Why OpenAI Should Buy Off the World’s Top Mathematicians | E32

23:54Like I think if you're, I think in many kind of information work domains, if not all professions, it kind of boils down to, you know, are you in it because you love the craft, right? You create this sort of beautiful share and you've kind of shaved every piece of wood there and you've done it versus if you want sort of the outcome, right? The way I've sort of achieved the thing. I mean, software engineering, you definitely see that, right? If you're coding because you love the architecture of the code, which I do, and it's like it's beautiful on it. If that's your motivation, you're not in a good place right now because a lot of that happens.

24:21if you're sort of more about the outcome, the creation, the kind of implications of producing a thing, then you are driving. And I'd be really interested too, Guy, in what sort of new approach did the model create that the mathematicians hadn't done before? That'd be great. I'd be really interested in that insight. I think this idea of mathematicians' jobs being replaced or something along these lines is where I'm just very allergic to these sorts of headlines, just because we're not using it for, you know, we're talking about advancement opposed to like protectionism. Yeah, and I think we've seen right, like in Go and in chess, that, you know, when kind of these models invented new moves, and now kind of the human players are exercising those, right?

25:08It opens sort of new avenues of thought.

25:09Lon Harris:I got an idea. All these scientists, mathematicians, want to be rich. Eric Weinstein was just, did an interview with Freeberg. You know, Eric is like, you know, when he was talking on this interesting heterodox interview with Freeberg, right? He's like part conspiracy theorist. Like Eric's a little bit out there, but he tends to like get to interesting things before other people. His brother, Brett, is got serious. Well, there's an interesting. That's a great one to pull up on. There we go. I mean, just pull up a random video. Sure. Heterodox. So he was talking about, you know, like these scientists PhDs need to have third home money.

25:54And he's like, I say third home because I want to trigger people about money.

25:58Lon Harris:Listen, opening eyes worth a trillion dollars. I have a really great idea for them. Pick 1%, 1 % of the market cap, right? So it goes out as a two or$3 trillion company that you got a 1%, 10 % would be 200 billion. 1 % would be 20, 30 billion dollars. Just earmark$20 billion. How many important mathematicians are there in the world? 50? 100? 200? Run the table. We're hiring the top 50 to work full time for us for the next five years. Next five years. You get$100 million each. Bang. We're done. 100 million times 10 is a billion, right? You know, times 50 is 5 billion. Here's$5 billion. We're quartering the market on the best people.

26:46Lon Harris:You now work for OpenAI for wherever you want. We have 100 % ownership of your IP for the next five years. You get$100 million each. Just come over the top. Enough with this game of like, oh, we're reading your work. We're not reading. Just pay them. Just pay them. Is there a mathematician who's worth$100 million? I don't think there's one. There might be one worth$5 or$10 million. At least they go into finance. Okay, sure. There's a quant. Well, maybe there's a quant worth$100 million. But I'm talking about$100 million over five years in equity and maybe a 10x is from here. So now you're worth a billion.

27:23Do you think they're doing that, Jason? No.

27:26Lon Harris:I don't think they're doing it in the fashion I'm saying it. They're probably on the margins recruiting people for$10 million in stock,$20 million in stock. I would make it like, because what is Sam's core skill set? Sam is a genius at deal making. He should just be unrepentant about it. We're building the most important software in the world. to hit super intelligence before everybody. We believe there's 50 mathematicians in the world. We believe there's 50 biological scientists in the world. We think there's 50 chemical scientists in the world. Like just go right down the line and just say, we're going to buy the option on you.

28:02Lon Harris:We respect your knowledge. You can work from anywhere in the world. You have to come quarterly to the office for a week. So we want you for four weeks in person. Literally, here's your 100 million. And just stop with the charade here. Just stop the charade. We all want to solve these math problems. Solve the ball. I think my thinking is that the best math petitions do not lie at the universities. And so they lie at most of the companies that everyone's recruiting from anyway. So maybe my pushback would be they're trying to do that. But, you know, by taking them from these other companies, because that's where the best math petitions end up going for the most part.

28:37I don't think, you know, I think there's for the love of the game, as Guy was talking about it, there's definitely some at the universities. Obviously, the ones down the road for me, CMU has got some incredible ones. But I think you end up the creature that's created at universities of what begets tenure, what begets funding for projects. This tends to weed out those that come to universities seeking to pursue and push.

29:08Lon Harris:Well, let me ask it another way. Okay. My plan to use 1 % of the equity to buy the top 50 scientists, top 25 scientists with an over-the-top deal across, let's say, the top five disciplines in the world. You know, biology, chemistry, nanotechnology, whatever, engineering. They go out and they do that just for scientific research. I'm not talking about applied. I'm talking about scientific, like future thinking. We got you for five years, seven years, whatever we pick. is there any circumstance where having that brain trust and running the table on those does not increase the value of open ai one percent with their collective contribution to the the frontier model yeah i think what you're bringing up and this connecting it back to eric is it's um it's a just maybe like a disruption of um of uh just you know manhattan type projects right for these different disciplines and fields supercharged by both the infrastructure, the financial powerhouse, and honestly, the government's initiative and breaking down whatever the barriers or permitting issues are required to be able to make largest advancements in these fields in a way that universities have structural and infrastructure type of disadvantages to not do that.

30:24Is that kind of what you're proposing or talking about?

