Loading summary
Transcript137 lines
- [00:08]
Marlene Gabauer
Welcome to the Geek in Review, the podcast focused on innovative and creative ideas in the legal industry. I'm Marlene Gabauer.
- [00:14]
Greg Lambert
And I'm Greg Lambert.
- [00:15]
Marlene Gabauer
This week we're joined by an old friend of the podcast, Ed Walters, Chief strategy Officer at Velix. Ed, welcome back to the Geek in Review.
- [00:22]
Ed Walters
Who are you calling old?
- [00:24]
Marlene Gabauer
Yeah, old friend. Not old.
- [00:26]
Ed Walters
Let's say a friend of long tenure.
- [00:28]
Marlene Gabauer
Okay. Because that sounds better.
- [00:31]
Greg Lambert
That is so true. I, I knew Ed before there was a fast case. 50.
- [00:36]
Marlene Gabauer
That's right.
- [00:37]
Ed Walters
That's true.
- [00:37]
Marlene Gabauer
That's right. Don't date.
- [00:38]
Greg Lambert
You're dating yourself to have you back on the show. So before we jump into the big topic this week, which is all of the news, this, this dropped with the Velex and your new releases. Actually, I don't think you knew I was going to do this, but I'm going to do it anyway. Wanted to have a little banter. So one of the things that we've been seeing a lot from the AI tools, the commercial ones, OpenAI's Geminis, Anthropics and whatnot, are the launch of the reasoning models over the past few weeks, which I've thoroughly enjoyed. I've really liked diving in with, with the deep research and asking it questions and kind of watching it think and, and humanizing the whole models by, you know, by saying that it's thinking. But I really wanted to kind of pick your brain on this and that is, you know, what kind of uses are you getting out of it? And are these models eking their way into some of the legal research tools as well? What's your thoughts?
- [01:42]
Ed Walters
Yeah, great question. I agree. I think they're totally fun and really the work product that comes out of them is fascinating. I also like to watch the progress and the kind of humanizing, even though I know it's phony. Right. It's like, you know, when, when Deep Seek inside of Perplexity says, hmm, I guess I'd better pull the European Union regulations for privacy. Now I'm reading those. That's interesting.
- [02:06]
Greg Lambert
You can see it's scratching its beard, right?
- [02:09]
Ed Walters
Like, I know it's phony, but I like it anyway. We're absolutely going to see these in the legal AI tools for sure. I mean, I'll just tell you, we're with them extensively inside of Velex. I don't think it's a big trade secret. I mean, we're benchmarking them against each other for different tasks. I was doing it this morning, as a matter of fact, working with our Tech team. And so I think there's a lot of promise for it. This reminds me a little bit of that jump between GPT 3.5 and GPT 4 where there was a pretty big jump in the capabilities and a lot of people were kind of innovating around that. So I, you know, it would surprise me a lot if we don't see that coming up all over the place in legal AI.
- [02:54]
Greg Lambert
So I, the thing I don't want to see is the legal research tool saying do you want me to email you after I've thought about this to let you know that then I'm going to conduct my research and then email you later? So that's right.
- [03:10]
Marlene Gabauer
Well, I was, I mean I was like, like I was thinking about like what being very practical, like what's the potential impact on knowledge work? You know, I feel like a lot of times legal research, you know, is, is not billable often and this may have a big impact in that space in terms of sort of reducing non billable work and allowing researchers and analysts to sort of move more quickly regarding these types of tasks and maybe even get some more insights.
- [03:38]
Ed Walters
I think that's right. And let me just say like, I think the other thing that's interesting about it, way beyond research, right, There's a lot of tasks that are kind of multi step that we've not really been able to trust agentic AI or whatever you want to call it to complete. And there hasn't been transparency, we haven't been able to see the steps. But I do think that there's interesting ways of saying look, here's the beginning of a corporate transaction. I want to know everything about the opposing party. I want to know everything about the opposing law firm and lawyer. I want to look at all of the underlying documents that have led up to this. I want to do some discovery about the context of the transaction that would be great for a lawyer, a team of lawyers to start with. And I think we're getting to the point where these agentic tools can break it down into multiple steps, maybe use different tools or different databases. I mean we're looking at this inside of ELEX saying you might want to look at statutes in one area, but then dockets in another area. Maybe you want the patent portfolio and you know, here's a slice of expert witnesses that we're going to want to do some work on at the beginning of a litigation matter maybe and the ability to do these multi step transactions and sort of bake that into a get me ready for litigation Tool get us ready for this corporate transaction. You know, we've just closed the asset purchase agreement. What are the 10 steps that our law firm does when we close a file like that?
- [05:04]
Marlene Gabauer
Just sort of capturing that.
- [05:05]
Greg Lambert
What you just said there made it sound like legal research is really hard.
- [05:10]
Ed Walters
It is hard. But I'll just say like I, you know, I always think people, because of the heritage of fastcase and Velux, people think of our tools and vincentai as like kind of a legal research only tool. And we're great at it, but it's like three or four of 23 different workflows that we're doing right now. And so I just want to make sure. I mean, yes, I think it's going to be amazing for legal research, but the applications are way beyond legal research too.
