The Fast and the Curious

After a two month unplanned hiatus, in this episode, Damien and Horace wrap up 2025 discussing the latest trends in legal tech and AI. 

From encoding morality to vibe coding by legal professionals, we cover recent developments that we should be watching. Along the way, we tackle the transformation in law firms towards AI adoption and debate the future of legal services in the context of accessible and open data. 

00:00 AI and Human Values at the Vatican
01:32 Building a Moral Knowledge Graph
03:02 Recent Developments in Legal Tech
05:22 Joining Clio and Industry Insights
06:51 AI Strategy and Law Firm Transformation
20:42 Trust, Intuition, and Integrity in Legal Practice
27:41 The Ethical Dilemma of Paywalls
34:08 The Future of Open Source and Open Access
41:29 The Philosophical Debate on Open Source and AI

Transcript

0:00

Terrible, terrible thing, Damien. We haven't released an episode for like a month and a half, two months. Shame on us. Mea culpa, mea maxima culpa. Slap, slap, slap. Partly that's because we've been traveling so much. So last time I caught up with you, you had just come back from the Vatican. That's true. Yeah. It's, it is, it's a weird thing that a friend of mine said, do you wanna talk AI to the Vatican? And I was like, uh, yes. I think the answer to that is always yes. And so we, we talked about the idea that Sam Altman has said that we need to "align AI with human values." And the discussion was, Hey, the Vatican has a ton of "human values" that are locked up in the archive and library. Maybe we can open those up so that the AI knows more of those "human values." And so for the last month or so, I've been working with two fronts. Front number one is to maybe open up a lot of that data so that the AI can ingest it to be able to be trained on, you know, world's history of Thomas Aquinas and Thomas Moore and Aristotle, and all of the best thinkers on what it, what it is to be a moral and ethical life. But then also to be able to build symbolic AI. To be able to think through these things. And right now I'm building something that is kind of like I did with SALI, but with moral and ethical things. You know, "human values" differ between Horace's human values, and Damien's human values, and other people's human values. And we kind of replicate that through law. That is, that law is the best approximation of us being able to collectively say, okay, these are our values. Don't murder. Don't steal. Yes. Yep. right. But of course German law is different than Texas law. Because under German law, Nazi paraphernalia has to come down. Mm-hmm. unlawful. But in Texas law, that has to stay up. Because that's free speech. So whose "human values" do you align with? Do you align with German human values or Texas human values? Yep. So really what we're building now is a way that you can use symbolic AI to be able to deterministically not probabilistically, but deterministically, almost have a SALI-like, way to be able to say what is ethical, what is moral? And be able to pull from the corpus, that is the Bible and from, uh, the Torah. And to all sorts of other things and be able to say, okay, this is the thing that we all can collectively agree upon. That is the, that is moral and just. And I'm working with Mike Bommarito, who's a friend, uh, you know, of both of us. And Bommarito said, oh, that's a project he can get behind. So he's gonna be extracting all of Oh wow. from the Bible. And to be able to build a knowledge graph, to be able to say that, you know, theft is wrong, murder is wrong, right? And to be able to build that symbolically. And then do that same thing for Islam and do that same thing for Judaism And then map them all together. See what overlay. them all together. Right. And so, and I think that this, number one, we often, you know, in our social media, we often think about all the ways that we're different. But this is a way to be able to say, no, this is all the things we can agree on. And this is a way to be able to say, this is the least worst option. That is most of the Germans and the Texans can agree upon these things, and then we just have to figure out, okay, for those that we can't, you know, the Nazi paraphernalia has to either stay up or come down. We could be able to use maybe different corpuses than just the law to be able to figure what is moral and ethical. That's so fascinating. The principles behind building what you are just describing is the same in law as it is in religion, as it is in, in all the different kind of human endeavors. Over the last couple of years, we've come to become more familiar with these building processes. Um, which is an interesting segue because in the last couple of weeks... there's so much to talk about, so, so much to talk about... um, but in the last couple of weeks you, you might have seen on LinkedIn, there are now lawyers who are vibe coding their own tools like Jamie Tso in, I think Singapore or Hong Kong. Yeah. Doing great things. Yeah. And, and he was able to replicate, uh, the, the table review tool from Harvey and Legora in a span of like a couple of afternoons. He was able to, uh, or not him, but someone else called Anton, Antur, I, I, I can't remember, uh, was able to do the, the Spellbook feature in the span of two days. And he's open source that as well, like the creative understanding, and process, and the ability for people to build is exponentially improving, and that's really exciting. It, it's true. And it makes you think about — we, within the startup community talk a lot about moats. Mm-hmm. What is the moat of any particular startup? If your clients, your customers, that is the law firms are competing against you, then what is your moat? This is a build versus buy question. And this reminds me a lot of Darth Vaughn from Ford, he spoke publicly, so I'm not speaking out of school, but he said, " Hey, law firms, you have a new competitor in town. It's called me, as an in-house counsel." You have to show me what you can do that I can't do with tools that my AI backed tools can do. So it's not just enough that you're using the tools. What value are you providing on top of that? Because I'm your biggest competitor, so show me what you can do on top of that. So I would say that that's for the lawyer client relationship, but what you described is the law firms, you know, essentially replicating these tools. That is for the vendor law firm relationship. To be able to say what can the vendor provide that is above what the law firm can provide itself by vibe coding. And we talk about this a lot, but one of those is proprietary data. Do you have the cases, statutes and regulations. That is one set of proprietary data. Another set of proprietary data is for contracts and transactions. What is the knowledge? Have I built out a knowledge graph of the things that matter for this type of M&A deal? Or this type of real estate deal? I think that proprietary data is the answer to the question "what value can you provide" above. But I would say precious few legal tech companies have that value on top. Yeah, and it's, uh, it's hard, right? Like we talk about in the, in the legal vertical, there's really three companies that have that public data. There's Le LexisNexus, Thomson Reuters, and now Clio. And since the last time we talked, uh, I'm now a Clion. I'm officially a part of Clio. A Clion, that's a cool name. Do we, they've had that do we, do we have to greet like this? maybe, maybe I have little green, uh, antenna that come up. Yes, That's fantastic. How are you finding it by the way? It's great fun. Yeah, they're really smart people and they're, they're Canadian and so I'm, I'm from Minnesota, which is also known as South Canada. Uh, so we, we, Yes. get along great. They are very lovely people and, uh, one of the, one of the companies I respect a lot in the legal, in the legal space. Um, so congratulations and congratulations to Clio as well. Thanks. It's really exciting. But it, but like you said, there's, there's only three companies that really have yesterday's case, yesterday's