Tech Brew Ride Home - (Bonus) Me On The "Securities" Podcast Talking AI And Regulatory Capture

Episode Date: November 18, 2023

Links: "Securities" newsletter "Securities" podcast Danny's twitter Learn more about your ad choices. Visit megaphone.fm/adchoices

On April 4th, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. Hello and welcome to a special joint podcast episode on AI regulation. I'm your host, Danny Crane, and I'm the host of Securities by Lux Capital, a podcast and newsletter devoted to science, technology, finance, and the human condition. Co-hosting with me today is Brian McCullough, the host of TechMem's Ride Home podcast, which focuses on the day's news headlines, context, and conversation.

Also joining with me today is Matthew Lindley, the founder and writer of Supervised, a Substact newsletter devoted to the machine learning, AI and business space. And finally, joining me is Lux Capital General Partner Shaheen Farshi. Today's podcast is focused on a huge range of AI regulations. There's just been a surge of attention from U.S., UK, and European policymakers around the threat of artificial intelligence, their concerns about it and the risks existentially and whatnot around AI technologies. Over the last two weeks, we've seen President Joe Biden introduce his massive AI executive order. we saw over the United Kingdom, Prime Minister Rishi Sunak, announcing the Bletchley Declaration of principles around AI development. And the European Union continues to strive towards a law around

AI regulation that has been going on for about two years, but has made a little bit more progress over the last few weeks. And so I want to talk about today just why are regulators so focused on this subject, what they're trying to do, and also some of our concerns about how that might squelch the innovation potential for AI startups. AI has been around for 60 years, but it's not something that felt immediate and present to folks in government especially, but folks in society in general, till what? It's not even a year yet, because when did GPT come out? It was like, I think, November 30th. We're almost to the one year anniversary. Yeah. So, you know, it's it's one of those things where everybody's been trained for the last 30 years of what's this new thing? What's this new thing? What's this new app? Should I be on it? You know, and so, you know, because what, 100 million people? People are on Open AIs, various products. This is one of those times where folks are not going to be caught short, I guess, or surprised by innovation.

And Cheyenne, you're about to literally get into your car and drive to the responsible innovation labs as launch event. What's happening there? Because that's sort of the latest update in this sort of regulatory process. We have a lot of investors coming together. And you have folks from the White House there as well, or Secretary of Commerce there as well, and the goal is to hear out the investors and hear out how the White House thinks about this. Obviously, we want to be able to support innovation. We're a for-profit entity like the other

investors that are around the table, and it's important of them to understand where the government is likely to take regulation and how that could affect our future investments and how that could help our investments get ahead. And obviously what we're most interested in is seeing this technology succeed and ultimately penetrate into all industries as soon as possible, obviously before our rivals get their hands on it and do the same. And let me ask you, Matthew, because you've obviously focused on a lot of the big tech companies as part of your supervised. You know, how did everyone sort of lay down their chips on the table? Some of the companies are being open, some are closed. They're taking different

regulatory approaches, kind of give us a scan of the map right now of where everyone's standing. When we're thinking about the launch of chat TVT and thinking about the sort of proliferation of large language models, there's kind of two things happening simultaneously. The first one is that it worked. You know, we've had six, seven, eight years of like false starts in technology that were kind of like flash in the pan, or not necessarily flash in the pan, but you know, you had like Uber and Lyft, which was, you know, venture subsidized capitalism. And then we had crypto, which had sort of debatable applications and things like that, and it's still, the jury is kind of still out there. And then when you, when we had launched chat TPT, it literally worked. Like, you could use it

for coding. You could use it for writing C plus B minus grade articles and things like that. There was an immediate reaction like, oh my God, it's great. And then the second immediate following reaction was like, we have no idea how this works. What the issue was is it sparked this challenge around transparency. And the language models themselves are naturally not particularly transparent. I feel like part of what's happened with some of the activity around regulation is just the, both just the technology naturally doesn't have an element of transparency to it. And so these companies have to figure out a way to add layers of that on top of it. I think what's so interesting about AI, you know, so much of it is a black box.

