Conference Season Rewind: Our Best AI Conversation with Filevine’s Dr. Cain Elliott

Episode Notes

Transcript

Gyi Tsakalakis (00:34)

Welcome to Lunch Hour Legal Marketing. I’m Gyi from AttorneySync and I successfully defended my championship chili cook-off championship. And I got a trophy to prove it. And here’s my one from last year because I’m so proud of my chili cooking.

 

Conrad Saam (00:57)

For those of you who are counting, this I believe is the fourth time Gyi has referenced his chili cook-off win.

 

Gyi Tsakalakis (01:05)

That’s how proud I am.

 

Conrad Saam (01:06)

That’s how proud you are. And I am Conrad Saam from Mockingbird and I love high-end mechanical pencils.

 

Gyi Tsakalakis (01:14)

Do you have a high-end mechanical pencil at the ready for our YouTube friends?

 

Conrad Saam (01:19)

No. This podcast is not sponsored by high-end mechanical pencils dot com. But I think they’re awesome. And it’s about as nerdy as you can get.

 

Gyi Tsakalakis (01:29)

Well, we will try to find an example to put in the show notes, but we don’t just talk about chili, unfortunately, or Conrad’s love of high-end mechanical pencils. We talk about legal marketing. Conrad, what will we be talking about today?

 

Conrad Saam (01:46)

We have definitively the best, most super and in-depth conversation about AI in the legal industry after we do the news.

 

Gyi Tsakalakis (01:56)

Let’s do the news.

 

Short news day today, but in the news is research from Nozzle about poisoned Google SERPs. Short version. Since September, Google has been poisoning, at least according to the research, right? It could be before that, but this is when Nozzle picked it up.

 

Conrad Saam (02:24)

This is great.

 

Gyi Tsakalakis (02:25)

Google has been poisoning the SERPs it serves rank trackers. Completely normal-looking pages whose results have been swapped out.

 

Nothing on the page gives it away. Each keyword’s own history does. They have a breakdown. You can go to nozzle dot io to learn more. Conrad, what does this mean to the average listener of Lunch Hour Legal Marketing?

 

Conrad Saam (02:45)

This means nothing to the average listener of Lunch Hour Legal Marketing other than it does. No, no, hold on, hold on. This is the punchline, dude. You tore the punchline away from me. The

 

Gyi Tsakalakis (02:57)

Okay, sorry.

 

Conrad Saam (02:57)

point being, Gyi and Conrad have been telling you to get over your obsession with rank tracking for a multiplicity of reasons. Here’s another reason to add to the multiplicity heap. The d— I, by the way, I love— Google’s really been pissing me off recently. This is awesome.

 

They’re basically serving up bullshit to scrapers and rank trackers. So the data that you get out of scrapers and rank trackers is less valid. I love that. You should only be looking at whether or not your internet activities are actually generating business. And if you are looking for up-and-to-the-right bullshit, there are plenty of ways to find it. Google has just made it even worse. I love this.

 

Gyi Tsakalakis (03:41)

And if you’re sitting there thinking, well, of course, Conrad and Gyi, we don’t use rank trackers. We use AI visibility scores. Guess what? You too are now relying on unreliable data. Even more unreliable.

 

Conrad Saam (03:56)

Yeah, I

 

was gonna say, you are now relying on unreliable data?

 

Gyi Tsakalakis (04:00)

Yeah. You’re relying on even worse data than you were before. And so for all of our worthy rivals who are out there in the marketplace talking about visibility, and I see your Instagram posts and I see your LinkedIn posts and I see your ads, go send them the Nozzle research and ask them how they’re accounting for poisoned search results in your visibility score.

 

Conrad Saam (04:26)

And as always, the words “worthy adversaries” are really “mean scumbags” from Gyi.

 

And Gyi, our recent listeners will remember the case of the AI-generated GBP spam campaign brought to you by Alexander Shunnarah. And those profiles are not there anymore. They’ve disappeared.

 

Gyi Tsakalakis (04:51)

Well, that’s great to hear that those came down and hopefully everybody learns from this experience and does not create false and misleading profiles, reviews, and AI testimonials. Sadly, I believe we’ll probably see this kind of behavior from other lawyers in the future.

 

Conrad Saam (05:14)

Hey, I’m wondering, since we put that episode out, have you received an influx of requests to knock down profiles from listeners?

 

Gyi Tsakalakis (05:24)

Well, we get a lot of— I’m in some communities where that’s what those communities do and I haven’t noticed a change, but I’ve not gotten a personal request. I think you’re the one who’s really the lightning rod for this one.

 

Conrad Saam (05:38)

I

 

I have gotten a large number of requests for me to help clean up the interwebs.

 

Gyi Tsakalakis (05:45)

Clean up the internet, Conrad.

 

Conrad Saam (05:47)

I am the janitor of Google Business Profiles.

 

Gyi Tsakalakis (05:51)

All right, in other news and also related to our first segment today, Conrad and I will be at Great Legal Marketing in DC, potentially the last Great Legal Marketing, we’ve announced that before. And this might drop while we’re there or as we’re getting set up. But that sounds about around then. So

 

Conrad Saam (06:13)

Be on a plane somewhere.

 

Gyi Tsakalakis (06:14)

if you’re listening to this from your transportation to Great Legal Marketing, please

 

do say hello. We’d love to hear from you. We will also be doing a Lunch Hour Legal Marketing live in San Diego at LEX Summit closer to the end of the month. And so if you’re going to be at LEX Summit, if you’re a listener that will be attending LEX Summit, come watch us live and bring your questions because we do this in front of a live audience. And the quality of that episode always depends on our awesome engaged audience.

 

Conrad Saam (06:49)

I’ll say those LEX Summit episodes have always been solid. I unfortunately am gonna miss the yacht party on the Monday before. Gyi, I believe you are also not going to have your boat shoes and captain’s epaulets on. Is that correct?

 

Gyi Tsakalakis (07:04)

Yes, I unfortunately will not be at the yacht party. I will be in Boston for ClioCon and if you’re in Boston, I’d love to hear from you too. I’ll be doing hopefully some interviews that we might use for the pod. So hit me up there. Let’s take a break.

 

Conrad Saam (07:24)

So Gyi, CallRail is kind of the de facto tool that almost all of our clients use. I believe almost all of your clients use it as well, correct?