- [05:37]
Greg Lambert
Until we get AGI and then it just takes over all of our workforce. That's it.
- [05:41]
Ed Walters
Yeah, it's not emailing you, it's emailing the client.
- [05:44]
Marlene Gabauer
It's like, I'll, I'll take this, I'll take this one.
- [05:46]
Greg Lambert
Yeah, you got this. I'll, I'll bill the client separately. Bob Ambrogi had their reported on the AI smackdown that was hosted by the Southern California association of Law Libraries, or SCAL as we like to call them, which was held recently. And Vincent AI was reported to have great answer to the question that was posed at the smackdown. So the test was between Lexis plus AI, which I think is Protege now Westlaw Precision AI, which they branded. I think the whole concept is co counsel now is the package. So branding, I guess is important and Velux's Vincent. And is there, is there a different way to brand that or is it still Vincent AI?
- [06:34]
Ed Walters
We're keeping it super simple. It's just Vincent AI.
- [06:37]
Greg Lambert
Thank you. Thank you. So, but the, the panel of, of librarians still said that, you know, AI assisted research is, is a great starting point and to use, you know, and use of the traditional sources are still necessary. So why, why do you think that they still look at going back to the core research? How is Valex trying to change that?
- [07:02]
Ed Walters
Yeah, well, I think that they did an awesome job in this review. I mean, not just because we did really well, but I think it was.
- [07:08]
Greg Lambert
I love seeing it hurt.
- [07:10]
Ed Walters
It doesn't hurt, but I love seeing law librarians in these kind of benchmarking tests because they are so rigorous in the methodology. They really understand well how to gauge quality and to do that in a kind of a quantitative, objective way. And so I thought Skull did an amazing job of this. And I think that the tools were very powerful. I think, you know, all of the tools did really well. I think it was pretty clear from the, from the conclusions, not mine, but theirs, that Vincent was, had the most depth of the answers and got some uniquely right answers that the other tools didn't. And as proud as I am, I'll be the first to say we, our whole team always says these are first draft answers, not last draft answers. If you had a superstar associate in your firm, your corporate legal department, and gave them a research task and they came back with like a just rock solid answer, you still have to go read the cases and the statutes to make sure that they say what they're supposed to. You have to verify even the best human lawyer work. And I think the same thing is true about these kind of legal AI tools. You should trust and verify. And my hope in this is that, well, you've heard me say this before. I don't think that legal matters are a single event. They're a chain. Right. And we might be able to replace the most boring or problematic or difficult links in that chain. But I always think that the last links of that engagement have to be human judgment. And there's no AI tool that can or should replace that human judgment at the end. And so maybe we can take like the first two days of an engagement and throttle it, you know, and get like an 80% answer in the first 10 minutes, five minutes, I don't know. But, you know, very quickly get some of the early blocking and tackling work done so that lawyers can do more. Maybe they bill the same number of hours, but they just do better work in that time. Maybe they don't have to file the extension that they typically file because they're able to really crank on that work in the first couple of days. That's kind of my aspiration for Vincent AI. I think that's what we're really hoping.
- [09:12]
Marlene Gabauer
Yeah. It's noteworthy that Bob actually tried the test himself and he was using Deep Research and he said it reported very admirably without access to paywall legal research content. And I, you know, I've seen some gen solutions that don't have that type of behind the scenes access. And, you know, they perform okay, but not apparently as good as Deep Research did. And I'm curious, like, do you have any thoughts about how they managed to accomplish that?
- [09:44]
Ed Walters
Well, they're using public data sets like Justia. Right. And so that's great. I mean, I think the questions I would ask are, where's the citator. And how are they making sure that they're pulling from the majority opinion and not the dissenting opinion? The majority opinion, not a concurrence in a Supreme Court case where four justices are joining part two and three are joining part three, but parts one and five are the majority opinion. How are we making sure that that part is being treated as governing law? And so this is a super hard problem. We actually created the cert citator at Fastcase and now Velex to help address that issue. We worked with the Judicata team when they joined Fastcase to parse the entire judicial opinion library and break out the majority opinions from the concurrences and the dissents. The first 85% of it is pretty easy for the large language models to do, but the last 85% is essential for lawyers. And you really have to have custom built legal AI tools to do that. And not just any off the shelf. We're lawyers too tools. I mean, this is really hard. We spent, I don't know, four years, $5 million building the search citator. And we had, you know, the Judicata team who worked on that with us in house is like some of the brightest people who have ever thought about legal tech. So I just want to say I have huge respect for the deep research tools. They're amazing. I use them all the time for legal work. The last 15% matters. And that's the part that the Vincent AI Velix team sweats all the time.
- [11:20]
Marlene Gabauer
Yeah, that's a really good distinction to raise.
- [11:24]
Ed Walters
And is it current? Right. I mean, we're updating these things like, you know, every day. We don't really know when the last time those models are trained or what data sets are pulling from or when.
- [11:35]
Marlene Gabauer
Those public sites have been updated as well.
- [11:37]
Ed Walters
Right.