statute, and yesterday's regulation. So that is certainly a moat for now. But of course there are other companies that are building Mm. capability. And so one can imagine that that's moat number one, that we have the cases, statutes and regs. Moat number two is the business data. Yep. That is the business of law. We have, you know, correspondence with the client and we have the client's interviews and that kind of thing. And all of that is context for the context window to be able to build better things. So I would say that that is pretty proprietary too. You know, it's a lot of privileged information that I, as a lawyer, can then use to be able to then build out my complaint. Or build out my contract. Or build out my documents. So I would say that the smartest people in legal tech are amassing that proprietary data, whether it's public law or private context. And it's gonna be a race to see who can be able to connect those corpuses to useful products. And then utilize that. And this, by the way, leads us down to multiple different paths. Uh, I wanna cover a few. One is who's gonna be really, really good at tapping into the internal data? People like DeepJudge for example, who, uh, you know, are pushing this enterprise search platform that they've got. And the context window. This idea of context engineering, which we have not ever really discussed on this podcast, but it's becoming more and more important as people realize that no longer are we talking about just simple prompting like, like writing five, 10 pages of descriptions instructing the machine to do certain work is no longer enough. It's about building all the engineering to plug in the relevant data as you are using the machine, maybe even through multiple rounds, um, in order to get the output. And I know you work on this a lot, but I don't think we've ever discussed context engineering and what that's evolving into. And also just to talk about generally what's happening in the market, the, the connection of business of law data with the, the practice of law data, which I know Clio has been doing for years. Um, and others are starting to catch up as well. So, uh, you might have seen Harvey announced a connection to Aderant. Question if it's a deep connection or a shallow connection. But there is a connection now between the business of law and the practice of law, even at the likes of Harvey. Um, so, so the landscape is shifting fast and, oh gosh, this leads me to, to a certain thought. So you might've read a blog post I put up yesterday, about. Yes, so, so Damien's giving me the thumbs up. So for, for, for the listeners who, who might have missed this podcast, while Damien has been traveling and I've been traveling and I haven't been able to let go on this podcast, I've been like, ranting my thoughts on, onto a blog and the blog I released yesterday was talking about how... I think I called it, um, "AI strategy should not be focused on AI" or something like that. And, and it's this idea of law firms right now have been in a race to adopt AI and tools and rolling them out. And adoption is the key success metric. And it shouldn't be. It, it should come back to what is it that the law firm want itself to be in the next five years. So, all of this, Damien, let's, let's cover some of this in the next, in the next like 20, 30 minutes. What do you wanna tackle first? Well, let's talk about your article first, which I've pulled up right now. It's, "Your AI strategy should not be an 'AI' strategy," which is a really good title for a really good idea. There's the idea like, how are you using AI? And that question is as dumb as the question in the 1970s saying, how are you using electricity? Right. Uh, because electricity runs my typewriter, electricity runs my fax machine, right? It's a dumb question because of course *everything* is using AI. At least everything *should* be using AI. So the real question is, what are the goals that I want to achieve? What are the business processes I want to achieve? What are the client goals I want to achieve? And based on that, you could say, okay, how are you improving your business? Which happens to use AI. And I think that's the right way to think about it. And it always comes back to use case. Knowledge management and innovation people have always said, what are the use cases that we can do between the people, the processes, and then you apply the technology to that triangle. I really liked your post. I think the year of 26 is going to be... we thought about '23 as, you know, everyone is "deer in the headlights." That was '23.'24 was more of, "okay, gosh, what do we do with this thing?" And then '25 is, okay, let's build. And we've been building, we've seen a lot of building all over. I think '26 is going to be okay, let's think through how to not just go around the edges of what we do, but really think about how is this going to transform things. And we are already seeing a lot of boutique law firms jumping ship from the biggest law firms and saying, we're building an "AI first" law firm. Yep."Because the billable hour model is not going to work with AI because it's gonna shrink our profits and revenues."" So AI is not compatible with the billable model," say these law firms."So we're gonna create a boutique that is AI first." And that is going to reward efficiencies because that's what contingency fee lawyers have done for decades. To say, the less time I spend on a thing, the more money I make. So contingency fee lawyers reward efficiency. So these AI forward law firms similarly reward efficiency. So I think we're gonna see more of these law firms jumping ship and being AI first. Not just with the bigger firms saying, what can we do around the edges, but instead, how can we totally transform the client experience to make it better because of AI. Because this is a transformative technology. So I, I think your blog post was amazing in the way to be able to say, yes, let's think about how to do things differently. So that was topic number one. Topic number two. You talked about the business of law and the practice of law. Also, I like to say "substantive law" because a lot of people say "practice of law," they often put in the businessy things along with the practice. So I, I would say "business of law and substantive law." And, like you said, Clio's been doing that for years. Thomson Reuters used to have Elite, but they do not have Elite anymore. Right. And query whether LexisNexis, what do they have on the business of law side, right? And so when you think about who can actually *do* this. There's maybe only one company so far that can do that. With Um, company with all the pieces. With all the pieces internally, and then this goes to your point about Harvey. You know that Harvey is saying that we are, you know, working with Aderant. But query how much data Aderant is going to give Harvey wholesale. is it gonna be like the LexisNexis relationship where they get like a pinhole? You know, sipping through an API straw. And, uh, then you have the garbage in, garbage out. And not to cast dispersions, but of course if on one end of the straw somebody provides you garbage, there's not much you can do to be able to fix that garbage once it comes in. Because you don't have the underlying case, the underlying statute, the underlying regulation, and the thousands of them. Because you only get the "true positives" that are coming through the straw. You don't have all of the "false negatives." That is the true things that are in there that the straw didn't give you. So that's true for the case law and statutes and regulations. It's also true for the business of law. That is, if you don't have the entire corpus of the client data, how much can you actually get, to be able to do deep analytics and to get deep insights that are not just sipping through an API straw. We started this conversation talking about proprietary data, whether it's cases or whether it's business of law data. And we have been talking for years about, number one, the substantive law. But I think we're gonna be talking in, increasingly with

