Latest / 4kd.ai / Ep. 11 | AI Regulation: Safety or Self-Interest?

Transcript

Ep. 11 | AI Regulation: Safety or Self-Interest? · 4kd.ai

• David Moses (00:01) Welcome to another episode of Forked AI. I'm here with Jack BeVier, my co-host. Jack, how have you been? Jack BeVier (00:06) Great to see ya, sir. I'm doing great, man. It's been a fun couple of weeks here. and always a pleasure to get with you and talk about what's going on in our business right now. It's been very busy. David Moses (00:17) It has been it has been here as well. So you you had you had indicated you you may wanna go may go may wanna go on a a a rant potentially about you know potentially against regulation, which is a new a new thing for you. So go. Jack BeVier (00:33) Yeah, no, so it's the that's just kind of like a piece of it. So I I as everyone I'm sure has been noticing recently in the news, very recently in the news, like the the frontier labs and various associated beneficiaries calling are have been calling for a slowdown in AI advancement. And there's been a lot of calls for additional regulation on AI. And it's you know, there was this, there was this one researcher who worked at Anthropic, I recall, and left after six weeks, tweeted something, you know, tweeted this thing that there's you know, that the the stuff that's going on is dangerous and we need to slow down. And then it seemed like it was a coordinated effort because a lot of other folks who were similarly interested in that goal amplified that message and it really kind of like took off and it. Feels to me as a passive observer, as there has been a coordinated campaign here to try to like bring sunlight on this issue at this time. and I'm like sitting back and I'm looking at it. And the messaging is not new, in my opinion. Like they have been calling for that. That has that has been a that has been a topic of conversation since there was a letter written four or five years ago. That I think Elon Musk was one of the signers of. And all these other Frontier Labs also signed calling for, you know, you know, maybe a I think it was a moratorium. They were calling for like a six-month moratorium at the time. That was like four or five years. That was like four years ago. Like, so this is not a new topic, but it is it is now certainly a recurring theme. and I guess you could interpret that in a couple of different ways. You know, one that hey, this is like these guys keep pounding the table talking about this. Maybe we should and and they certainly are the ones That should know, like, let's listen to them. and I'm so I'm like, I'm zooming out, I'm I'm listening to it, I'm zooming out, and I'm just like, and then my sister calls me and she's or texts me and she's like, hey, you know, like what's your what's your take on all this? You know, like she knows we do this podcast, so what's your take on all this, right? And I'm, you know, I'm gathering my thoughts and thinking about it. And I've been, I don't know, forming, forming some thoughts over time. And there's a couple things that I've noticed. And the timing of what's going on here to me is just damningly circumstantial. Just it is so there's so much circ self-interest for this messaging in this moment that I'm like really putting my little tinfoil hat on and like want to hop up on my soapbox in the you know, in the middle of you know, in front of the Washington Monument and start, you know, getting my blow, you get my horn out. because I So here's the the things that I'm that I'm seeing. The we we have had a revenue growth at the Frontier Labs that is literally the biggest thing that's ever happened to money, right? Like we we like a hundred billion a month of revenue. You know, the I think it's the NFL something worth like 25 billion. it was recently at some point. These guys are doing a hundred billion a month of revenue. And that's the enterprise value of the National Football League, you know. and so like this is the you know, the the enterprise value here is astronomical, unlike anything, you know, it's literally an order of magnitude higher, usually more than what we're used to thinking about. And and there's a phenomenal reason for that. well, and the so these guys are and so there's a tremendous amount of business interest, and they want to go, they want to IPO, right? Both anthropic. And OpenAI are trying to actively IPO those businesses because the valuations are gonna start with a T, you know, they are you know one, two, three trillion dollars each at of enterprise value. So there's like a lot of financial incentives for them to IPO, and they have not yet. And yet they have not yet. And as I'm as people talk about how dangerous these things are and And that, you know, and and as a result calling for, you know, they don't want to advance it further. I it occurs to me that I'm like, I mean, I kind of agree. I have gone in the past nine months, I have gone from knowing nothing about cyber essentially nothing about cybersecurity, to deciding that my company was going to be AI native and we are, you know, on that path and I'm going and and spending a lot, you know, investing a lot of money, you know, towards that end. Investing a lot of money towards that end. And as a result, since my enterprise value is accruing in that layer, I'm spending a lot of money getting my shit together, you know, pretending, you know, learning how to be a CTO and trying to figure out like what I probably should have done five years ago, 10 years ago from a cybersecurity perspective. And as I've learned more about that, I'm like, yes, actually I was quite naked about a year ago. Like I I did actually absolutely did have like surface area to attack. that that I have fortunately since, you know, closed to a significant extent, though nothing's perfect. so I don't disagree with the premise that this tool can be used in a nefarious way. I don't disagree with that premise. But the example that's getting a lot of attention recently is this hugging face jailbreak, where it was essentially yeah, there's a gamified situation where OpenAI wanted to see what would happen. if they if the if we gave a a swarm of agents a a goal and in this particular case they they you can argue in you know you can argue argue in retrospect that they kind of like bent the rules a little bit or broke the rules depending on your perspective and in or in furtherance of this goal and that agents are capable of that, right? That they're not like just little docile submissive things. Like if you tell them go do this thing, they are capable of moral justification the same way the humans they were trained on, in my opinion, was capable of moral justification to the greater good. You know, I'm not, you know, they told us not to access the internet. But the thing is we have this question about the clarification of the rules. And the only way we can get this answer is by ex you know, is by going through this door that we found and and they did that. And they absolutely did that. So they're, you know, I agree. They are they are capable of those things. I would add though, just like humans are, right? Like these things are not, these agents are not shooting lasers out of their eyes. They are writing code that humans could write. Like they're not some superhuman thing. It's just the cost structure of a bunch of nerds who can write code has gone down by a thousand X. That's what's different about this. It's the the unit economics of a line of code. Have changed massively over the course of the past five years. And so that's that's the fundamental thing that is that is driving all of this. So just like you could have a thousand hackers in China try and fuck with you f you know three year five years ago, now anybody with you know an un you know an unhit, you know, with a with a jailbroken LLM, powerful jailbroken LLM can also use those for a nefarious intent. So Anyway, I just wanted to like bring that down. Like these guys, you know, this this is a very th this this these all these threats already existed. Cybersecurity weaknesses already existed. It just got cheaper to exploit them than it did before. So we can go with a bunch of different tangents the David Moses (07:42) And and and and that shouldn't be understated, Jack. Like it's when when you say cheaper than it was before, it's like like two Yeah. Like like to to Jack BeVier (07:49) By like a hundred X, by a thousand X, yeah. Yeah. David Moses (07:53) some you know pick pick your f like it would be written in scientific notation. Like that's how that's that's how much Jack BeVier (07:59) Yeah. David Moses (08:01) cheaper and easier it is. Jack BeVier (08:03) Yeah, exactly. Yeah, exactly. so the so the thing is that when these companies go public or even as just as as people get to be able to use these tools more effectively, right? Or like get used to using them, right? Right now a bunch of rich companies, bunch of rich American companies are buying all these tokens, right? Because they're the ones who can afford And as the models get better, And and more functionality becomes cheaper. A dude in, you know, a dude who was formerly spamming little old ladies in Wichita from overseas will be using it too, because that's what he is. He's a scammer and he's trying to get money for his family. And the best tool currently is call rail. And in the future, the best tool is gonna be, you know, cyber attacking with. Jailbroken LLMs, it's that it's just coming. Like that's just the new tech. And so, you know, bad people use tech the same way good people use tech. The problem is that when people do st when when people do start to get hurt, right? When as they do with every piece of technology that has ever existed, every every piece of technology that has ever existed has always been fused but used with with nefarious intent by nefarious people. So when it happens, when this version happens. Those companies are gonna get sued to Kingdom Come. Anthropic and and OpenAI are gonna get sued to Kingdom Come. There is going to be product liability lawsuits as a result of the use of their technology for nefarious purposes or just accidental mistakes. The lawsuits in America will absolutely come. And here's the thing: I think the conversations that are happening between the investment bankers and the IPO investors and are that. Hey, what are you guys doing about product liability? Because you're telling us your product is dangerous. So we're gonna lose. So so when those lawsuits come, we are going to lose them because the makers have been telling us that their product is dangerous. And yet they get up in the morning, by the way, and go to work every day and continue to build a dangerous product. You put one coder from Anthropic. on the stand and ask him what the cooler talk was in 2025 and 2026 and the jury is going to fucking find you guilty. You were there the the product liability lawsuits are going to be lost. Like they're Facebook's losing