yeah, beginning actuaries. Ouch. And so people always say, well, what will they do? My response to that, and I’m really not being flipped when I say this, they’re going to be removed from this horrible drudgery of having to copy paste, copy paste, copy paste, and hopefully be able to get a lot more creative. Let’s take the actuary. If they could spend say 40% of their time not doing grunt work and copying and pasting and trying to memorize tables, they might have a lot more time to think up different ways that we can use an actuarial table, different approaches that we haven’t had literally the time to think about.
Dave Nadig: Other ways of managing risks.
Jim O’Shaughnessy: Exactly. New models for risk are going to be emerging from this across all fields. Now, the one I’m most familiar with is asset management, which I think there will be a massive impact. For example, my early interest in AI was when it was still called machine learning because I viewed it as the next frontier for quants. The dirty little secret of quants is if you look at all of our underlying models, they’re pretty similar. And they’re pretty similar because we all had the same data sets.
If you’re talking to my friend Cliff Asness, you know what he trained his data set on, it’s exactly what I trained my data set on. And so your models tend to replicate. Now, of course, they’re different in many, many ways. For example, my old company, O’Shaughnessy Asset Management uses composites of factors as opposed to single factors. Some quants still like to use a single factor.
Dave Nadig: But that all comes down to implementation, right?
Jim O’Shaughnessy: Exactly.
Dave Nadig: I mean, the math doesn’t change that much.
Jim O’Shaughnessy: Exactly. And so I think that you’re going to see an explosion of research using machine learning techniques and AI techniques. In fact, that was my first goal in forming O’Shaughnessy Ventures was going to be a company called Gray Swan. Now why don’t you have that firm right now, Jim? Well, because the compute costs of trying to put that type of firm together were going to be prohibitively expensive for me.
Dave Nadig: Today you could probably do it.
Jim O’Shaughnessy: Today, I could probably with the cloud, yeah.
Dave Nadig: But still not free.
Jim O’Shaughnessy: Not free.
Dave Nadig: And a crowded field. I mean, the asset… People always go where the money is, right? So all of this money in AI has been chasing a lot of finance ventures along the way.
Jim O’Shaughnessy: Totally.
Dave Nadig: And we’ve got great examples. I mean, there are firms out there doing great work on the Kaiju Capital. There’s a whole bunch of folks out there doing interesting AI for asset management work, but your response to your great reshuffle thesis has not been to lean hard into more quant stock picking stuff.
Jim O’Shaughnessy: That’s right.
Dave Nadig: You’ve actually gone in two completely different directions, one being more generative AI work and the other being sort of reinventing what it means to be capitalist through a patronage model. Let’s start with that last one. You talk about this rise of digital natives being a big part of the great reshuffle. What you individually, I believe are now just handing fellowships out to interesting young folks that you think are the future. Talk to me about that model. It’s pretty unusual.
Jim O’Shaughnessy: Yeah, I’ve always, and unfortunately, I’m going to have to reveal my nerd sources here, but I was rereading Francis Bacon’s novel Utopia. And in it, as you know, it covers Atlantis, the mythical kingdom of Atlantis. The part that grabbed my attention was that they always sent out 12 explorers into different reaches of the earth to see what they were missing, to find out new knowledge, et cetera. And I love that 12, that’s why we started with 12 fellowships.
Dave Nadig: I love it.
Jim O’Shaughnessy: Historically, a brilliant innovator, a genius could be born, live and die without ever knowing that they’re a genius, without ever knowing that they have this skillset. And that really for the first time in history is no longer true because of the interconnectivity of the world, because of the tools that we have to be able to search for these people. We can find them and we can fund them. And since I believe passionately that free markets are the way to improve world poverty, to improve world health, to improve the material situations that people not living in the United States or Europe are trying to move towards. I think that since we can find them, we must find them and fund them.
