0:01

Mr. Damien Riehl! Mr. Horace Wu, how are you? I'm doing well. And how are you? I couldn't be better. I guess I could be better, but I'll say I couldn't be better We're gonna stay optimistic, but that's usually how we end, not how we start. that's right. And I guess, all in all, life is really good. I work for a great company. My wife and children are doing amazingly well. I just bought a, of all things, I just bought a Tesla Model Y that has full self-driving. What? my wife said, "We are not gonna give money to that asshole." So I didn't buy it new, I bought it used. And, so I didn't give money to that asshole, but I did want… i'm on the Minnesota Governor's Council on Autonomous Vehicles. I'm its co-chair, and I felt bad about every-- all of my colleagues saying, "Oh, and my full self-driving Tesla does this" full self-driving. And so I'm like, I should probably be walking the walk or I guess in this case driving the au-autonomous vehicle driving. I bought it and it it is… i've ridden in Waymos, but I've Ya? Ridden in a full self-driving Model Y. It is just remarkable. Really remarkable. For the last week have not driven manually and it is, it is doing everything that I would, except better. That is that it's a better driver than I am. Huh? my wife we were driving to dinner and it was the first time that she didn't complain about my driving. Like that I was going too fast or that I was cutting someone off and I thought, "This is maybe a way that my marriage could be improved through full self-driving." So anyway, all that's to say is that life is pretty good when when you have a full self-driving car. Wow. Look, you've just put on an ad for the man that we shall not name and that's… Oh, I don't know. I don't know. There's a talk right now that's like a conversation going on right now that Twitter, or now known as X, is where all the legal folks are going back to. And I'm like, "No, why? Why are we doing this? Why X?" Yeah. Jack Newton just had that and Jae Um, a friend of the pod was just mentioning, he's trying to lure her back, and I think that, my X has never been a cesspool like everybody else's X has been a cesspool. My X has always been 100% legal tech. The only thing I follow is legal tech, and and the only thing that the feed gives me is legal tech. So anyway I've not left it because it's a lot of good signal, that is, signal-to-noise ratio is very low. So I am not an X user. I was n- I was not a Twitter user. I tried it and I just didn't have enough to say. I feel like I was one of the wallflowers in the corner of the room, and it felt awkward for me. So maybe I gotta try it again, I, I'm not sure j- be sure if you do it, just f- just follow legal tech people and follow like Andrej Karpathy and the, the people from Anthropic and the people from OpenAI. Just if you l- leave your feed curated to just all that it's not as good as LinkedIn, I think, for signal, but it's really good for s- signal. Interesting. I'll give it a go. I'll give it a go. But that i- interestingly might lead to our number of topics for today. So, we haven't caught up properly for a little while, and I actually used Claude to go through the last month of news articles and releases for legal and legal tech and AI and stuff, and it curated for me a list, and then I culled that down to about seven or eight topics that we're gonna be talking about today. So, you don't know any of these topics. We're gonna be doing rapid fire. Awesome. And it's gonna feel a lot like a Twitter conversation, I suspect. So shall we get started and cover the news for the last month? Yeah, All right. let's do it with the character limits and everything. I'll try and keep it under 140. Is it 140 or is it 280 now? they expanded it, so I don't know what it is, but We'll keep it short. Let me keep it short. as they say That's absolutely true. So let's do the first one. So the first one is and this is probably a little bit old news now because it's a month ago, OpenAI launched Astra for Law. And this comes on the heel-- the heels of Google and also Claude, and pretty much like all the major labs are now in the legal vertical. And we saw this kind of news article, I think maybe a week ago, where Phil Saunders, the CEO of Relativity, said that they're gonna prioritize working with Anthropic, OpenAI and Google over Harvey and Legora. And, he's saying that this is because it's more likely these companies are gonna have longevity. Never mind the potential of like a giant financial crash, but, this is a really interesting statement, right? Because if Relativity is doing this, are we gonna see more and other legal tech jumping on board and saying, "We're gonna prioritize the labs over Harvey and Legora"? That is fascinating. I had not heard that. And it's also not surprising. If you look at the Legal Tech Hub, our friends over there, there are one thousand one hundred legal AI companies. And so one can imagine that if you're Relativity or somebody like it how do you prioritize amongst that one thousand one hundred who you're gonna go with? And even the shortlist of the usual suspects, like you mentioned and otherwise you even have to say, even with our resources, like we can't do… do we do ten? Do we do five? And maybe you just keep it to a handful of those that you think really have legs. And of course the odds of Google going away tomorrow are pretty slim. The odds of OpenAI going away tomorrow or Anthropic going away tomorrow, the- these are all pretty good bets, if if you were to say who's gonna be around in five years. I think when everyone's thinking about how many arrows in their quiver are they gonna be putting together I think that you put the arrows that are gonna have some legs, to mix my metaphors, uh and OpenAI and Anthropic have some legs. Like what do you think? Is this gonna be the culling of the one thousand to say that a lot of those are just gonna be either subsumed by other companies or just killed with OpenAI, Anthropic, and Google? look, there there's a couple of factors going on here. One, it's in the interest of the labs to do as much as possible themselves. Like everybody wants to partner with them, everybody wants distribution, but why? Why would the labs open themselves up to partner with you? You gotta provide something that's more than what they can do themselves. And we've seen just this week that vibe coding can do a hell of a lot. I don't know if you saw, like someone vibe coded the entire Photoshop, like the Adobe Suite The Adobe Suite, yeah, a- along with all the things, yeah. It's ri- remarkable. And it's probably legal as reverse engineering. And so if someone can vibe code that, what can a software vendor bring that is not already possible just by AI vibing it out? So, like