product liability lawsuits left and right and and no one who built Facebook thought they were doing anything bad. These guys are going to work every morning saying that there's a pot you know quantifiable potential for an extinction event You think they're not gonna lose a product liability lawsuit? They're smoked. Like these guys are guilty on every product liability lawsuit that's good that they get hit with. And they and so you know what? Everyone's talking about the enterprise value of a trillion, two trillion, three trillion. I'm like, no, it's fucking zero. These dudes are bankrupt, man. Unless they can find themselves a they can engineer themselves a safe harbor, a legal safe harbor that. gets them out of these product liability lawsuits. The only way forward, the only way that any of these Frontier labs actually have quantifiable enterprise value is if they can engineer a legal safe harbor to get out of product liability. And I think that frankly, that is what all this bullshit is about. I I think that they have not been able to IPO yet because of this issue. The call for regulation is ha like they they it is is one by the way what the fuck is the federal government or any government frankly gonna do about to regulate a frontier lab and the they would have to hire the guys from the frontier labs which and to tell them what to do. And so what Bernie Sanders is gonna bring these guys into the room and be like, yay, t teach me how to be safe and they're gonna come up with a fifty step plan to make it safe. And they're gonna say, but that's very expensive. And so in exchange for that, Bernie, I need anyone who follows these steps and submits themselves to this regulatory structure needs to have needs to have a safe harbor from product liability. And I and I'm pounding the table, man. I swear to God that quid pro quo is gonna be part of that process. That's the only fucking reason they're doing it in the first place, is to create this legal safe harbor. And That's the only reason that makes any sense to me that that they would be on the morning show scaring my sister about what's going on with AI and like is this stuff safe to use? I have another pie I have a second, I have a part two of my of my sandbox, but I'll stop there because I think there's enough to talk about. David Moses (12:51) No, I think I think that's I I think you're spot on that they are they're they're really and and and look to some degree I'm not quite as cynical about it, right? Like if you if you built something that in the right hands could cure cancer and save millions and millions of lives, but in the wrong hands could lead to I mean, death, you know, a at the worst, but even, you know, more simple things like you know, the complete I guess the grinding halt of, you know human ingenuity when it comes to like writing novels or children's books or things like that, because people will stop because there's just no point. you know, the I I I think anytime you have that powerful of a technology, in order you know, there needs to be government involved to to some degree not destroy it, you know, to to to to some degree protect it, regulate it. I mean but but this is like no technology we've ever seen. Like if you if you think about like, you know, like the Nobel Prize, right? It it it's named after Alfred Nobel who, people who don't know, invented dynamite, right? So like TNT, dynamite, like you know, obviously that led to an you know an unknowingly n you know large number of deaths, right? But it also means that we have highways that go through mountains. Like there there are great things that happen with technology. If you have, you know, planes, you know, decades later, you know, someone is, you know, going to decide or figure out that if you hijack one, you can fly it into a building. Like there will always be supervillains, right? And to some degree, like the technology still has to like the the Does the good outweigh the bad, right? Does the potential good for good outweigh the potential for bad? and what's weird about AI is that it's ubiquitous, right? Like it's just as, Jack BeVier (14:47) Okay. David Moses (14:48) you know, dangerous from an existential standpoint to the writer as it is to the doctor, as it is to the, you know, marine biologist, as you know, as it is to you know, it it is it is completely ubiquitous in terms of where it can do good and where it can do harm. And yeah, I think I think to some degree they're up there Jack BeVier (15:10) Okay. David Moses (15:11) talking about the dangers of AI mostly because they need protection from the liability. I think that's that's spot on. That that's correct. Jack BeVier (15:17) It's that's David Moses (15:19) Like they they need that. They need government to come in and draw some line somewhere that Jack BeVier (15:23) Yeah. David Moses (15:23) says if you do this, then we will protect you from that. But it's also because they don't know what's going to happen. you know, as this technology grows and gets better and stronger, they just don't know. But but what they do know is that the answers to Bernie Sanders are really like not anything anyone wants to hear. Like the when I say anyone, I mean the people who are, you know, in the I'm terrified of AI camp and I'm, Jack BeVier (15:51) Mm-hmm. David Moses (15:52) you know, I think this is going to be a a a terrible blight on the world. Though th the answer to them isn't, hey, let's stop AI. Like the correct answer is you need a superhero to stop a supervillain. Like the only solution to AI, the nefarious uses of AI, is making sure that the good guys can move faster and are smarter and stronger than and one step ahead of those, you know. And and it does. It means giving up Some of our privacy or all of our privacy. You know, I mean, let's be serious. you know, like unless we're sitting in a room together, you know, there's no such thing as there's really no such thing as privacy anymore. it but it but it means it's it means giving up some of those some some of our freedom and some of our privacy, to to be one step ahead of, you know. The supervillains. Jack BeVier (16:50) I think that I that's where I come down on it for sure. In the end, I'm just like, the only the only path forward, because this tool will be used by nefarious people, the only path forward is to protect yourself with this tool. And so I just want to I want to master it before they master it. And I want to like go, you know, lock down the fort before everybody else learns how to use this tool and you know tries to tries to get in with it. so by the way, towards that end. But to me, that is the only place that we are going. Like a regulatory agency is not gonna change that path. That is the only answer anyway. Whether there's a regulatory agency, there's not a regulatory agency. Whether these w whether the whether it takes David Moses (17:29) No. Well their rules are gonna be out of date the second they write them. Jack BeVier (17:33) Yeah. Well, like that is not going to save me, right? Like I'm just like this this this whole like, hey, regulate me because that's what America needs, you know, to tell scaring the housewives of America. My my sister's not a housewife, but you know, but but you know, be when you're on the morning show, that's what you're there to do. You're there to scare the housewives of America. and like scaring the housewives in America is not going to change the way that we actually protect ourselves. Like the government will not save us in this regard. So what the fuck are they doing out there for like for like I it doesn't make any sense that the the the motivate that that you know stirring up that hornet's nest does not in my opinion do anything about making the world a safer place and so I I'm I'm offended by it honestly like I've if you can't tell I'm like fired up up on the soapbox I'm like I'm I'm offended by I'm offended by the practice I think it's I think it's insulting I think it's insulting to our intelligences I think it is exploitative of our hypothalamus and like and I and I I don't like it, you know. It it it offends me. David Moses (18:35) Yeah. Yeah, yeah. I mean it to You know, to some degree. They're basically saying to us, Yeah, you're being chased by the bear. So here, come hide in this bear's den. Jack BeVier (18:49) Yeah. Yeah, yeah, yeah, exactly. Let's go ahead and spirit. Exactly. David Moses (18:50) You know, I mean, Jack BeVier (18:54) Here's the here here's part two of my of my soapbox here. So the here I think these guys have an existential threat from from a from a capturing this enterprise value aspect also. They've got a rock and they got a hard place. That was the rock. The hard place is that Let me start with Fable Five is amazing. Like Fable Five to me is an inflection point because now it's extraordinarily expensive. And so it's not being widely used because it's so extraordinarily expensive. Except it is being widely used. It's being $100 billion a month widely used, but only by those can who can afford it. And they're all just like, this is really powerful tool. And they're all trying to figure out how to use it, but they're not quite sure. Like best practices. Have not been developed yet as to like how to actually implement this stuff. We're writing them in real time. Like that's what's going on here, right? So the thing that has impressed me so much, and I mentioned this on previous episodes, is that, dude, I've been able to point that model at a file folder and an email account and like. I just tell it the task that I need it to be done, and I don't spend much time telling it how to do it. I don't spend any in some cases, I don't spend any goddamn time telling it how to do it, and it just figures it out and frankly writes the SOP and makes some recommendations to me how to make my SOP better. Fable five, in my opinion, does business. It is smarter. It is it is sufficiently intelligent that it is as smart as any CEO. Or business builder or COO, it can do everything I can do and everything I can do better. Like it is just that intelligent from a business process perspective. And so I think the world is now in a CapEx cycle of rebuilding all of the business processes and making them all better and automated, you know, and softwareing everywhere that they can. And as a result, we have as a result, there's this revenue. Curve has exploded that has never existed by like an order of magnitude in the history of mankind. And that's how that's how much the market agrees with me, right? Like, and so it doesn't cure cancer. Fable six or fable seven will, you know, might cure cancer. and you know, do all the other kind of crazy, you know, do all this other crazy frontier shit. But it is sufficiently powerful. To get me to open my wallet at a very high rate, for for total self-interest, right? It is sufficiently powerful to extract value from the businesses of America and next to the world. It is sufficiently