And we set up the fellowship as a completely strings free. So we award 12 fellowships a year for a hundred thousand dollars paid monthly. But it is not simply to founders or entrepreneurs. What we’re looking for are people who are doing fascinating, interesting work. And by the way, yes, they tend to be young, but we will accept and hopefully fund, like if some 70-year-old comes up with a great idea and they need the funding, they’ll be able to get the fellowship too. But the majority of the people that we’re finding are younger. One of the things that you know when you study things like math and the harder sciences is that there’s a real clock on those people. People in those types of endeavors and pursuits tend to do their best work when they’re young.
Jim O’Shaughnessy: Well, certainly if we look at physics, mathematics, I mean very, very few folks are making their big breakthroughs in their fifties.
Jim O’Shaughnessy: Exactly.
Dave Nadig: They’re teaching in their fifties. They made their big breakthroughs in the patent office.
Jim O’Shaughnessy: Exactly. And so we thought, wouldn’t it be cool if we could remove the constraints on a brilliant young person, fund them and fund them without strings to do their work? So for example, William Zang was one of our first fundees. He’s looking to open source quantum computing, and he’s a brilliant guy and strong proof of work already. And the entire thing was… I love that idea. I love the ability to give a young person who’s shown a lot of incredible intelligence around something I think is a very important thing for us to be trying to look at.
And yet, I spoke to you at lunch about the other one that we funded, contemporaneous with him, was a couple who have five children under the age of 10, and they were dissatisfied with the public school options where they lived, and they wanted to do an investigation of alternative education for kids, and they wanted to make a documentary about it. And it just dropped. It’s on YouTube.
Dave Nadig: What’s it called?
Jim O’Shaughnessy: Kids just want to be free. It’s Nat & Martha Sharpe.
Dave Nadig: Kids just want to be free.
Jim O’Shaughnessy: We can put it in the show notes.
Dave Nadig: But that’s such an interesting model because in those two examples, the gentleman you have working on quantum computing, very easy to see how, oh, okay, in six months, a year, five years, this guy’s going to land somewhere and do some amazing work. Or he is going to go build something and he’s going to need funding. And in all of those cases, I can see a role for you as an investor participating in that process, whether it’s just helping connect people with financing, whether it’s funding things yourselves, whether it’s being part of boards.
The other example though is really about putting this piece of information, this piece of… And it’s not really entertainment, it’s more knowledge out into the world from the perspective of two parents trying to solve a very real world problem that every parent has. There’s no payback on that. So do you consider this charity, do you consider this sort of loose investment? How do you think about it?
Jim O’Shaughnessy: So I don’t consider it charity. I consider it an investment in knowledge. And I’ve been incredibly fortunate in my life in terms of my ability to build companies that did very, very well. Now I find myself in a place where I get to play any type of game I want to play. And the game I want to play is to only advance win-win deals, and to only frankly do business with people who I think are ethical and want to play positive sum games.
We’re kind of back to the great reshuffle. We’re at an inflection point, and it can go one of two ways. It can go more than one of two ways, but two directionally different ways. One is open and one is closed. And I have five grandchildren. Back to AI, I want the world to not be run by a panopticon. I don’t want a few people who say they have my best interests and your best interests at heart to say, trust me, I don’t want to trust them.
We want OpenAI. I want AI for everyone because I think it’s a remarkable tool that can be used both for good and bad. Let’s be very clear about that. But if we directionally get it right and make it transparent and make everyone, or give everyone the opportunity to have access to these tools, that’s going to make for a better outcome in my opinion.
Dave Nadig: Let’s take that hard pivot towards AI.
Jim O’Shaughnessy: Sure.
Dave Nadig: You brought it up and said, can be used for good or ill, I know you’re not a doomer on this stuff. If anything, you may be a little bit Pollyanna on it. Let’s talk about both sides of that argument. The doomer arguments are to me largely about control or they’re about AI is actually developing consciousness and free will, which I think is not worthy of discussion. Because I think we don’t understand what’s going on here yet. We have ways to go, but the control issues are real. You are now involved with, I believe, more than one AI ventures at this point. What do you think is a legitimate concern for the average person in the public to have about AI, what should they be concerned about? What should their attention be focused on?