 

Gyi Tsakalakis (07:31)

Absolutely, except the ones who object to call tracking.

 

Conrad Saam (07:34)

Those silly, silly firms that are hamstringing themselves. But you know, there’s so much that CallRail can do in terms of bringing call data into CRMs really, really seamlessly. I think most of the major intake management software and CRMs have been able to hook into CallRail. So it makes that really, really easy. And the ability to hook CallRail to advanced artificial intelligence like Claude to start mining information that you get out of those inbound calls, invaluable.

 

Gyi Tsakalakis (08:01)

Absolutely. You wanna know what the client’s journey looks like to finding you, what their pain points are, and do that seamlessly and to increase your ability to provide an excellent intake experience, you know, getting that CallRail intake data directly into your CRM is a really powerful tool.

 

So go to CallRail dot com

 

and tell them how much you

 

love Lunch Hour Legal Marketing and are so grateful that Lunch Hour Legal Marketing sent you to CallRail dot com.

 

Today’s show is sponsored by Eve Legal, the leading AI platform for plaintiff law firms.

 

Conrad Saam (08:39)

Eve helps firms with both the practice of law and the business of law, with AI tools that dramatically cut the time it takes to produce casework. And Eve OS, the operating system for plaintiff firms that lets managers and firm owners answer any question about their business.

 

Gyi Tsakalakis (08:54)

Over seventeen hundred plaintiff law firms use Eve to take more cases, reduce time on desk, and increase settlement values.

 

Conrad Saam (09:04)

You can learn more at www.eve dot legal.

 

That’s www.eve dot legal.

 

Gyi Tsakalakis (09:15)

We are back and we’re highlighting an interview that we did with Dr. Cain Elliott of Filevine. In my opinion, it’s one of the best conversations that we’ve had about AI. And even though this was recorded— this video I believe was recorded, I think it was October of last year.

 

Conrad Saam (09:39)

Close to twelve months ago.

 

Gyi Tsakalakis (09:41)

And we thought that, hey, we’re gonna be at Great Legal Marketing, we’re gonna be at LEX Summit. Why wouldn’t we air this outstanding interview that we did at Great Legal Marketing last year with Dr. Cain Elliott of Filevine? With that, roll the interview.

 

Gyi Tsakalakis & Conrad Saam (10:00)

Dr. Cain Elliott, welcome to Lunch Hour Legal Marketing. And in the category of smartest people in the room, we are so grateful to have you here. Tell the people of the world exactly what it is that you do at Filevine.

 

Cain Elliott (10:15)

The best job in legal tech. Wow. I genuinely mean that. I run a skunkworks team. We build products that are outside the mainline of the product. We try to find new ways to do things, try to build things that are a few years ahead of their time and then down the road get to see those things in Filevine. So whereas I’d say the focus of the product team has to always be on the next six months to a year, what they’re building, our focus has to be on three to five years from now, what we want to see in Filevine.

 

Unique background where I was a professor teaching philosophy, working in the

 

Gyi Tsakalakis & Conrad Saam (10:51)

Another philosopher. We’ve had all philosophers here.

 

Cain Elliott (10:54)

Including you, you are our fifth philosopher on the pod today. Some people have good ideas about things.

 

Gyi Tsakalakis & Conrad Saam (11:01)

I had more.

 

Cain Elliott (11:02)

But yeah, I was doing that, I was teaching, there were some accidents in the lab and I wanted to go out and build things instead of just working— and my buddy said, you gotta come work in legal, running a law firm, and I said, you’ve gotta have your stuff together. Why’d you need my brain in legal? He said, you have no idea. We need your brain. So I ended up working as the CIO of a firm out of Philadelphia and trying to remake that in my own image for three years and working with that firm of two hundred and twenty people out of nine offices.

 

Right. And then when I had fulfilled what I wanted to do there with the managing partner and our kind of vision for rebuilding the firm— Yeah. You know, Filevine wanted to get a hold of my brain. And I’ve been doing that for about six and a half years now.

 

So you talk about coming up with ideas that are— what’s your fail rate or success rate? Fail rate is super high. Yeah. Super high. So whereas in the mainline of the product where we have, you know, seven thousand firms on now, I think it’s something like that, we can’t do that. So we’re able to build out smaller tools, try out ideas. I would say fail rate is something like a traditional startup in the sense of about a ninety percent fail rate. I think most of the ideas have to fail and we try to fail fast with them.

 

So we know what’s working or not, where the tech is, what we can do, how far we can push it or not. And then like I said, honestly the most important part of that is not just trying things and failing at them, but it’s then how do we act like good translators, to be good citizens for the rest of Filevine and translate that information out, disseminate our learnings from that so that other development and product teams can learn from that and keep building the most interesting things possible.

 

Gyi Tsakalakis & Conrad Saam (12:44)

That might be the hardest part.

 

Cain Elliott (12:46)

Really that

 

is— I think everybody finds that if you’re doing well in legal tech, you’re doing well as some kind of translator, right, between everything that people know as expertise in their domains and then expertise in the technology, and this is just ramped up to another level. So I keep the team very small and very tight and then ultra-focused in order to be able to pull that off.

 

Gyi Tsakalakis & Conrad Saam (13:09)

Awesome. So speaking of domain expertise, for folks again that might just be hearing your name for the first time on Lunch Hour Legal Marketing, what were you teaching? What was your background? I know it has to do with AI, and AI is new for a lot of folks, but not necessarily new for you.

 

Cain Elliott (13:25)

Yeah,

 

so

 

I taught just like I like to work, so all over the place. The expertise generally speaking was twentieth-century Jewish philosophy. Okay. And then a little jurisprudence mixed in there and like I said, some STEM. And then I started getting into AI because I was working on catastrophic risks. So that’s how I was also in nanotech. So different things that could pose catastrophic risks for the human future is how I got into that side of philosophy. And then to be able to understand whether the technology was risky or not, you need

 

to know enough to know what’s dangerous or if it’s dangerous. And so that’s how I got into that. So I was working on that, teaching classes about it at Penn. That was more than ten years ago.