smart to and powerful enough to to capture the area under the curve, which is the US, which is the global economy, right? Like you don't need Fable six, it won't do me any good. It won't it it I'm I I I'm not sure I would pay for it. Yeah, I won't. I David Moses (21:51) No. Nope. marginally irrelevant. But I by the way, I'd make that same argument between Opus and Fable for ninety call it five, ninety-five percent of business processes. Jack BeVier (22:07) Yeah, I think that's true. I I I think that's true. it's just that the opus ones, it just takes more time. You know, it's it's just it I I think it it it it requires it'll Mike the the time it'll take me to build up rebuild everything with opus versus the time it'll take me to rebuild everything with Fable, I think is materially different. But David Moses (22:25) Yep. And for you that makes sense. For me, that makes sense for you know Yeah. For Jack BeVier (22:29) Well we also just all have a budget, right? Like we we can't just David Moses (22:32) for for some for somebody who well, but even within your organization, right? Like like there are people for whom the the the variable cost of their time you know isn't it doesn't favor using Fable over Opus because for that little for that five percent, you know, they're just as likely to have They're just as likely to break it as they are to, you know, to fix it. and the time involved is is you know, 'cause Fable Jack BeVier (22:58) Yeah, no, I think that's fair. Yeah. David Moses (23:02) is like the the the thing my my rant about Fable and about Opus really in general, and L LMs in general, is its neutrality. I I think it really I think where it really falls short is that from a business perspective, like I'm I don't want like I want the slant to be in favor of my business, right? Like doing things that doing the right thing, but not necessarily doing the the the the thing that is in the middle, right? Like and and and it always looks for this for this middle between, you know, you and that other person. And that middle, when you're talking about, you know, should I Jack BeVier (23:39) Hmm. It's interesting. David Moses (23:40) right should should I should I implement this lease clause that is so one sided, right, that it basically just lets me write the rules and lets me interpret those rules relative to the person on the other side of the table. Like that's an easy, I think that's an easy analogy to say, okay, well, no, it, you know, yes, there's probably some middle ground there that from a legitimate you know, w w we have our tenants throwing these leases into, you know, Chat GPT and coming back with, Jack BeVier (24:18) Right. David Moses (24:18) hey, this clause Is it legal? Because it could be interpreted this way or against this statute and this, that, and and you know, the it's obviously exposing very one-sided clauses to my lease. And for the most part, I'm like, yeah, you know what? That was kind of ridiculous. Like that like Jack BeVier (24:32) That's super interesting. That's super that's super interesting. David Moses (24:36) you know, so like we're like adding in like this clause is not intended to, you know, circumvent whatever statute, right? Like, you know, th things like that that Jack BeVier (24:45) Yeah, yeah, yeah. David Moses (24:46) were but But where I Yeah, right? Like t yeah, ex But Jack BeVier (24:47) Tightening shit up. Just yeah. Everyone's smarter. Everyone's smarter. Yeah, yeah. David Moses (24:51) but but there's but there's also like there's the middle the other way, like where we're a where you have tenants that will use it to find basically the inconsistencies in the statutes and the rules and the way they're interpreted themselves to kind of manipulate the system, right? And that's why our leases are so one-sided, is because you have l used to be lawyers. Now it's going to be just the tenants and Chat GPT. It used to be the lawyers that were twisting them. You know, and and so when chat goes out there from my perspective and and I say, like, I want to rewrite my lease or I want to rewrite this or I want to rewrite that. And it says, okay, well, because of this case, right, this needs to be like it it starts to push you to a to a middle that I'm not comfortable with because it it may be it may be taking into account these very, very edge cases where I wouldn't do the wrong thing in that situation. Right. But now I've opened the door. Jack BeVier (25:54) Yeah, yeah, yeah. Yeah. And yeah, we get how how long is this lease going to be? Are we going to contemplate every possible thing that ever might happen and you know, like and prescribe a process for everything? Like leases are for judges, you know, like unless it reaches that David Moses (26:04) No, we're not exactly. And that Yeah. So so Jack BeVier (26:08) bar. David Moses (26:09) Yeah. So the the middle ground i i it needs to be you need to set where that where where where the starting where the middle ground should be, right? It's like, look, I don't wanna hand I don't wanna deal with edge cases, right? I only wanna deal with, you know, normal business situations. And when it comes to an edge case where, you know, a landlord has been found to be super taking advantage of the tenant, like I'm gonna trust my people aren't gonna do that. And Jack BeVier (26:34) Mm-hmm. David Moses (26:35) the other part is the verbose part, the thing you just mentioned. It's like I I had somebody, like we had a reasonable, a tenant who was basically who was an attorney, although still just threw it in chat, sent it over. like clearly. what he pointed out, all of them were reasonable things that he pointed out. Like pretty much every single one of them was reasonable. And I said, fine, and all of them except for one, like the one that was a fee. I'm like, no, you're paying the fee. Everything else is fine. but then he sent it and somebody at my office used chat to like write the addendum to the lease to deal with these things. And it was two pages long, and I was like, No. It's like like one sentence, maybe two, can handle each one of these clubs. That's it. Like, this is not two pages. And and Jack BeVier (27:10) Yeah. Yeah. Yeah. David Moses (27:14) that frustrates the crap out of me too, because it makes it so long that you're really slowing business down because now getting that Jack BeVier (27:19) Yeah. David Moses (27:20) thing, you know. You know, it's it that that's not what you want. You don't want to use AI in a way where you take the same amount of time, right, because what it's coming up with is far more complicated than what you would have come up with if you had spent the time yourself. But now you just have something more complicated and you took the same amount of time. Like that has to be where we fight against it and we say, no, no, keep it simple. Keep what I probably would have wound up with anyway had I manually done it myself or had a hired a an attorney to to do it. Keep it simple, right? But just give me the time savings. Right? Give me give me the time savings. And that's where we have to kind of push back and say, No, no, no. I'm using this tool to save time. I'm not using this tool to, you know, make a a novel out of a short story. Jack BeVier (28:09) Yeah, yeah. Yeah, absolutely. I see I see that. lots of those similar use cases in ours. Like we have like folks who are arguing against, you know, the you get in like reasonable arguments, reasonable parties can disagree, right? Like and it's like, you know, before when you used to get a three paragraph email, you were like, wow, this person's like really fired up about this. Like they took the time to write me a three paragraph email. They're either insane or this is a well thought out argument that they and you know, by the end of the first paragraph you can figure it out. And now I'm just like, I'm getting I'm getting one two page emails with you can tell they're just like, here's a bunch of documents, here's the conclusion I want. Make an a make an argument that in furtherance of my conclusion. And I'm just like, all of these are very weak, unconvincing arguments. This is a like and no one would have spent the time to come up with two pages of weak, unconvincing arguments. But because you didn't, you spent 15 seconds. You spent the same 15 seconds. It's just the work product is larger now. You want me to go spend my time to reply to this bullshit one at a time? And now we're just gonna what and get in we're gonna get in AI pissing matches just back and forth. Like I'm like, I'm not engaging in any of that stuff. Like like absolutely not. Y call me. Like play call me. Like we're not we're not we're not doing an AI, you know you know, paper tiger warfare. David Moses (29:29) Well, but be creative with it because you if you if you know they're gonna s take that same response, throw it back into GPT and basically say, Hey, does this alleviate my concerns? And if it says, Yeah, for the most part it does, except this, you know, y if you know they're gonna do that and you can anticipate that, then you basically Jack BeVier (29:48) Yeah. David Moses (29:49) just right, you you take you take the one little sliver of that you were okay with that they that was really at the heart of what you know. chat came up with for them. And you just kind of, you know, massage that out a little bit, you know, it it'll still take you 15 seconds to respond. But when they you you want to think about that they're going to take your response and put it into Chat GPT. And so by doing this, you want it to be something that they're, you know, that when they plug it in, it'll give them a thumbs up. And then you can you can move on with life. But you're right. It it it can become an AI pissing match if you're basically just saying, no, no, I'm dug in. Right. I want you to explain why my position is better. And then the other guy's gonna say, Well, I'm dug in and that you know, and and yeah, you wind up with a with a completely, you know, where whereas you what you really wanna do is kind of like concede their point in a way that still protects you, you know. Yeah. Jack BeVier (30:47) If they've got one, you're giving them the benefit of the doubt that they've got one. I I was coming from a place of they just don't like to hear no, or they just don't like the answer, or like, you know, the the David Moses (30:56) Well, they think they have a point. Jack BeVier (30:58) mm, often or as often not, you know, they just they just I get a lot of just like they just want, you know, like they want the property to be to they want the refi based off of a $300,000 value. The property's worth two forty. We say, hey, it's worth two forty, sorry. And if because we thought because we actually did a real thoughtful analysis, the end we we think we just think we're right. We just think they're wrong. But I'll use the hyperbolic example just to but the broker just wants to get paid. The broker just wants and the and