number two: 12:58

business of law. I think you're right. Like the, the move I'm seeing from law firms is people are starting to wake up and realize that in-house counsels, especially with the aid of Gen AI, can start doing a lot of work away from law firms. No longer do they need to brief out as much work as they used to. And we've covered this, uh, many times in our conversations. Um, but it's picking up the accel. This is an accelerating trend and from my sort of observations of what Gen AI can do, there's sort of three categories of what is enabling. The first is it's enabling the removal of lawyers completely from certain workflows. Highly commoditized stuff that, you know, lawyers can just write a script for and gen AI can go approve, disapprove. Lawyers are completely removed. No longer can they charge time for these tasks. Second category are where you used to take five hours to do something. Now you can do it in one. Um, and so clients are not gonna tolerate paying that extra four hours, so it's speeding up your work. And the third is Gen AI is enabling certain tasks that no humans had done before. And not that they weren't capable of doing it, but they just didn't have time or couldn't do it, or for whatever reason, just wasn't possible. Um, and these three different categories is forcing law firms to rethink how they price. How they engage. Um, and what is the product or service that they're offering? So like talking about Harvey, again, Harvey and A&O Sherman have been offering this kind of like commoditized, um, I forget what they call it, but essentially like a, a, a portal, to use a Legora term, to have a portal to access the lawyer's expertise and workflows and, and their clients will pay a subscription or whatever the model is, and A&O Sherman essentially like make money from what is a commoditized stream of work. And I know other law firms are exploring this as well. So I think part of 2026 we'll see law firms releasing a lot more subscription tools. Um. And what else? What else they gonna, are they gonna do? Right? Like the whole model is changing. That's right. And, and to that point, our friend Kyle Poe, who is from Legora, Yes. you and I had, uh, had drinks with in New York just a few weeks ago. He was just on Ted's podcast Yes. Theodopoulos and, and was just talking about, the portals. And Kyle made the really good insight that I as a law firm want to be able to provide that kind of subscription service to my clients. To be able to say, here's all of my law firm expertise, maybe over, uh, decades or centuries. To be able to say, "here's all of my expertise that you can have at your beck