Jim O’Shaughnessy: I think that the quotidian, for lack of a better word, uses for ill, we spoke about the spoofed voices that older people are getting. AI can perfectly recreate my voice, your voice. We both have very easy access…
Dave Nadig: There’s enough of us out there.
Jim O’Shaughnessy: To get the voice, I have, I don’t know, 178 podcasts.
Dave Nadig: Probably get it from there.
Jim O’Shaughnessy: They can train it on. And so there’s going to be a lot of scams. And then think, don’t think just voice. Think if you got a FaceTime call and it looked just like Dave, or it looked just like Jim and it sounded just like Jim, and you’re 90 years old, you’re going to probably fall for that. So I think that in the shorter term, again, back to this cultural leg, especially for older people, they’re not quite as familiar with the trends of what’s going on in AI. And they trust their senses. They get a phone call from their nephew, Dave, who says he’s in trouble. Their urge to help overwhelms. I think that you’ll also see a lot of nefarious uses of videos that are fake.
Dave Nadig: Deep fake.
Jim O’Shaughnessy: Deep fake.
Dave Nadig: Like the politics.
Jim O’Shaughnessy: Yeah, politics, corporate espionage, court trying to destroy a competitor. There’s going to be a lot of things that are going to happen on the negative side. Now I think that several of the companies I’m involved with are actively searching for countermeasures against these bad uses of the technology. I like to remind people that fire is responsible for our prefrontal cortexes. I think you know we didn’t cook our food. And so when we started cooking our food, guess what? We got this whole new part of our brain that gave us the ability to invent AI. But fire is incredibly dangerous. And so we didn’t try to ban fire. We instead created fire alarms, fire departments, fire exit.
Dave Nadig: We stuck it inside car engines.
Jim O’Shaughnessy: Exactly. And so I’m not in any way Pollyannish about it or Panglossian about it. I am, I think very realistic that the good uses of AI vastly outweigh the bad, but there will be bad and there will be problems that we have to deal with. And I think trying to deny that is a fool’s errand. It’s like then you’re not a serious person if you are out there saying there’s not going to be a single problem with AI. There’re going to be plenty.
Dave Nadig: It sounds like you think most of those problems are actually problems with people, not problems with AI.
Jim O’Shaughnessy: Exactly, very intuitive. Technology itself is neutral. It is not good or evil. It is the human using that technology and nuclear weapons. They have one purpose of a nuclear bomb. The only purpose of a nuclear bomb is to kill people and destroy property. That’s it. To me, that’s a very evil purpose. And so when you see comparisons of AI versus say nuclear weapons, it’s completely missing the point. That is a single purpose tech, if you will, that is for bad things, in my opinion.
Dave Nadig: Well, I think the example I got into this argument with Stuart Russell about this, the argument he makes is more… It’s around human cloning because that one, it’s very difficult to say that is purely an evil technology. It will only be used for evil. You can come up with [inaudible 00:23:02]embryonic genetic manipulation and come up with nightmare scenarios, but it starts turning into Marvel movies with super soul and stuff like that. But again, he would argue that while we have international agreements about banning or severely restricting the use of human gene lines and certain applications, and that should be the appropriate kind of model as opposed to the nuclear where it’s just like, well, we just make bombs.
The manipulation of embryos and human cloning, you can come up with a lot of very positive cases. Yet as a society, we’ve sort of made the call to put the brakes on that kind of research in certain circumstances. Do you think that there’s an appropriate example like that in AI or do you think that that’s too far of an approach?
Jim O’Shaughnessy: I think for right now, AI has not risen to that level where we’re actually trying to redesign or create new forms of our species. That’s very different to me at least than what we call narrow AI uses right now. And I think that should we reach a place where those types of questions need to be asked and answered and solved, I want to be part of that conversation.