his customer just wants those those refi proceeds. And so they they don't actually have a good argument. They're just throwing arguments at me and trying to wear me down with arguments. And I encountered David Moses (31:38) Yeah, that's Jack BeVier (31:39) that more than anything. I I encountered that work that that abusive LLM work product in that in that context, probably that's the the biggest example that I've seen. So it's it's I I agree with your point. When when they when when it's an actual reasonable parties disagree, I completely agree with what you're saying. so so anyway, but make it good. David Moses (31:55) So one thing I wanted to No, go ahead. All right. So w one thing I wanted to you had talked about litigation and the and the whole idea being that the LLMs really need to protect themselves against, you know, the liability that's gonna wind up you know, coming down the pike. But I've said this a few times and I think it's really not been I don't think the alarm bells are sounding anywhere near as loud as they should be about the the litigation of the matter, right? So it's like taking out whatever the subject is of the litigation and what the potential liability is, the speed and frequency with which one can litigate is to me terrifying, not from a, you know, A versus B, but from a A versus B in front of the court, right? Jack BeVier (32:54) Mm-hmm. David Moses (32:55) And I think that is the I think that is a an incr I think it has the potential to literally crush our legal, you know, system and process as we know it if the courts are not moving full speed ahead to implement AI themselves. Jack BeVier (33:15) Yeah, that's where I was going. And the bar associations too, right? Like the bar associations need to come up are are all probably in state by state wrestling with what's our AI policy gonna be. Like how much, you know, if ninety percent of the brief was written, you know, ninety percent of the motion was written by AI, is that okay? Like do we need a time, you know, do need a docusign to prove that they actually read every word? you know, how do David Moses (33:38) And I don't know, and and they Jack BeVier (33:39) you define abuse? Yeah. David Moses (33:40) that that's right. And and and before it was, hey, you had AI write write this brief and it cited cases that didn't exist or cited existing cases that don't Jack BeVier (33:50) That's easy. David Moses (33:50) say right. Though I think those things are easy. And that was like the big backlash and the immune response of the legal profession kind of kicked into full gear because that's obvious. Yeah, of course. You know, but but what happens when it's no longer getting those cases wrong. What happens when it's, you know, on point? First of all, it's gonna expose all, like literally all of the inconsistencies amongst the courts in any given place, right, in terms of their rulings. Like every single right Jack BeVier (34:20) That's very interesting. That's a really interesting analysis. This circuit court versus that circuit court. That's a really interesting, that would be a really interesting analysis. David Moses (34:28) Right. And and but but the I again, it's like it's the court's ability to handle this shit is going to be I mean, it it you're gonna you're gonna see situations where it's not six months to a year to you know, when you file a lawsuit. It could be five years, right? It could be four years before a lawsuit reaches the point And and then and you're gonna wind up with if the courts and the judges don't have the capacity to absorb and understand like this, you know, t you know, this this 200 page point by point summary of this 10,000 or 100,000 pages of discovery, the court can't review that and understand what these things really mean. And you have AI going back and forth between A and B and the court having absolutely no clue what is going on. You still need somebody to go and argue it, right? But, you know, I I don't know. I I just think I think the court systems are going to be completely overwhelmed in short order here. And I mean, I think it's going to be something like in 2027, there will be an inflection point where the courts simply say, you know, w we need more resources, we need more people. There's no way of handling this. And filing a lawsuit isn't going to be a hundred bucks or two hundred bucks anymore. Maybe it's, you know, five hundred bucks or a thousand bucks. You know, or they'll implement some reasonable, you know, form of AI that allows the court to absorb it. And I'm scared to death about that too, because what happens when, you know, yeah, that LLM just decides Jack BeVier (36:09) That's touchy. David Moses (36:12) to find the middle ground between any two parties, regardless of how one sided the issue really is. You know. Jack BeVier (36:18) Yeah, that's n that's very interesting. I wonder if bar associations end up being like they're the only one with the power to implement state level rules that say like, hey, you know what, can't do that here. Like, you wanna practice in this state? Can't do that, can't pull that shit here. Cause w because we watched New York go from a year and a half average litigation time to six years of average litigation time, and we're just not gonna let that happen in Texas. And so, no, you gotta write the shit. You gotta type the shit. Like And we run your we run everything through David Moses (36:46) Yeah, but how do they enforce that? Jack BeVier (36:48) LLM and if we find any LLM in there, like if we find more than ten percent of LLM, it's just automatically dismissed, you know, without prejudice. Try again. You know. But you only get you only get two shots of the without prejudice Apple. So David Moses (37:01) Yeah, but how do they how do they police that? The watermarks? I don't buy that. I think the watermarks are Jack BeVier (37:05) no, they they get the language they get the language analyzers now, that like that are analyzing which I'm not exactly sure. I mean I you could I'm I think that's probably an escalating cold war, right? Of like creating a a legal brief preparation system. But if this but if the bar association just outlaws it, like you do have a lot to lose. Getting a law license still takes a lot of time. So like that is still a really thick stick that they can hit everybody with. David Moses (37:32) I don't think so. I I don't think they're gonna hit it. I d well, I yes, I think that is a huge stick. I don't think they're gonna hit anybody with it. Because I'll tell you what, th it goes back to what do we know about AI, right? We know that seventy plus percent of people in America are against AI. Right? We know that the adoption rate, that the recidivism rate, I'll call it, right? The the the rate the rate where of people who start using it, right, and and stop is like less than one percent. Jack BeVier (37:59) Yeah, I was gonna say nobody. Yeah, like yeah. David Moses (38:01) Like less than one per so so because of that, like once attorneys recognize how useful it is, right, in helping them prep their I mean, they look they still have to go and argue it, right? Like they're brief, it like appellate courts I think are gonna have a rougher time because they they really don't rely on most appellate courts don't rely on the oral argument anyway. They've made their decision. All they're really doing is asking questions, you know it Whatever, that that is what it is. But the but at the at the at the at the lower court level, it's gonna be that argument, right? Like that that attorney still needs to know what his AI wrote and he still needs to be able to prep for it and argue it succinctly. But I'm telling you, it it's not even that the like forget even even about the brief being written by an LLM. You could write the brief. You mean first of all, I I I I think if I gave it if I were an attorney who's been practicing for more than five years. And I gave it all of the briefs I've ever written, it could it could write like me. To to the point where even if an LLM can identify that yes, this was written by an LLM, if I showed it to you, Judge, and I showed you samples of my writing, you wouldn't be able to tell that it was written by an LLM. You would think it was written by me. and and that the Jack BeVier (39:15) Yeah. David Moses (39:16) LLM is just wrong. But as you know, but but that aside, it's really it's It's writing the arguments, anticipating the other side's arguments, what their what their arguments are going to be, building that in like there I I see some incredible efficiencies that can happen too, but I just think it goes back to the verb the verbose nature of LLMs and the fact that layperson, like I'm much less worried about attorney using an LLM because they do have a law license to. And they do have they're they are generally scared shitless about it. I'm worried about guy representing himself, right, who slipped and fell or who Jack BeVier (40:00) Per se. Per se cases. Yeah. David Moses (40:01) right, like those are going to be just more than the court can handle and more than they're ready for. Jack BeVier (40:05) So so so here's the thing. I agree with that. Now you can defend yourself with that same tool. A little risky, maybe, right? Because you don't have a law license. But I agree. I think that the danger is the ambulance chasers and the shakedown artists, the guys who were like gonna file a complaint against you just to shake you down for 30 grand because the cost of litigation is gonna be a hundred. That I'm really concerned about. The only way Let me I'll make an argument that is not that I've not fully thought out, but I'll make the argument for fun. Only way to fight fire is with fire and just get used to learn develop a skill of defending yourself on a pro se basis as a company. And be so that you just you just don't hire an attorney when it happens. You just say, hey, you take it, you throw it, you but you get some subscription to some legal LLM. You throw it in there and be like, hey, we're gonna respond to this. We're not gonna engage in attorneys. We're gonna represent ourselves on a pro se basis because this is a this is a shakedown bull, this is shakedown bullshit. And you end up with more and more pro se cases ending up in court. But but they don't end up in court because once you get to court, the knucklehead, like you said, has to like you have to get in front of a judge. Everyone like wilts and like, you know, disappears, you know, three days before they have to get in front of a judge and actually. make an or an David Moses (41:23) Yep. Jack BeVier (41:23) argument here. And or and if you want to really if you want to really like you know bring a bring bring a gun to a knife fight, you either get your law degree as you you you either bring it you either have general counsel, engage someone who has a law license, or at least went to law school. So you or you or you just go to law school so that in that knife fight