- [11:38]
Marlene Gabauer
So in addition to the skaal testing, val's just released their report on legal AI AI. And so this study is, is apparently the first systematic attempt to independently benchmark legal AI tools against a lawyer control group in this instance. And so they use, you know, I guess, several real world tasks and they got it from AmLaw 100 firms. And apparently there were some instances where the AI outperformed the lawyers. So I'm wondering, should we be worried?
- [12:10]
Ed Walters
Well, I mean, this was certainly true with Ediscovery and law firms made a ton of money from eDiscovery sending young lawyers to warehouses in upstate New York and New Jersey. Strangely specific example. Yes, I was one of them. And you know, at some point software got Better at that, better at accuracy, better at recall for discovery tasks than human lawyers. And there are more lawyers in America after ediscovery than there were before. They tend to be happier. Fewer paper cuts, fewer associates quit in the warehouse in that.
- [12:47]
Marlene Gabauer
Sound specific, too.
- [12:51]
Ed Walters
Strangely specific example, Ed. But. So I think we may be at that moment right now, like the big watershed event for ediscovery was a report by Morric Grossman that said that the tools were had higher accuracy and higher recall than human review, which has its own error rates. And I feel like this VALS report may be one of those watershed moments. Look, I deal with lawyers all the time and law students who say legal AI is great, but these tools just aren't there yet. And when I ask, like, what do you mean by that? They say, I just don't feel like they're quite good enough. And this is the first attempt really to quantify that. To be fair, the lawyers did great, too. These are very good lawyers. But in many cases, the AI tools as a whole did measurably better than lawyers.
- [13:43]
Marlene Gabauer
Are there areas that you would expect that they would do better?
- [13:46]
Ed Walters
Yeah, sure, absolutely. Take a giant mass of documents and create an accurate summary. Sure. I think it's clear that large language models will have an advantage there. You know, I think lawyers did really well in complex synthesis of Sec Edgar documents. That doesn't really surprise me. I was surprised, though, that the AI tools as a whole did as well as they did. I'm really psyched about it. I'll just say directly, the tasks that val's AI were studying are not super strong tasks for Vincent AI. You know, our strength is really this giant mass of structured legal documents for tasks like research or analyzing the legal implications of an asset purchase agreement by looking at Michigan law. You know, and these tasks were kind of extractive. We're going to give you a document and then pull all of the key terms out of the document. We do fine at that. Right. But I mean, our strength is really in more complex tasks. And Vincent did great. I mean, in four out of the five areas where we were competing, we were within the margin of error or measurably better than the lawyers doing that work. And again, this was blind grading. It wasn't. The people who were evaluating it weren't looking at the answer key. They couldn't see which came from which service, whether it was created by a lawyer or somebody else. The other takeaway, by the way, that I thought was really interesting was at the slowest, the AI tools were six times faster than lawyers at the fastest. They were 80 times, not 1-880-80-times faster than lawyers working on the same tasks and better and, you know, measurably higher quality answers. That's amazing.
- [15:30]
Greg Lambert
Yeah, it is. Yeah. Well, I mean, that's the thing with technology is that you use it where it makes you more efficient. If you can do the triad of better, cheaper, faster, then let's try to do that.
- [15:45]
Ed Walters
Absolutely, yes.
- [15:47]
Greg Lambert
So, Ed, one of, one of the features that you just released that I'm really excited about is the ability for Velux to do multimodal type interactions where it can analyze audio and video files, which is something that I've been excited about since, since OpenAI came out with vision. And you could take a, you know, take a picture of a guy standing on a ladder that's on top of a forklift and ask what OSHA violations there, there are here. So, you know, and I think that really has a really good use case in litigation, probably others as well. So you mind kind of giving us a little bit of the details of how it works, how you got it implemented, and what are some use cases that you think people could use it for?
- [16:38]
Ed Walters
Sure. And would you like to take a look while we discuss it?
- [16:41]
Greg Lambert
You want to jump in on that part, Marlene? Let's do it. Let's do it.