I don’t think we’re quite there yet, but just speaking on that issue, what’s going to happen when CRISPR can cure sickle cell? Are people going to argue against that? That’s the challenge. These are thornier questions than people sometimes realize at first. There are all sorts of horrible diseases in the world that could be eradicated through CRISPR technology and gene manipulation.
That’s something for the most part that I favor. Do I favor unconstrained, unregulated use of these technologies? Of course not. I think that they should be heavily regulated. I think that they should have very bright line rules around what you can and cannot do with that. But I think that we’re nowhere near that yet with AI. I think the best way to think about AI right now is as an incredibly powerful tool for the human user. And I like to use two of the use cases that I find the most promising. The first is AI tutors.
We all have very different learning styles. Well, the beautiful part about AI tutors is they train on you. And so they are much quicker to iterate the lesson into a format that resonates with that individual. And we’ve already seen remarkable things. Stability AI got to remit to educate a third of the schoolchildren of Malawi and the early results, let’s be very clear, they’re early results and we haven’t had a formal study done yet, but the early results are incredibly encouraging. The kids who are being trained on the AI tutors are way ahead of the kids in the traditional classrooms.
Dave Nadig: And we know this about education in general, right? Individual tutoring shovels off something like five years over the course of a normal educational lifespan because individualized instruction works. We know this, but we can’t do that to everybody. AI theoretically allows us to create individual tutors for everyone. Khan Academy is already starting to do some of this work. I mean, that strikes me as a very now application. What else do you think we’re underestimating in terms of how quickly this is going to have an impact on our lives?
Jim O’Shaughnessy: I think that the unlock here is going to be when there is an AI agent on your cell phone that you get to determine how much of your business that AI agent can…
Dave Nadig: Doesn’t know my calendar, doesn’t know my contacts.
Jim O’Shaughnessy: But it can, if you want it to, it can. And when that happens, everyone is going to immediately enjoy and see the benefits of AI.
Dave Nadig: That feels like it’s tomorrow. That does not feel like it’s next year. I mean, what’s your over under on when your SIM gym in your pocket is going to be over?
Jim O’Shaughnessy: I don’t think it’s tomorrow. I think it’s maybe this year.
Dave Nadig: Okay, so it’s soon.
Jim O’Shaughnessy: It’s soon.
Dave Nadig: It feels very soon like all the pieces are there.
Jim O’Shaughnessy: Right and the other thing that I like about it is most of us are used to dealing with Alexa or Siri, but just imagine Siri actually competent.
Dave Nadig: Being good.
Jim O’Shaughnessy: Being good.
Dave Nadig: And actually able to do things.
Jim O’Shaughnessy: And do things, right? Here’s a great use case. So you and I have gotten together just for fun, and we take a long walk in the woods, and I, as one often does, when one is walking, you think of something that you’ve completely forgotten about and oh my God, it’s my sister’s birthday today. Can you imagine how nice it would be to pull out your cell phone, say it’s Eileen’s birthday today, would you find out a perfect gift for her?
And have the agent go out because it knows Eileen, and Eileen has recently taken a course in calligraphy and you know because you talk to her a lot that she loves it. And so the AI comes back and it says there’s a brand new course on calligraphy in St. Paul, Minnesota where she lives, and there’s a wonderful restaurant nearby. Would you like me to call the restaurant, have a birthday cake made for her and then enroll her in the class?
How cool is that? And so that type of use of AI, I think is going to be the most ubiquitous. The incredibly useful time-saving things that anybody with a phone can use. And I’ve always been an advocate for AI for everyone in everything, and let’s take those apart. For everyone, I think that is absolutely critical. I do not want a, we already have incredible disparities in income, in assets. We do not need an intelligence or applied intelligence disparity to emerge.