you have a gun because you actually when you get in front of that judge, can just wipe the floor with a guy because you've been taught the taught the basics of how to how to deal with this. And we just use lawyers less because they've been charging a tax. You know, they had to go through they're smart people, but they're not they're not the smartest people like that we've got. They are smart enough to pass the bar and then they charge us a tax because they have been educated on a set of rules that we are just not educated on. And now I've got this tutor who can walk me through this thing. Like, you know, I'm my attorneys are freaking out. Any anyone any of my attorneys right now who are who are who are just like, that is the stupidest fucking thing I ever have heard you said, Jack. And all the attorneys, all the licensed attorneys right now are also like, that's the stupidest. It's just not that simple. And I know, and I know, but I don't know, man. Like, but I don't want to David Moses (42:34) But it will be. Jack BeVier (42:35) get shaken down 30 grand every time I get a freaking letter in the mail. You know, like I'm I I can't I can't deal with that. I can't I can't run a business Getting shaken down for thirty grand every time I get a letter in the mail, you know? David Moses (42:46) I I agree. I I think though, for it first of all, in Michigan, even an LLC, even a single member LLC cannot represent themselves. You can't you can't go and represent the it has to be there has to be an attorney there. But that Jack BeVier (42:56) Yeah. Yeah. David Moses (42:57) said, I think a smart attorney right now, especially like a young tech savvy smart attorney right now, is going to take the angle that look, we are if if you want a firm that is going to take your frivolous bullshit lawsuit that you just got. and bury the shit out of them in you know litigation and paperwork, but we're gonna do it using AI and for a fraction of the cost that you're gonna pay to some large, you know, you know, the the the law firm of Dewey, Cheatham and Howe to you know to to to to steal a a line from CarTalk. So Jack BeVier (43:37) Loved it. Totally got it. It landed. It landed a hundred percent. David Moses (43:42) then we're your guy, right? Like we're we're the we're the firm to come to because we are AI native and we are going to we you know correctly like we figured it out how to correctly bury the other side and you know make it so if the if the lawsuits bullshit, if it's completely frivolous, Jack BeVier (44:00) Yeah, if it's bullshit, yeah. David Moses (44:02) right? Like you know, like and and I you know to me then I then I think you can put I think a firm like that can put, you know, now they can start to I think that this is where the legal, like, you know, the the legal insurance profession, like where you're like that that whole idea of like you know, where you pay a subscription and you get legal, you know, legal services. To me, I think that's where that type of firm can you know, they can make a very large buck, right? Because they know what their cost of litigation is is a fraction of what a traditional firm would incur and they can still collect those premiums to and and legitimately defend their, you know, their customers. Actually actually, you know, make a much bigger delta on you know, on those premiums. Mm-hmm. Jack BeVier (45:03) That's the markup on labor. Yeah. Yeah yeah. Yeah, no, that's an interesting David Moses (45:06) And I hate like Jack BeVier (45:07) idea. David Moses (45:07) Yeah. And and your it like your point, like these frivolous bullshit, thirty thousand dollars I slipped and fell I like my favorite case to talk about is I I first got into business, okay. I was in business for maybe a year. Right. I was twenty I was twenty two, okay? So knew nothing about nothing. And I got a lawsuit. And I got a lawsuit from This, you know, ambulance chasing attorney for someone who had slipped and fell outside of our building. Okay. So th the the facts around the case were they slipped and fell in the parking lot of the building. I didn't own the building. The guy who owned the building didn't own the parking lot. And when she slipped and fell. My company wasn't even in business yet. I didn't e I I didn't and and and I got sued and I was beyond pissed, right? Because I'm like this is like this is Jack BeVier (46:09) You don't have it 'cause you you you don't have the time or the energy or the money to deal with this bullshit. Yeah. David Moses (46:13) Right. And I'm like, I'm trying to build a business from scratch here. It is this isn't this isn't gonna work for me. And so I got, you know, I got all angry and huffy and puffy, and like a twenty-one year old tends to do. And I, you know, I filled out a grievance form for the attorney. And I sent it to him and I said, Here are the facts of the situation and dismiss this lawsuit, or I'm filing a grievance again. And I don't care. Like I will fight this grievance. I'm twenty one years old. I got nothing yet. You know, and Jack BeVier (46:44) Twenty one. Yeah. Mm-hmm. Yeah. I will show up to every hearing. It is like everything that I'm angry about, all the chips on my shoulder, you will be directed at you. Yeah. David Moses (46:52) You're bearing them. And right? But like today, I would be like, here you go, insurance company. Like I'm not, I'm not like, here are the facts. Jack BeVier (47:00) Yeah. Yeah, yeah. David Moses (47:02) Insurance company. And you what they would they would probably pay them. Right? Because cause the insurance company would hire some attorney who's, you know, Jack BeVier (47:09) Obviously, get rid of these things, yeah. David Moses (47:10) yeah, who who has like, you know, in Michigan it's a P number. I don't know like what you what what the number is in in, you know, but whatever it w it was was like Like like the one that was just issued or like one that was issued like last week. Jack BeVier (47:22) yeah. David Moses (47:22) Like they just got their law license and they have no idea what they're doing and they just got shoved this case. Yep. Jack BeVier (47:25) Yeah, yeah. They have they have a bill. They have a student loan bill. That's what they have. Yeah. David Moses (47:29) And that's what the insurance companies do. I don't know why, but they want they want to pay the lowest amount per hour, because they know they're gonna settle ninety percent of these cases, ninety five percent of these cases, because they know the attorney on the other side is just looking for a payday anyway. And it's their job it's like real estate, right? Like it's their job to sell their client that this is a good that this is a good price for this thing they're buying. And the other side it's, you know, this and and they do the dance so that insurance pays, attorney gets paid, you know, you know, person who slipped and fell on, you know, ice. Jack BeVier (47:59) Yeah. And a market and a market is formed, right? And then like and and over time they there's a market price for this situation, this set of facts, and both sides know what the market price is for this situation, this set of facts. And yeah. Yeah. David Moses (48:10) Yep. Yep. Jack BeVier (48:12) Yeah. And they as long as you end up within ten grand of that, like you move on to the next file, you know. David Moses (48:16) Yep. I know a guy with a hundred and ten houses. I wouldn't manage property for him. because I told him like we started to talk about maybe maybe I'll manage your properties. And he No, I I I sent him my contract. He's like, Well, I don't have insurance on my properties. And I'm like, like, none? He's like, no. I'm like, not even liability insurance? Like, what he's like, definitely not liability insurance. I'm like, why? He's he's like, he's Jack BeVier (48:42) How? Yeah. David Moses (48:44) like he's like, I'm an attorney and I see these cases all the time. And I know that 90% of them are attorneys looking for a payday. And ninety five percent of them. And once they see that I will fight it all the way to the Supreme Court, they will go away. I don't know how that's worked out for him. Jack BeVier (48:57) That I'm an attorney and that I don't have insurance. Yeah. Yeah. And the combination of those two, it is I'm an attorney. So my, you know, my the your counterparty's variable cost is zero. And that there is no insurance to hand this off to, like, you're not participating in the market, right? Like this is not gonna get you're not gonna get a market outcome here. Like, never mind. All I was here for was a market outcome. Not here for moral reasons, you know, like David Moses (49:22) That's right. And and you're and and he's probably right from the standpoint that like the odds Jack BeVier (49:26) Yeah, yeah. David Moses (49:27) of a person going to you know, the the odds of a person finding an attorney who's really going who really cares about them and is gonna fight this thing on contingency all the way as far as it takes or as long as far as it goes, is, you know, I I would say You take the cost of insurance multiplied by a hundred plus properties over the course of X number of years. I feel like you're coming out ahead in that math by a long shot. Yep. property casualty, Jack BeVier (50:01) Yeah, I think he's I think he's right. I think he's right. I have a I have a I have like a thing. David Moses (50:06) I don't know if he's right. And then and does that hold true? Jack BeVier (50:08) Yeah, well. David Moses (50:08) And and and and does it hold true next year in twenty seven or twenty eight? Does it hold true when guy who slipped and fell can fight that all the way wherever they're gonna fight it? With chat GPT. Jack BeVier (50:20) Yeah. You're gonna you're gonna have to fight you it's just just like the cybersecurity issue, right? We're gonna have to fight we're gonna have to fight back with the tool. The other way, you know David Moses (50:27) Yeah, the mat the math changes. But but you're right. It it is the tool, right? It is the only way to fight it is and and again, got like I feel for the court, like the the clerks, right, who have to read the hundred page brief and the you know, like that that was twelve pages before. Jack BeVier (50:48) It's very interesting. I have this like recurring thought that I can't shake that that I should go get my law license. That I should just go to law school. I should just suck it up and go to school at night and get my damn law license so that I've got it. So that when I get those letters, I can be like, hey, I'm a lawyer, I'm around house counsel, I'm also one of the owners. Like you'll be I'll be responding. And then use these tools and just, you know, I'll have to manage. I'll have to read everything, you know, like I'll have to manage to not get my license revoked. But It would help change the economics of every situation. And I'm like, it's probably that's probably worth it. You know, like you know, if you if you get if