We don't know explainability. We don't actually know how these models work. We train them and they function. We put in inputs, outputs come out and sort of under the concept of the Turing test, as we sort of get those outputs, we're like, it seems kind of human. We're asking questions that's answering them properly. We know those are sort of the correct answers, but I feel like that's intermixing with science fiction. Because one of the most notable stories that I saw over the last couple of weeks was this anecdote in Time magazine about Joe Biden, who was at Camp David,

watched Mission Impossible, Dead Reckoning, Part 1, which is about an AI takeover from a sort of AGI agent that sort of takeover to the planet. It was like, oh my God, that's so dangerous. We've got to regulate this. And that's actually what's spent up the exact. order. And so part of me is questioning, like, is the challenge between this black box that's, we're sort of filling in that black box with our own fears and our own science fiction of what artificial intelligence is possible. Because at the same time, we're trying to regulate and block everyone from using it. I can't get it to order food for me. And so there seems to be a huge gap between the capability to what is actually capable of doing. And to me,

I see this unbelievable that the U.S., the U.S., the U.K., the European Union, private industry is all coming together in like two weeks on a topic that on other issues, which we could have absolutely used more help. They've been absent for decades. I mean, you know, history hat would say that that would be, again, after 30 years of feeling like they were behind the eight ball in terms of technological innovation, regulators want to at least present that like they are ahead of the technology this time. You, the sci-fi analogy that I would use in this debate, because obviously there's been, you know, given when the movie robot first came out and the term was, that's like almost 100 years ago, there's been 100 years in fiction of fear of making machines that are smarter than us or more capable than us. The science fiction analogy that I would use for the opposite side of this debate is that this is the compute and the computers that in our best imaginations for the future, we always imagined. happened. I'm talking about what we imagined in the 1950s and that sort of Jetson's version of computing. But the real science

fiction analogy would be the Star Trek computer. So on the one hand, you have the fear that the robots and the computers are going to take over the world, enslave us, or whatever. The other side of it is what Captain McCard does on the enterprise is computing that is just make it

so, right? It's, I want this done. I don't have to click an icon. I don't have to enter commands. I don't have to even deal with a file menu or anything. So in a way, you've got the debate that this is technology that could blow up the world. On the other side, this is the computing sort of utopia that we have always wanted to have.

You know, Humane, was it this week or last week, came out with that AI pin. And so we've gone through everything. We've had the pads, like became the iPad. Ever since we had flip phones, we had the communicator. And now they have the com badge. the ComBadge requires the Star Trek computer to make that a reality. And so that is the computing utopia of I don't need to know anything underlying.

I don't need to know various apps. I don't need to juggle any balls in the air. I just need something done. I ask the computer to do it and it gets done. And so in a lot of ways, I feel like this debate is what everyone's deepest primal fears have always been about computers versus the sort of promise of computing that we always were sold but never actually delivered. We've been dealing with file menus for 40 years.

We've, you know, going back to the command line, you know, like this is the, this is the sort of computing that, because no one's talking about yet, AGI would be computers that are smarter than us. What we're dealing with right now for probably the foreseeable future is computers that aren't smarter than us in the sense that they make decisions for us, they just obviate all of the messiness of us making the decisions. Yeah, well, first off, mid journey is a command line, so let's not forget about that. Yeah, yeah.

To get into the infrastructure side, because I'm an infrastructure geek, you know, one of the things in analytics is there's this idea of that, you know, if I can query my data with natural language, like, you hear that described as the Holy Grail, right? Like, can I communicate with the data that I have, whether I'm like an enterprise, company that's like trying to figure out like, oh my god, I'm going to lose this customer. Or if I'm just like a normal user, you consistently hear that described as like the holy grail. Like can I communicate with the information that I have personally or new information out there with just natural.