Dave Nadig: So let’s jump right into that. How do we make sure that there isn’t a haves and have-nots? AI is expensive, right? I mean, you look at what’s the estimate? It’s like a hundred million dollars a month or something like that, that OpenAI is spending on compute power just to keep ChatGPT-4 running. The idea that you’re going to do that for 6 billion people on the planet for free doesn’t seem tenable. So how do we create a world where the profit is accruing to the people who’ve made the investments like good capitalists. We want that to happen, but by the same token, we’re using the power of some of these tools in a way that’s benefiting to all of society.
Jim O’Shaughnessy: So like anything we are already finding that constrained AI, in other words, much smaller training runs, much smaller models are actually going to… Like the model I just described for you. That’s actually a tiny model that we already can fit on a cell phone, it works on a cell phone, and so that doesn’t have that a hundred million, that’s what we call big iron. You’re hitting big iron for those really complicated in the cloud type solutions.
I don’t see that as the ubiquitous way forward for AI. I do see a lot of, in fact, we’re looking at several potential investments right now in companies that are doing evolutionary style training of AI models, which does not require the heavy firepower, much lower compute costs. And there are some other experimental ones that we’re also looking at. Now, I’m not saying that all these are going to work, but like anything, what you’re seeing is first, the attention is all you need, that paper.
Dave Nadig: 2017, Google.
Jim O’Shaughnessy: Opened up everything. Interesting note there. None of the authors of that paper work at Google anymore. They either have their own company.
Dave Nadig: That doesn’t surprise me at all.
Jim O’Shaughnessy: Or working at a startup. So that unlock happened. But now we’re seeing a whole host of other ways of training, be it neural models, be it evolutionary models. There’s a whole host, and I don’t want to get into the nerdy weeds here, but they’re cheaper. They are more robust in small tasks. And just to be clear, the task that I outlined for getting the dinner and gift from my sister, that’s a small task. That’s an agent task.
So I definitely think that these models and these much lower cost training will emerge. I also think that it will become many of the uses for people like the AI tutor. I could see that quite easily being something Apple would love to have on their iPad that you’re going to give your grandson or your nephew or your niece or your child. So I definitely see the larger companies like the Apples of the world, specifically Apple or Android, offering those as features.
Dave Nadig: Just giving them away.
Jim O’Shaughnessy: Just giving them away.
Dave Nadig: As part of a package.
Jim O’Shaughnessy: As part of a package. If you look at Japan, for example, their system there is very different. The telecoms control most apps and they compete on the amount of apps available on their particular phone, the quality, how next gen it is, et cetera. I wouldn’t be at all surprised to see that happening, like AI simply being part of an operating system.
Dave Nadig: So it sounds like the real immediate future here is about intellectual property, not capital spend. So good ideas well implemented and marketing, frankly, getting it into the right number of people’s pockets. So those seem to be the big wins. Neither one of those requires owning all of Taiwan’s fabs.
Jim O’Shaughnessy: No.
Dave Nadig: I mean this is not necessarily the boom for Nvidia stock everybody has been thinking it is. If that is in fact the case, if it is intellectual property marketing, that is the next leg up.
Jim O’Shaughnessy: I do think though, the Nvidia example is a good one to bring up because there are use cases that I’m incredibly bullish on that do require a lot more compute.
Dave Nadig: Medical.
Jim O’Shaughnessy: Medical.
Dave Nadig: Medical. Yeah.
Jim O’Shaughnessy: So that’s the other one that I am incredibly bullish on. The mRNA vaccine, horrible thing to call it. It’s not a vaccine, it’s a technology. It’s a very powerful technology and it can be repurposed to other uses. So for example, right now we’re very close to a inoculation against malaria, against a variety of other diseases that whole continents suffer under. And I, for one, am incredibly bullish on that.
The reason I bring that particular platform up is we humans tried to fold proteins for 60 years, never got anywhere. It did it in like six months. And so it’s those tasks where we humans are not designed to look at an array of a hundred million data points or a billion. It doesn’t really matter. Humans are not designed to understand the exponential function. It’s pretty much…
Dave Nadig: We’re very linear people.