- [16:45]
Ed Walters
All right. So this was something that a lot of law firms and corporate legal departments were asking us for because they deal with an intake interview that they videotape, where they have the audio transcript of a deposition or maybe like a video of an oral argument, as we have in this example right here. How do we deal with that? Like you finished the oral argument and you want to report back to the client, what do you tell them? You can't give the oral argument and take notes at the same time. Being able to take the transcript of a deposition and say what parts of this deposition are inconsistent with the documentary evidence we have before. And you can't deal with that if you can't transcribe it. There are plenty of services that transcribe audio and video, but they're not secure. Right. And so Vincentai is SOC2 type 2 compliant, the most secure platform certifications you can get, because Marlene told me that we had to and we did. And so now you can deal with audio and video files in this very secure environment that meets all of your clients requirements as well. So this is it right here. In this example, I have uploaded kind of a long 30 minute MP4. This is a video file of an Ohio Supreme Court oral argument. It involved a judge who is reprimanded for speaking about politics in court. And so immediately when I upload it, Vincent creates a transcript. And so we can see right away the entire transcript of the oral argument. Now I can copy that, paste it another into a Word document or something from that, just like other places you've seen in Vincent. AI Vincent reads that transcript and says, what do you want to do with this? I mean, you can ask something open ended if you want. I think this is really a strength of Vincent. It says, like, here are the kinds of things that lawyers would do when you have this kind of document. The transcript of an oral argument, for example. But whatever it is, and so you see these very specific suggestions, create a concise summary of the defense arguments presented by Mr. Funk on behalf of this judge, focusing on the First Amendment issues, create a timeline, draft rebuttal points, and then at the bottom, these are legal questions. So if you want to do research and find out what the legal precedents are for First Amendment protections for judges in Ohio, you can select that. And of course, we'll pull that from the Velex collection. What are the standards for judicial discipline in Ohio? These aren't baked in. These are suggestions that are created by reading through the transcript of that oral argument. And so you can see when I create a timeline, you know, it puts it in a nice table. I can download that table in Excel. This is new, so in the last column it says source, and the source is from the transcript. And if you mouse over the site, it will show you what in the transcript created that element in the timeline. But if I click it, it will open the transcript and jumped right to the part of the transcript where that came from, allowing you to verify for yourself. That should be in the timeline or not. And let me just say one thing about this, so I'm pretty excited about this individual element. A highlight of Vincent in the past was the transparency of it. You know, being able to see what sources go into making an answer, have hyperlinks to them, open the statute, open the judicial opinion, open the regulation, confirm it for yourself. This allows you to do that same kind of transparency with documents that you upload. So if you upload, you know, a hundred documents and say, analyze all of the legal risks in these documents, it can create a table or a list and have the citations to those documents. These are all from the same one. But in that case you would have, like, the name of the document and the place where that risk was identified, and you can Mouse over it and see where it comes from, or click on it and jump right to the spot in that document. This creates that same kind of transparency that Vincent is known for with primary law, but that transparency extends to documents that users upload, which is really nice in this case, like a transcript or an oral argument or something.
- [20:57]
Greg Lambert
Now, I know one of the things I think right now what it's doing with. With the video is that it is extracting essentially the. The transcript and. And using the text.
- [21:11]
Ed Walters
Do you.
- [21:11]
Greg Lambert
Do you see in the very near future that it can go just directly off of the video itself so that it's not just essentially creating text out of the. Out of the voices, but can leverage some of the nuances that are in the actual images of the video itself? Because that's something I think we're not quite at yet, but hopefully getting to.
- [21:38]
Ed Walters
Yeah, I love that idea. And what I imagine is like maybe the top of the frame being the video and the bottom being the transcript, or the top being the audio and the bottom being the transcript. And when you jump to the point where the written transcript exists, you'll jump to the same point in time in the audio or video file so you can listen to it and hear it for yourself.
- [21:58]
Marlene Gabauer
Very cool.
- [21:59]
Greg Lambert
It's interesting times.
- [22:01]
Ed Walters
It's amazing.
- [22:02]
Marlene Gabauer
Very cool. Very cool.
- [22:04]
Ed Walters
So this is part of the winter 2025 release of Vincent AI. We're moving to a kind of a seasonal release schedule. And there's just so much in these. You know, when we launched autumn 24, there was like, a lot in that release. In the Winter 25 release, similarly, there's a lot going on here. People who subscribe to Vincent get all of this for the. Included at no additional cost. It's included in the price of the subscription. I mean, it's a pretty startling upgrade. And I'll just mention, like, all of the benchmark tests and everything that were done, the Scal and the Val's AI benchmark where before we launched winter 2025. So there's all kinds of new skills and abilities in Vincent now too.
- [22:49]
Marlene Gabauer
So another part of the winter 25 upgrade, I guess, is Velix is now integrating docket alarms, litigation data to produce litigation workflows, which is really great, I think. I mean, it's like we've been focused for so long on transactional workflows in big law and innovation there. So I think it's nice to see, as a former litigator, I think as an. Nice to see litigation get its due. But in using these workflows what are the ethical considerations lawyers need to be aware of? And also, I don't know if you can show those or not, but if you can, great. We'd love to take a look.