And we have been so far off from that. You know, you talk to people like, you know, Tristan Handy at DBT who's been in the architecture space for a long time to say like, oh yeah, like this has like been working on this for like 15 years, right? And so I think that just part of, the other part of it is just like it always felt like so far away. It always felt like it was like 50, you know, 100, 500 miles away.

And then all of a sudden, Chachapida came out and there'd never been a level, like a level of a cultural zeitgeist attached to it, where the only comparable, I think in my head is like, you know, people love to call it like the iPhone moment. But there was also like Facebook where there was just like this immense, like cultural impact crater that happened after its launch that immediately created this, immediately created this like debate around everything almost like instantaneously.

If I can make, sorry, one more history analogy here, because I think this will lead us into the idea of the regulatory capture if that's an issue. Again, going back to when computers were invented, what were they good at that they were better than humans? Number one, calculations. they were designed to, you know, drop bombs and shoot missiles and things. And what took 100 people in a room, you know, weeks and weeks now took days or seconds or hours. And the second thing was storing data. You know, a computer will always, in theory, have that data and remember it. Humans can't do that.

The third thing is, as you're saying, with the Facebook analogy, is communications, which weirdly for decades no one thought that computers would, you know, obviate phones and things like that. But so if you follow those three things that computers were better at us, This fourth thing that at least this version of AI right now is doing is we've spent the last 30 years putting all of human knowledge onto the web and other places. So storing it. And now this is a technology that allows it to pattern match it and regurgitate answers based on that corpus of information. Right. So in a sense, people always use the analogy of, you know, the smartest intern you'll ever have or, you know, the personal assistant of your dreams and things like that.

Yeah, yeah, yeah. Yeah. So, but in a way, if you think of that as the fourth sort of type of compute, again, with my analogy of you just have a computer. Let's use Jarvis. Jarvis is your personal assistant if you're Tony Stark. The question is, will the largest model always win, will it always be the best and most sophisticated? In which case, will there only be one winner? Will there be one or two or three?

Will there be only one Jarvis or will there be multiple Jarvis? And so then that gets into why some people like Ben Thompson recently have been saying, isn't it interesting that some of the people asking to be regulated are the people that already have the Jarvi out there? I think Brian and Matthew touched on a lot of this, but for me it's, there are three things I like to call the three eyes that have made AI what it is today. You know, one is imagination, and a lot of that has to do with science fiction over the past 100 years. No other technology has captured our imagination, and by design, because fear sells tickets, you know, end of the world, that sells movie tickets, has captured our imaginations and drawn our attention as much as AI.

Think about the Internet. Think about the cloud. Think about mobile. Sure. I mean, to your point, Brian, all of these. technologies were kind of introduced to us in Star Trek and other science fiction shows, and so we kind of expected them. But then how or the Terminator was something that we hope would never happen. And guess what? Something that resembles what was introduced to us in science fiction that ended the world is now supposedly happening. We're going in that direction for a lot of people. The second is the interface. So, you know, our grandparents probably could interface with personal computers the way we on this call did back in the 80s. When I was coding basic, you know, when I was in the fourth grade, my, you know, my mom had no idea what I was

doing. Whereas today, when I show my mom that she can, my mom likes to sing, when I show her that she can create a song with like, you know, by creating a prompt consisting of four words, that absolutely blows her away and she can now interact with it. Whereas she couldn't really interact with PCs in the past. Who could interact with the cloud? Who knows what the hell the cloud is? But we can interact with AI. And the third, which I think is the most impactful, which is what we're talking about here, is our imaginations. Human beings are capable. One thing we're really good at is letting our imaginations run wild. And none of these other technologies, PCs, mobile, cloud, crypto, none of these other technologies in recent history have catalyzed our imaginations

to run wild the way AI has. So these three things together really have catalyzed a lot of these discussions that we're having right now and the concerns, obviously, from our governments and obviously society at large as to how this can impact us. I mean, Shaheen, I don't know if you've met some crypto bros, but I'll tell you, when it comes to imaginations running wild, some of the folks thought the tokens were just going to completely transform the fundamentals of human existence.