Jim O’Shaughnessy: We’re very linear. Our math skills are just native naive. Math skills are 1, 2, 3, a lot. But AI is trivial. And so it can look into what’s called liminal spaces. And that is incredibly important for things like medical discovery of new drugs, combinations of things that we wouldn’t have thought of that can have very wonderful and healing effects, sometimes curing effects. So that’s going to still require a lot of compute as are the huge kind of national level models. Our anticipation is that every country is going to want their own national model. And we’re finding that, for example, it’s Stability AI right now, and those are going to require a lot of big compute.
Dave Nadig: Tell me what you mean about a national model.
Jim O’Shaughnessy: So a national model, for example, Stability AI just released the first large language model that is entirely in Japanese. And one of the common complaints from the world community has been that AI has been very English centric, which I think is a fair criticism. And so one of the things that we undertook at Stability was let’s make sure that these speak Farsi and speak French and speak Japanese and speak all of the various languages in the world.
Dave Nadig: But it’s much more than just language. I mean, that criticism of certainly ChatGPT-4 has been as much about the cultural issues as the language issues. Translated into Spanish it’s still very Amero/Eurocentric in its references and how it thinks through things because that’s largely what it’s been trained on. When you do a Japanese model, are you relocalizing the training sets?
Jim O’Shaughnessy: Yes. Very well, great intuitive jump to that. That’s what a national model is.
Dave Nadig: Got it. Okay.
Jim O’Shaughnessy: A national model trains the large language model or the stable diffusion sets on a very different data set, a very different set of cultural norms, a very different set of cultural taboos. A variety of things that are unique to that particular culture.
Dave Nadig: So does that imply a world where 10 years from now they’re going to be hundreds and hundreds and hundreds of these models, and you might be using the official Japanese model, but you might also be using somebody’s renegade Japanese model?
Jim O’Shaughnessy: Maybe is all I can say right now. I tend to see this as there will be all sorts of… The part of what you just said that I actually like is I think that there will be very specific agents, if you will, that are designed for very specific tasks. And so there’ll be thousands of those and tankers will make them on their own and we’ll keep them on their closed systems. They won’t be on the internet. Companies will try to commercialize them and essentially get them to be adopted either through an app type sales store or some type of program for a laptop or a desktop computer.
So I think that in the normal kind of market sense, you’re going to see thousands. I don’t think that that’s going to necessarily be what I mean when I say a national model, when I say a national model, that would be kind of Euro version of the official Japanese large language model that hopefully they’ll make available to all of their citizens.
Dave Nadig: And perhaps other people bake that into then their applications.
Jim O’Shaughnessy: Exactly. Yeah.
Dave Nadig: Yeah, I can see that.
Jim O’Shaughnessy: And they call it getting hooks into the model when you’re designing something that makes use of that model. I suspect that you’ll see quite a lot of that.
Dave Nadig: Well, we’ve already seen that just in the US, in the sort of entrepreneurial bubble around ChatGPT. There’s now a hundred and however many thousands of businesses that are effectively just… We do a thing and we put hooks into ChatGPT-4 to give us an interface.
Jim O’Shaughnessy: Those are called wrapper companies.
Dave Nadig: Yeah. I don’t mean specifically the wrapper ones, those are a little bit smarmy, some of them.
Jim O’Shaughnessy: Yeah, a little bit.
Dave Nadig: But no, but there are other whole…
Jim O’Shaughnessy: Oh, I know what you mean.
Dave Nadig: But even companies like Adobe are using this technology to make their text and video editing better and those kinds of things.
Jim O’Shaughnessy: Absolutely. And we think that’s going to be the largest part of the market actually, and we think that open source is going to be the solution that large corporations prefer because large corporations are going to absolutely demand 100% transparency into the model. And if it’s a closed model like ChatGPT, the corporation simply is not going to trust it.
Dave Nadig: Well, yeah, they can’t. I mean…
Jim O’Shaughnessy: Exactly. They can’t. And so one of the opportunities that we feel that we are uniquely set up for at Stability AI is we are 100% open source, but we can still go behind the corporate firewall because it’s their model. All we’re doing essentially is making a bespoke model multimodal AI for that particular company. So it’s a bit more like a red hat model, if you’re familiar.