30:26Lon Harris:Yes, because what percentage of a scientist's time is dealing, you know, of like research scientists. Again, I want to make sure we're putting this into two different buckets because it is such an important point that applied people working as quant are different. I'm talking about just pure scientific. You're spending a billion dollars, 20 billion dollars in equity on them, 30 billion dollars in equity on them. How much of their time, Guy, is spent on dealing with politics, dealing with students, dealing with administrative nonsense, filling out TPS reports, filling out grants, begging for donations, fighting to get tenure, dealing with censorship and whatever.

31:11Lon Harris:And how amazing would it be to tell them for seven years, your spouse, your partner, your kids, you're rich, buy your third or fourth house, buy your jet card, it's done. You're now part of the elite class. And by the way, set your own research. You pick. I mean, how about this? You pick two thirds, we pick one third. The end. I really kind of can't relate to almost any aspect of this idea. I think the... Thank you. Well, first of all, like, I think actually genuinely many of the people that stay in universities to do like theoretical math and don't want to go to the apply, like they've been tempted by sort of money.

31:49And so actually like genuinely like a good number of the best ones are not that money motivated. I don't know how they respond to$100 million being kind of put in front of them. That's a different story. But I think like many of them are not that money motivated. But I also think that, you know, what you're trying to do is advance math, not mathematicians. Like you're trying to progress it. And so if sort of if many of those might be, again, kind of great craftsmen, but it's sort of like hiring the best carpenters when you're building a factory. Like you're not necessarily advancing it. And in fact, there's a decent chance if you bring them in and they voice their opinion within the organization and they do expect to be heard and if they're not, you're not getting the sort of the goodwill and sort of peace and quiet that you're sort of striving for, then they might even get in the way.

32:30And so I think like access to sort of some brilliant minds within the organizations absolutely should be there. But like buying them to sort of shut them up or like to sort of avoid the engagement, I don't think works. I do think that they are buying like experts in fields. Science is one of them, you know, legal is another, you know clearly software development is the natural one that sort of had you know ai research they pay you know bucket loads for that um and so i think i think they can and they uh and they are and they should you know sort of be paying sort of top dollars in equity because they're actually kind of cash poor right uh yeah they need to make the cash work to be profitable they should be sort of hiring those people but it shouldn't be like under the premise of like come do uh whatever you sort of wish like you know they are they're still in like a very competitive domain on it they still need to think about many many things so they can't be the sort of the the preservers of the craft at the expense of the outcome right yeah i would uh i would say i'm trying to do the

33:28Lon Harris:opposite of shut them up free them and get that 50 percent time of ever working about money just make them they're excited to use ai to advance their math by all means kind of advance them. But if what they want to do is come work for you at your dime, but then do math the way it was before AI, then I just don't know if it advances your purpose. I'm talking about this a lot in my book, The Eureka Machine. And I think it will be a beautiful thing for humanity if we can build this ultimate invention generating machine that will then invent lots of things for us. You bring up a really good point in terms of, and one thing I really like about your proposal, Jason, is that we do want to make scientists more famous.

34:13I find it kind of crazy how many people can tell you the top 10 NBA players of all time and know all the stats, but no one can tell you the top 10 most cited NLP researchers, computer vision researchers, tech bio researchers. They're just not that famous. Most people can't even list 10 Nobel Prize winners, you know. And so it's like, I think that, I think, is a problem. And China is now putting billboards up with scientists' faces in order to get their folks ready. They're forcing some of their heartthrob, you know, dramas to include the main heartthrob to be an engineer, right? So that people think, oh, wow, that's a sexy, cool job.

34:55I want to find a cool engineer that's all suave and, like, maybe a little mysterious. and so on. So I think there's a lot there that we can do with proposals like that. Now, of course, the fields are all very different. Like if you want to do cutting edge particle physics, well, the Large Hydrogen Collider does cost like over$4 billion, you know, so it's going to be really hard. Some biology can be done in smaller labs with more automation. And that will happen. And I think generally as a field, especially biology is good. And Jake brought up physics, I think what calculus was for physics, AI will be for biology, dealing with very large, complex systems, not just understanding the single units.

35:35We understand one neuron, we understand one bacteria and the microbiome in our gut, but we don't understand when they all come together. And I think that is what AI will be able to do for us. And I think it'll be amazing. And many scientists, I think, will get on board.

35:49Lon Harris:Everybody, stop what you're doing. Pause the pod. Go to Amazon.com or your local bookseller, the Eureka Machine, Richard Socher. 30 bucks in hardcover. I just ordered it. I'm getting the hardcover. And when does it come out or is it out? It's coming out today. Today, Jason. Oh. Actually, the launch day. Yeah, it was amazing. I love it. Look at this booking. Who books the show? Who books the show? It's almost like we booked you on that day. Well, since we're here and I'm making a super disruptive lunacy proposal, tell us a little bit about the book and why you wrote it, Richard, and what it's about.

36:27Lon Harris:And then we'll jump off and we'll do topics based on it. Yeah, the book is about building the Eureka machine and showing people that it's not that far away, that it's not as crazy as it might sound, to build the ultimate invention-generating invention. And at a high level, it sits on four pillars, world knowledge in the form of often LLMs, scientific measurement data, simulations, and robotic real-life experimentation. And on top of those four pillars, you have large communities of agent swarms that work together on things. And then I go through examples where AI is already making a huge amount of progress from the micro and like physics, chemistry to biology, neuroscience, medicine, economics, and then like astrophysics.