- [23:23]
Ed Walters
Yeah, happy to. So let me just say, first of all, thank you because I think that the docket alarm materials are so helpful. It's a gold mine, right? There's so much information in the complaints and briefs and pleadings and motions before judges. The docket alarm collection is more than 860 million docket sheets. Briefs, pleadings, motions, and each individual document's great. You can use it as a precedent. If you're about to file something, it solves the blank sheet of paper problem. But in the aggregate, it's so useful. The information in here about individual lawyers or judges or law firms or parties, if a party never settles, that will be something that you see in their docket filings. It's not going to be a judicial opinion or something. Right. There's not going to be a statute that you can conjure up that says Google tries its cases instead of settling them. But you can see that in the docket information. And so one of the cool things about it in this winter 25 release is that we have released a whole bunch of new AI workflows that leverage this docket information. I think this is unique. I don't know if anyone else who has legal AI tools that are building on top of this kind of giant volume of structured data. And the other thing that's cool about it is you can ask like open ended questions like, show me all of the data protection cases that deal with children. You know, there's not a flag in PACER or something or in California dockets that mention whether the case involves children. But you can find that with AI. And so in this case I asked, give me a profile of a lawyer and Vincent in Docket Alarm, or Vita as we call it, went into the docket databases and found a bunch of the cases litigated by this lawyer. Not like one document. It's not like pulling the complaint, it's pulling the entire docket for the case. Some of these cases aren't even completed, Right? And so we have a complete summary of this case, what her role was, the status of the case, the background of the facts, what happened in the motion practice, what the outcome of the case was, how complex it was, what experts were in the case. And if I click on it, you go straight into that full docket in docket Alarm. So you can see for yourself, you can Run the timeline to sort of see what happened, happened in this case. I mean, these are amazing. This is really cool. And then so this pulls like, you know, lots of these major cases for this lawyer and summarizes them all and then rolls them up into an answer and says, here are the types of cases this lawyer has handled. False Claims act, rico, securities litigation. Here's how effective she is in her motion practice. Here's the average case duration, here's the most significant cases. Here are the major clients who she's worked for. So this is, I don't know, it's not the first time we've profiled lawyers ever, but this took like maybe 10 minutes. Right, right, right. And so it's almost crazy not to do this at the beginning of a case. Like, why wouldn't you want to know everything about the opposing party, about the law firm who works on the case, about the individual lawyer who's going to be first chair? I think of it almost like shepherdizing the lawyers and the judges and the opposing parties before you start a case.
- [26:34]
Marlene Gabauer
That it's going to offer you, you know, insight in terms of, you know, your litigation strategy. How familiar is this person with, you know, the type of, of claims that are being raised? You know, do they settle, do they not settle? All of those things are going to influence what you know, how you approach it and how you advise your client.
- [26:51]
Ed Walters
And I'll just point out I didn't run like a Boolean terms and connectors search here. You know, this is, this is the way you would ask like a research associate a question. And Vincent figures it out, says like, here's the kinds of things I would want to pull together. I think this is the most agentic of the winter 25 workflows. It's doing like 50 different things behind the scenes and then bringing them all back together and then summarizing them in this beautiful comprehensive fact based report.
- [27:20]
Marlene Gabauer
I feel like you could sort of use this in combination with other things in doing V. Lex and sort of, you know, put together a package and sort of deliver that as a product, you know, as opposed to just kind of answering a question.
- [27:30]
Ed Walters
It's a great insight. Yeah, I think this is, this is one of the ways forward for law firms, by the way, if their work becomes more efficient and the number of hours dips a little bit because of that efficiency. Why don't we do more for clients and refill the bucket? Let's, let's do a few, let's create new products for corporate legal departments that would be super helpful for them that they would want to see. They would want the extra information and the insights that we can create. And that's, you know, these are products that we can create inside of law firms and make even more money.
- [28:02]
Greg Lambert
Yeah. And, and Ed, one of the things. Do you mind just going back real quick to the, the one you did with the Illinois lawyer, hers. So I just wanted to, to read that because the, the, the actual prompt was just please profile Illinois white collar litigator. And in this case, Lisa Knoller being the name.
- [28:22]
Ed Walters
You know, she's a friend from law school, so I picked on her.
- [28:25]
Greg Lambert
You want to see what she's up to these.
- [28:26]
Ed Walters
She's a litigator famous.
- [28:30]
Greg Lambert
I'm, I'm just wondering, you know, because we're seeing a lot of tools like, and I think perplexity was kind of the leader in this of also being able to essentially search the Internet as well. I'm curious, do you see at some point that you could augment some of these and just say, you know, also go out and search the web and augment the information, or do you think there's issues with that at least at this point?
- [28:57]
Ed Walters
I love that idea. I mean, especially if it's, if it's from like a firm website or something. I think that would be super useful. I want to put guardrails on it because the information we're using is all public documents. It's all verified, it's all been filed in courts that there's a little bit more verification I would want to do on the public web. I'm not going to start pulling YouTube comments and threads from Reddit or something.
- [29:22]
Greg Lambert
I would just say juicy information as it.
- [29:25]
Marlene Gabauer
You might be able to find some stuff in there.
- [29:28]
Ed Walters
But I do think that there is the potential to weave those kinds of agentic workflows into these products where you have guardrails where you can verify. But at the very least, there is such a goldmine of information in these kind of publicly filed documents in state and federal courts. The big data implications, when you combine it with AI allow you to ask very simple questions and get quantitative, very specific, well substantiated information back.
- [29:57]
Greg Lambert
Well, thanks for walking us through that.
- [29:59]
Marlene Gabauer
That's great. That's great.
- [30:00]
Greg Lambert
So since we're talking about security and making sure that we're verifying the right information and we're being careful with it. You had talked earlier about Elex becoming SOC 2 type 2 certified. You'd already had, I think, the ISO 27001. And so, you know, what does that mean for your users? Does that mean that we can now put our information into Vincent and feel, you know, that that's safe? You know, did it help?