 

Gyi Tsakalakis & Conrad Saam (14:07)

Okay. And I think that’s helpful background for folks because again, for most of the world, except for like sci-fi and maybe some dabblings in machine learning ideas, this AI stuff’s brand new. It is, yeah. And you’re building a product that, from what I— we were at LEX Summit, which was awesome by the way.

 

Cain Elliott (14:26)

It

 

was an awesome one. The vibes were really— the only downside was when it started raining. Yes. When everyone was ready to go outside for a beautiful event. I have to say our poor organizing team, they were so panicked about the rain and I was like, everybody’ll be fine, it’ll be fine, but they— you know, I watched them put their heart and soul into it and they just wanted everybody to have

 

Gyi Tsakalakis & Conrad Saam (14:46)

Well I was surprised you didn’t manipulate the simulation to stop the rain. That was my follow-up question, right?

 

Cain Elliott (14:52)

What did we do? They don’t let me do that when their guests are involved. Yeah, yeah.

 

Gyi Tsakalakis & Conrad Saam (14:55)

That’s still in the skunkworks program.

 

Cain Elliott (15:01)

Yeah. Manipulation of the weather and stuff, all of that, I probably don’t wanna incur the kind of, you know, tort that’s gonna involve. That’s gonna be a mass tort.

 

Gyi Tsakalakis & Conrad Saam (15:10)

That’s right.

 

That’s right. So Filevine, a lot of announcements, but one of the big— and maybe this wasn’t so much, it was probably a little bit before even LEX Summit, but it seems AI is really interwoven throughout the entire product. Yes. And you know, a lot of lawyers are still kind of learning about AI and trying to get comfortable with it. I’m sure you’ve been talking a lot about this. We see the headlines of, you know, case hallucination stuff. How do you talk to lawyers about—

 

Well one, let’s start here, because I’m getting ahead of myself. As best you talk about translation, what’s the best way that you can describe how exactly this AI technology works?

 

Cain Elliott (15:51)

So

 

couple ways to think about it, but the best way to describe it I can steal from Geoffrey Hinton. So if anybody listening to your show has ever heard of him, he won the Nobel Prize in Physics for his work in neural networks and helped create the kind of foundation of all this stuff. So about 1986 he was having first successes with this. And the way he explains it, which I really like, is let’s say you were trying to do this

 

instead with a computer, and all this— we think about what is a large language model, we know it’s a neural network, we hear that word, but all those words, they mean nothing. Just—

 

Okay, regular humans hear

 

Gyi Tsakalakis & Conrad Saam (16:33)

Regular humans like us.

 

Cain Elliott (16:35)

all this and it sounds like a lot of different technologies. How would you explain this? And I think the way he does it is best. So he says, okay, let’s imagine we were trying to build this thing by hand. So instead of coding it out with a computer, we’re just trying to teach a machine how to figure out what things are. Let’s start with a simple example. So we’re gonna say, I wanna teach a machine how to figure out that this logo of yours is a hamburger. He uses a bird, but we’ll use—

 

Gyi Tsakalakis & Conrad Saam (17:00)

Thank

 

you for being on brand for us. Appreciate it. Yeah. Well this is why he’s the smartest guy in the room.

 

Cain Elliott (17:05)

So pandering is fantastic. I do my best. We

 

pander as well, so we recognize it when we see it.

 

Gyi Tsakalakis & Conrad Saam (17:12)

Game recognizes

 

game.

 

Cain Elliott (17:14)

I see this hamburger

 

and I’m trying to say, how do I teach a computer that this is a hamburger, right? There are a lot of ways to approach that, but the best way to approach it we found is not to go through and write out all of the attributes you would ever find of a hamburger. Take forever. It’s an endless process, right? Because if I tried to explain to a machine, first I have to explain that this is not a disk, right? This thing is not meant to be a Frisbee. It’s not meant to be, you know, an oblong stone. That process would never end.

 

Right? And humans don’t really think like that. So if we want to teach a machine to learn what a thing is, an object, or to recognize it, we could do a couple things. But one easy thing we could do first is we’d start by saying, well, let’s have a set of basically digital neurons, just like the things in the middle of your skull and mine, and we’d say, I want you to look first where there are edges. Edges are something our brains find and something computers can find too, because pixels or receptors on your eyes—

 

the same way of saying, find out where there’s darker pixels on one side and lighter on the other, find those edges. So first we have a bunch of these neurons pinging off, as he says, and they’re showing us what’s edge versus inside an object. So then we can start looking at the different components of it. We’ve got those. And the thing we do with a large language model and neural nets is basically when like

 

to like items start pinging, you start pinging more. So imagine that you’re trying to make this machine a little bit better at recognizing a hamburger versus, let’s say, a square sandwich. You’d say when those edge detectors see round things, well all those little round detectors, they start pinging. And then the other detectors for things inside here, like the rounded meat, they start pinging at the same time. When all those things are pinging at the same time, you’ve probably got something pinging that’s saying it’s a burger.

 

As opposed to it’s a Frisbee or some other kind of disk object. If you think about that, you’re getting much closer to understanding how these systems work. So I think it’s good for people to go through that exercise, think about how something like that works, that it’s just things that, when they resemble other things or have characteristics of other things, they start pinging together, and the more they ping together, the more they go down and figure out what the object is, and they’ve got all of this massive data that’s trained into the system.

 

And that data and the processing power means it’s able to cascade through all the possibilities of what this thing might be to arrive more and more often at the thing which it is rather than another thing that’s like it, but not.

 

Gyi Tsakalakis & Conrad Saam (19:52)

That was amazing and this very much helps frame the conversation. So you talk about possible, and it gets more data, it gets better, it gets better at— it’s more probabilistically the thing that it’s trying to identify.

 

Cain Elliott (20:09)

With

 

talking just with an AI, the large language models we’re using, those are usually next-token generators, which means they’re trying to predict the next thing. So for example, if we start with rounded object, yellowed and browned on the outside, some, you know, discrete colors blended together on the inside, it’s trying to figure out next what would be the most likely or probable thing that you or I would say or it should say in relation to those. So mostly

 

the LLMs we’re using now, those are next-token predictors. So they’re trying to figure out what’s the next best word that should come out of my face, digital or analog. And the other reason I think that’s really useful for people to think about is because that means these are computers that are much more like us than other computers we’re used to using. Regular computers with algorithms that, you know, we have known outcomes for what they’re looking for, don’t do a lot of that kind of just predicting.