I think the th- 1,000, 1,100 companies all want distribution. They all want attention in this really starved economy. But why would anyone wanna play ball with them is the big question mark. And then there's the, the other side of it, which is I think like for all the companies and we're hearing from Legal Geek today in London, it's all about MCP now. We are hearing that everybody's now recognized the way that you get to be used is by fitting into a workflow that lawyers prefer, and that is a chat user interface. So like all of this push and pull and people wanting attention, I think it's gonna create a really noisy sort of system. And I honestly don't know who's gonna survive. I personally think there's gonna be a giant financial crash coming, and I'm not sure OpenAI and Anthropic are gonna come out of it unscathed. Yeah. I think you're probably right. The MCP connector idea is, i've been thinking about this a lot and, if only there were a way to allow interoperability between systems in a way of unique identifiers in the form of SALI and and FOLIO. And and I've vacillated. So I say that on the one hand, and then on the other hand, I wonder with MCP, how important is it, to be able to have unique identifiers when the systems themselves can reason as to say is a motion to dismiss the same as or different than a demurrer in California?" And because the large language models know demurrer and motion to dismiss or know merger agreement and know force majeure clause and act of God clause are probably similar, right? Because the AIs know that, then maybe MCP might be all you need. Maybe you don't need those unique identifiers to push and pull the data. But you're shaking your head. What do you think? Look I'm gonna push against that just to say the one thing that you do kind of risk is inconsistency. Like a model today is gonna say,"Yeah, those are the same things," and the model tomorrow, same model same everything, will go, "hang on. Maybe in this case it's not." And now you end up getting different results from the exact same chats. That's exact- And so I feel like I'm Abbott to your Costello or you're the Abbott to my Costello; where so you've just thrown me a li- And yes I think that you could have option one is just have the neural nets shake it out, and then it has the inconsistencies that you showed. And, the probabilism says that it's unreliable. But of course FOLIO and SALI are deterministic, therefore reliable and consistent. And what if we have both of them together? What if you could be able to say, "Here is now a knowledge graph saying that a, a merger agreement with a force majeure clause is probably the same as a merger agreement with an an act of God clause"? And then you could be able to say "Here is now a skeleton a deterministic skeleton as a knowledge graph upon which the neural net, the probabilism can reason with that deterministic skeleton." So that's a best of both worlds, what's called neuro-symbolic AI. That is the neural nets of AIs and then symbolic AI, the knowledge graph. So I think that's probably like peanut butter and chocolate. Those two tastes taste even better together. Or banana and peanut butter, depending on kind of what sandwiches you want, right? true. Let's not talk about Vegemite though. Vegemite goes with nothing. For those of you who know me, you will know that Vegemite for me is at the bottom of the bin. But that's an aside. So, I know. I know. I just lost my Australian citizenship, guys. I'm so sorry. Oh, that's fantastic. I've have had, I've had Vegemite only once, and I will emphasize once. That is all. Yes. Sometimes you can't dial it back to zero. Um, Yeah. So anyway, so I, yeah, I think that, so I vacillate on but I think that the secret is going to be create a knowledge graph of the things that matter, and then be able to do that for your specific domain. And that kind of harkens back to your question earlier, that if you've you as Horace and you as Da- me as Damien, if we've created knowledge graphs that are something that is better than OpenAI or Anthropic or anybody can do because we have the subject matter ex-expertise that they don't have, then maybe there's some value that we provide that they don't, and that's maybe the way that we stay alive. this has become a sort of a flavor of the month since the Kirkland announcement with Palantir. They're constructing a, an, an ontology or a knowledge graph of private fund stuff and, everyone's been talking about it in the marketplace. And y- you might have seen a couple of other companies, including DraftWise, have come out and said,"Oh, we can do ontology too." So like everyone is pitching this idea of selling forward deployed engineers to create ontology specific to an organization, which then allows an AI sort of a scaffolding to reason upon. And all of that I think is very rational as a construction, as an architecture. The part that blew me away, a- and Damien, I don't know if you've heard this in the marketplace, is the cost of one of these engagements is minimum seven figures. That's ridiculous. I know I think my socks and pants blew off when I heard that. That seven figures is really remarkable. It's that, And I guess the question is, do you need a Palantir to do that? Or can you can you jam with the AI and say, "Build me an ontology"? Have… Build an AI to be able to interview all of your lawyers and say,"What is it that makes it go this way? What does it make it to goes that way?" Every time a lawyer says, "It depends," have the bot say, "It depends on what?" And then be able to fit that into the ontology. That's what I did with SALI and FOLIO, is that, that came out of the lawyer's mouth after depe- it depends I put into SALI-FOLIO. So you can imagine a bot then doing the same thing. And maybe we can have autonomous ontology building that isn't gonna cost seven figures. That's I guess thing number one. Thing number two is that speaking of ontologies just today my friend and colleague, Joshua Lenon shared this with me. That there is a legal ontologies foundry. This legal ontologies foundry is saying, "Hey, maybe we could create an ontology of of legal ontologies." And here is me with my comment saying, "Hey, you might wanna review SALI and FOLIO. It has 18,000 legal concepts donated by Thomson Reuters, LexisNexis, Microsoft, and some of the biggest firms in the world. No need to recreate this 18,000 spoke wheel." And so anyway, this is this is something that ontologies are getting cool again. And I think everyone's waking up to them. And I I think people are going to have to rediscover the beauty of the ontology that is SALI and FOLIO. I agree. And I also believe there's no such thing as like, a universal ontology. Really ambitious people have tried