- [30:33]
Ed Walters
In a word, yes. Yeah, I mean, yes. So. And we've had these very good security practices all along. We. I think that's table stakes. It's very, very important. Yeah, I agree. It just, it takes a little while for SOC2 to complete the certification, to, you know, validate that you are doing everything correctly. SOC 2 Type 2 is like one of the highest security certifications you can have. Law firms have to have it for their own internal systems. And what it means is that lawyers, law firms, corporate legal departments, can rest assured when they upload documents, whether it be something for an M and a transaction to do due diligence, or, you know, collections of documents, they're working on litigation, or if you're uploading a contract draft or a draft of a complaint or something, they can be assured that that is never used to train AI. It's not being sent out on the public web. It's being kept within the secure environment for that law firm. And at the end of the session, all of it is scrubbed. So you don't have to worry about, you know, data breaches or something. The information put into Vincent AI is kept in the highest security vault you can possibly have. And so I love having the certification. I think that's very useful. I think there's a lot of law firms who trust us in the first instance, but their clients insist that you have to have that kind of third party validation, which I totally understand, and I don't blame them. Maybe one other thing I would love to call out for this release, although I'm a little sheepish about it. So I think we've done a pretty good job of launching things instead of announcing a roadmap. It's a pet peeve of mine with legal tech products when they say, you know, today we're announcing version two of this thing. It will be released at some point, you know, in the next six months.
- [32:21]
Marlene Gabauer
It'll be released in six months.
- [32:22]
Ed Walters
Yeah, yeah. Or like, you know, here, here is the next generation of our tool. This is a PowerPoint that shows you what it might look like at some point.
- [32:31]
Greg Lambert
There's a wireframe. Yeah, here's a slick advertisement of something that we're thinking about doing.
- [32:38]
Ed Walters
Look, and I get it, I shouldn't knock them too much, but I think at least the Velux team has done a very good job of building first and then announcing it. And Releasing it. All this stuff is live. Everything I've shown you has been live in the platform for several weeks. The best firms in the world are already using it to get a leg up. So this last part I am just a little shy about. I'm announcing with our team, Vincent Studio, and we're opening the beta to it right now. Anyone who's interested can send an email to betaelects.com for consideration for Vincent Studio, that puts the firm's own ideas, their own tiles on the home screen. And so I would just say, like, the Vincent Studio is a makerspace for the world's most innovative law firms to create their own workflow tools. They can use Velex data. They can use Velex tools, but then use their own expertise, their own use cases, their own data if they would like to, or their own AI tools, but they can do it all in the secure environment. And instead of us saying, we're going to send you a copy of the, you know, Texas code, you can just use the Texas code inside of elex, which we're updating every day, and build on top of that. And so when people build these tools, their workflow tools will appear on the home screen right next to ours, and there will be like a, you know, a collection of the firm's workflow tools that leverage the firm's expertise. If a firm has a particular strength in oil and gas or securities offerings or, you know, venture proffers or something, then they can use that expertise and the firm's own documents, supplement it with ours, and create their own workflow tools on this homepage. I think that's cool. And we've had a lot of law firms come to us and say, I would love for you to create something that's very highly bespoke to us. And I think our response has been, look, we can't build everything that every law firm wants to do with AI, but we will co develop it with you. We've got this brilliant team of people at Velux Labs who are really good at product specification and project management and design. And they know all the tools, all the APIs inside of our house, all the data sets. So this is an opportunity for law firms to create their own legal data tools. One thing I love is I think that there's a lot of cool innovation projects inside of law firms. There's really good ideas. Sometimes I've heard, like even some firms are taking their new associates and having not quite a hackathon, but like kind of an idea a thon where they pitch an idea for a new tool. I would love it if the winning project were submitted into the Bitson Studio beta. And they would then by winning have the opportunity to work with our team and to build it in the real world. You know, I. Sometimes I feel like the law firm innovation initiatives will have like, you know, 90% of what they need. That last 10%, like you don't want to update the US code every day, you know.
- [35:40]
Greg Lambert
Yeah.
- [35:41]
Ed Walters
That last 10% is something that we can help with. Maybe it's the last 40% or I don't know how much of it is going to be from the firm and how much it's going to be from us. But if we can help firms get those innovation projects over the line using our expertise and our data with some co development in Vincent Studio, I think that's going to be a real winner. I'm excited about that.
- [36:02]
Greg Lambert
Interesting, because I, I know like yesterday I was working with a, with a group of attorneys and one of the, one of the projects that they had pitched was like they're, they're the. These were finance folks and like UCC3, which I didn't even know about was. Was. Was a thing Shows, shows where I am but you know, very.
- [36:27]
Ed Walters
Did you know There was a UCC2?
- [36:29]
Greg Lambert
No. What.
- [36:31]
Ed Walters
Just want to make sure we were catching you up. Yes.
- [36:34]
Greg Lambert
I just remember when it was you so. But you know, but one of the things that was kind of holding back that project was the fact that well, yeah, you know, we can create like a KM system that will have this but as with most KM systems, as soon as you develop it, it's already, you know, it's already becoming obsolete. So I can see where something like this would be really, really interesting for those types of groups for KM projects where you know, it's very important to keep things up to date. No. And know as they change. So I think this is a brilliant idea.