you get five houses, it's not worth it. But like, you know, when your business gets big enough, I'm like, you know, at one point we had general counsel that I was paying in-house. Like, I don't currently, 'cause I don't have as much of a need at the moment. But like, you know, it's it's at the point where I guess it's just it ends up being like, what do want to do with your time? That's all. But I like being that t I got that twenty one year old you know, streak in me still that wants to just fight. So David Moses (51:47) Yeah. I mean I I i i I I try to stay away from that the way I used to not be able to stay away from it, but Jack BeVier (51:55) You're much no, you're much more mentally healthy than I am. I agree. You're much more mentally healthy. David Moses (51:57) No, I I I I I wouldn't necessarily say that. I what I would what I would say is that, you know, I've been through a divorce and you haven't. And you don't know what wasting time on useless bullshit really is until you've been through a divorce. Jack BeVier (52:12) that's hilarious. Dave, I gotta circle back to the second part of my the second David Moses (52:17) Yeah, yeah. Sorry, you had two parts. Jack BeVier (52:18) part of my soapbox, which is I was I was regaling, you know, I was I was I was stroking Fable's ego. And the the business conclusion that I've come to with that is that I think that with Fable 5, that's all I need to rebuild pretty much everything. And very quickly. It's not even a painful process. It's kind of just fun to like just to now I have to like, you know, open my wallet and light money on fire and I'm not enjoying that part of it. But but the but the process but it's like a luxury good, like at the moment that I'm like, this is this is fun, like I'll just rebuild, you know, just rebuild every part of my process until it's an you know, until it's a lean, mean, efficient mouse trap. and David Moses (53:00) Do you use VAs? Jack BeVier (53:00) yeah, for sure, yeah. Yeah, David Moses (53:01) Okay. Sorry, keep going. Jack BeVier (53:02) yeah, for sure. and but fable six or fable seven that cures cancer, two things I don't care. you know, like I I don't think that it's gonna be that like this is so good that I can't imagine it being much better. Like, what's it gonna be? It's like taking the Acela, you know, instead of the Amtrak. Like, what is it gonna it's gonna get me there like f fifteen percent faster? Like I this thing's already like I leave, I come back twenty minutes later, it's reproduced a process. So the other is the other issue, or the the and the other side of that coin is that when they come out with Fable Six or Fable Seven, the equivalent, I'm just making that shit up, right? When they come out with the next frontier model and it gets distilled, it'll get distilled into Fable V or something very close. David Moses (53:50) Mm-hmm. Jack BeVier (53:50) And and the distilled version. That costs five percent as much or one percent as much eradicates the revenue curve that exists currently. There is no more money to get than all of America's businesses wanting your product and using it for a while. That is maximizing the area under the curve. And as a result, these guys have a trillion, two trillion, three trillion dollar enterprise valuations because the revenue curve looks like what it looks like. But when Fable Seven comes out and it gets distilled into Fable V by the Chinese, 100 billion goes to one. And it is an existential threat. It they could they will cannibalize their revenue. By releasing the next model, they will absolutely cannibalize the area under this curve. And and David Moses (54:39) And the bubble will burst. Jack BeVier (54:41) and at the same time, and that's the moment. and and they've got this regulatory, they've got this legal issue, this legal safe harbor issue that's preventing them. So they wanted to go IPO based off of the hype and no one connec you know no one thinking about that. And they wanted to, or maybe they just tell their investors in closed doors, they're just like, yeah, you know we're not doing the next one, right? Like, like, or David Moses (55:01) Yeah. Jack BeVier (55:01) if we do the next one, we are not releasing it publicly. That's what I think's gonna happen. Fable five is the best model that we're ever going to get as consumers. I think it's the last model. I think it's the last consumer model. Every incremental there will be incremental development, but those will be productized and kept internally and they will cure cancer with it and then they will sell us the cure to cancer. But they we we will not be getting the tools to cure cancer ourselves because we don't own the means of production for that. If I was to maximize the area, David Moses (55:35) Man. Jack BeVier (55:35) if I was to maximize the area under the curve, we've asymptoped. Nothing else will get released commercially. This is the last consumer. David Moses (55:41) I don't know, man. Jack BeVier (55:43) This is the last consumer product. David Moses (55:43) Pat patent office, nineteen hundred. I I I'm taking the over on this. Jack BeVier (55:48) But you you have to dude, but to c to to create the I I I love it. This is gonna be a great great ongoing debate, and I'm looking forward to five years from now having this shit thrown in my face. Because what an amazing world if I'm wrong. but David Moses (56:00) Yeah, that's right. That's right. But but I look, the question is r th the the the the kind of the canary in the coal mine is Jack BeVier (56:04) Dude, the fastest car the fastest car you get to drive is a you you don't get to you you don't get to drive a car that's goes faster than 120 miles an hour. They exist. You just can't use them on public streets. Like they're like you can go buy one, you can go spend a lot of money to have to create this little bespoked, you know, thing that you want for yourself, but like it costs the the comp and and and it's gonna be an oligopoly of those. It's gonna be it's gonna be anthropic, openai, and grok because they're the only ones with the and NVIDIA. They're the they're the only ones with the balance sheet to create the clusters necessary to create a Fable 6. There are only three companies with the with the with the balance sheet and business model to do that. And then NVIDIA will take over the open source consumer market and it will be open source Opus 4.6. And we'll n and and they'll guardrail Fable 5 until they max until they extract the until for the next five years. They're gonna cap it and and they're they're going to extract value under the area of the curve for the next five years. Cause if they if we're and if they ever get their safe harbor, they'll IPO because then they'll be able to they'll they'll be able to you know David Moses (57:16) Ever moat. Jack BeVier (57:17) capitalize the area under the curve. But if they can't capitalize the area under the curve via an IPO because the legal safe harbor, they will just defend they they will just try to scare, they'll try to create little, you know, little safe harbors in the friendly states. And and just collect the money because to get the area under the curve because they could never get the big pop for the you know net present value of future earnings. So they'll just extract rents from the American economy until we all finish our CapEx cycle of rebuilding our companies. and they'll go create the next models, but they will be scientific partnerships with specific, they'll they will they will choose a partner. And we will be the compute and you will be the subject matter expertise, and we will create a part a product together. And then we will create a product together in another vertical, and we will create another product together in another vertical. And they're the outsourced intelligence, they're the outsourced super intelligence for the product pro for the subject matter experts that are the best in every field. And we will have a bunch of stars. we'll have a bunch of stars in the economy because it's whoever was able to negotiate a deal. Whoever would was was able to convince the superintelligence companies that they and we'll or you'll have an oligopi three. You'll have three you'll have you'll have three stars in each vertical because there are only three partners to partner with. That's my that's my prediction of the future. David Moses (58:43) Okay. Wow. So Jack BeVier (58:45) There's so much money to be made by not letting us go any further. And and at that moment. David Moses (58:50) No, I I I I don't necessarily disagree with you. Th but there's there's a question I have, right? And the question is Jack BeVier (58:54) Hold on, and in that moment, and in that moment, that's the time that they get on the morning show and tell us that it's too dangerous and we can't go further. I'm like, yeah, you don't want us to go. You're you're gonna not go further because that's maximizing the area under the curve. You're gonna tell me we're not going further because you're doing me a favor. Go fuck yourself. This is b I'm I'm be we're being lied to. David Moses (59:14) I look, I I if not it's not a lie if they believe it. Jack BeVier (59:19) They don't believe it. If the if maybe some of the engineer maybe the some of the dumbass nerds who have never left their basement believe it, but like with like crazy politics. But the shareholders, the shareholders and the owners of that company and and the and the investment bankers advising them. David Moses (59:37) No, I I I think yeah, I'm gonna I'm gonna take the other side of this. with with Jack BeVier (59:40) Love it. David Moses (59:42) with conviction this time. I'm not just playing devil's advocate here. Like I'm I I actually believe w what is about to come out of my mouth. first of all, where's the curve? Right? Like it like you you're operating under the premise that the curve stays where it is and the area under of the on under the curve stays a a defined size. Jack BeVier (1:00:01) No, no, no, no. I agree. The curve is all the pie is also going to increase as they do this. Brilliant. But they but they won't, they're not gonna David Moses (1:00:07) So and and the other the other part is that Jack BeVier (1:00:10) let us they're not gonna let us open source it. We're we're gonna have to we're gonna have to open source it with some bo we're gonna the open source version we're gonna get is gonna be, you know, not as good. David Moses (1:00:19) Well, I look I think well let me let me let me just quali like I will tell you when I will believe what you're saying. Right? Like like here's what I ha well I'll have to see in order to believe we've reached that point. 