 

In fact, that’s why people get frustrated with things like— you said jumping ahead, but it’s not jumping ahead. It’s directly in line with the story, which is next-token predictors can get really, really good, but it means they’re still based on statistical probability rather than something where we’d say, you know, A plus B equals C, a given outcome with normal computer algorithms we’re used to, where we always get the same output. That disturbs some people, but if you think about it, it’s much more like how you interact with other humans, where all kinds of stuff comes out of their face.

 

Hallucinated or otherwise, right?

 

Gyi Tsakalakis & Conrad Saam (21:41)

That’s right.

 

Cain Elliott (21:42)

Because I always tell people, I say, you know, you’re worried about AI hallucination. Part of this is just the learning curve that culturally people are on of getting used to working with systems that work in a different way, that actually work more like the humans they’re around than the machines they’re used to, rather than there’s something fundamentally wrong with the system.

 

Gyi Tsakalakis & Conrad Saam (22:01)

Okay, this is fantastic. Thank you. So let’s talk about Filevine. Because there’s a difference between— well maybe there’s not, you can explain this to me. Explain to me some of the differences between, like, if I was just to upload stuff into ChatGPT versus using a closed system that’s my data with Filevine.

 

Cain Elliott (22:25)

So

 

couple different things, a couple different ways we can take that. I’ll start with the most interesting things and then go to what I think are equally important but boring things. So one of the most interesting parts is, if it’s from your system of record, obviously we’re able to organize the data in such a way to try and get those probabilistic outcomes of what it’s coming up with to be closer and closer to the expectation of what you’d be looking for. So we’re trying to organize all that information so that the system will be really

 

looking at the things that matter as opposed to the world of all possibilities. So as an example, when you open up, you know, ChatGPT, Grok,

 

whatever it may be, that LLM, first of all, it’s running a massive system because it has to try to embody the mind of each of us and all of us all at the same time. So it’s gotta be good for you, gotta be good for you, gotta be good for me. And

 

Gyi Tsakalakis & Conrad Saam (23:22)

And for all the things.

 

Cain Elliott (23:24)

for all the things and for all of the people. Because remember, just like a person, when you’re trying to figure out— if I’m trying to figure out what’s the next best word to come up with, well, I chose hamburger to go with your logo, right, instead of using bird. Well, part of that is me trying—

 

trying to model other minds of what would appeal to you both in conversation.

 

Gyi Tsakalakis & Conrad Saam (23:42)

Mm-hmm.

 

Cain Elliott (23:44)

So it’s got to do that for all minds. Now, when we have it in a controlled system and for your firm, your organization, really we’re giving the LLM the opportunity to try to model out the mind, essentially, of what’s in your firm. So that’s one thing, right? So we can narrow the scope of concerns, interest, and background knowledge it should be looking for that’s relevant, right? That means things like—

 

hopefully we’ve loaded up, if you’re keeping all of your documents with us, right? We’re loading it up in ways that it can look for items that maybe your team missed or didn’t think of, and that’s there, as opposed to needing to figure that out and upload it separately to a system. So those kinds of things matter. The other thing is we can make decisions and switches on the fly about what models we think are doing what work best, so that, you know, over-educated people like myself can help make that decision, and then you’re just getting the best output of all your—

 

Gyi Tsakalakis & Conrad Saam (24:37)

So clarify it.

 

Cain Elliott (24:38)

—without you needing to be like a consumer of every model and then switching all of your workflows all the time if somebody comes out with something much better. So if I’ve got better, you know, search of relevant facts available in the larger web, and let’s say at the moment Gemini is doing it best, well you’re gonna switch all of your work over to Gemini, and then, well, I really like this drafting tool, so I’m gonna switch this model over to that. We can handle all of those switches

 

and traverse that for you, and hopefully that’s where the premium is coming into play, that, you know, Filevine is taking care of having people like me who are sitting there reading all of the background research on these different things and looking at all the baselines and benchmarks to make sure we’re giving you the best, as opposed to you’re rooted down to one provider and then there’s a huge switching cost when you’re running a firm to move all your stuff over, all the stuff you’ve been uploading. So that’s one thing. The other thing is, simple facts that I don’t think change because it’s AI, which is, and I tell everybody

 

listening, I don’t care if you’re using Filevine or not, I’m not on the sales team. My job is to build amazing products and teach people about them and make sure they feel like they can make good decisions. Don’t use something free online.

 

Gyi Tsakalakis & Conrad Saam (25:54)

Good

 

advice for any kind of technology context.

 

Cain Elliott (25:57)

Not new to AI, so I—

 

was told, hey, hey, just because it’s AI on the label doesn’t mean it’s safe. I went into a firm, and I obviously won’t say which one on a pod, but I went into, you know, a substantial-size firm, you know. Tom, the new associate, found this, you know, Chrome plug-in AI. I’m looking

 

Gyi Tsakalakis & Conrad Saam (26:17)

And I—

 

Cain Elliott (26:18)

at this thing and going, where the hell is your data flowing to? And would your clients like to know that it was going to some service that’s in Thailand? And honestly, right, I know all the model providers,

 

everything I can about this space. I had no idea what the system was and I had no idea what it was hooked up to. And they said, but it was a Chrome plug-in, surely it was okay. Right. And I said, just please

 

speaker-2 (26:36)

Thank you.

 

Cain Elliott (26:37)

—get it out. Like, you don’t have to buy our AI, but for God’s sake, don’t do that. So normal rules apply with that kind of stuff, where we have, you know, enterprise-level agreements where there’s zero data retention, that all of your data is getting obliterated from the systems. And we can also keep up the hygiene with that, where end users don’t have to worry about, did everybody in their organization

 

delete all these. So the one example I’ll give is this. So even if you use a free product, right? If you’re using something like OpenAI,

 

if you or your firm are uploading a bunch of stuff there, as long as they keep it there, it’s still hosting not just the documents, but the embeddings, the mathematical model of how a system can figure out very easily what a certain clump of words is versus another. And you’ve got to go through having the overhead of someone to make sure everybody’s deleting all this. So the same rules apply as with other IT things, but I’d say, yeah, make sure it’s paid and make sure you’re keeping hygiene,

 

just like we would, or pay a provider who’s taking care of those kinds of things, whether it’s us or somebody else. So those are some of the— and the other thing is that we have the resources to push really hard to move the use cases forward, and we hope in ways that are most meaningful and impactful for our attorneys, as opposed to people trying to find new ways to use this stuff on their own, which is a productivity cost the economy in general

 

beyond legal is encountering right now. It’s expensive just to figure out how to use new tooling. So hopefully we’re doing some of that groundwork, and teams like mine are doing that in advance before it hits you.