and failed, so maybe someone is gonna get there. But I do believe you do need a subject matter expert to map your use case to a specific ontology in order for it to work. Whi- which brings us to the rise of knowledge management again as the primary driver to capture value inside of law firms, and we're seeing this more and more with the ridiculous sums of money that have been offered to the AI and innovation teams that are going around the Am Law market. So they… These are all public, but you've got 400,000-plus for as a salary for a director level at some of these law firms, and I'm like,"That is crazy." You've never seen that at, at, at law firms before. Yeah. What's old is new again. We as nerds have always said that knowledge management has always been the, the secret to unlocking all sorts of value. So it's it's both I guess maddening to say that "Hey, why didn't you hear us before? We've been saying this for decades." So it's maddening in one sense, but the other sense it's heartening to be able to say that finally people are walking down the right path. yeah. And look I'm happy that people are finally getting paid what they're worth. But you just gotta have, have to wonder, like, how long is this market gonna sustain itself? Because I think law firms are really paying to try and differentiate at the moment, because I don't think the partners themselves know how to stand out in this really crowded market when everybody has the same AI tools. What's your edge? And I feel like a lot of these hires are essentially the partners offloading their own responsibilities and say, "Here, Chief AI Officer, you figure it out for me. What's our edge in this law firm?" So I'm g- I'm gonna get a bit philosophical. What is the please What is a, what's the role of a lawyer, and what is the role of a knowledge management person? And part of it is a lawyer is to be able to be smart with the law, but part of it is just to be a counselor, to be able to give life advice. You're getting a divorce. Here's… Don't be an asshole, right? Your your business litigation maybe just, don't do this, right? So anyway, so the part of it is being a counselor. In the same way, being a knowledge management person, part of it is being a, almost a counselor too, to be able to say,"hey lawyer, I know your life is hard. Tell me what's making your life hard so I can make it easier for you." And then you can be the, the, the person the lawyer says, "Yeah, this, this part of my job sucks because this happens." And then, "Oh, I, as a knowledge management per- person, can make your life better. Here's h- how I can vibe code a thing to be able to go." Anyway, AI could probably do that, but it's probably a lot better, at least in the near term, for a human with good human touches and good s- soft skills, even though they're really hard, good soft skills to be able to help these lawyers into the next step of their careers through knowledge management tools, vibe coded or otherwise. So anyway, so just like a lawyer's, a big part of the lawyer's job is being a counselor and maybe helping them through in ways that are not really legal. I wonder if a knowledge manager's job is to help people through technology in ways that aren't necessarily knowledge management, but just helping them learn technology in the way that they can practice next. 100%. Like I think the role of the knowledge management, by the way, today is very different from the same role 10 years ago. It has morphed. It's much more not just about the words on the page or collecting precedents and templates, but also dealing with the changes in the technological landscape so that all of that knowledge can actually be used. And what is possible now with AI to capture tacit knowledge wasn't possible even two years ago. We had a chat with Dr. Megan Ma like maybe a year and a half ago where she's working on that project to capture the persona of the lawyer. And we were reading how this had been rolled out at Vorys, and it was at Dechert, and I think it was at a couple of other firms as well. Like that's amazing. We've finally gotten to the point where you can now essentially have something like a digital twin that's able to do work and act like you, even when you're sleeping. Yes. Building on what you just said is that, now we can be able to have, essentially knowledge collection at the speed of speech, and almost at the speed of thought. It, it-- because you can literally speak and then it becomes that knowledge artifact that could be then used later. So that's point number one. Point number two is that you can also build software at the speed of speech. I was walking home from from coffee with a friend yesterday, and one of my friends who works for an attorney general's office, he, he does antitrust things. And he he said, "Damien, there's software that I've always wanted to build." He said, I've got this hair-brained idea that if we were able to collect where people are moving from one job to the next job, say through LinkedIn or otherwise and then we are able to collect maybe salaries for such things and be able to put those into an AI, you might be able to tell for where there are depressed salaries, that maybe there's no poach agreements between companies to be able to say that, 'Hey, we're keeping the salaries down by keeping people from poaching each other.' Wouldn't that be cool if we had a software to do that?" I said, "Yeah, it would be cool. How about we take out my Claude Code on my phone, and how about I record literally you speaking what you want the software to do?" And so we started recording with Claude Code, jamming with the idea, and I used a harness called Compound Engineering, where it has a brainstorm function, where I said "Now brainstorm with us and ask us questions like you you might wanna ask." And Claude came back and said, "Okay, where do you want the data to come from? Do you want it come from LinkedIn? Do you want it come from o- other repositories of job postings, et cetera?" And we said yes and yes, and then they said, "Okay what about anti-scraping laws? LinkedIn has sued a lot of people for this, so how can we get around that? Do you wanna pay for the data or not?" So anyway, so w- we jammed for, maybe six or seven minutes on these kind of questions that are the same questions that an expert software engineer would ask, And so really the, the-- now my friend, it's literally in the background, there are now a dozen bots building what my friend wanted, and I'm gonna make it free and open source. And so really there's the value of a team of top builders within law firms. That value's kind of gone away because a knowledge management person can do the same thing I just did, saying,"Hey, let's record the Zoom call