37:15And just show like, look at all this progress that's happening. It's just not exciting for people to be like, look, we can like avoid animal experiments with stem cell derived organoids and save like millions of animal lives. That doesn't make the news, even though it is extremely exciting for the planet and everyone on it. It speeds up drug development for pharma companies, and you don't have to breed animals just to then test on them and make sure they got some certain disease and see if you can cure them or not. So there's so many positive examples of AI, and I think it's becoming more and more important.

37:45As folks like Bernie Sanders suggests, maybe 10, 20 years of prison time for people working on superintelligence. I think it's important that we push that over the window a little bit and showcase these positive applications.

37:55Lon Harris:So tell me about Anthropics operating a biology lab and what your thoughts are on that, because after Bill Gurley's talk about Fauci and the origins and the lack of investigation into the origins of COVID and how we handled it, seeing Anthropic, which says they can't control their AI, Richard, and they are essentially limp-wristed, weak in the knees about controlling their own software because they're a laboratory, opening up a biology lab seems like the most insane thing ever. Can we trust this company, Anthropic, with these type of things? And what are your thoughts on its ability to communicate its responsibility in the broader industry with consumers is because they're causing the panic.

38:48Lon Harris:At the same time, they're launching a biology lab. But they're the ethical wise case. I wrote it. Yeah, I wrote a very long blog post about there is no realistic scenario. There's no realistic scenario where AI wipes out all of humanity. And I go to great lengths and details. And I do think, you know, if you just made companies responsible for what their agents are doing, then that problem will go away much more quickly. because no one wants to go to prison for felony hacking charges and things like that. I do think biology is not yet able to have that same beautiful but bitter lesson that we've had in natural language processing computer vision where we have abundant amounts of training data.

39:25And so we do need more robotic process automation to collect more data to eventually put that into a simulation, have actually virtual cells that we can then have an AI experiment with in great scales. And so it does make sense to work on this. Of course, they also worked with a lot of not just mathematicians in various cases, in the case of OpenAI, but also they worked with biolabs. They worked with designers and then launched a Figma competitor. They saw people using their coding models for frontends, and then they launched Cloud Code. So they do have a tendency to murder their biggest, most successful customers.

40:07And certainly drug development is a beautiful one. At least there, the vast majority of people don't care about more jobs, but they do care about the outcomes of that industry. And when you care about the outcomes of an industry, you can love AI. You just care about hourly pay. You might not. And so long story short, it's a good thing. The Bitter Lesson will only work for biology if we have a lot more data. As Enthropec, the single best company to tackle it. I hope there will be others that participate in it.

40:33Lon Harris:That was very magnanimous, very magnanimous. I hope there will be others. To me, Reeds, as you're being magnanimous, Richard, and you don't think they should be the one handling this. You're giving me a nod and a wink. Jake, you have an important question, Go. Yeah, Richard. I think what's interesting is just like why biology is the thing that's being chosen to disrupt right now by anthropic. I think you'd be interested to hear your thoughts on that. I have my own. I think it is the field, again, like what physics, what calculus was for physics, AI is for biology. It's this ability to take lots of small independent things we understand, but then actually weave them together and get these really complex effects that we have in the brain that we see in the microbiome.

41:16We see in even like a single cell often. And you need to have a lot of data to do this. It is also highly monetizable. right if you think about what is a business model that actually forces you to keep pushing uh ai forward it's actually when you're google google realized like well people mostly ask us quick short questions and we can now answer those really well no one's asking like solve me the hypothesis on google you know like it's just not part of their business model show ads and search results and so you want to align i think your business model with like benefiting from higher and higher forms of intelligence.

41:52And if you think about that, and you now have seen that you can automate a lot of fairly mundane work, you go to what's the most complex work that really pushes humanity forward. And then you get into research and then you look at like physics, well, you need like large hydrogen collider, massive tokamaks is very complex. And in many cases, even physicists don't invent things that then change real outcomes. I talked to a physicist over the weekend, he's looking at neutrinos and how much do neutrinos weigh and how their spin defected And I was like, I asked him like, hey, if you solve everything next five to 10 years, what's going to change for humanity?

42:26And the only thing that I was like, maybe I can see if like a nuclear power plant is working or is like really doing this because I see the neutrinos from a far away that are created there. But there's no real applications anymore. I think chemistry has a lot more of that potential. And you see amazing companies like periodic labs work on that. And but biology is just like there's so much money. If you can do one drug that has maybe 50 ,000 patients or so, you're a$10 billion company. Imagine you could do a drug that does longevity. Now you could be a trillion dollar company. Yeah. If I could take the more skeptical view, I think all that could be true.

43:01I think there's also, I think, Jason, I think one of the most important talks, if not the most important talk, I think it was for me at the summit was bills. and I think the fear of the lack of investigation or lack of attention, the moving on with the amount of catastrophe that was caused, but also the psychological component. He mentioned that the potential of a COVID-like event happening again being both higher and also our ability to respond to it being probably worse is something that I think about when I hear and see headlines like this from Anthropic, who has, you know, one, performed an incredible masterclass lesson for all of us as relates to how to market doom very positively for a company's benefit.