 

Gyi Tsakalakis & Conrad Saam (28:18)

It’s awesome.

 

Cain Elliott (28:19)

In your— completely tactical— one of the announcements,

 

I think this was two weeks ago, lead scoring. Yeah. Did that come out of skunkworks? We do— did that? Not my team. Okay. Yes. But that did come out of an internal skunkworks project, but not with my team. But not with your team. Yeah. I’d love to take credit for everything. You should.

 

Gyi Tsakalakis & Conrad Saam (28:46)

We appreciate your

 

integrity as well. It’s very, very noble. That’s right. That’s exactly right.

 

Cain Elliott (28:55)

So I was gonna ask you to tell where that came from and how that came about. So it’s just such a— as a marketing team, this is such a valuable thing that has been missing. It is one of those things that we’ve, you know, heard for years from teams that are working in mass tort matters, but it takes so much time to triage and go through and to figure those kinds of things out. It also turns out that a lot of the AI that’s been out for a long time, and I would say some of that’s just what we

 

used to call good old-fashioned machine learning. Right. Where that’s some of the stuff that’s been most rigorously developed. Because things like fit— does this fit this type, what’s the affect in something or the way people are responding? That’s some of the stuff that social media companies have been working on the longest. So there’s a pretty big repository of material there for us to work with to try to make that. And then also we have a really valuable and important, I think, data set from our clients

 

themselves of what they end up taking and not taking. So you’re able to use the data you’ve had from LeadDocket, for example, our intake tool, to score what those kinds of outcomes are, where you didn’t know you were scoring on them the last few years, but you were essentially doing that by giving a thumbs up and thumbs down to which leads you were taking and which you were dropping or which, you know, you were referring out. And a lot of that requires that input, right? Yeah. And without the input, it’s almost impossible to achieve. Sort of.

 

Weird— that’s a really good topic because it’s strange. I want to keep going deeper on this. Okay. I’m fascinated by that specific problem and the size of the firm. Okay, so being interested in that topic of marketing, it’s weird. So there is a way to overfit the model. If you’ve ever heard that term. Okay. It is exactly what it sounds like, in the sense that you focus too much on what you’ve done in the past and distort good probability

 

by overcorrecting, overfitting a model for what you’ve done in the past. Right? So it’s the same kind of thing where you tell somebody, just because you’ve done this over and over, usually that’s a good rule of thumb, but don’t do that forever because you’ll destroy your business and your livelihood. And, well, you can destroy all kinds of things in your life by just following patterns over. I think that’s rule number one in, like, therapy— they tell you following the same pattern forever

 

Gyi Tsakalakis & Conrad Saam (31:13)

That’s right.

 

Cain Elliott (31:16)

doesn’t mean you’re winning. That’s not the way to— Charlie Sheen-style life.

 

Gyi Tsakalakis & Conrad Saam (31:19)

I don’t know.

 

Cain Elliott (31:22)

So I see why.

 

Gyi Tsakalakis & Conrad Saam (31:24)

Yeah.

 

Cain Elliott (31:24)

So

 

but, okay, so there’s one thing of like overfitting a pattern model. So another way I explain this to people is, let’s say your work is just transactional. All you ever do is contracts in your whole career. If I take a large language model and just send it to law school on contracts, it’s a pretty bad lawyer, right? All those other things you’re bringing in from life and experience make you a better attorney. So if I do that with a model, same thing. Firms are like, why don’t you just train a model just on our data? It’s gonna be terrible.

 

Promise you don’t want to use that. Because the firm data in and of itself is— or because it was

 

done by humans? No. Although that’s an issue. Okay. That it can be very messy. But no, it’s because one, you started to say this already, which is statistically it may be too small to really matter. That’s what I’m really going to— because, sorry, keep going. No, so statistically it may be too small to matter. But the other thing is that the other clues and hints you’re getting as an embodied living human being, all those other hints you’re getting from years of experience of, like, hearing the way someone spoke to you and seeing if they’re really interested

 

or they’re fishing for, you know, maybe they’re interested in your firm or not, right? But they’re just hitting out and they’re calling every firm locally, right? Those kinds of nuances you would hear or see in their language, if you just train on the data set of what was, the model would learn to ignore those things, right? Machine learning is called that for a reason. Just like when we talked at the beginning about if it’s pinging more or less, your brain starts hearing certain words and starts pinging more, saying this person’s not really interested,

 

pinging less and saying, yeah, ignore that lead, right? Same idea with these models. If you give a lot of pinging noise in one area, you start to drown out the other signals your brain would pick up, even if they’re not as heavy in there, right? Really fascinating thing though, some studies are coming out now where you can take the models, and if they get big enough and large enough, right, you can start to figure out buying preferences without ever needing to really poll the groups. So that’s a weird thing you’ve seen now, where you can run— if you’re trying to

 

profile someone based on something like demographics, where we have a huge data set about preferences. I mean, if nothing else in the last 20 years, the United States has gotten really great at profiling everybody in the—

 

Gyi Tsakalakis & Conrad Saam (33:44)

Hooray.

 

Cain Elliott (33:46)

—but there’s a lot of data. So if I say, you know, just take myself and I say, you know,

 

a white heterosexual male living in the Intermountain West in his forties, and this is his background and sets, what are likely to be the preferences? There’s a reason why Amazon can put things, you know, near my doorstep that I’m more likely to order online before other things that it thinks I’m not gonna order. And that comes out too in preferences about essentially knowing what people are gonna buy before they buy it. And that feels a little bit uncanny, but I

 

think it’s a good lesson, and I’m one of these people who believes pretty strongly— this is not universally shared both within Filevine and the wider world— but I believe fairly strongly there’s not such a hard divide between wetware and hardware, between human brains and non-human brains. I don’t think that divide’s real, and I think actually we’re gonna get better legal practice, better laws, better consumer protections and things by allowing for some of that blur. So for example, I think it’s really bad if

 

consumer protection laws about AI right now get really focused on, you have to let the consumer know if they’re talking to an AI or a person. But for example, that leaves this blurry space of: the human went, asked the question of the AI, got a response, copied and pasted that in the chat back to a person and clicked it. Well, now it’s not a problem.