lawyer and tell me what, what would you-- if you were in an ideal world, what would you dream of building? What's the dream tool you would want?" And then whatever comes out of the lawyer's mouth gets recorded, and then I just jam with Claude Code to build the thing. I think that this is a remarkable time in human history where we can give people magic very easily. So I was having this conversation with a, a VC this week, and the question really is do we really think that everybody is gonna become a vibe coder? Do we really think that everybody's gonna have the skills to be able to do this? Or is it just gonna be a specialized percentage or fraction of the industry who'll be able to do this and do it well? I don't think we need to have everybody be a vibe coder any more than we need everyone to be a prompt engineer. Prompt engineering was a thing for about 20 minutes, and then it's not, because the AI just does all the things, right? And so I would say that like I know how to use Claude Code, and I know how to use compound engineering to do, to be in the park with my neighbor friend and do the thing. But that's gonna take 20 minutes. And then he's just gonna say to his AI,"Hey, I want software to do this." And the AI's gonna know, "Oh, he probably wants me to build software for him. Let's go into the Claude Code thing, and now let's ask him questions." And then after that, it builds it for my friend. So today, in this kinda liminal stage, he needs me because I'm an AI whisperer. But that's not forever. We can argue about timelines. How long is it gonna take before the AI just gives him software without even knowing that he wanted software? And maybe next year. I think just like prompt engineering had a very short shelf life, I think the, vibe coding is gonna have a self, short shelf life because we're just gonna build… AI's gonna build software for us without us even knowing that we've asked it to build software. See this is literally a blog post I wrote yesterday. And the blog post was entitled something like "There are only two valuable things left to build." And it was data was one, and the other was skills. So tacit knowledge. Everything else you can vibe. And so then we-- you and I have been talking about data forever, right? Yesterday's case, yesterday's statute, yesterday's regulation, right? There's only that it's scarce therefore valuable. But if-- the irony, of course, this is also public law, right? So something that is public is also scarce seems weird. anyway but for now we'll ride it out. So yes, data, and then skills. I don't know, I don't know how scarce skills is gonna be because that tacit knowledge is so easy to become untacit and go into artefacts. Yep. Yep, yep. It is, but like it depends on how deep you wanna scratch. I, I believe it's Okay, so let's say the skill is the ability to stand up in court and argue a case. Okay I agree. That can probably be generalized and captured from multiple people. But what if it's the skill to know a particular judge in a particular context for a particular kind of a case? That's gonna be a lot harder to capture. And now you can drill down even more in saying a particular judge for a particular case in in a particular day or particular hour. All of those things become harder and harder to capture as you go down to the nuance. So like anything, I believe there is a certain line that you're gonna have to draw to say, " Okay, anything beyond this kind of general level, we're g- we're willing to pay for. We're willing to pay for the expertise. We're willing to pay for that extra layer." So I think that's a spectrum that we're gonna be facing. I love that you mentioned, I know the judge because that's something that, you know, that, that is going to be, Essentially what you're saying when I tell the client, "Hey, I know this judge. I've appeared before him a bunch of times," essentially what you're telling the client is that I have data that you don't have. My data is scarce, right? Yes. Have these data points back in the chambers, that there's no recording, there's no, Where this, this judge beat up the other side for doing this bad thing, right? So essentially, I'm telegraphing scarcity to my client by saying that I know the judge. And so there's a, a real question, what-- how do you get that data? And one can imagine there could be a within a law firm saying, "This judge hates it when you do this, because I heard him say that, in chambers once." So that, that can now be an artifact that is no longer scarce, at least within my chosen sharing. So that's kind of point number one. Point number two is the fact that I know the judge is doing what we as public speakers and we as writers have known for millennia, and that is know your audience. And so as we think about we need to figure out the personas audience. I think I've mentioned that I'm on a, a group that is talking AI at the Vatican. We did that back in November. We're gonna do it again in one month. I'm a part of a, a taxonomy group and an ontology group that is debating whether AI has consciousness or not. And and you can imagine there's some pretty strong feelings in Rome as to whether AI has consciousness or not. And they've convened so a lot of philosophers are essentially doing the steel man for the fact that it does not have consciousness. But then they've convened a, essentially devil's advocacy group, my group, to be able to throw rocks at it and say, no, "AI does have consciousness and here are the reasons why." And this is essentially to bolster the no consciousness group's arguments. So we're gonna take all the no consciousness people and then essentially say why they're wrong. Okay. like you don't like that. No I do like that. I actually want to hear if you can give like a bullet point summary of why you think AI has consciousness. Oh, mostly we're steelmanning the argument. Okay I, Damien Riehl don't think so, but I want to be able to have the as strong-- I want the group to be as strong as possible, so I want to throw arguments at it to be able to bolster it so they can counter those arguments. Let's-- If that makes sense. So I, Damien, don't think that is conscious, but for example, they, So steel, steelman, the, the argument from the church is gonna be, "Hey, there is no consciousness 'cause there is no soul, there is no there's no l-lifeblood," et cetera. My response as a steelman, again, devil's advocate, to be able to say, "Okay, then how about a cat? Does a cat have consciousness?" Is it-- if it-- Probably, but it what kind of consciousness is it? And what are the indicia of consciousness that a cat provides? And then, okay, now how about somebody who is maybe in a coma? Do they have consciousness or not? Right? Are they no longer