So you clearly are not going to the right parties. That's my conclusion from that whole thing. But I want to combine the group together on this because I think we're opening at the core question, which is our imaginations are opened up, right? So chat TVT, the fastest growing app from zero to 100 million in history, nothing else comes close to it. Not TikTok, not Instagram, not Facebook. Nothing else has grown quite as quickly.

So there is this sort of like a hit very fast as of now. It's only been 50 weeks since it was launched. But in that 50 weeks, you suddenly had every single person, every congressman. has talked about it. Politicians all around the world are having a discussion about it. But then the question becomes next. I want to go back to Brian's setup,

which is, you know, which model ends up winning here? You know, in the last couple of weeks, we've obviously seen a couple of new models. Matthew, I think you were just writing about one in your most, most recent newsletter was yesterday. There was a new model that was just released. It had like 73 versus 68 on Lama.

Was this mistral or? It might have been mistral. It's so hard to keep up with this. It's absolutely crazy. It's every day there's just like something new popping up at this point. And over in China, we saw 0.1.a.i, so Kai Fu Li's AI startup, it looks like sort of a reposition of Lama, but has also done extremely well, particularly on sort of low memory situation.

So it's particularly efficient at its performance compared to some of the other models out here. But I think there's an open question of when it comes to the capabilities, we just talked about humane, we're talking about iPads, or talking about the Star Trek world. all this is miniaturized. We had these technologies. We had an iPad 20, 30 years ago, if you remember the Newton. It just didn't have all the capabilities we wanted it and it's come together.

But the question is, what is going to drive the AI innovation over the next five years? Is it going to be open source model? Are they going to be efficient open source model? Is it going to be the world's greatest multi, multi-tillion petabyte scale model? What wins in this context? Because I think the question is, you know, so focused on the regulation, if that if you think the world is coming in the open source world, it'll be very hard to regulate. It's going to be very hard to control. It's going to be hard for any government to do that. Whereas if it is about

these huge data sets owned by a couple of major companies, that's a world that's much more regulatable in the first place. I went to the first ever AI engineer's summit last month. And the word of the week was open source, because these engineers and these developers, they want stuff that they can, you know, fiddle with themselves. If everything's a black box and it's proprietary, and again, at their event recently, OpenAI did not open source even GPT3, which lots of people were hoping they would do. The engineers at the summit, they wanted things that they could run on their own devices, that they could have a greater control over sort of the secret sauce that's made about this.

So one of the things that everyone forgets is open AI is called OpenAI because people first saw, you know, this new type of compute in a world of the cloud could be a, a service where you just pay one person for the Jarvis, right? One entity for the Jarvis. And that's originally why Elon and other folks found it OpenAI. And the problem is, is that all of those engineers also understand that until they can get to a technology where it's not as expensive as eye-watering to create these large models, as Sam Weldon famously said, you essentially have a scenario where the only people that can afford

to do the cutting edge, the bottom. most sophisticated stuff are the people that have billions and billions of dollars and essentially endless pockets of money and compute to allow these models. So that's why you see, you know, Google, is it Google going with Anthropic and Microsoft with Open AI? And so you're seeing, I said on the show that weirdly, if this is not even a year into this AI moment in inverted commas, it feels like there's already incumbents because the cutting edge models are already sewn up, at least in partnerships with the major tech companies

that we're familiar with. And then like, you know, the meta is going open source because that's sort of their wedge to do the other. And then you have the VC class, which, you know, some of us are a part of, that obviously the, if there's not, if there's only three or four winners, then this is not an investable space. So you have this. you have all of these different incentives where if you feel like, if you're open AI and you feel like you're already ahead of people, then you can build a moat by, number one, getting as much money as you can and iterating as fast as you can.

Number two, asking for regulation because that's another way to build a moat. Because no, no, no, we don't want small developers being able to run these models on their laptop. That's too dangerous. And then the other, one of the things when people think about this, like Ben Thompson's piece, as I mentioned, is talking about how, isn't an interesting that? that these incumbents, as I'm calling them, are the ones calling for regulation. Everyone forgets about Nvidia.