 

Gyi Tsakalakis & Conrad Saam (35:13)

So—

 

Cain Elliott (35:13)

So I find things where it’s based on a really hard distinction to be really ill-fitted to the world we’re going to be entering, we’re already in and will be living in.

 

Gyi Tsakalakis & Conrad Saam (35:24)

Well

 

it’s funny, because even outside of the context— I guess it’s tech, well, you could tell me— but like even autocorrect, like, do you have to disclose that you’re using autocorrect?

 

Cain Elliott (35:33)

Grammarly is—

 

Gyi Tsakalakis & Conrad Saam (35:36)

Right.

 

Cain Elliott (35:38)

—is doing that now, it’s ramped up even for— actually, I don’t, you know, I don’t talk ill about other things, but I would worry about using Grammarly inside my phone because it’s pulling every word off of your—

 

Gyi Tsakalakis & Conrad Saam (35:49)

Right. Another theme— so we’ve been going to a lot of conferences this fall, and a theme that I think is a concern for lawyers is it feels like, and this might not be intentional by any of the legal technology companies, that we’re moving into maybe this walled garden type of space, where we have one platform to rule them all. Obviously from a product standpoint, and I know this isn’t your department at Filevine, but from a product standpoint it makes a lot more sense. It’s a lot more sticky when you’ve got all of your—

 

you know, you’re supporting the firm through the entire client journey and all that kind of stuff. Talk to us about your view of walled gardens and integrations and APIs. I mean, you know, the firms that we work with, a lot of times they’re Filevine, they’re using LeadDocket, but they’re using a lot of other systems. It’s really important for us to be able to get that data out. How do you think about that kind of stuff?

 

Cain Elliott (36:42)

Yeah, so there’s a balance to be had. I think we are moving from an API tech economy to an AI tech economy, and there are different considerations there. So I’ll put them out in terms of the way I think about it. And I think that our customers and clients—

 

I would understand that they might have different considerations, but I think it’s important to be open about what I think those are. So data flowing through other systems— one is that, like I was just alluding to and we’ve been talking about, having more data about the process itself is advantageous in this circumstance. And so obviously aggregating more of that data, not because of what initially—

 

and it’s good you phrased it that way. Initially I think there were a lot of people who were thinking about this in terms of, are these people trying to scoop up my data or something? And like I’ve said, I tried to explain, the economics of just trying to grab your particular firm’s data, there’s a lot of reasons why that doesn’t actually serve your firm.

 

speaker-2 (37:50)

Yeah.

 

Cain Elliott (37:51)

Does it mean that if your data is all flowing through my system, can I offer better AI offerings? Yes. Yes. So that’s one thing, right? So you’re trying to differentiate the offering, make it better. That’s one thing. The other thing, too, is that, look, getting these results that are high quality means you have to do a lot of things with the data, transform it, store it in other ways that get expensive. So being just a pass-through for somebody else’s data to get that experience, that can become expensive. And the other thing is in terms of product—

 

at least how I think about it— pre-AI, the way we would phrase this at Filevine is too many inputs for attorneys to deal with. So you got emails coming in, texts coming in,

 

faxes a lot, you know, we’re around this wonderful event, lots of PI attorneys, and faxes are still used, and telephone calls and everything else, too many inputs. So at Filevine we were trying to solve for too many inputs. I think in the AI era the way I think of it is we’re trying to solve for what will become, for a lot of our clients, too many outputs. Everybody wants to build a widget on top of something else with AI baked in, and then our clients have twenty platforms to check for that.

 

Gyi Tsakalakis & Conrad Saam (39:00)

Not cheap either.

 

Cain Elliott (39:02)

Not cheap either,

 

but it’s also a different kind of problem than the knowledge distribution. So you gotta find some kind of balance between those, where you want to offer the best product possible, but also you want it in a way that your users want to consume it. And like I said, if there’s a widget here, here and here, that’s less advantageous. So not speaking to how it’s dealt with externally maybe as much, but internally we’re trying to unify those experiences as much as we can within the Filevine platform.

 

So that, you know, our announcement of what we’re calling LOIS, the Legal Operating Intelligence System, is really to have a unified semantic data layer of artificial intelligence across the platform, across the different tools we have, so that wherever you are, you’re getting the full intelligence of what you use in Filevine. You have a very kind of user-centric mindset for what is otherwise very heavy.

 

Yes. Where does that come from? There’s the Hegelian phrase from philosophy. You know where they say Geist, which is mind in German, right? This is during the age of phrenology: the spirit is a bone. Yeah. The mind thing with the bone. So I always think you wanna be in the clouds mentally and be in the rubble and the mud and the ground. So I think our best work at Filevine comes from being

 

really obsessively, all the time, almost in a perverse way, involved in what our clients are doing day to day, and I think a lot of that comes from the fact that we hire a lot of people in our product and tech teams who have some kind of background in legal. So a lot of us feel like it’s our duty to not abstract away from their, what I call, mini-murders.

 

Gyi Tsakalakis & Conrad Saam (40:55)

It’s tough that you just—

 

Cain Elliott (40:56)

—awful pains every day, so we try to stay really close to that and then stay at the forefront of it, and yeah, that’s how I would say the organization approaches it. So you have a vision into the future, what might come out of Filevine— I don’t— yeah. And we’re happy to hold you to this if we can. What should we be looking forward to? Look, I think legal tech, I would start—

 

yeah, this, I’ll speak— this is Dr. Cain, not the whole organization necessarily— but if you were to ask me what does this start looking like, I think legal tech starts looking a lot more like Spotify than what you would think of as a traditional CRM. I’ll explain what I mean. I like this. So think about all of the workflows you do every day, all the things you do for your clients, all the calls you’re making, the emails, all these things. Think about them more like playlists instead of just checklists of to-dos.