human, right? And then it-- you come back to, okay, what are the indicia of consciousness? And I I'm speaking to you right now. I'm responding to you. I'm, I've s- I'm smiling. I'm really animated. These are all things that the bots can do today. Horace, you don't know if I'm conscious or not. All you know is my indicia of consciousness. And so really, how do we know whether the AI is conscious or not? I got a call just before this from a number I don't know. And I rarely pick these up 'cause I usually send them to like Google Screen, and it will just, do its thing, and it'll tell you, oh, the person hung up. And I decided to pick it up'cause, I was cooking food and I was like, "Yeah, what the hell?" And as they were talking to me, I couldn't tell if it was a person or a bot. There, there were enough ums and ahs to make me go, "This, this pattern sounds like a bot." But then I don't wanna call it out. I don't wanna say, "Are you a bot or not?" So I asked it some really weird random questions and it responded in a way that made me think, "That was a little too smooth. I think you are a bot." Yeah. And that's the reverse Turing test almost. That it's testing you to see if you know that it's a bot. The But yeah, so like we're in a weird world that, yeah, as, as the AIs get even better, it's gonna be harder to tell the bots from the humans. And so then if our human indicia of consciousness now merges, how do we even how do we even know that there's no consciousness? So anyway, so this brings us all back to I know the judge, which you're saying know the judge. So as part of the Steelman group and this argumentative group know your audience has always been important. So I said, what we should be doing is building personas. That is building personas because the arguments for what we're calling the Catholic Turing test that the arguments for an atheist are gonna be much for- different than the arguments for a Muslim, which are very different than our arguments for a Jew, different than arguments for, a Lutheran, et cetera. So as I said, we should be building personas so that argumentation can be able to be fit for our audience. And so really the I think that is true for things of AI consciousness arguments, and it's also true for, of course, legal technology. We can now build personas for our particular user. And in the past, we might had 20 or 30 personas of our ideal users, but now I think we can have almost infinite personas. To be able to say, is this an M&A lawyer that sometimes dabbles in funds right? Or is this a, a personal injury lawyer that sometimes dabbles in bankruptcy, Yep. so now Yep to have essentially all, a mixture of personas to really fit the individual. So instead of broadcasting personas, we can narrowcast personas. The implication for product building is really hard because now you have to figure out what do you tailor? What don't you tailor? What are the, the experiences you want to create? I, I was listening to this interview with Alexander, Alexandr Wang, I think, from now Facebook. Yes, exactly. And he was talking about that this release of Muse that they've, released last week, and it's last week before and now it's growing like crazy and everyone's adopting it to become their personal agent. They had to exercise a lot of opinions in creating this. They didn't want this to be a, an agent created by a committee and then be in this middle ground like nobody wanted. They were very particular about how to build it and he was saying something like to build this, what is probably one of the biggest launches for Meta this year, they used a team of 200 engineers in an organization of about 70,000 people to, to do this one very important project. And to me, that was like this, this crazy sort of product building thing, almost Steve Job-y and I'm sure Alexandr Wang would be very happy to be compared to Steve Jobs. That it, it was like, okay you just had to predict what your users want. So to the point about narrowcasting, i-in a sense, you're almost saying we as the product builders are gonna offload some of our decisions onto an AI for an AI to narrowcast for a specific individual. And our job is to put the parameters and the guardrails and the framework around this so that the AI doesn't stray beyond what's permitted, and that's hard. It is hard because it really requires taste. Because those, those few people within Meta of those, you know, 70,000 people, the few people on this project were probably chosen because of their good taste. And… yeah, you're g- you're offending 600 and, sorry, 69,000 people at the… is it like SEAL Team Six? There can only be so many people in SEAL Team Six, and Yes. the rest of the, Navy Navy SEALs. So anyway, so the I would say that to be able to say, okay, we're… How many did you say, 40 people or how big the… two, 200 people out of 70,000. Sorry. So the, those 200 people were probably chosen to be able to say that these are people that have good taste that can wrangle the AI in a way that is not gonna be unwranglable. And to, so I, I think that this is kinda why people hire Rick Rubin, the record producer. He said I'm a shitty musician, but I've got really good taste." And and he, Rick Rubin also just released a book, about AI and vibe coding to be able to say that, "Hey, vibe coding is what I've been doing for the last bunch of years. I just didn't call it vibe coding." But it's his good taste that can be able to, wrangle the AIs to make good music. And so I would say that this is maybe the 200 people within within Meta's good taste to be able to make their Muse such a I guess a pretty good success for every- everyone, so far. it. It's, they've heard it's amazing Yeah. Yeah. So w- we'll watch and see 'cause I think this might be an indication that hiring the right people absolutely makes a difference. And I think that's true. So speaking of hiring people I think the, the hottest jobs going at the moment i- in the market are still coming from, our friends Harvey and Legora. And you might have seen that I think in the last two days they both released the same tool for in-house teams called Intake, which I thought was hilarious 'cause the first thing I saw was a Spider-Man meme on LinkedIn. Right. It's true, yeah. And then the… I think that lesson number one is that trademarks are hard. That is because I've been, I've been talking about, wouldn't it be stupid if one company got the trademark on something called "co-counsel" when co-counsel has been a thing for, 100 years? You talked about co-counsel, but then now all of a sudden that's a product? So and then the fact that Intake is now two companies, that's indicia if if you're doing something else that something else, someone else is doing, then that's an indicia