- [37:15]
Marlene Gabauer
Yeah. I mean thanks for. And you know, as long as it is easy for, for them to use. That's the key.
- [37:23]
Ed Walters
I think that's really the strength of AI. The old way we did this the kind of Boolean terms and connector searches where you get back thousands and thousands of irrelevant results and maybe not even all of the important ones. I feel like that boolean searching maybe isn't dead, but certainly shouldn't be the default anymore. If you can ask the questions in very clear language.
- [37:47]
Greg Lambert
Have you know that hair on all the law libraries.
- [37:49]
Marlene Gabauer
I was going to say we're going to get a whole bunch of comments on this.
- [37:54]
Ed Walters
Yes. But I also know that, you know, a lot of that hair was lost in boolean searching.
- [38:01]
Greg Lambert
That is true.
- [38:04]
Ed Walters
You know, I really feel like that that is one of the strengths here, being able to ask the question like you would ask a colleague. And, you know, there's. As you remember from the Autumn 24 release, there is this feature inside of Vincent called Prompt Assist, where it will, like, read the question as you typed it and said, hey, you know, there might be a different way of phrasing that, or it looks like you're asking two questions. Do you want to answer one or both? You might get a better answer if you break this up into a couple of different things. I think that's amazing. I think that's really useful. And instead of getting back, like, you know, no results found because you misspelled a word, having to come back and say, you know, you misspelled this word. And by the way, there's a bunch of expressions about this that you don't even need to worry about. They're kind of related in a, you know, AI sense, and we'll just go find them for you.
- [38:50]
Marlene Gabauer
Well, I want to change topics before we. We wrap it up. We just saw in the news, recent news that Rev Ventures, which is a corporate venture capital fund funded by Relics, has invested in Harvey. And I'm curious since. Since you are a futurist and, you know, a big thought leader in. In the space of legal technology, what message do you think that's sending to the leg?
- [39:17]
Ed Walters
I don't know about a futurist, but maybe I'm a gossip. I don't know.
- [39:22]
Greg Lambert
Aren't those really the same thing?
- [39:27]
Ed Walters
I'm fascinated by this question. I would love for you to ask that question to somebody at Reed Elsevier or Harvey. It's very hard from the outside to see what might be involved in that. I assume that Relics is also investing a ton of money in Lexus's tools as well. I hope so. Maybe this is like spreading your chips around or something. Yeah, I don't know. It's kind of hard to say.
- [39:56]
Greg Lambert
Are they giving you any money, Ed?
- [39:59]
Ed Walters
Rev? No, Relics is not an investor in Velex.
- [40:09]
Greg Lambert
All right, well, Ed, we're at the crystal ball point, and I know we asked you just a few months ago your crystal ball vision of the future, but what do you see as driving change over the next year or two, and what do you think the industry is going to look like?
- [40:27]
Ed Walters
Yeah, I'm cringing thinking about what I might have said the last time you asked Me this, we should have looked.
- [40:31]
Marlene Gabauer
That up before we did this.
- [40:34]
Ed Walters
Some accountability here. So let me just say this, let's go.
- [40:38]
Marlene Gabauer
We didn't look it up Ed. So you can just say whatever, right?
- [40:40]
Ed Walters
So I, let's establish that I was completely right with my predictions last time, as you asked. As always, let me just say this is a little bit self interested, but I'm observing it in the world. One thing that's really important right now in legal tech is how defensible people's positions are. Things are changing so fast. There's so many new companies coming in. The foundation models are getting better. The kind of bigger players, the platform players in legal AI are constantly expanding their offerings. And I think one of the biggest questions for the next couple of years is, you know, are these tools defensible? Are people building something that, you know, co counsel couldn't just release as a feature or that couldn't be coming out in the next version of OpenAI's tool tools, you know, and so it's going to be really hard, I think, for point solutions or you know, companies who have like a single product offering to be able to defend that territory when the foundation models are getting so much better, so much faster. And I think it's been part of our strategy at Velix, like we said when we merged with Fastcase and Velix, that the foundation models would become something of a commodity. You know, do we care whether we're using Claude Sonnet 3.7 or you know, GPT 4.5 deep seq or something? We don't really know at some level where whether that's going to be a differentiator. But structured data is absolutely a differentiator. Like it took Fastcase and Velex, Velex around the world, fastcase in the US like 25 years of boiling the ocean to build that database and there's so much that happens behind the scenes to keep it updated, to build the cert citator to make sure that we're pulling the right parts of judicial opinions, that we're updating statutes with recently passed acts. It's insanely hard. I mean that's a, that's something that you can't just roll into the next version of Gemini. It's something that you can't just throw a lot of money at and build in any short amount of time. And so I do think those differentiators in the market are going to remain important. That was our thesis two years ago when we merged and I think it's maybe even More true. Now, if I can make a slightly controversial prediction.
- [42:53]
Greg Lambert
Please. Those are the best kind.
- [42:54]
Marlene Gabauer
We like those.