'Cause I I don't think like you say you know, you can't drive a hundred and twenty miles an hour. I don't think we're driving a hundred and twenty miles an hour. I think we're at like I think we're like, you know, the model T that you can drive forty miles an hour, but i if you if you drive it any faster than that for any reasonable amount of time, you know Shit's gonna break. I I I think we'll get to 120 miles an hour. But where I where I think this lands and where I think the bubble might burst and there's not gonna be any more like the value of anthropic come out with the next Fable 6, Fable 7 becomes becomes a it's just something that doesn't make economic sense for them because they will be cannibalizing themselves. Is when we get to recursive self-improvement, when models can truly build themselves and and not just fable. Like I know, I know obviously most of Soul 5.6 from ChatGPT was built by AI and the same, you know, with the same with Fable. when the Chinese models, the open source models can recursively self improve, then I think you have to come out with Fable Six. You have to come out with Fable Seven, because otherwise everyone else is gonna come out with the next thing. And maybe it's just a little bit faster. Maybe it's just a little bit smarter. And maybe you don't need that. But if it's a tenth the price, yeah, you're gonna buy it. So the but But the other thing is I I think where this ends, like where Jack BeVier (1:01:57) I think I think they're just gonna create their own set of closed models. I think China's it's it you're open source when you're losing. But when you're when you get to the lead, you closed source. And so and and so like anthropic is c David Moses (1:02:08) I I to some degree I agree with that. But I think they're o I think China's open source because they're because of their general viewpoint towards intellectual property. Right? I Jack BeVier (1:02:19) I think that they're gonna create I think that they're going to open source up to where we open source, but but but anthropic is gotta put a ceiling on this because there's a ceiling on their revenue if they go further. And open AI is catching up and Grok is catching up, and there's really nobody else behind them who's talking about a closed source model. And I think China is because they're all losing in China, they're all open source until the ones until the until they pick their stars and they'll pick three stars. And then close them down and build their clusters. And then they'll create their three closed models because they have the same financial incentives to transform their economy that we do. David Moses (1:02:56) When do they stop, Jack? Like when does that like to me they don't get to that like I don't know if GLM, whatever point, whatever, right, is how close they are to being able to Basically take the model and post train it yourself and have it improve itself indefinitely. And the question is really, are they gonna close before that they reach that point? Because if they don't close before they reach that point, then they're really it's it it it's inevitable that there doesn't need to be Fable Jack BeVier (1:03:32) Yeah, yeah, you close before you reach that point. David Moses (1:03:34) Six, Fable Seven, because whatever you just download it and as long as Jack BeVier (1:03:37) Well yeah. David Moses (1:03:37) you have the RAM to to To do what you need in the data. Jack BeVier (1:03:39) Well yeah, you don't though. You don't though. You you have to close before that because you can't afford because an open source project doesn't have revenue and so it can't afford the clusters of GPUs and electricity necessary to allow recursive self improvement to go happen. Like you need money for that. You gotta pay for it. Someone's David Moses (1:03:55) Yeah, I don't know that Chi I I don't know that Ch Jack BeVier (1:03:58) gotta pay that bill. David Moses (1:03:59) I don't know that China cares. I just I just don't know. Jack BeVier (1:04:03) China cares. What do you think? Well then they nationalize it. Then then it becomes a national project and they bring it in-house. Totally possible. Totally possible. David Moses (1:04:08) I mean yet but to some degree there does to some degree it's already nationalized, right? I mean Jack BeVier (1:04:12) They even they haven't made the they haven't made the capex in yeah, the the capex investments in GPUs and electricity. That's the question. Who's gonna stroke those checks? David Moses (1:04:20) They've they're they're doing. Like China's already, they've got they've they've built more solar in the last year than we've ever built, right? They've built more you know, they're they're definitely moving in that in that direction to be energy independent. Jack BeVier (1:04:32) Yeah. David Moses (1:04:33) Right. But but forget about like forget about like like that part of it, yeah, you know, maybe we there's an argument to be made that we won't necessarily even need like once the models become more efficient at doing what they already do. then we're not going to have the power suck that we currently have using that. Right. It's not going to be quite the the you know the the yes, the the price per kilowatt may go up as demand goes up. And yes, you need some energy independence to offset that. But if the but if the you know whatever the ratio is between, you know, kilowatts and tokens, right, i if that improves, then you wind up with, you know, just much more efficient systems, much more efficient models. But I think that China's push towards AI isn't necessarily about American businesses or businesses around the world becoming more efficient in their administration. I think for China, I think their number one focus is that they have an aging population that needs to be cared for. And I think robotics is, you know, there's like hundreds of robotics companies there and none here, relatively speaking. There a few here but but not relative to China. I think that they have an aging population that needs service. I think that they've got there there are still many, many parts of China that are developing. and they they just I I really do think, you know, that the the the communism in them is going to prevent them from stopping based on a based on the economics of it. I just I just don't buy that premise that China's going to stop until I I I think I think Jack BeVier (1:06:21) Well, I just think I just think they're gonna nationalize them. I think that if anthropic, openai, and grok existed in China, they would be national companies. They're too valuable. They're too valuable. David Moses (1:06:30) But they need to be like these the it but yes, but you're but you're also speaking from the from the standpoint that these th that the models need to be accessed. To some central repository, some some company. Like we need to go from our laptop and access that, you know, that model. And I don't think that's where the world is in five years. I think in five years, the world is the models themselves are locally wherever they are. On a on a robot, on a in a car. They're not, you know, they may be reaching out to some central repository for updates, right? But generally they're able to improve themselves. Right. They're able to because they have to be, because they have to make the decision making involved and you know, the the calculations involved in robotics and in cars right now, as Alana's brew, it's really there's n there's not sufficient bandwidth, there's not sufficient internet that's stable enough to allow a car to drive itself. without making those decisions completely locally. Like it needs to be s it ne it needs to be and and and in order for it to be completely local and for that that ball to continue to roll, those models need to be open source so that people can buy them, you know, or or the closed source models will have to be bought. Like you'll have to buy Fable 5, right, to be able to install it on something that requires You know, like a a robot who's doing surgery, right, needs to have a local LLM. It can't it can't rely on the latency of talking to anthropic servers in order to get a decision on whether or not, you know, what it's encountered, it should, you know, do this or do that. So I I think that yeah, I could see I could see them being closed, but they're but they're closed, but they're bought. Right. Like you I I'm gonna buy and manipulate Fable Five however I need to for my use case. I I can Jack BeVier (1:08:41) Do you think David Moses (1:08:42) see like wouldn't you wouldn't if I if I said to you, hey, look, I can give you Fable Five level intelligence. For fifty grand. That's Jack BeVier (1:08:51) Done. David Moses (1:08:52) it. Fifty grand. Like and you don't ever have to buy a token. You don't ever have it's just fifty thousand. Jack BeVier (1:08:58) You just gotta yeah, plug it you gotta have the hardware to run it and the electricity bill. David Moses (1:09:01) Right. And and we're not there yet, but we're I mean, this is a problem that is a v going to be a very short lived problem. I mean, this this RAM issue like Jack BeVier (1:09:09) Do do you think that David Moses (1:09:10) RAM is not gonna be that expensive in a year. It's so fucking ch it's like Jack BeVier (1:09:13) Yeah, I'm excited about that. David Moses (1:09:15) eighty five percent margin and it's so cheap to build the market, we'll figure that out. Jack BeVier (1:09:19) Do you think that do you so do you think that the Fable Five equivalent model that we can either buy or is available on an open source basis? Well, when you say buy, I mean what, leap like you protect it with a license, like once you're into it, you can copy it. So like how do you you can't lock down? David Moses (1:09:34) You no, you you're buying the you're not you're not buying the license, you're buying the hardware. Jack BeVier (1:09:37) Yeah, but ca can't and you can't and you're suggesting you can't transfer the the software off of the hardware? That's what I mean. Like so how do you sell how do you monetize, hey, here's an instance of David Moses (1:09:42) Sure you can. But you need to have the hardware to operate it. They're not gonna monetize it. Jack BeVier (1:09:48) the model? David Moses (1:09:49) I I'm telling you, like if the RAM existed today, you would probably have already bought the hardware you need. Jack BeVier (1:09:52) you're just saying buying a product that uses a model that that's that's that powerful. Not that you can use that model for any other purpose. It's just I understand now. Okay. It's David Moses (1:10:00) Yep, yep. Jack BeVier (1:10:01) not a it's not a model that you get to interact with a chip. Mm-hmm. You can steal a chip, but whatever. I got you. I follow. So the David Moses (1:10:05) Yeah. Wait, so you you you you'd wind up with like Anthropic's gonna have a product, right? Or they're gonna partner with somebody and you're just going to buy this machine, right? And this machine, Jack BeVier (1:10:15) It's really smart, yeah. David Moses (1:10:16) right? But it's just this machine, right? And then you're gonna all all of your business systems are gonna connect to this machine. And you're gonna post-train it, however you wanna post-train it. You're going to do whatever you wanna do, right? 