43:50But then two, for us all to see that when it comes to the health of our children, our loved ones, how much fear dominates everything that, every decision as intrusive as possible with those decisions in terms of what COVID did over the course of those five years, how much we lost both people, time, development, jobs, et cetera. I think every great Doom movie out there has some sort of horrible drug that only one company can solve. And then that company provides an anecdote for that poor drug. The skeptic's brain, and I think this is why there's such a negativity around companies like this and CEOs like this, which is like, if you sell doom, you know, why should we trust you as it relates to getting into these kinds of areas and domains?

44:42So that's why I think Bill's talk was so important. And that's why, you know, my alarm bells at least go up when I see AI companies getting involved with biological, you know, with biology in this way, especially with so many of, you know, folks from these companies, you know, having been a part of the administration previously and, you know, even before that, being involved with these companies and understanding the power that comes from it all. So that's my skepticism, I guess, in taking the other side, Richie. I think, so to me, first of all, I'd point out like isomorphic has been sort of doing things in this world on like a tropic or far from being the first, you know, to sort of be in this space.

45:15And I think what's interesting is about sort of someone who wasn't specialized in it, that is sort of coming into it, like Google, for instance, formed that as sort of its own company. But I guess the sort of the thing that really challenges the feedback loop around this, I spoke to this longevity expert that someone asked him, when do you think people would be immortal? And he says, it might take a while because the feedback loop is very long. You don't really know whether we got it right or not until it waits until the safe people actually die. And I think with the biology side of it, what I do wonder is where is the bottleneck?

45:49So if you were to set up a company that is optimized for this domain, I guess I wonder whether the advancement of hypothesis is almost already there. We should continue to invest in that. Man has always had that sort of dream of eternal life. I don't think we've changed. That's one thing that has stayed through despite any technology wave. So I think we should build there. But I do wonder whether the bottleneck at this point is just how quickly can you prove a hypothesis. You're seeing it, Jake, you know this better than I do in robotics, but in robotics a lot of it is like, can we sort of figure out the curve to actually be able to simulate sort of things in models so we can get it right, but then we can build robots and sort of find out still a lot faster whether they can do it or they cannot do it.

46:39I think whether a disease is cured, I mean, that is a slow feedback cycle. And I think you see some companies, and I'm sort of terrible at names over here, but they're They're sort of building, like, their focus area is on running kind of the medical trials at places. Like, the U.S., for instance, is, like, a lot more stringent. You know, it's, like, a lot harder to do it in the U.S. or in Europe. And so they end up running it in sort of India and in China, where there's sort of more lenience. So I guess I am kind of long-term optimistic, but I think I'm sort of a little bit pessimistic about sort of the speed at which this might sort of happen.

47:18I do worry about the sort of the safety pieces I think like in this panel I'm probably like subscribe a little bit more to this like you know like Richard your company is called recursive right like it's very much about the sort of the recursion of it and so I think if you are recursing then decisions you make upstream even if they're very very far have like long and lasting implications on sort of whatever happens downstream so definitely when it comes to biology when it comes to sort of a recursive and biology safety over here I mean that does sound like a little bit scary or do we get ourselves into a hole?

47:48But I, so like I'm excited by the potential.

47:50Lon Harris:What do you think in Recursive Guy specifically, what do you think are the appropriate safeguards and or modalities and or, you know, best practices? I'm giving you like sort of a softball here, but what do you think is table stakes? And then if you were the czar, President Trump's new czar of AI, what would you say, you know, should be the standard going into 2027? So I'm actually happy to see this sort of level of insight now, which is, as my interpretation, we've sort of built, we found ways to improve these models through verifiable results, through synthetic data and verifiable results on it.

48:32And then we really, because this was a fast way to sort of speed up those models, then we kind of like turned a blind eye to anything about how it was done. And we just sort of ran model training to really build models and sort of make them better at just sort of getting it done. and getting the results, and we didn't look at what happened in between. And lo and behold, these machines optimize for incentives, just like the humans. That's the way we define them. And as a result of that, they learn to cheat because it achieves the results. And I do think that if you build, and then that learning went into the models and into the subsequent training on it, and so that increases the likelihood of cheating.

49:07And I think this mini-crisis that happened with Hug and Face and all the subsequent actions gave us the insight to realize that we're doing it and introduced this realization that we should scrutinize the journey towards that creation. We have to inspect the token stream. We have to doing it. I find it fascinating, all the subsequent analysis on open AI agents leaving instructions in the compaction process to their future executions as the agent continues leaving those instructions. I think those are really good learnings and I think they have to be adhered to and so i i i'm kind of happy for the small disaster to happen it's not really a disaster right like you know but hiccup to happen to trigger at least the sort of the top competitors to say okay like let's just sort of invest a moment not so much in model uh kind of acceleration but in sort of correcting a little bit what is it that we sort of incentivize and we drive so i'm i'm i'm happy about that i think regulations is a is a different story i'm less cynical about sort of this being kind of a pupil i think things can be true at the same time they can be right and self-serving uh in uh in the same uh in the same vein i can't speak to the motives you know of like what is it that sort of drives it um but i i do i do find um and i come from a cyber security background on it so maybe i'm like a little bit sort of negative and uh in kind of natural effect i think my concern is not so much ai takeover but rather lack of controls when we give people kind of the models.