and call: at 3: 15:32

00 AM you could go into my portal and be able to get these insights. Of course that can be a service the lawyer provides, but the lawyer's not gonna provide the secret sauce. That is the corpus that this is "ragging" over because that corpus of RAG is going to be the most important aspect that maybe is the only moat that a law firm has— that won't allow the clients, that is the corporations that hire the law firms. The clients can do a lot of things themselves, but that RAG corpus from the law firm is maybe the last bastion of "this is my value" that I'm providing as a law firm. So I think you're right. And the real question is what will law firms do with your three categories, which I love. Number one, remove lawyers altogether. You know, things like NDAs. And, you know, approve, disprove. And then number two, speed. It does things faster. And number three, do things that weren't possible before. I really like those. So as you think about the RAG corpus from the law firm, how far up the value chain does that corpus provide? It'll certainly do the, you know, NDA approved disapprove. That, that's, that's table stakes, right? Make it faster. Maybe, right? But then that's going to eat into your billable hours if you're a billable-hour thing, right? But then, you know, it does things that weren't possible before. Well, now we have all of a sudden expansion of the total Pie gets bigger. The, the TAM — the TAM gets larger. And so, there's a real question as to one: how the law firms are going to use that corpus that is going to do those three categories. But more importantly, number two, how do you incentivize lawyers to contribute to that corpus. And we've talked about this a bit in the past. But if that is the crown jewel of your law firm, that is the knowledge that you've accumulated over decades or centuries, how do you have that lawyer that was thinking about jumping ship to your competitor, how are you gonna get that lawyer to contribute to that firm that he's thinking about leaving? Because what's stuck in that lawyer's brain, is also that lawyer's value. That lawyer's moat. And so am I going to give my lawyer moat to you as a law firm, even though I'm gonna be jumping ship and going to your competitor next week? So that stuff in your brain has monetary value. So is the law firm going to pay the lawyer to be able to contribute to its corpus? A subscription for a lawyer individually. That's hilarious. Maybe like if, if you want my moat as a lawyer, you're gonna have to pay for it, law firm. Because otherwise I'm gonna jump ship and create my own boutique, which I'm gonna keep my brain, thank you very much. I'm gonna put it in my own corpus. Thank you very much. And then I'm gonna be able to say to my clients, "Hey, I'm the same lawyer that I was 20 minutes ago before I jumped ship from this law firm, but I've got the stuff in my brain, they don't. You can hire me. I'm AI first, and I've got this stuff in my brain that I didn't wanna give to my law firm, hire me instead. You pay less money and I still give you my brain." This is maybe the path that is gonna have to say, "Hey, law firms, you need to wake up and say that your corpus is your knowledge base. And how do you incentivize the lawyers to contribute to that knowledge base?" Well, exactly, like, I mean, the, the, the whole, the whole wave of workflow tools is allowing us to capture the sort of process information that historically has been really hard to capture through documents and other sorts of data formats. And I love this idea of like, beyond a salary or wage, you can license out your workflows to a law firm, so you get paid for so long as a law firm is using your proprietary workflows. I don't know how that's gonna work in, in practice, but I love the idea of it. Extracting the stuff from the lawyer's brain in the past has been quite expensive. Uh, because you have to hire a knowledge person, an innovation person to be able to say, "let's extract this data." And then you have to be able to take the interview, and be able to say, okay, let's put this into some expert system. Some way to be able to say an "IF / THEN statement to be able to make it this." So that's the old world, but you can imagine law firms, building out a system like, I'm about to show you. Maybe we could have an interview with a bot, to be able to have the knowledge management bot. To be able to say, you are an expert knowledge management person who is doing an interview of me, a lawyer. Ask me questions that will be able to extract the stuff from my brain to put it into a knowledge-based corpus. And then transcribe all the things. And then take that transcription and put it into a tool to say: "now extract all of the SALI tags or the FOLIO tags from this corpus, and to put it into IF / THEN statements." Old world, it was quite expensive to pull all that data. But in the new world, maybe it's quite easy. It's just a conversation that your bots can have with dozens or hundreds of lawyers. I think, uh, Dr. Megan Ma, who we absolutely adore, um. She's great. She's awesome. I think she's been doing that with Dechert. And they're presenting on that at SKILLS in a couple of weeks, um, in January, which is gonna be fascinating to watch. This is where, you know, she's her worked of decade to capture human expertise and create these, I think she calls, she calls 'em simulations. Um, yep. And, and it's this fascinating idea of like, well, if you can create a facsimile of your top experts, and they can independently assess what is essentially like artificial judgment, it frees up the human to do a lot more. It, it, it's, it's a force multiplier. So it's a, it's a fascinating model, uh, and again, changes how law firms are gonna be structured. That's a hundred percent right. I've thought a lot about what value a lawyer provides, to the clients. Jordan Furlong, who I respect and like a lot, he's said that there are three things that, as the AIs do a lot of the document production. That is, we think we sell widgets as in documents, but AIs, like I'm building and your building, are gonna create a lot of those documents. So what's left for the lawyer is, number one: trust. That the client trusts you, and is able to say that, your problem is now my problem. Number two is intuition. You're not gonna have ChatGPT speaking into your ear in front of a judge or in front of a jury. Nor are you gonna have that, in the room as you're negotiating an M&A deal, right? So you need intuition. Number three is you need integrity. You need the judge to say you're not gonna screw him or her over, and you need the other side to know that you're not gonna screw him or her over. So I would say that those three things, trust, intuition and integrity are the human qualities that are really necessary, going forward, post AI. And I would say that I actually have a personal, since the last time we talked, my neighbor wanted to build, wants to build, a fence right up against Yes, I remember the story. So, it's really dumb for all sorts of reasons. And, and at this point, we almost have it settled. So, by the time this actually gets to air, it might be settled. But I've brought in a friend of mine who I've known for 25 years, who is a litigator. We worked together at Robins Kaplan. And I was thinking about the value that my friend Jake provided to me — and is still providing to me. And uh, you know, I have Vincent, so I have the cases, I have the statutes, and I have the regulations. And I can run all sorts of things through Vincent to ask the legal questions about the Minnesota law, about how close you can go to the fence and the building fire codes that might be violated by such things. And so I have all the law, you know, unlike most clients who don't have the law, right? But I actually have that. So what value is Jake providing to me? And he's providing me that trust. And he's also providing me the intuition of saying, "Hey, have you thought about this tack? Have you thought about finding this public document that might be in the records in the Ramsay County Courthouse: this is a way that you could be able to prove these facts that you got through Vincent." Hmm. I said, that is a great idea. And so I went to the Ramsey County courthouse, went through a bunch of microfiche to be able to find a bunch of facts that are actually gonna help win my case. And then I went to the Ramsey County Records Department to be able to find a bunch of facts that are ultimately going to win the case. I didn't think of that, but Jake is really smart and helped me think of

that: 22:50

his intuition was able to help me ideate to be able to do things better. So really, I'm now in a better position with my fence dispute because of Jake. Not because he had the cases, the statutes, and the regulations. But because he had the intuition. And because I trust Jake. So I think that that's maybe a metaphor in a microcosm of the macrocosm to say that people are still gonna wanna hire lawyers like Jake. Because they trust them. And they have intuition. And Jake has integrity. What do you think of... I know I come from the litigation side — does that resonate on the transactional side too? it a hundred percent resonates on the transaction side. And, and like I would use medicine as a sort of proxy here because I know nothing about medicine and, and I, I was traveling for a few weeks, which I haven't told you much about. And yes or no. talk about that. Yes. Well, no, uh, After. I'm not sure. I'm not sure. That's su super fascinating. Like some people take holidays, others have holidays thrust upon them. I, I'm in the latter category. And, and so during the holiday I had to look up something about, uh, one of our toddlers who, uh, was vomiting. And I, I looked it up and, you know, I, I, I have no medical, uh, background or training. And so, uh, GPT was like, well, you know, if they vomited twice, you really should take them to, to the hospital, uh, and, and get 'em checked out for a concussion. And my, my gut sense was like, I'm pretty sure he's okay, but I told my wife this and she immediately 2am in the morning drove the kid to, to the hospital, to the emergency room.