 

You start thinking about it like that. You start thinking about, well, you know, if you have your playlist, I have mine, maybe I’d like to riff on yours. Take a listen to what that sounds like. Can I learn from the best tracks in your playlist? The other thing that I like about this metaphorically, but I think it speaks really loudly to people in 2025, is, you know, when you’re following a colleague on Spotify, if you’ve ever shared your Spotify with anybody on any account and they can see what you’re listening to, you can do that in a really unobtrusive way. So

 

you know, I can see what my lead developer is listening to in the background. I might want to riff on that, play some of it too. But I can do that without pinging him and writing him an Outlook message and saying, what are you listening to right now, Chris? That’s disruptive, right? But the idea with playlists is you can weave in and out of your own things you’re crafting, your playlists and tracks and your greatest hits, you know, the MSJ you’ve got that you think is just kick-ass.

 

You know, you can riff on that, but also your colleague could listen in and see, that’s pretty hot. I like the way that sounds. Let’s hit that too. So you can kind of toggle in and out. And I think the audio metaphor works too, because just think about any kind of smart headphones right now, where you can do transparency mode, total noise cancellation. You have something, you know, transparency somewhere in between, noise cancellation to get everybody else out, or totally off where they’re just there but unobtrusive. I think that’s how

 

people would like to engage with the work of colleagues, in ways that’s either totally focused on my work, listening into their work, or absorbing all of their work in a way that’s more natural than how we’ve had it in a lot of workflows in the environment of the firm, where it’s context switching that’s very expensive, that’s also painful in terms of, you know, your time, your energy. So I think people want to flow more in and out of playlists within the firms to find those tracks.

 

Gyi Tsakalakis & Conrad Saam (43:59)

Making better music.

 

Cain Elliott (44:00)

Right. And also, here’s an amazing thing about the digital life. Unlike what I used to do making, you know, mixtapes when I was younger, that also had switching costs that were pretty expensive compared to digital track lists that you’re sharing back and forth. So that’s the way I think about our workflows, document assembly, these other components that we can put together digitally, where hopefully you can mix and match the best of your firm, the best of what the AI is putting forward, and then roll out newer and better playlists for your clients. Does this change the firm?

 

Like, what you’re really describing is I’m moving away from a centralized standard operating procedure to more of a fluid, I’m working out what works for me in the moment. Yeah, but where that’s coming out of, hopefully, is your standard operating procedure, right? So not just out of the ether, or maybe if you’re getting started maybe it is coming out of the AI giving suggestions at the outset, right? But I think for most people, if you have established practices, it’s coming out of that standard operating procedure, but giving you more flexibility than you would have had historically to do that, again with

 

less switching costs and pains of, like, I have to rewrite the whole interface of what I’m working with or change to a new software system to get this new vibe, as the kids might say, that I’m interested in listening to, right? Yeah.

 

Gyi Tsakalakis & Conrad Saam (45:16)

So this is

 

kind of like out of— you can be like, not my wheelhouse, but I’m just curious about your thoughts on this. So you know this stuff, you know how this technology’s working. In our world, because SEO vendors have been doing this forever, is this new shiny thing comes out and there’s promises of, we know how to hack it, you know, we know how to cheat it. Some of it’s even like, if we talk to it,

 

the more we talk to it, we can influence the system’s responses. So is that a thing?

 

Cain Elliott (45:56)

Here’s—

 

Gyi Tsakalakis & Conrad Saam (45:57)

Like, can I log into— can I be in ChatGPT and teach ChatGPT to tell— right. Thank you. Thank you. So you heard it here on Lunch Hour Legal Marketing.

 

Cain Elliott (46:10)

There’s a lot of different— you can get better at, like, for example, results, ways you would get results in things like Gemini, for example, what it shows. Not all of that though is related to new skills, and not all of it’s going to be dependent on things that someone can just go in and change, right? So you can’t, like, tweak a dial on an LLM like that that’s already trained. You could tweak the output you have so that maybe the next model picks up

 

more of your pages and backlinks like it did before, right? But on-the-fly changing, you’re not influencing Gemini, too. You can’t go in— you could talk to ChatGPT for the next year and it wouldn’t know more the next time it opens about your firm. And anybody who tells you otherwise, let me give you the definitive proof that that’s the case. The machines we’re talking about with LLMs are stateless. So a stateless machine does not open back up

 

with memory that came from the discussion. Those memories all come from the training database. They’re all stateless. So anybody who says they’re manipulating that, they’re talking about manipulating an algorithm that would be running on the fly, but these are stateless systems, so that’s not possible.

 

Gyi Tsakalakis & Conrad Saam (47:25)

Thank you.

 

Cain Elliott (47:26)

You can make live search better and other things with these systems, but that’s still using search inside of the AI models. But change the models themselves? Not possible. We owe you dinner. We’ve been talking about this. Just a beer would be—

 

Gyi Tsakalakis & Conrad Saam (47:40)

We can do that. And I—

 

again, you’ve been very generous with your time. A couple, just two more thoughts that I just wanted to get your reaction to. Did you follow this Vals AI, VLAIR? Yes. So, and again, this is not my wheelhouse, I’m just curious your thoughts. But it seems, my perception is, that they’re basically trying to benchmark some of the major legal—

 

Cain Elliott (48:07)

Legal, finance, a couple other things.

 

Gyi Tsakalakis & Conrad Saam (48:08)

Yeah. And I didn’t recognize— ’cause I’m not— I think they tracked a little bit more Big Law and what Big Law is doing, maybe. Research.

 

Cain Elliott (48:16)

Research, yeah,

 

so a lot of the articles that came out were about legal research specifically. Which was

 

Gyi Tsakalakis & Conrad Saam (48:22)

Yeah.

 

Cain Elliott (48:22)

also a little bit misleading, I think, for a lot of people who read it and were like, so AI just hallucinates for legal work, which, that’s a little bit different. Like, legal research is its own piece. Right. So yeah.

 

Gyi Tsakalakis & Conrad Saam (48:34)

So anyway, where I was going with that was, one, it looked like both the legal-specific models and the general models outperformed the lawyers, the baseline lawyer number. But amongst the general and the legal-specific, there wasn’t a huge difference in performance between the general and those that were grounded in legal. And it seems like there’s this bet out there that grounding things— and again, I’m probably not even using the language properly.