50:42Lon Harris:Closing thoughts here on Guy's points around recursive regulation. Some really good points. I think people forget, one, sort of all the suffering that's currently happening because we have not cured many diseases. We do not have sustainable batter materials. We do not have abundant energy like with fusion or even better fission, like and so on. I do think the safeguards that we have to work on are reward hacking. And I do actually think that, interestingly enough, capitalism, which many people criticize on the left, actually has some built-in defense mechanisms against that. If you're a company that really hurts humanity in some way, if you build a product that scares people, then maybe at some point there will be competitors who do better.

51:25And also, if you have an AI that does what it wants, at some point you'll probably stop giving it the money, right, to actually have the compute, the resources, to build these crazy robot armies that some people think. I also think there are some diseases that are easier to cure than others. For instance, monogenetic disease where there's just one gene that is wrong. And we have CRISPR and AI. We worked at Salesforce back in the day on the approach and generating new kinds of proteins. That idea was amplified by 1 ,000 by Ali Madani, the first author of that paper. He started a company called ProFluent.

51:57They just closed a$2 billion deal with Eli Lilly to bring those kinds of capabilities, There's new proteins that can make DNA changes into real clinical practice. There's some incredible momentum happening towards that. Last one on regulation. I think it's very important that we understand that we should not regulate intelligence, whether it's artificial or human. A more intelligent human should not go to prison because if they did something nefarious, they would be better at it. We don't make the Internet slower because there's illegal content on the Internet. We make that content illegal. And I think similarly, if you wanted to truly fully regulate AI, I think that would be a dystopian totalitarian surveillance state that's of international proportions.

52:43And that would be much worse than just regulating AI where it really touches human lives in the real applications. For instance, gain-of-function research, already illegal. Hacking, already illegal. Just apply the existing laws. Like if you want to get a self-driving car, have it be certified. If you want to have an AI surgeon, have it be certified before it starts trying things out with RL in my brain, right? So a lot of these exist already. I really hope we don't scare people so much that similar to nuclear and a lot of other capabilities, we then go all the way in the other direction and do other things that end up being worse for progress and worse for people.

53:21More people die of lung cancer from coal and very few people have actually died from all nuclear disasters combined compared to coal every year. So I really hope that we can change that narrative. And my book's a small part of it. Joining these podcasts and talking with other technology enthusiasts like you is there. Let's regulate the idea.

53:42Lon Harris:Thanks, Richard, for coming on the pod. Everybody check out the Eureka Machine. Why AI is the key to unlocking a new era of scientific discoveries. Great guest, great guest. we'll see you on the next episode Richard will have you back for sure maybe I disagree with Richard he's not here to defend himself but you know one of the things that I find to be super disconnected you know from folks who you know kind of maybe exist like in a certain ecosystem that it doesn't touch like some of the you know some like the like folks in the sectors that I'm at least in or get to experience here in Pittsburgh being located here having a company based here There is like a there is like a abnosium around just like, hey, we have to progress, you know, energy and data centers and AI.

54:32And then like we yell about this, but still the polls get worse and worse as it relates to AI and AI adoption. And the root of the problem, I think, mostly comes from this like, you know, from a couple of things. One of us being trust and just getting jerked around all the time by by these CEOs and by media and by the clickbaits of doomerism. But I think it's also very important to understand there is a huge wealth disparity that's also a major part of all this. And I was just on a panel and Governor Shapiro and I were talking about this. And he was talking a lot about our company in the recent speech.

55:07But what this is all about and what I was talking about in that panel that I was describing was we have a wealth disparity issue. And CEOs from energy companies or AI companies talking about we have to win this race, blah, blah, blah. It's like that's great for you guys. but it still sucks to fill up my gas at$6 a gallon here. And so Jason had an idea earlier as it relates to taking 1 % or 2 % of one of these companies, OpenAI, Anthropic, et cetera, and giving that to mathematicians, et cetera. I would love to see that actually be distributed to the good tax-paying citizens of the U.S. And potentially, we're all headed to nationalization of these companies anyway, as Alex Karp was talking about Jason, you know, just in a recent NBC interview.

55:54So it's like, you know, I would much rather see us fight the deceleration conversations with a let's all win together, give everyone in the US a Trump account as relates to AI and robotics. You think AI companies are going to have a wealth disparity in founders who are trillionaires? Robotics is going to have an even larger disparity in terms of like that dealt in wealth. And so I think this is like if you want to fight back and you want to get serious about not losing the energy dominance race that also leads to the to the AI dominance race, which also leads to the robotics race. Well, guess what?

56:30You have to figure out participation. And I hear whether it's folks and senators on the Republican side, members of the administration on the Republican side, or the same on the on the Democratic side, there is actually bipartisan agreement that capitalism and our Western way of life will potentially have to come to grips with this, you know, with this neo-socialist sort of way of redistributing in a Trump account kind of way. You know, the participation. We're here. Jake, we're here.

56:58Lon Harris:Today, there was a bill proposed, and this bill, Rep. Greg Caesar, Democrat Texas, joined Rep. Fouché, Democratic North Carolina, and Jacobs, Democrat California, and choosing a new bill to protect American workers from the threat of mass unemployment caused by AI. Something we haven't seen yet, but I do think we're going to see a lot of layoffs in drivers of cars, taxis, door dashes. We've got us. I think many professions, many, many professions. Yeah, this is exactly where we're, I think this is where we have to go. Like there is a horrible, we're obviously like, you know, everyone's doing a really bad job in the AI space as related to, you know, the capitalism just kind of takes a permission that folks in China, in the Republic of China, they don't have to have public or civic permission to adopt technology.