Dave Nadig: Yeah, it’s very much like the early open source movement.
Jim O’Shaughnessy: Exactly. And we think that’s a massive market because corporations are going to need this technology. They’re not going to trust anything that’s closed, that their cybersecurity professionals don’t have total understanding of. And so we think that there’s a great business model here as well. But you’re going to see… Will AI also create a cottage industry? I think so, absolutely. In things like we were discussing over lunch, like I’m an investor in a company called Wand. What Wand does is it’s an actual tool, a wand for graphic artists, and it hooks into their AI resource and they can use any of the commercially available wands. We have preferences as a company for them. Obviously stable diffusion type models would be our preference. But the thing that’s cool about it is it’s a tool that graphic artists are quite used to using.
Dave Nadig: And it’s physical.
Jim O’Shaughnessy: And it’s physical, but what’s really neat about it is you draw your drawing, the AI then iterates on it, and then you have essentially a partner, the AI, who iterates your own work with you backwards and forwards until you get to something that you as the artist find resonates with the message that you’re trying to do.
Dave Nadig: And that seems like that… We started this by talking about how the best use cases is going to be as a tool for a human. That the cyborg is center model of this where you man and machine together. That’s obviously a great example of it. We’ve already seen it in things like in Adobe, you can do context to where fill, which just all of a sudden makes a background for you with the sailboat in it that wasn’t there, but the picture is still yours.
It creates this really fascinating discussion about generative ownership and creativity that probably way past where we’re going to get to be. Before I let you go, because taken a ton of your time already, I got to ask you one thing. You have your fingers in a lot of pies. You talked to a lot of people on sort of the front end of a lot of what’s interesting in science and culture in the world. What are you most hopeful about right now? What gives you the most hope for the future?
Jim O’Shaughnessy: I think the most hope for the future that I have is that we are at a critical inflection point as a species, and we finally have a suite of tools where we can create what I call the human colossus. And that means an interconnected, loose network that any human can join, add to, hopefully, using the tools that we’ve been discussing to unlock untold human creativity, untold human ability to innovate. We are removing much of the drudge work that humans still have to do, even in knowledge industries.
And that will be filled by the ability to generate new ideas, new thoughts about the way things should be organized, et cetera. So I think that we are right now on the cusp of being able to unlock a creative tsunami that will make the Renaissance look like a walk in the park. And I’m very hopeful that this is a universal occurrence. Bad news just sells better than good. We’re designed by evolution to pay much closer attention to bad news, with good reason. When the bush is rustling, run away. We are the descendants of the the people that ran away.
Dave Nadig: Figure it out if it’s a bunny later.
Jim O’Shaughnessy: But I definitely think that given the not great uses that people might put some of this technology to, I think a number of use cases for the flourishing of not only human creativity, but also innovation, discovery. Back to medicine, I think we are on the cusp of solving a lot of problems that have been beveling humanity for most of our history.
And the other thing that I’m incredibly excited about is that back to the fellowships, we can now find people in Bangalore, India that I would’ve never had an opportunity to know about, meet, have any interaction with. Plus, not only can we find them, we can fund them. And to me, that ability just lets all of the boats rise as the tide rises. And so I’m like… The other night I was having a wonderful conversation with my grandson, and I just looked at him and I was happy about something. He goes, Papa, why are you so happy? And I said, because this is the best time ever to be alive.
Dave Nadig: I love it.
Jim O’Shaughnessy: And I really believe that.
Dave Nadig: Well, thanks so much for the time, and I hope you’re right.
Jim O’Shaughnessy: I do too.
Dave Nadig: Dude, this was so much fun. Really appreciate it.
Jim O’Shaughnessy: Yeah, it was great.
For more news, information, and analysis, visit the Artificial Intelligence Channel.