You know, invidia is sort of the resource that everybody needs to move it forward in this moment. Do you think Nvidia wants there to be only two or three big winners? They want a thousand flowers to bloom. They want an ecosystem of tons of different models, of tons of different size companies and things like that. So it's interesting that everybody is racing to create a moat for various reasons. Some of them, monetary, some of them, the technology is still so compute intensive. But that's why it's sort of everyone's trying to, you know, firm up their angles right here. I don't look at this as a zero-sum game.

I feel like the opportunity is absolutely massive. And you look historically, technology has always disseminated to the masses. If you look at compute and semiconductors back in the 50s and 60s, you had to be a government to make computers. And then the 70s and 80s, there were a small handful, probably half a dozen companies. Fast forward to 2004, and I taped out my own chip as a broke rat student, part of my research project. So historically, technology has been able to land into more and more hands. Now, does that mean that the prior and company, are going to be completely obviated, maybe, but I absolutely see a future where the

anthropics coheres, open-a-as of the world that have their proprietary models that are accessible through APIs, have a market, have a giant market. And there are going to be those use cases, which I expect to be more enterprise class use cases, where there is compliance and explainability and safety requirements where they have to be open source. And there's going to be, in my opinion, many companies that are going to be working on this. I think it makes complete sense for companies because of the reasons that Matthew and Brian brought up, that Nvidia, the hyperscalers, anybody who has anything to do, whether they're making chips or offering cloud compute, or even companies like Cloudflare, to offer any kind of infrastructure to make these open source models available to their customers to use within the context.

of those services. I think that will, that's going to be a right of passage at that point. And so this isn't VHS versus Betamax. You know, this isn't because some, you know, this isn't Phillips CDs versus whatever. The, the discs we had back in the day, I forgot what they were called, the Sony discs. But like, at the end of the day, it's, again, it's a giant opportunity. The applications are practically endless. There are going to be many companies, and there will certainly be room for these giant, quote-unquote, incumbents that we call them today, and the long tail of startups that are going to be enabled by open source. And I think it's our responsibility to make sure that we protect the open source movement

and make sure they're part of this. Go ahead. And this isn't even talking about the data abstraction layers either because there's data bricks and Snowflake and MongoDB where all this data is actually stored on top of that. And there's, and like it's important for those guys to also like not pick a winner. To your point, right? because they need to, they, they, they're incentivized because, you know, the house always wins, like more compute is better, right? And as long as you're doing, using my service for

compute, I'm making money and so on and so forth. And so as long as I, as long as I cultivate as much as physically possible, like, all I'm doing is like more compute and more compute, more compute, which is great for me. Matthew, let me ask you about Open AI specifically, because some of the developers that I've been speaking to since, since the keynote, you know, they were surprised by some of the things that OpenAIA already offered them, but they're also incredibly skeptical about, they feel like some of the tools that were even announced were intentionally underpowered. So the sense that I got is that from the developer space, folks are still wondering, to what degree is Open AI holding things behind their back because they want to be

the winner and have the product? Like the GPTs and things like that, that's the typical platform play that we're familiar with, where here, do your thing, develop it, we'll share revenue and stuff like that. But I'm wondering to what degree, because Sam Altman is openly saying, that's not our product. Our product is going to be AGI. So I'm curious if you're hearing from developers some of what I was hearing, which was we don't know necessarily where to go because we don't know where Open AI is going

to be a competitor to us in the future, obviate a startup that we're working on because they'll just make that their product? What are you hearing in terms of how folks were reacting specifically to Open AI? Part of the challenge with AI for developers, too, is we're still onboarding. You have people that had expertise in neural networks, their PhDs that left to

go into industry. I think Trey Dow is a good example. He's the creator of Flash Attention 2. He works at this startup, which I think is a Lux company Together AI. But you're so good, Matthew, at getting those in there. We didn't even ask you to do that. Well, it's like, speaking of like, what are developers talking about?