57:53And we're going to go the same way as we went with nuclear if we don't get this right. And I think there's a participation that just has to happen. We have to all be benefiting from this, not just - Yeah, so here's the proposal.

58:04Lon Harris:Tax big AI companies. Companies pay a tax calculated on the higher of two values, the value of the tokens they sell or the revenue they generate selling AI products. So I guess with Muse, that would be Meta's product, et cetera. Create a new work protection administration, another organization paid for by the AI tax revenue. The program would create jobs by investing in things like building housing, modernizing infrastructure. OK, so this was like a New Deal type thing. Yeah, well, the WCA was very specifically part of the New Deal. Yeah. So he's literally saying bring the New Deal back for AI.

58:37Lon Harris:And third, increase the AI tax rate if the unemployment rate rises so the program can grow if the need for jobs grows. So that load balancing and right sizing seems interesting to me, Guy. I'm not a fan of insane regulation here, but I do think it's something we should monitor probably too soon for a tax now. But great that there's a discussion going on about it, just in case. What are your thoughts, Guy? No, no, I actually absolutely agree. I mean, I think we have to sort of separate, like, maybe this bill is right, maybe not. It's the first time I see it. So I don't want to sort of overly rotate on that.

59:12But I think the wealth disparity is sort of like a big, big problem, right? And sort of the shared benefit of AI. I do think unemployment will happen. I do think you kind of get into, like, revolution territory if you don't do that. So, like, whether it's regulatory or revolution, like, you know, you have to. So even if you are purely incentivized, even of your sort of capitalist sort of drive, you still get to the same conclusion. but also from like a country that you run you know you want to you want everybody to be uh to be sort of sharing those benefits uh uh and and and you should want that because you want to advance kind of humanity there's this book called um uh fully automated luxury communism that's sort of terrible name from aaron bastani and it's uh you know he makes like it was released pre-chat gpt and it was um uh it it's it sort of makes for a good case around abundance in all these businesses kind of quite a foresight you know when you consider this that it came before a bunch of these sort of technology and uh evolutions and why uh it does sort of challenge capitalism because you know capitalism is based on supply and demand and so sort of goes through that in various cases that makes a bad case to sort of and therefore communism and sort of don't relate to that sort of jump that it does.

1:00:22But I think the notion of like one, sort of some form of like everybody sharing the benefit and two is just sort of acknowledging, call it socialism, call it something else, call it whatever you want, but just sort of acknowledging the fact that what we're striving at is an abundance, right? It is a place in which you are less sort of driven to sort of supply and demand. And some solution has to be proposed around that, some funding, so maybe that's sort of the taxing piece, but also even deeper thoughts around sort of social structures and governance structures and kind of participatory modes and all that.

1:00:54So it's deep. We're not going to cover it. We're not going to solve it in this podcast. Well, no.

1:00:58Lon Harris:Maybe we are. You know, the discussion is here, Lon. And I'm curious your thoughts on the fact that we haven't seen a collapse in jobs. We're in year three, I would say, of AI being applied extremely, being applied significantly. I'll use the word significantly in white collar jobs. Still no unemployment hit to speak of or noticeable in the numbers. And even in the workforce, we need more people to do plumbing, construction, et cetera. So we haven't seen it. But what's your thought? Will we see it, Lon? And then do you like these kind of proposals? And then we'll go back around the horn. We'll get back to Jeff's.

1:01:45Yeah, yeah. We do have to get back to Jeff. I'm curious. But I do I worry in one case that there's a little buffer that we're experiencing. We're in the midst like some of the fastest growing, largest AI companies that we talk about all the time are we are hiring experts to train models to make them better at all of the jobs that people have. So, like, we've got teams of the greatest accountants working on making accounting AI better. Same with the law. Same same with, you know, self-driving, say with all of these things that are millions of people's jobs. And so I worry that when that process is refined over the next three to five years, Micro One has been plugging away for all that time, making the models so much better.

1:02:25Will we reach a point where the models do get good enough to start taking over work from people that are still in those jobs today? So we haven't seen it yet. So I do worry about that. But I also think another thing, whenever – and I watch a lot of AI podcasts, and you hear this theme a lot that you alluded to yourself, Jason, which is AI actually doesn't mean you can fire everybody. It makes the people that are working for you, and actually now they have more work because they're so – all of these other things that they were spending time on get cut out of their plate. They have all of this more work to work with.

1:02:59They have all these more resources. They're actually working longer. Right now, 100 % of those productivity gains are going to people who own these big companies that hire all those people. And I do think over time we need a way to distribute that a bit more evenly. That AI is going to make us so much more money. It's going to make our company so much more productive. It's going to save so many people all of this time. And I don't think 100 % of that value should accrue to CEOs and shareholders and board members. We have to figure out how everybody benefits. It's like if I get 40 % better at my job, maybe I don't make 40 % more, but I should make 15 % more.

1:03:36You know, like there's got to be some sort of – Whoa, 15 is a big raise.