that you're on the right track. So Right They independently did the same one means that it's probably a pretty good name. And I think that's true, but I think more interesting is both of them are making a move from law firms to in-house on the same day. I think it's a sign that a couple of things are happening. One, everybody in the market knows that every law firm has put together a plan B to move away from Harvey and Legora. I am not spilling any secrets here. And because they've put together plan Bs, Harvey and Legora now have to figure out what is it that we're gonna do to maintain these astronomical valuations that they're gonna have to sustain?" So that that's like pressure number one. And pressure number two, and I'm hearing this from different folks, that the VC for Legora and Harvey are pushing them towards in-house because that's a much, much bigger market than going after law firms. And now it's creating this tension of who, who are they serving? Are they serving the outside counsel? Are they serving the in-house counsel? Is it ultimately gonna serve the business and therefore all lawyers can be bypassed, and legal services and advice can come directly from the AIs themselves? So I think this is like this… the beginning of some trend that we're gonna see play out in the next six to 12 months. so I heard from someone I know at one of the big two, I'm not gonna say which of the big two, Harvey or Legora, but at one of those companies who said that they thought that the top of the market, that is enterprise law firms is saturated. That they both have essentially divvied up who's gonna go with whom. So the top of the market is saturated, so they are instead going to be going to the mid-market and small market. And that is mid-law firms to smaller law firms, all the way down to solo smalls to be able to now go beyond that saturated market. I-- We'll call it a y-axis. And then the x-axis is is going directly to the consumer, that is the in-house counsel like they did with Intake. And so there is going down market and then going in-house, in-house is of course logical. Because down market because you need to sell to somebody and then in-house because that's where the money is. And really, the consumer I talked to so many in-house counsel. I spoke at the Product Liability Council in Denver a couple of weeks ago, and so this is all in-house people who are saying "We are, we are now… A lot of us are just bypassing the lawyers and we're doing things that the law firms haven't or won't do, and we're doing it cheaper, better, faster." So I, You've been, saying this for months. You've been saying this, the, the phone that doesn't ring, Damien? that's exactly right. And so it seems obvious that, of course if you're going to expand your market, then then go for the in-house, go where the money is. And th- then the next question is how soon before Harvey and Legora go to consumers? That is if I'm a, if I'm a personal injury plaintiff, can I use something to be able to bring in a lawsuit using one of those tools? Or if I'm a, if I wanna be able to, do a small business deal, can I just go directly to go the LegalZoom route? You might have seen Harvey's already buying TV ads, and TV ads are definitely targeted at consumers, 'cause who watches TV now? Yeah. It US Open, right? The, these consumers. Yep, yep. So I think we're seeing the beginning of this, and I agree with you. The, the top of the market is saturated. The top of the market have all started their plan B strategies. So, something is gonna happen. It's a question of time. It's a question of what are the labs gonna do directly? What the, what do the market opportunities look like? All of those are not predictable at the moment. So we watch and see. Oh, we do. And on the topic of plan Bs and being able to build maybe what you had to buy in the past, i've mentioned in the past my my lawsuit against my next-door neighbors Yes inches from my house. And I said, "Hey, let's find a reasonable distance," and they said, "No, it's going in," so now we're in a lawsuit. So in preparation for mediation we… I said I need to show that I either have, at the best case an adverse possession of the side yard between the houses, that I own it now, or at at worst case, I have an easement where you can't build a fence because I walked the… I've walked there for the last sixteen years. And the more I walk there over sixteen years, the more I prove my easement." I thought of I was talking with my friend and lawyer, Jake, and Jake said we probably need to we need to be able to do a mediator statement so the mediator knows what's going on." I said I could do it in the old way, or I could build a web app for the mediator to be able to show them what I'm doing." So here is a Whoa! of my side yard. So this is my house on the top, and this is my this is my neighbor's house below, and here's a heat map of all of the times that I've walked my trash to the recycling and to the travel every time I've mowed, every time I've cleared leaves I've cleared my ice dams. And then I thought it's one thing to be able to speak it in a brief, it's another to show it in here. And then I thought, "Wouldn't it be better if that that map here moved?" And so now, here I can click on this, and I can click play, and look at this. This is now, people walking and me taking my trash out. And this is my predecessor, that is the person who I bought the house from, that is now going, and it's… At this point, there are now 1,100 uses back in 1994, of the side yard. And each one of those uses further shows that I have an easement or I have adverse possession of this. And now we're up to 3,000, and then soon we're gonna be up to 6,000. And think about how much more this demonstrates the use of the side yard. This looks like Grand Central Station on a time lapse. And who's gonna say that that my having now 10,000 different uses is only occasional? And that's the cases that have lost have said if it's only an occasional use, it's not gonna do it." But here we are now at 22,000 uses over the last thirty years. And what judge is gonna look at that and say, "Oh, that's too insignificant to show that he doesn't have an easement"? So anyway, so th-- for those who have are watching this on YouTube, which I would recommend you should watch it on YouTube to be able to see what I've just showed what is more persuasive, me writing a brief with words or me just showing you a time lapse of all these uses? And then me to be able to say, "Here's all the times where I just walked," or, "Here's the, all the times where I took out the trash," or, "All the times where I mowed." So this web app to be able to do this is I took a weekend to do, and it's much more persuasive than anything