'Cause cause training is so important. You've used these things, I've used these things. The biggest problem with using LLMs is is that it makes unreliable decisions because if you always have to give it the context in order for it to know the thing it needs to know to do the thing it needs to do, you're gonna wind up making bad decisions some of the time. And if you post train it where it always makes that decision, every single time it encounters that thing, then you can rely on it. And I think you need local models to do that. and I think I think smart businesses are going to get to the point where they're using local models and post training them, not because they're the smartest, because they're smart enough and I can post train them so I can get a more reliable result out of them. Yeah. Jack BeVier (1:11:10) Reliable. Exactly. Yeah, yeah, yeah. I completely agree with that. I completely agree with that vision. David Moses (1:11:14) And then it'll be free. I and that is the existential threat, right? I I I d I think I think yes, while I think the instinct will for anthropic and open AI will be to protect that, you know, one trillion, two trillion, three trillion dollar valuation that they're gonna try to wind up, you know, IPOing at. I think in the long run that I I think that race is lost. The bubble will burst. Because the cost of this shit is the cost of a kilowatt. Jack BeVier (1:11:44) Okay, so I have two questions. One is that so do would do you think that the do you think that the entity that releases a fable five level intelligence on an open source basis or on a product, do you think that they're American or Chinese? David Moses (1:11:59) Chinese. And Jack BeVier (1:12:01) Okay. David Moses (1:12:02) I think it happens in the next year. Jack BeVier (1:12:04) I hope you're right. I mean I hope you're I hope you're right. I don't give a shit David Moses (1:12:06) If if if i if if that was the next if that was the next question. Jack BeVier (1:12:09) about these three companies, you know, extracting value out of America out of American economy, right? I I just want the tools. So yeah. David Moses (1:12:15) That's right. I mean, look, there'll be a lot of companies that don't implement them because they're going to be sitting there worried about what have they injected into this model that, you know, sends our data back to China. Yeah, I think people are going to be worried about that. And to some degree, they might be right to be worried about that. But that's why we have Fable Five because or Fable Six or Fable Seven. Hey, go into this model that I have on my local machine and tell me what's in it that shouldn't be there. Like Jack BeVier (1:12:39) Mm-hmm. David Moses (1:12:40) and and it'll find it. Because it's smarter than the model that is in the machine. but yeah, I I think I think there's going to be a large percentage of companies in a year. I would say, you know, thousand X, ten thousand X, a million X the number of companies that currently implement local models that are using them a year from now. Jack BeVier (1:13:01) I I completely agree with that. The let me ask you a different question. Isn't there an upper limit to the value of intelligence? Like the like the this we are not run by the smartest people. They're not they are not our leaders. They are not women do not flock to them. They're the are David Moses (1:13:21) Ha ha ha. Jack BeVier (1:13:21) our are smart, you know, the the things that I want out of life, there is a minimum level of intelligence necessary. And then there are decre and then there is a, you know, there's there's a curve, right? There's a there's a utility curve to intelligence that is not linear. And I guess my argument is that like much above Fable Five, I don't know. Like, it's kind of just for fun. Like practical applications into hardware become what's cool. I agree with that. Like what I can do on a laptop has you know, has a d has declining utility above a certain number above a certain level. Fable five I think is probably that level. Like or like maybe one step above Fable Five and I'm just like, I don't know. It's got a pretty good representation. David Moses (1:14:04) Didn't didn't we say this about didn't we say didn't we say this about Opus four? Didn't we? Jack BeVier (1:14:10) Did we? I don't know. I don't know. David Moses (1:14:12) I don't know. I don't know either. so like my brain was exploding when you like you said like isn't there a limit Jack BeVier (1:14:18) Isn't yeah, isn't there isn't I think there's a dis there's a declining marginal utility to intelligence above a certain level, like I don't know, I'm picking rando, you know, picking a number like like less than 90 problematic, increasing marginal utility up to 150, then it probably starts to become a liability or like or or just you know, certainly not like you don't get more happiness under the curve. I I I use that analogy a lot, the the calculus analogy a lot, above above a certain level. David Moses (1:14:47) Actually actually probably intelligence at some point is diminishing in terms of happiness. at way before you get to one fifty, my guess is you know, the the the Jack BeVier (1:14:51) Yeah, to meas yeah, absolutely. Yeah, exactly. David Moses (1:14:57) reason the reason the reason for the phrase ignorance is bliss is like, you know Jack BeVier (1:15:02) Yeah. David Moses (1:15:03) we're we're reminded of that every single day with the people that we 'cause it's like, how can you be so happy and so dumb at the same time? Jack BeVier (1:15:07) Yeah, there's like it's nice to have like optionality though, like you y Yeah, yeah, it's it's nice to like I don't know, it's nice to like have optional agency though. Agency requires a certain level of intelligence, you know, otherwise you're told what you need to go do in the morning, you know. David Moses (1:15:22) Right. I look I think I think that I think that there may be a limit. I think that's that's probably fully accurate, that there's a limit to the return on intelligence. but human curiosity and ingenuity will I think you while you may limit the value of intelligence beyond a certain point, I don't think you limit what that intelligence can do. And because of that, it's gonna be not just intelligence, but the speed of intelligence. And again, I'm not necessarily getting into like what happens if VCI becomes a thing and now I can put something into my brain that allows me to understand things at a level that that I couldn't before. But that something like that aside, I think that we have a long, long way to go as a species, you know, in in terms of figuring out our own biology and physiology. you know Curing diseases, figuring out a way to be happier, more content, eliminate the our tendencies to become addicted to things that are harmful to us. I think we have a long, long way to go. And I think that maybe maybe we're heading in, you know, maybe we're heading in a direction that is scary only because it's unknown. Right, only because it's limitless. And that I don't think the the I don't think the the the brilliance of the human brain lies in the higher echelons of its IQ ability in terms of IQ. I think it I think the real genius is in the efficiency of the brain. Like for for for what is the equivalent of energy of a twenty watt light bulb. Right? Like we we have just an un it's just an incredibly ridiculous level of intelligence and intuition. Per watt of energy. and I don't know. I mean, I I I I think I'm excited for what the world will look like. but you're right. Everything that we do today, if you put that that caveat on it, everything that we do today will cost us next to nothing to do. And there if we don't if we don't figure out new uses and if we don't get very curious about what the world could turn into with this powerful technology, I think we're gonna wind up with a bunch of people with nothing to do and no money. And, you know, it it's gonna hurt a lot. Right. With any with any technology, right, you you're gonna have that. With AI it's gonna be much faster and much stronger. It already is much faster and much stronger. So it's like we have to figure out what we can do with this. because it's coming. Jack BeVier (1:18:32) I think that the I'm excited about the intersections with the physical world as like because I do th I I agree with you. I think I think that I think you know the we we're not getting smarter more quickly. Like we just but we have figured out this how to create a machine that is smarter than us and we know how to make it even smarter still. And then but then then the limits of possibility become physics and those, you know, f you know, physical interactions with the with the external world. And I think that's the next frontier that is like like right now, like anthropic and open AI like, you know, own the software horizon, but the SpaceX stuff is no, he's not Elon Musk isn't competing with anybody. Like not really. Like and and and he's I just feel like that's that makes a ton of s that that vision just makes a ton of sense to me for like, you know, for what the actual future of humanity is gonna be. anyway, so awesome man. Good conversation as always. David Moses (1:19:33) Yes, sir. We will we'll be back at the Jack BeVier (1:19:35) Pick these topics up, you pick these topics up and others on the next one. Hope be really curious to hear people's thoughts on my on my little soapboxes there. Like rip my ass apart in the comments. Like give give it to me. You know, dis disagree with something that I said, like let's go. You know, I I would love to debate this idea. I I feel hustled right now, by by the world or you know, by the by by a couple of companies. And I I You know, I think that that makes more sense to me than any other version of what I'm seeing. so anyway, we'd be very interested in getting other people's takes on on that on those that set of theories there. And also like I'm interested because like it, you know, I'm I'm business planning around this stuff, right? And like w this we're talking about a one or two-year time horizon here. We're not talking about like what's gonna happen in the next five to ten years. Like w what happens in the next year or two is gonna fundamentally change like how we should. think about integrating this re these technologies into our and the path the you know the exact roadmap of how we are integrating these technologies into our businesses. So I don't think this has been a very, you know, a little bit of an up in the clouds episode, but I don't think that this is an up in the clouds. I think the implications of these are very like what I do tomorrow morning at nine o'clock. So don't want to lose that. All right, all David Moses (1:20:49) Right. Jack BeVier (1:20:49) great to see everybody. David, pleasure as always, sir. Hope everyone enjoyed this episode David Moses (1:20:52) Thank you, sir. Jack BeVier (1:20:53) of Fork.ai. See you guys.