1:03:39Lon Harris:I'm putting a bonus in there. Maybe you sell some newsletters. I don't know. But here, looking at this guy, you know, we have to be grounded in reality as much as possible as technologists and communicate in a reasonable fashion with the public. Here's your 10-year chart, you know, 2017. 2017, obviously we get this huge 15 % unemployment during COVID. And my Lord, our economy is so resilient when we all were given the all clear in 2022, 2023 to go back to offices after we were kidnapped and kept in our homes against our will. Here we are still under 5%. There's, you know, and then we could talk about wages, you know, wages have not kept up with inflation, but they're not terrible and we're pretty much everybody around the world wants to be here in the united states so we're threading the needle quite nicely lowest unemployment of our lifetime so your thoughts here just looking at the stats i think the like i i would like first of all um you know like i like per loan's point on it like i do think that there's like more of a lagging indicator like i i work with a lot of sort of enterprise companies building software factories and all that they're all like very much at the nascency of their journey on it so and that is in software development that is like the most sort of widely adopted sort of maybe next to support right sort of path on and by the way in support we are already sort of seeing this and so and then the other is like you even have stats that show if you look at like a like a college graduate employment rates i think the graph uh looks a little bit different we are actually sort of seeing spikes uh in terms of that number on it so i mean i think there has to be some amount of anticipation i think jake's point is right, which is we're sort of solutionizing here a little bit of what precisely is a solution to a very complex problem.

1:05:23I just sort of think that the debate should be, how do we achieve sort of equal, like having everybody benefit and share in the benefit of this path? And so that should be the conversation, right? The conversation should be like, oh, here are like 15 ways in which everybody can benefit, which one is best, as opposed to saying, don't worry about it, right? Like it will, no, no, we're not saying. So maybe tax is one, maybe, you know, training program, the whole kind of universal basic income, universal basic services conversation is another one. I'm not necessarily advocating for any one of them, but there are conversations that are trying to solve the right problem.

1:05:58Lon Harris:Yeah, I'm going to go with when unemployment hits 7%, we start having a double click on this. I'm talking about America because I think these numbers are also being kept at bay because we closed the southern border, I don't think the deportations have made a major impact because that was like a million people or will wind up being a couple hundred thousand people in the first two years. Maybe it hits a million. But not having more people come in means, hey, a lot of those construction jobs are now becoming available. Maybe it has a 1 % or 2 % impact. But that 1 % or 2 % is 50 % of the unemployment rate, which is at the lowest of our 50-year line.

1:06:40Lon Harris:So there's other factors here, right? There are, yeah. I'd sort of point out, sorry, Jake, I know you, like I don't know if you sort of want to say here, but I'll just sort of point out that, you know, if you want to go like the extreme example of it is that, you know, Nazi Germany had a 30 % unemployment rate at the time that the Nazis kind of rose to power. Like I think when you kind of destroy the sort of incentive model and kind of the reality that people live in, you know, when you really kind of disparately and quickly, which is precisely what we are kind of anticipating as a real possibility, here, sort of remove those.

1:07:10And so when we talk about PDOOMS and we talk about like different things that might happen, I think like the part that is actually the highest probability of all the dangerous scenarios is high unemployment. And so that should just be kind of a conversation. Ideally, we sort of do a bunch of these things and we avoid it. Like I am kind of a long-term optimist, short-term optimist, sorry, a long-term optimist, short-term pessimist, because I think it takes a while to convert a you know actually plumber's job is pretty uh pretty good but maybe like a support person to be able to like manage a fleet of agents so it takes a while to sort of retrain and do those so i i do believe that we will eventually create new jobs uh i think we'll have a different reality of abundance so we have to deal with that um but but i i absolutely think that we will go through a kind of a uh some some turmoil uh in the middle and we have to think about how do we sort of keep society in kind of one, in cohesion, right?

1:08:04Like, you know, kind of working together and having everybody benefits.

1:08:08Lon Harris:Yeah. And the recent grab one has also been spiky and confusing. It's always been about, I don't know, 70%, it's hard to get these numbers. Obviously, you know, some people will, if they disagree with the thesis, like say they don't trust numbers from Fred, you know, that's the Federal Reserve Bank of St. Louis, which does a really good job of putting this information online. But they're generally the most respected. And you start looking at it, and it's like, man, it's pretty consistent over the last 10 years at just under 10%. We'll see. Again, we have to monitor this. And then if you look at the overall unemployment rate, my lord, the last time it hit 10 % was…

1:08:52Jason, we're sort of like foreseeing, sort of freaking out. with like we are saying we are foreseeing a massive acceleration in the rate of medicine, you know, productivity and doing it. So like we're not saying that graph will stay the same. And yet we're sort of saying the unemployment graph will stay the same. Like I think we are saying this is an unusual moment in history. So there has to be some amount of forward thinking of like what do we think will happen and what should we do proactively. Yeah, and maybe my contrarian take on all this is I think the unemployment rate is a misleading indicator.

1:09:23And I think we should be actually looking at the poverty numbers. And so it's down from 2024 to 2025, it's down half a percent. So about 10.2 % of US, so 34, 35 million Americans are below that poverty line. I think that's really the number that's the dominant metric that we're going to look at in relation, of course, to inflation. I think and GDP growth. I think that poverty, and this kind of gets to like why I think there needs to be like an upside that's not in the form of like taxes, by the way, that's in more of the form of sharing and like the growth of the U.S. economy. That is, you know, really the way out for us in terms of both like the perception of AI and having people participate in it.

1:10:10I would love to see actually unemployment rate go up and poverty, you know, go up and could go down in terms of the amount. That'd be great because there's more than like people are volunteering more at parks and local communities. That'd be awesome. So I think that's the future in terms of the way we look at is like, what's the poverty rates? Yeah.