that I could have done before. As we think about building versus buying, like this is something that just wasn't poss- possible before. Or you would have had to hire a, a software consultant to be able to build. And one more thing, It wasn't economical. You could have always done it, but it just wasn't economical It wasn't. And then I wanted to show that for them to win, they have to run the whole gamut to be able to show not only do I not have adverse possession but I also don't have an easement. And they can say, "Hey, the Thompsons didn't have 6,000 things. They only had 120." But then you still have to show that the Riehls didn't have it, and then you have to go down here. But then you have to do the same thing for the trash and do the same thing for the recycling. So they have to-- I, I built this web app to show that they have to go all the way down. They have to run the whole table to be able to win, because I still win under pretty much any scenario. so again, this is really… I feel like that my opposing counsel is bringing a knife to a gunfight, that, he just doesn't know that the things that I'm showing are gonna be far more persuasive, and they just can't, they can't compete. It's massively impressive, and I can see your inner litigator kind of coming to the surface. But one thing, that demonstration, that video, that heat map you showed was only possible with data. How did you get the data? So the data came from the testimony from the person I bought the house from. She has an affidavit that I'm literally holding up right now. And in this affidavit, I-- she said I walked once a day to be able to go from the front to the back, and I walked my trash three times a week. And so I took this affidavit here, and I said, "Now calculate that over the seventeen or eighteen years that she lived in the house. And then do the same for the-- when she raked the leaves." And then I take all of my affidavits and all the testimony that I had and put that in there too. So that's the data that it used. and then I said, "Okay, this walk, I actually walked next to the house, and this one I walked in the center of the thing." So I was giving it the data. So anyway, so there's, With any litigation, there's, there are facts and there are laws. And so the facts as the data really you can you can be able to provide as a client. But then the law, of course, are in the cases of the statutes and the regulations. And I s- I then took all of the elements of easement, and it took all the elements of adverse possession. And then what I did is I then I showed made one other thing that is that is fun, then I took all of the law that i-- that is what is needed to be able to prove adverse possession and easement, and then here is now the, my brief to be able to say "They bear the burden of proving that I was permitted and didn't take it myself. Here's the timeline, and then here is the every element of adverse possession. Actual possession? Cool. Here's all of my ways that I prove adverse possession both the Thompsons, my predecessors, and the Riehls. And then open and notorious? Here's the way that I did that." So I marched through every single one of the elements of every single claim to show how I win. And I did that also for the a bunch of other aspects to be able to say, "Here's-- You're gonna lose because the person who lived next door to me, who could have given me permission, died years ago. So she can't testify, and so-- and you didn't move in until recently." So anyway, all this to say is data is going to be king and the data largely comes from the clients. Either the client's factual data or the law, legal data. And again, that goes to scarcity of data. Only three companies in the world have that legal data. Now speaking of the three companies, one of the other things I found when I was doing research for this episode, and I don't know if you've seen this

yet : 38:51

Thomson Reuters versus ROSS. Yeah. So on the 29th of September, so only two weeks ago the Third Circuit, and I didn't read this, but I got a summary of it, issued the first federal appellate decision on the fair use of AI training, affirming that ROSS infringed on Westlaw's headnotes. Did you read this? I haven't read the case. I've read the, I- it's been on my to-do list to read it, but yeah, I've read the, the news coverage of it, and I will say that when I was a litigator there's a saying that bad facts make for bad law. And I think that might be the case here. There are very unfavorable facts that ROSS hired somebody to to essentially go through and copy all of the things because the third party had access to Westlaw where ROSS did not. Yep that that's just bad facts. And there's similar to the, Blurred Lines case when I think, Pharrell said, "Hey, let's make something that sounds like a Marvin Gaye tune." And then that's bad facts, which made for bad law, right? So I think Yes i- in the same way that the Blurred Lines said that you can now copyright vibes the musical vibes, that's bad law. And I think in the same way, this ROSS case to be able to say that that is not fair use I think that's also maybe bad facts making That facts. B- because the-- because really what was being copied was a description of the law. The law is a public document. And so you can't copyright the case, right? And then s-- you can copyright maybe commentary on the case. That is a copyrightable expression. But at the same time, if that expression hews really closely to the law that is uncopyrightable, then it's not copyrightable, right? So if it-- because the more creative I make the des- the description, the less useful it is to the lawyer, because now the editor has all of a sudden been creative, and now I'm gonna misstate the law because of the editor's creativity. No, the editor needs to hew very closely to what the case said, and the closer you to hew to this public domain case, the less copyrightable it is. And then now that ROSS has taken this pretty close to uncopyrightable description of the case and then abstracted away in 12,000 dimensional vector space, to be able to take the ideas, not the expressions of the ideas, but just merely the ideas away is now taking a almost not copyrightable description of the case and then putting it into 12,000 dimensional vector space. Is that not fair use? And so anyway. So I would say that if there weren't as bad facts as they were, we probably would come to a different result. But here we Yep. Yep. And I just think if a similar scenario happened today and someone didn't use the headnotes and instead they got an LLM to read over the case and produce their own headnotes, this wouldn't arise at all. 100%. And there's for two reasons. One is that, why would I need Thomson Reuters headnotes