But, but we're like, we're still like onboarding people, onboarding developers and to teach them how to use this stuff in the first place. Like the concept of using an agent like intuitively, intuitively is pretty easy. But like getting that started from scratch is not, not easy. But like there's there are all these like ramps that need to be in place. in order to get people accustomed to actually using these technologies. And Open AI, I think in particular, kind of gets that from like when you look at the the products that are releasing.

Like, Assistance API, I think is a great example. The most important thing is getting as many people like familiar with that topic as possible, because then again, like they're using my products and all that kind. So I think it's more about abstracting open anything. I think you're getting at something that I think is, it sounds obvious, but I actually think it's much less intuitive than we think. which is we assume that open source, you know, is going to be harder to compete against a closed source product. We assume that these incumbents, which are mostly closed source,

want government regulation. They think the government regulation will sort of lock in their monopoly. And I think a lot of this is actually contradictory. So to your point, Matthew, you know, we're still bringing on developers into this community. People don't really know how to code AI systems. It just got started. You know, LLMs aren't that old, BERT's not that old, open AIs, APIs are not that old. And so open systems tend to do really well against closed systems when it comes to onboarding developers because it's much more intuitive and easier to understand and use these systems. Second piece, though, is at the government regulator side,

you know, as we're talking, the big United States v. Google court cases underway, and Google runs the largest applied AI model in the world, which is Google Search, certainly the most profitable AI model in the world. And part of this trial, if you're sort of following the coverage, is just how much we're learning about how Google has maintained position digital advertising in Google search, both on the algorithmic side, how it prioritizes its own rankings,

how it builds traffic with other sources like Apple and Firefox in order to sort of maintain its dominance. And so one of the things that kind of comes out of this, if you follow kind of Lena Kahn's track over the last 10 years, is as we sort of opened up the black box, we learned there was all kinds of things that were going wrong behind the scenes. And in some ways it was actually the open technology is that were much easier to regulate

because we can actually evaluate them, see them, understand them, as opposed to the closed source technology. So I actually think it's ironic because in some ways, the largest companies may actually be inviting their worst enemies sort of inside their tent by saying, well, we're very safe, we're trying to protect everyone, and then you're going to find out that they're not. And then regulators are going to feel, you know, kind of betrayed. And I could imagine a world in which actually this was a terrible strategic error on their front, you know, this year that comes back to bite them in the, in the rear end on this parentally safe channel.

From a perspective of if you want this technology to evolve intelligently, regulation in theory is one way to go that leans on the safe side, and there's tons of great reasons for that. But the other part of it is, and again, developers screaming for can you open source these models, can we fork these models, can we do all these things? That's the best way for the technology to develop quickly and to develop in ways that will be creative and will be fascinating and will be creating all sorts of new tools and new jobs and new. Again, coming back to the idea that Open AI was called Open AI because folks did not want this to be behind the sclerotic sort of huge walls of companies that are trillion-dollar companies. 25, 30 years old. I think that the open source way is probably the best way and the fastest way to get us a Google, for better or worse, of this era,

of the next 20 years, 30 years, the next Facebook, the next Apple, whomever, of this space. And so that's another thing to think about is one of the arguments for open source is companies tend not to continue to be innovative the next generation on, 20 years on, after they're founding, 40 years on, there are obviously some exceptions like that.

But if this is a brand new space, you want brand new players that have no legacy systems because they'll build the ground rules in the way that makes sense for the new era. So, I mean, that's an argument too, is that open source is just the way to get this technology better faster.

Let's say what we will about, you know, Twitter, sorry, X's the platform formerly known as Twitter's AM model that came out. You look at like some of the stuff that they're doing, and it's like, when we were talking about, like, can I build some, what does it look like if I built it from scratch? It's built on Jacks and Rust, and these are, like, very new technologies.