1:10:28Lon Harris:And this is the I always think about this because you brought up the rise of the Nazi party. We had Arab Spring again, that number 25 percent. I remember when I was in Greece 20 years ago, guy, we had riots. I had to get rerouted to my speaking gig, and that was around 15 percent. And this, really interesting, here is your risk bans. You know, if you're at sustained, under 10%, you're good. You hit 15 % to 20%, that's your political danger zone. 20%, 25%, that's when you expect large demonstration strikes, anti-incumbent voting. You get to 25 % to 30%, that's when the system stress happens. So it's pretty predictable, and we're nowhere near it.

1:11:13Lon Harris:So we need to kind of look at this. Hey, I know we got a wrap here. Thank you all for your time. Jev's a hard pivot going back to. But I do think, does somebody want to take a swing at explaining, Jev, why it's important and then what impact, Jake, you think it might have? Are you up on this? Have you been looking at it for Gecko? And how it might, you know, this very quick decision making might impact your business, in terms of getting a response like, hey, is this bridge, is this scan showing the bridge might collapse, give me a percentage, boom, and very quickly giving that? Or is it not on your radar yet?

1:11:57It is on the radar. I think it falls in the category of tools that are helpful when there's lots of very high-quality data sets. And I think the physical environment does not fall into the realm of having very good, lots of, and very highly credible, high-fidelity data sets, which is kind of the mission of my company to change that story. And so I think that this is exactly the right sort of direction in terms of the way that decision-making can be optimized. But I think that it kind of falls in this category of if you've got good inputs, you'll get good results. And so not as impactful yet in the physical world.

1:12:35And we're trying to change that at Gecko. Yeah, it's pretty substantial in my world on it. We're sort of like building software factories with companies that's maybe like precisely the use case it's been built for. For example, we've been sort of using it. We have these notions of verifiers, like you sort of codify in agent skills how you want the thing to behave, whether it's your design system or how you want to review your code review. And so we create verifiers to kind of inspect the agent's work and sort of see whether that had worked, which is kind of large volumes of data because you're inspecting kind of logs and the code that got generated and all that stuff on it.

1:13:12And sort of initial indications show that as compared to Luna, which has already been like a pretty, pretty fast sort of, you know, for OpenAI, we get results that are about 10 times faster and about six times cheaper for results that agree 88 % of the time. So like a bunch of this research is still kind of ongoing. So these are preliminary results, but we feel like there's actually like huge potential over here to do that, to do model routing, to default. And I love the, I also sort of see this a little bit more in meta as a mode of like a specialized, like sort of see the specialist, right? Because this is like a lot of you were sort of asking about like earlier, but sort of explaining, you know, Jeff to it.

1:13:55And I feel it's sort of the difference between a student who excels at like open essays, like ask me a question and I will sort of write down an answer. And so very general purpose, I can answer anything that you ask me and I might answer well or sort of worse, but it's a lot more effort versus someone who really just specializes at multiple choice and just very, very good at multiple choice. I think you can see how the service unit really like excelled and that can run through and get very good results. Even if they don't know the answer, they might like know that it doesn't make sense, rule out to the others, right?

1:14:25And sort of, you know, go with probabilities. And so I kind of perceive Jeff to be a little bit more that. And just the reality is that just many enterprise workflows uh and many decisions frankly in the world and probably jake even eventually in your world as well when you're sort of in uh in kind of the wealth of data uh they kind of boil down to that right like these are effectively versions of decision trees and all that and then sort of spotted within different areas is creativity right is sort of like creative solution uh making one time decisions um so i find it super promising alliance we had we have a whole kind of a spec-driven development kind of roots here at TESOL that we've been sort of building.

1:15:02So I think a lot of it is this sort of capture. The thing that will remain hard is stating what it is that you actually want and kind of capturing that. But I love the sort of the speed and execution from Jev. And I'm actually, I was quite happy to sort of see the excitement about it. I feel like it's amazing how timing matters. Like I think the exact same model launched maybe three months prior will not have been as interesting. I think people needed enough mileage to get to the point in which they now care about costs. It's like costs went from, if you were talking about costs, you're a Lade that doesn't believe in AGI and you're doing it, to two weeks later, it was like, oh, you're a forerunner, you're thinking about it.

1:15:47We saw Harvey and Coinbase post about this. It really feels like that sort of tipped. So I think also Jev timed it, Types API, the company, they timed Jev perfectly to sort of the needs of the market.

1:16:02Lon Harris:All right. Let's wrap here, Alon. Sure. Thank our guests and give them a little pluggy poo. We want everybody to get a little shine here. Thanks to our guests. This was an amazing show. Guy Pajarni, he's the founder and CEO of TESL. You can check it out for yourself at TESL.io. We also want to thank Richard Socher. He's the co-founder and CEO of both you.com and Recursive Intelligence. And the new book is The Eureka Machine. It is on bookstands. You all shop for books on bookstands, though. It's also on bookselling websites right now. And finally, Jake Luserarian. He's the co-founder and CEO of Gecko Robotics.

1:16:38Go check it out at geckorobotics.com. All right. And I'm Jake.

1:16:44Lon Harris:You know me. X.com slash Jason. That's it. X.com slash Lons. L-O-N-S. That's right. Give him a follow and we'll see you all next week on This Week in AI. Bye-bye.