Um, and she stayed there till like 4: 24:13

00 AM in the morning because emergency rooms, all of it world are overworked. And they came back, it's like, yeah, they said it's fine. This is that sense of intuition, right? Like if, if you had a doctor that you could just call and say, "Hey, what do you think?" Describe the circumstances. They would just tell you, sleep it off, sleep it off, see what it is in the morning, you'll be okay. But because we didn't, and the machine doesn't really have taste, doesn't really have that sense of like, okay, let's assess the judgment and all the circumstances, and give a reasonable opinion. They err side of overly cautious probably because it's part of a system prompt. I get why the machines have to do that, but also it's not helpful to the recipient of the service. So using that as a proxy, I can totally appreciate why your friend Jake is so valuable because it just cuts through these standardized, rote answers that the machines are gonna give you. I really like that example. That's a great example of a false positive, where the AI said, this is maybe something you should be concerned about. And then the doctor saying, no, you shouldn't be worried about it at all. And then there's also a lot of other examples of *true* positives. Where the people have said that the AI found this thing that, turns out, was cancer, but my general practitioner had said, oh, don't worry about it. Uh, just sleep it off. Right? But then I went in and the doctors would say, gosh, if you'd waited another day, you would've been dead, right? So anyway, there are these two like,"oh, don't worry about it": option one. Or option two, "my God, the AI found the thing that is, uh, really, uh, it is so amazing." So really, that option one general practitioner that said, sleep it off, it'll be fine. And then the person dies. That option one physician just had intuition. That is they said, you know, in my years of experience, I've never experienced this thing before, therefore it's probably fine. Yep. that medical practitioner was really going through that physician's rifle shot of experience, rather than the entire corpus of all the data. Mm mm. So something I've been wrestling with a lot is to what extent is "intuition" and "judgment" — which is something that says, "oh, that's what we're gonna need going forward, is intuition and judgment." To what extent is intuition, judgment, and taste just*data* — that isn't *yet* in an artifact. That's a big question. right. That doctor number one that got it wrong and said, sleep it off, and the person died, that was because that lawyer lacked the data of the thing that it actually was. So there's a real question as to"what is the value of intuition?" And is "intuition" and my "personal judgment" merely just my analysis within my "brain large language model" of the corpus of data that I have. But it's missing the corpus of data that I *don't* yet have — that isn't in an artifact that I can access. You know what, like, let, let's, let's kind of like extrapolate that and see if it's even possible inside of legal. Like I think for medicine it's a different, it's a very different fish, kettle of fish. Um, like. Medical data, broadly speaking, unless it's prohibited by HIPAA and other sorts of like privacy rules is publicly available. You can train models on that. In legal, confidentiality is built in by design. Unless a document, unless a case is designed to be public, everything's confidential. And so like, is it even possible for us to give enough data to train a machine to be able to have that intuition? I would say that there's two clarifying points to what you just said, which are mostly true. But the ways that it's not true is that for physicians, journals put things behind a paywall. Mm. Pub Med — and things like that. Mm-hmm. So that data is not publicly available.

And so there's a real question: 27:41

societally, do we want that data to continue to be behind a paywall — and prevent the scientific advances that AI can generate? The AI can't be able to access the things that it has to pay for. So the question is, do we as a society wanna say, "well, if you want that federal grant, it needs to be open." Or if you want to do any science, that science should be open. So that's point number one with physicians. And there's a related point number two, with law. As you said, yes, client confidentiality is baked into the process. But there's a lot of parts of the law that should be open, like cases, statutes, and regulations. And we've talked about in the past, but to download PACER, which is all the cases in the federal, it'd be about $2 billion with a B. Yep. That's $2 billion that you would have to pay to the federal government to download all the motions, briefs, and pleadings. We've of course, downloaded a bunch of that. So we have to recoup our costs. And so we have to then charge our clients that number. But who can't pay me is the poor person. That wants access to justice. And so this is a, a true societal — not just a legal — but an ethical question. Do we want to give rich people a different justice than poor people? And do we want the court system to accommodate that injustice? Because I tell the courts that I speak to, I've said this is an access to justice problem, judge, that you've created. You can't point the finger to any anybody else: it's you that are charging me $2 billion. And making sure that that poor person doesn't get the same justice as rich people.

So I, I guess two points: 29:00

physicians, some of that stuff is behind a paywall.