- [42:55]
Ed Walters
I think that there's not going to be any solid middle ground for improvement of the kind of general foundation models going forward. The jump from GPT 4 to 4.0, the jump from, you know, kind of 3.4, 3.5 of Claude to 3.7. These are all relatively linear. The makers of these models seem pretty confident that we're going to achieve artificial general intelligence, or AGI sometime during the next four years. And I'll just say, like I'm pessimistic about that. So either that's going to happen and all bets are off everywhere, or there's going to be a kind of a plateau. You know, we run out of words. There's not enough words to get like that kind of step improvement in the models and the ability of the foundation models kind of plateaus a little bit. And that's okay. That doesn't crash the market or anything. There's all kinds of stuff that they can do better. There's a million great products. People can continue to build on top of those foundation models, even if the models don't improve, like the products built on top of them can continue to get better. But I think the kind of continuous improvement we've seen over the last couple of years might be hard to sustain. It might either plateau or go completely vertical within the next couple of years.
- [44:13]
Greg Lambert
Interesting. Yeah, we'll see. Because I do think I'm. I listened to interviews. There's one of the guy that runs Anthropic that I listened to this morning and he was basically saying that they're. What did he, what did he say? In a, in a finite number of time units, we are going to release this, this new model. But because they just released 37 which was a step up, and they're saying they're adding Internet, which they said was a miscalculation on their part for not adding that in right now. But he was pretty certain that there are four which was going, which he predicted, which would be a huge leap forward was, was coming out soon. So it's going to be interesting to kind of see what, what happens and what that means, if it's a significant leap or if it just means, you know, we're adding that, that video functionality that we talked about earlier. That's big league.
- [45:12]
Ed Walters
Well, maybe my last thing can be very practical, which is I really do think that law firms and corporate legal departments have significant untapped data assets. And I think the KM departments are going to show amazing gains in the next couple of years. I think the combination of public data with private data is going to be amazing. Maybe in Vincent Studio, maybe through our integration with Imanage, But I think the combination of that proprietary, unique, differentiated law firm or corporate legal department data, and the insights of the huge public data set, you know, inside of Velix and Thomson Reuters and Reed Elsevier, if they open the walled garden, I think could be incredible. And so we're putting big bets there on the Imanage integration on Finston Studio to try to build that next generation of applications.
- [46:06]
Greg Lambert
And, Ed, while you were talking, I looked it up and apparently we did not ask you a crystal ball question the last time. You didn't what happened?
- [46:14]
Ed Walters
So I got nothing wrong.
- [46:16]
Greg Lambert
I looked up the time before, which was when you were on with Sonja Ibram, and you predicted that I would be wearing a Louisiana State University jacket at Double Link, which came true.
- [46:29]
Marlene Gabauer
He was right. He was right.
- [46:31]
Greg Lambert
So you were spot on.
- [46:31]
Marlene Gabauer
See, he was right. He was right.
- [46:34]
Ed Walters
I'm batting a thousand in the crystal ball.
- [46:38]
Marlene Gabauer
Well, Ed, thank you so much for joining us on the Geek and Review. It's been really great having you on the show. And I think you're one more closer to the tiara.
- [46:48]
Greg Lambert
Yeah, you would look good in a tiara.
- [46:51]
Marlene Gabauer
I'm sure he would. He would.
- [46:52]
Ed Walters
I would wear it.
- [46:54]
Marlene Gabauer
And thanks to all of you, our listeners, for taking the time to listen to the Geek and Review podcast. If you enjoy the show, share it with a colleague. Would we love to hear from you on LinkedIn?
- [47:03]
Greg Lambert
Yeah. Or Blue Sky.
- [47:04]
Marlene Gabauer
Or Blue Sky.
- [47:05]
Greg Lambert
Yeah. I need to start putting that in the outro.
- [47:09]
Marlene Gabauer
I will.
- [47:09]
Greg Lambert
So, Ed, we'll make sure that we put things on the show notes, but what's the best place for people to reach out and learn more about what's going on at Vitalix or connect with you?
- [47:20]
Ed Walters
You can find me on bluesky@ejwalters and you can find all of this information on the velix website@velex.com Vincent and we have a ton of documentation there, all the announcements, and if you want to reach out for Vincent Studio, betaelex.com. all right.
- [47:46]
Greg Lambert
And is that just. Is that anybody or is that current subscribers?
- [47:49]
Ed Walters
That's anybody.
- [47:50]
Greg Lambert
Okay. All right. I know what I'm doing. Right after this call.
- [47:52]
Marlene Gabauer
Me, too. And as always, the music you hear is from Jerry David De Sicka, who has dropped a new album. So go out and check it out.
- [48:01]
Greg Lambert
I ordered mine yesterday, so. Thanks, Gary. All right.
- [48:04]
Ed Walters
Thanks, guys.
- [48:05]
Marlene Gabauer
Bye.
- [48:11]
Ed Walters
That the devil's back on the ball hey, hey, don't take me away I could walk on by the North Star But I fail to notice how still daylight time the devil's back on the ball the devil's back on ball and the devil's back on ball.