because I can just do it myself? So what are the-- what's the value of headnotes? Like pretty much nothing. And then I would say also, what's the value of secondary materials that comment on the cases, statutes, and regs? And the answer is next to no value. And that's been the, that's been the moat of Thomson Reuters and LexisNexis for decades, is that, oh, they have the secondary materials. But with AI, who cares? Who needs Yeah. materials? you need is the f- the primary. As long as you can get the primary materials in your hand somehow, what's the difference? So the-- And that's point number one. Point number two is that if I make an LLM summary of a case and you steal my LLM summary, no copyright infringement. And the reason for that is 'cause if Yes office has said that if AI generated, then uncopyrightable. So essentially, you've stolen an uncopyrightable thing that is also in the public domain. Yep. Yep Option one is that it's AI can do it myself, so I'm not gonna need to steal from you. Option two, you can steal my AI-created thing freely because it's not copyrightable in the first place The, this, this argument came about when do you remember when the Claude Code harness was leaked and people started copying it, pasting it everywhere, and an argument was "hang on, Anthropic doesn't have copyright on this because they publicly said most of this was vibed out." That's right. And th- so that's one reason, yes. It's vibed out, therefore uncopyrightable. The other is that you cannot copyright something that is functional. That is if the only way a machine works is to be able to make the machine this way that design of the machine is uncopyrightable because that is a functional thing, not a creative thing. To be copyrightable, it must be creative. So there's a pretty good argument that you make, number one, is that one, it's vibe coded therefore uncopyrightable. But number two, it's functional. Therefore, because it needs that to function, there are only some- so many ways to be able to express that functional thing, then therefore it's not copyrightable Oh, I'm gonna just blow my own mind here for just a second. You know how we're talking about capturing the brains, the tacit knowledge of lawyers into skills files? A lot of this is done via LLMs today. So what the LLM produces at the end of that's actually not copyrightable. But that according to the US Copyright Office, ja- Japan has said that may be copyrightable, but US has said not copyrightable. I-i-in that case, what's the actual IP? Who owns it? What, what's gonna happen to the skills? Oh, and that's the, the biggest question. And I've been saying, I've been pounding this drum for a long time saying that if you work for a big law firm and you say,"Hey lawyer, put your tacit knowledge into our artifacts so we can use it," then that is only as good as that as long as I'm still within that law firm. I could jump ship and then take all that tacit knowledge and then take it somewhere else, and there's nothing to steal. Like there's the, the, because it's that tacit knowledge is not copyrightable in the first place. Just mind-blowing. Just mind-blowing. We've, we're coming towards the end. Are. And so let's talk about optimism. What are you opti- what are you optimistic about, Horace? What are you excited about? look from this conversation, what I'm optimistic about is that I think a lot of people are working on solving different problems from a lot of different angles, and I think AI is now at the stage where it's really competent. Beyond anything we expected in the last 12 months. Like I was a, not an AI skeptic, AI plateau-er. and now I'm like yeah, even if I am an AI plateau-er, the plateau is really high. And so we- we're now at the stage where all of us can potentially 10X our productivity, and a lot of people are doing that, and we're seeing that. So I'm really optimistic about that. Damien, what about you? so I was recently named to the board of trustees for a university. So I'm now one of twenty people that is gonna decide how these eighteen to twenty-two-year-olds are gonna matriculate and actually become better human beings. So it's a liberal arts university, and we had our two-day retreat and meetings last Thursday and Friday. And I was essentially brought on the-- as a trustee of the university to be able to provide AI knowledge and to be able to say, "As AI is maybe eating more computer science jobs, what are we gonna be teaching?" and I met with four students at a table, and I said, "What do you do?""Computer science." "What do you do?""Computer science." "What do you do?""Computer science." And the th- fourth one is a a occupational therapist major. And I said the bad news is three of you probably won't have jobs but the fourth one definitely will, right? The occupational therapist, like robots are not gonna take, pressing on somebody's legs to make sure that they can do it. That at least for the near term, that's gonna have that. I'm optimistic that during the course of the discussions, I said one of the, my fellow trustees, he made, tens of millions of dollars on, in business. But he said, "I was a French major," and he said, "I was a French major, a liberal arts major that was really good at business, made a lot of money." But then in the '80s and '90s, everybody said, "No STEM." You need to have STEM. You need to have return on investment to be able to do, science, technology, engineering, and mathematics to be able to get a return on your investment. That was the beating of the drum for twenty years, thirty years. But I said, "AI is eating math." You have a, you have Millennium Prize problems, a hundred-year-old standing ma-mathematical problems just getting cooked. are we gonna be teaching mathematics, or are we gonna swing back to liberal arts to say, "What is this life all about as AI is doing more of our work?" What's the point this all? And I think that we're gonna be swinging back to the liberal arts, so that's my optimism, to say that the philosophers are gonna come back as we-- as robots do our, more of our work. We're gonna have more of the real questions in life saying, "How much do I-- should I work, and what really is life meaningful?" Only if we're allowed to wear togas, Damien. I, I want to see you in a toga. uh, No, Horace. I, that, that would- I, I think perhaps our next podcast should be in togas. We might be able to do it in Switzerland, which is something we'll talk about a little bit later too. I, I'm gonna speed run this episode and get it out there. But looking forward to our chat, I think next week too. I think so too. Horace, always good talking to you, man, and thanks Likewise. listening. So much fun I love you guys. I'll see you guys soon. See ya. Bye.