And then law: 29:04

yes, a lot of it's behind a confidentiality wall and a lot of the law that should be public that oversees all of us... it's not free....is not free, but should be. And so really, I think this is maybe one of the most important societal questions— as we go into our AGI future — is to what extent should the rules of the road, not just in the United States, but in the United Kingdom and the European Union and all 100+ countries worldwide. To what extent do we keep those rules of the road behind a paywall? Or do we open them up — for the AIs and systems that run those AIs — to be able to say, "Hey, poor person, these are the rules of the road for your country, for your state, for your city, and for your neighborhood. And the AIs will then give you that legal information in a way that is not behind a paywall." I think that's really the moral question of our time. It's a hard question to answer, right? Like I think there's a lot of self-interest, not just with the corporations and lawyers involved, but also with the governments, um, that. Not so much prevent, but makes it difficult for them to want to give this away for free. Um, I don't know. To that point, the listeners might or might not know that the Administrative Office of US Courts, the AO for the federal courts, have appointed being part of a, I think we have 12 people that are saying, how can PACER suck less? Um, and so we, uh, we meet periodically. We just met this past week. And they went through a bunch of things like, uh, should we do this dropdown easier? Or should we do this part of the venue? And I said, when we met in person in DC I said, "don't do any of that all, just do one thing, make it free. Because if you make it free, I can figure out what dropdowns I want, I can figure out what data I want to extract from it. Just make it free." And they responded, saying "that is out of scope for our discussions — because that goes to 'court funding.'" Yeah, Because the courts are funding the courts through PACER. This is essentially saying the quiet part out loud right. How do you exist? The courts should not be funding it through having the litigants pay for the courts. Instead, the Congress should be funding the courts properly. It's an arm of government. It is an arm of government! Do we want justice for all , you know, equal justice under the law? Do we truly want that? Then we as Congress have to pay for it and stop putting a burden on the backs of the poor people who can't afford PACER. I say to anyone with ears, the amount to run PACER, I think annually, is something, you know, the congressional budget office said something like$20 million or something like that. Elon Musk can be able to reach into his couch cushions and be able to pull out $20 million. Right? I think we need to look long and hard. What if we were to make PACER free, and how much would that help OpenAI, Anthropic, Google Gemini, and X.ai, right? They could be able to have all of that "human created high quality" data every single day for the low, low price of $20 million. And for a company that's valued in the hundreds of billions of dollars, that seems like something we should be able to lobby and get behind. Okay. Speaking on that for for one second, let's just say Anthropic does give$20 million and make it free, but why would they wanna make that free for everybody and not just themselves? So I think there's a lot of self-interest that prevents a company from wanting to do this, especially if they're serving their shareholders to increase their value. I do think, however, that $20 million is a really small amount of money in the context of all the funding that's been going on, and, and I am astonished that this has not already been made free. I think it's just because it hasn't been raised as a, even an idea, Hmm. The lobbyists haven't gone in. And so to, to your point about, you know, why would Anthropic donate $20 million? It wouldn't be Anthropic's interest. But what if there were a coalition of Anthropic and Gemini and OpenAI and xAI. Much like they're doing with MCP and the Mm mm. the kind of protocols. Here's how we're gonna be doing AI collectively. And that's being housed, I think under the Linux Foundation, if I'm remembering correctly. But what if Anthropic and OpenAI and xAI and Google were to donate to the foundation to be able to say, Hey, foundation will pay the 20 million, collectively. So maybe, you know, everybody chips in $4 million, right? And then the Congress, doesn't have to fund it. Everybody gets the data for free. And here we are. Bob's your uncle, as my friends Hmm. in the UK would say. But the problem is do we want private companies to fund public government services? And the answer is philosophically, it seems gross to me. We shouldn't have to have the Anthropics and the Googles and the OpenAIs funding, our government, like government should be funded because then you have maybe perverse incentives to be able to please the Anthropics and the Googles and the OpenAIs, rather than please the public, who are the true stakeholders. We know the goal, make the law free. But who's gonna pay for that? Private or public? You know, it turns out we have private people paying for the East wing of the White House. And that seems to be okay. So Which may never be built by the way. right, right. So that is, you know, to that end, uh, maybe to the end of making the law free, maybe I can hold my nose and say it's fine. Yeah. Yeah. Look, it's a, it's a bigger question that we can answer on this podcast today, but I think there's a trend at the moment, which I, I don't think favors what we want as an outcome. And, and that is, even if you look at what's happening in the software world, open source is changing. It's dying in a way. So pretty much everyone, everyone in the sort of SaaS space uses a, um, uh, a library called Nginx. Uh, NGINX is essentially like a router. You, you, you, you, you'd kind of log into this as your first port of call and then it routes you to the right place in the software, um, to do its thing. It has been open source ever since the beginning. And, as of March next year, it is no longer gonna be open source. And it's one of several very critical libraries, um, that are, that are gonna be going off open source. Like it'll still be there, but it's no longer gonna be maintained for free. This sort of trend away from open source in software was unexpected for me, and I think it's kind of a, a, an an indication of a larger trend that's happening in society right now, which means I don't know if we are gonna get free law anytime soon, Damien. I, I hope that you're wrong. I worry that you're right. To the point of open source is dying a bit, let me give you a counter example. please. I'm sad about Nginx, right? But uh, we've talked about Mike Bommarito in the past, and Yes. Uh, for our YouTube viewers, we are building out FOLIO and so we're doing a lot of these things Mm-hmm. free and open source. So here you can see GitHub repos, you can see XML files, you can see Python libraries. You could be able to see, here's motion to Dismiss, expressed as HTML, as RDF, as Markdown, as JSON. And so you can see all of these beautiful things are free and open source. That is a nice way to be able to see a motion to dismiss is a demur and also an MTD in a time entry and a motion for dismissal in some of these jurisdictions. Here's a definition. Here are the parents, here are the children, and then here's all the translations into various languages. And here's a hierarchy of a motion to dismiss I love this. This is so pretty. I. Isn't that pretty? So Mike Bommarito vibe coded this on a weekend. And then it's all free and open source. This does a lot of things that SALI didn't do. And a lot of things that I complained about. But then