Facebook was built on PHP. That was an 11-year-old technology, right? And so when you start thinking about, like, what are these things look like if they're built from scratch? One more thing historically, you know, one of the things people say about, you know, computer science has never had sort of its atom bomb moment or, you know, the reason we have a Nobel Prize is because

the person that invented dynamite felt bad about the fact that it could be used in wars and things like that. And so the Nobel Peace Prize is, you know, sort of an answer to that. One of the things, like, let's assume that AI is as dangerous as nuclear power and nuclear bombs. Yes, probably only governments should have control of that, but also that technology came about literally in wartime. And so one of the things that people forget is that the

regulatory regime in place was a wartime regulatory regime where you could, by Fiat, regulate entire industries because entire industries were being run by the government at the time. If nothing else, it's interesting that for the first time, governments have for the last 50 years basically been hands off most technology, especially computing technology, because the idea was, is this is how we're going to have growth in our economies and more jobs and things like that. It is interesting, and it could be just again because of the natural swing back and forth of things like, you know, now we're under the Lena Khan regime and things like that. They want to get ahead of it of the eight ball, like we said earlier in the show. But also consider the fact that this is

the first time I can remember that, A, governments around the world are trying to regulate a technology, before it's mature, but before it's really sort of even gotten on the scene in a full way, right? Again, not even a year into this specific AI moment. Number two, we're talking about things like regulating compute power, which is akin to only certain people can have access to nuclear technology and nuclear energy. But one of the things about these regulations, which people can correct me if I'm wrong historically, have we ever suggested laws that you can only make your computers so powerful. Now, we have suggested that you can only export computers that are so powerful, but we're literally talking about only certain, there's a threshold by which you can't be more

sophisticated in compute. And I, ironically, export controls to China. Right. Of Nvidia chips are precisely prevention of high-powered computing to certain countries. But now it's being suggested internally for your own domestic, you know, at some point, only one person will be authorized to have the best chip, sort of like the foundation series in reverse. Right, exactly. I think the only thing that comes to mind, and I think I referenced this before, and a securities article was sort of export licensing controls around encryption algorithms. There's a huge debate in the 80s and 90s to North Korea, to Iran and elsewhere that said,

you know, for RSA and a couple of other algorithms that you could not export those online. But again, at least in the West, every time that laws have been suggested to, regulate encryption and only the government can have control of encryption, those have been beaten back. Right. So, like, that's a, I'm just saying it's an interesting time that for the first time, and maybe, I feel like the analogy to the, this is the nuclear moment for computer science is the apt one, that essentially governments are again saying, like they did at the dawn of the nuclear era, this is too powerful for even private enterprise.

I'm not saying that governments take it over, but we need to have Gar-Rail's and place. And so this is computer sciences moment in that sense because the government is literally saying this far and no further potentially. And I will say, and we are running out of time, I'll wrap up by saying for those who saw Oppenheimer this number, you know, to my mind, this is the discussion around nuclear in the 1920s, not the 1940s, but when Nealzboer and others were just sort of figuring out the atom and how it functioned, you're starting to just, we haven't even done fission yet. We're just sort of getting the physics around and be like, we're going to just shut this down. And to my mind, it's so early to be putting in, even necessarily

guardrails, because exactly what the problems will be, exactly who will own it, exactly how much control you actually will have on it is very much an open question. In the case of nuclear, it was the fact that you sort of with the weapon, and then you went to energy. In fact, we actually tried to make it more peaceful and more commercial over time because it did start on the defense weapon side. AI is sort of on the opposite side. We've not started on, we don't have killer robots first. We have Roombos who are cleaning the floors haphazardly, I will say.

And so to me, there's just it's so early and a lot more it needs to be seen and people shouldn't just jump on it because it is a black box. But we have been here for the full time of our calendar. I want to thank all of our folks, Brian McCullough of the Tech Mean Ryan Home podcast.

Shaheen Farshi, partner at Lux Capital, Matthew Linley, founder and editor and writer and all around Jack in the box for supervised sub-Sex newslet our focus on AIML, data infrastructure. Thank you so much for joining us. Hope to see you again soon.