you also have this as XML. Everything I just showed you is here in XML. Everything I just showed you is here in Markdown for us to use in a retrieval augmented generation or in your AI. And here it is in JSON. So all 18,000 things that FOLIO has is here, free and open source. It's all in GitHub. I think that there are people fighting against what you've described, Bommarito and I are doing this. And along those lines, there's a new thing, as of three weeks ago called Open Gloss. This is an encyclopedia of 150,000 concepts. In contrast, the Encyclopedia Britannica has 40,000. This has 150,000, so almost four times as many encyclopedia entries as Encyclopedia Britannica. And so as each of these entries, it has a dictionary entry. For example, a murder is the unlawful killing of a human being by another person, typically with malice aforethought. So that is the definition. And then it gives a full encyclopedia entry, and it also gives the different senses because a murder of a person is different than a murder of crows, How is this different from Wikipedia? This is all bottom up created with a large language model. Um, and Oh. If we go through here, it's a movie director murdered the scene, and there are synonyms, like a synonym of murder is homicide, killing, and slaying. A child of murder is first degree murder, premeditated murder. A hypernym., a parent, is a murder is a type of crime. Mm-hmm. And then we also have a knowledge graph, 9.1 million semantic edges between the nodes connecting those Wait. have etymologies. This is all created by language models. Yes. So if you're listening to this, go to our YouTube channel and watch, because I'm now going to show you murder, the ontological basis of murder. There are five senses. You can say murder as a verb. I murdered someone and murder as a noun. I committed murder. Here's the long form encyclopedia entry for the thing that is murder. We also have the unlawful killing. A synonym of that is a homicide killing slaying. It is a type of crime, a type of felony. A type of murder is first degree murder, second degree. And here are examples in a sentence to disambiguate this murder from the second type of murder, which is a group of crows, which is a collective noun. And here are examples. And then the third type of murder is to present something in a matter that ruins it. For example, someone ruined, destroyed, or mangled. The director murdered the scene. The translation murdered the prose. So this is a way, if you're using a, a large language model or any tool to extract the thing murder, you're gonna want to disambiguate the director murdered the scene saying that's the wrong sense, but instead, and I don't wanna have the, you know, murder of crows. But I do want to have the murder of a human being. And so all of this, the way that you had asked how this is different than the Wikipedia is Mike Bommarito, the mad scientist genius he is, was gonna build this for his kids. His kids said, what does this word mean? And he created this. And then he said, okay, murder. I'm gonna highlight that. And a synonym is homicide. So after you've created the bunch of pages for murder, go on to homicide. Do the same thing for that. Yep. Then go on to manslaughter and then go on to premeditated killing, and then go all the way through. And so what he did is created 150,000 concepts that are all now free and open source on GitHub I love this. source. And free and open source on HuggingFace. You can go to HuggingFace and get the 1.2 gigabyte file. All of these things, all free and open source. And so what we're doing next is to be able to, to take this Open Gloss dictionary, these 150,000 things. And then we're talking about connecting them back up to FOLIO and SALI, because this is a superset of FOLIO and SALI. So we, we started this call by talking about like some of the impressive building that's happening in the legal tech space. This is impressive building by someone in legal tech that's not directly re relevant to legal tech. But a few questions pop up in my mind. First is, um, super duper cool who's verifying this, who, who's gone and verify that 150,000 entries are correct. The beauty is one can crowdsource this, and so if I like this definition, I can give a thumbs up. Here. If I don't like it, I can give a thumbs down and I can say why it's a thumbs down. I could be able to maybe enter saying You're missing "heinous" as a hypername of this. So the answer is nobody is currently doing it, but you can imagine having an open source way to be able to have people improve it. And so that's what we're talking about doing maybe with FOLIO, is to be able to say, here are all the 18,000 FOLIO tags as they are right now. We'll map those 18,000 FOLIO tags to the 150,000 open gloss concepts. And because FOLIO is probably a subset. Mm-hmm. and there's probably a one-to-one mapping on every single one of It should be. Yep. then what we're thinking about doing is taking all of those 150,000 and filling in the gaps of FOLIO. To be able to say, here are additional things that FOLIO and SALI have missed, but now are able to be counted. Maybe we'll expand the 18,000 to 20,000, to 50,000, to 100,000. And maybe that'll be fine because the humans aren't gonna be the ones tagging it, It built for machines. With, with all of the proper senses that a murder of crows is different than a murder of a person. So I would say that yes, some open source things are going away, but mad scientists like Mike Bommarito are actually doing some open source-y things that are actually doing better for the world. To tie a bow, along with the first part of our conversation, there's also the Vatican discussions. I said to Bommarito, Hey, how about taking the Bible, and being able to extract all of the FOLIO and OpenGloss, items, concepts from the Bible. Do the same thing with the Torah. Do the same thing with the Islamic texts. And he's, said, oh yeah, I, I would love to do that. So anyway, these are quite exciting times for open source do gooders that are looking to do good. And I think that, we do gooders need to outshine the people who are looking to do "not good." I, uh, it's, it's, it's uphill. It's an uphill battle, but I think it's worth it.'Cause legal is such a small, little vertical in the grand scheme of human endeavors and experiences. Um, it is all that we do every single day, but it is such narrow field. This OpenGloss project just makes me think like, is this the next iteration of Wikipedia? Like Wikipedia came along because Encyclopedia Britannica and, and, and, Encarta, I think it was with, uh, with Microsoft, um, just was not accessible enough to the general public. So Wikipedia came along and now Wikipedia is sort of morphing and changing for a variety of reasons that we're not gonna cover, but OpenGloss is here. And, and if you have a language model that's responsible for creating the entries in the first place, you remove some of the human biases and you replace it with a machine bias, but it's perhaps somewhat more predictable and less subjective. And that's right. It's, you can almost think of it as a consensus. Almost like the poly market. The wisdom of the crowds is essentially reflected in the large language model. So if the crowds are biased, of course it's gonna be biased, but maybe less biased than any one individual person. Uh, you know, that one individual Wikipedia editor that has, you know, an ax to grind, for example. So I think what you've said is right, and importantly, under the US copyright law, if a machine creates it, it is uncopyrightable. So For now, for now. well, right, maybe, but, but, but really, so even if Mike Bommarito wanted to copyright it, maybe it's not even copyrightable in the first place. And so really this, this kind of locking behind a paywall to be able to extract value from rent from the thing that we want to copyright. Maybe those walls are coming down because it's cheaper than ever to recreate a Wikipedia like Mike Bommarito has done. And so that kind of takes us to a separate but related topic that I've been thinking about a lot is that, and I think what you and I have talked about in the past, to think about software as a service is a one to many. That I've spent a lot of time building software as a service to be able to provide it to many people. Maybe we're gonna go from software as a service, one to many to software for one. That is, I vibe code my thing. And that's what the lawyers that we talked about at the yes. show, they, they're essentially building software for one. So to what extent do they even need any software as a service for one to many when they're just doing one-to-one. And essentially open gloss was Mike Bommarito's software for one. He recreated Wikipedia for him. Yep. And what's to keep somebody else from doing that same thing. And so I, I wonder if we're moving, as a paradigm shift of society from one small group of people needs to be able to build this for everyone to, "everybody can just do it this themselves." I, I have a, I have a philosophical issue with that, which is waste; wastage and chaos. Um. That's a scarcity mindset. Not an abundance mindset. What if, what if it doesn't cost anything? Then what are you wasting? That's Jevon's paradox, right? We thought that LED light bulbs were gonna save us a lot of energy, but they're so cheap we'd leave them on all the time. But that waste of leaving them on all the time doesn't really cost much. So the waste of Mike Bommarito — and guess how much it cost him to build OpenGloss in GPU costs over the weekend. The fact you said over a weekend kind of puts a time-bound, time limit on this. A thousand dollars. It cost him a thousand dollars in GPUs and electricity to be able to build this thing. So for a thousand dollars he recreated Wikipedia, right? And made it even better because now it's free and open source so you don't have to pay the Wikipedia toll. So was that a waste of a thousand dollars? We did. It would be, it would be if everybody had to do it for themselves. Right. And that's, that's my concern here where if everybody can vibe code their own software solutions and they do, that's a lot of repetition. And then what, what happens when, say 10 people in the same organization vibe code the same solution? Well. Which one do you use? Which one's the best? How do you maintain it? That sort of chaos is what I'm concerned about. I agree with that. And then I, I've, I've thought about that a lot with my day job and Mm, warned against having forking and being able to say, oh, this is my bespoke way right, right. Which is really the same as the person down the hall, but just with a different gloss on it. But one can imagine maybe fixing that to be able to say, let's cluster semantically similar things together. These things live in similar vector space. Therefore, if you want Jane's version of this, that's different than Joe's version of this, but they're all pretty similar, right? So maybe that "waste" can be accommodated with clustering. To be able to say, you know, pick your flavor: Jane's or Joe's. It's wasteful, but maybe helpful to Jane and Joe. And as the cost of intelligence goes down— in 2023, a million tokens would cost $50. And now that costs 40 cents. So it goes from $50 per million tokens to $0.40 per million tokens. Then how much waste, like as that continues come down, maybe being "wasteful" matters less. The possibilities are enormous. Um, and it goes back to if you can do all of this now with language models and, and the only thing stopping law firms and stopping people is creativity and the, the, the realization that it's possible. Where's the limit? What can we do? How does it change everything in 2026. A hundred percent. I also wanna be able to maybe think about what you talked about earlier, saying that law is such a narrow field rather than societal in general. I would maybe challenge that. Mm-hmm. My friend Bridget McCormack, who is one of the smartest people around. And I love hearing her podcast, which you should listen to. Yes. But she said law is the societal operating system. It is the operating system to decide how things get done in society. So I would challenge us as lawyers, not to say that we are in a narrow domain — law — but to say that we are building society's operating system to say how contracts are enforced. To say how laws are enforced, from the litigation side or the criminal side. So we can define how our societal OS is built. And we should maybe be able to say, what are the underpinnings of that societal OS. Maybe the law that feeds the OS should be more like Linux — free — and not more like Windows, behind a paywall. I. I like that. And it's, uh, it, it's somehow become the theme of this episode now to talk about free versus paid. And I think what we are gonna be, what we're gonna see in 2026 is legal services. A lot of legal services will become free. And, and I think what we are gonna pay for is gonna change its nature a lot. Um, so we're coming up to top of an hour, Damien, and, and I'm gonna ask you the question you always ask, which is, what are you optimistic about? And then I'll return it to you. Please. I'm optimistic, having worked with a bunch of do-gooders at the Vatican and I'm, I'm currently jamming with them on this CatholicOS we're calling it, right, this open source thing to be able to figure out what is the ethical and moral thing to do. And I'm optimistic about us being able to create in Symbolic AI, good old fashioned AI, the ethical things that we've been talking about since antiquity, since Aristotle and Plato and Socrates have been bouncing around these ideas. We can maybe operationalize those ideas in code and be able to then create a decision tree. This business decision I need to make, does it comply with EU law and UK law and US law? Of course, it can do those things, but then is it ethical? Is it moral? Is laying off all these people, even if this complies with all the laws, is laying them off just for laying them off's sake. Is that something we should be doing? I'm optimistic about us being able to put into a hard code, good old fashioned AI, something that Aristotle, Plato and Socrates have been talking about for millennia. What are you excited about? I'm excited about a lot of things, but what I'm optimistic about. Optimistic about, yeah. What I'm optimistic about is people are moving on from talking about buzzwords. In the last couple of weeks, I, I have not heard the words "agentic AI" or "agents" for a few weeks. Um, I, I'm, I'm hearing less and less about AGI. Um, and more and more I'm hearing people talk about the practicalities of, uh, change and, and, and how they themselves have to confront the realities of their work and their lives to approach change. So, so the sort of waning of the hype is something I'm very optimistic. There's been a three letter acronym, LFG, , let's F-ing go. Uh, and I think that, I think 2026 is gonna be the year of LFG. I think we're, we're gonna be buzz wording lesson and building more, Oh, it's so good. Man, I, I love talking to you so much. And I'm so glad that you've had a restful ish, uh, vacation and I'm glad that we recorded this. And thanks to all of our listeners. I, I, I hope you enjoyed it as much as Horace and I did. Likewise, and hey, uh, Damien, we haven't said the most important thing on the 19th of December, which is the day we're recording this: Merry Christmas for next week, and have a happy holidays. A happy holidays to everyone. I hope you have a restful end of the year and let's rejuvenate to LFG in '26. I'm looking forward to it. Bye everyone.