 

Well, I would— Dr. Cain, can you explain what I mean by grounding things in real—

 

Cain Elliott (49:13)

I like that. Well done. I wish everybody at the office would use that.

 

So usually when people are talking about grounding knowledge, what they’re saying is, what’s the data I get to reference when I’m answering the question, right? So when you’re talking to these LLMs, you have a couple of different things going on. One is their training data. So that’s exactly what you imagine. This huge repository of, like, all of the internet and known texts, a lot of it’s Wikipedia, right? By the way, you wanna influence the next set of LLM data, not live on the fly, but— get your Wikipedia page in there, right? Like, that’s a

 

huge thing. That’s even why Musk is trying to build his own, like, Grokipedia.

 

Gyi Tsakalakis & Conrad Saam (49:53)

Right.

 

Cain Elliott (49:55)

So that’s— you’ve got training data, like as much human knowledge as the system can absorb and has the size and bandwidth to do, and enough energy to do so. Then you’ve got live information, like a Gemini and then

 

xAI do a lot of work of going to search, looking on the internet for live stuff while they’re doing it. And then you’ve got user inputs, kind of right at the moment you’re interacting with the bot. And so those second two are kind of transient, and then the first is hard-coded in the system. That’s where you get all these things when you hear about weights. When people are talking about weights, remember when I said ping more or less? That’s what the weights are. Those are just telling the system ping more here, ping less here. That’s what weights are. So the other thing you can do is

 

you can say, when you’re going out and doing that search, that second stage, you can formulate kind of what are the sets of information or knowledge you’re allowed to go out and look at. So a lot of people are saying, if you’re doing something like legal research, you’d want to say— I think the phrase that’s often used in that space is people say, I want an AI that went to law school. So they’ll be talking about, it has access to case law, it has access to statutes, code, you know, codices, depending on what domain of law you’re working in.

 

And all of those kinds of things, and we’re talking about grounding, meaning when you go out and look for stuff, look within this set, right? At Filevine we’re more focused on: look within the set of data that belongs to your particular case or your set of cases. It’s more like that grounding input along with the user input and training data. With case law, it’s pretty hard because they’re non-deterministic systems, like we talked about. And you take a non-deterministic system and you get a non-deterministic output.

 

The only thing, like to your point here, the part that I get a little bit crazy about, is when people say, like, we cited this wrong case. A lot of that’s coming from how systems are used. And I’m not saying it’s user error. Because if the company promises one thing and the output is totally off, there’s a mismatch on both sides usually. So it’s not just a problem between, you know, me and the keyboard, but there’s a problem with the way the vendor pitched it.

 

Gyi Tsakalakis & Conrad Saam (52:07)

Yeah. It’s understanding what the tool can do.

 

Cain Elliott (52:11)

Like when I write a novel—

 

AI— I write stuff. It’s like a page long, with also a document about me in advance, so that it knows the way I’m expecting to get references right. And I don’t think our users want to write that out every time they’re opening a new AI, which is part of what we’re doing.

 

But pretty light on the information. I always tell people, like, would you let a first-year associate draft and submit something without checking it over?

 

If you would, okay, cool, they must be very talented. But I wouldn’t do that with AI and I wouldn’t do that with a first-year. So that’s one thing. It’s like, what’s the expectation going in? But also legal research, that’s a pretty difficult domain because, one, Shepardizing isn’t an exact science. You just have a lot of people working on that, but that itself isn’t always set. And part of the thing everybody does when, you know, they’re writing up a motion or going to court is arguing over that exact

 

thing and the interpretation of it. So part of the work is actually making sure it’s indeterminate. So I’m kind of thinking actually the systems look more like you would as a lawyer in a certain way. So that’s a really, really tough domain, a hard one to tackle. And I think, you know, more and more data will help, but also, like I said, there’s a learning adoption curve where you have to be careful with that. Also,

 

benchmarks right now, I think it would be really hard if it’s not your, you know, day job like it is mine. It’s really tough, right? Like, there’s a new model every week or every other month. In legal tech I think, you know, when I used to track everything everyone was doing in legal tech, if I checked once a week, I’d probably have all the information I needed. Good luck keeping up with that today, right? Down the hallway, right? Yeah. So that itself, you know, would be a full-time job if you tried

 

to take that on. And I think it’s a bit early for that in terms of how we’re using it. And there are a lot of differences in the back end of how those systems are set up that we don’t see when we just give a generic benchmark. So I think it’s great work to do. And I’m glad to see it. And the more transparency we have on all these things, that will certainly bring more light to everybody. But I also think, like I was saying at the beginning, and I really like that you gave me the opportunity to kind of explain how the systems work,

 

I think we need a lot more of that in terms of fundamentals education. Not because we have to say, we’re lawyers, we don’t know how that works— that is people, you know, abandoning their responsibility to explain to clients how things work and not taking the time to figure it out. It’s not true that it’s very sophisticated, hard technology and therefore can’t be explained. I don’t believe—

 

Gyi Tsakalakis & Conrad Saam (55:14)

You just did it. Yeah.

 

Cain Elliott (55:16)

But you do that

 

with any domain, and I just don’t buy that. I think that is skipping over the work, one, because either they don’t know how it works, right? They’re just plugging into something and have no idea. Or two, they don’t feel they need to take the time. And so that’s one of the things I’m lucky with in my job, that I get to take the time to try to do. But we need a lot more of that for people to get more comfortable with the systems, to feel like they’re interacting with things where they know what’s going on, rather than black-box tech, which doesn’t work well for either side. I think actually that

 

people get pissed off with vendors when things are black box, as they should. Users get experiences that are degraded, I think. So the more we have general education, other than people saying, I like to use ChatGPT to do this, I like to use Anthropic to do this— no, but what is the tech, what’s actually going on? It’s longer-term education, but it’s certainly worth doing because it’s not going anywhere. Like, let’s say there’s an AI bubble right now.

 

Gyi Tsakalakis & Conrad Saam (56:13)

I’ve been saying

 

that.

 

Cain Elliott (56:14)

So you mean like twenty-five times?

 

Gyi Tsakalakis & Conrad Saam (56:17)

What do you think about that?

 

Cain Elliott (56:19)

Here’s the thing. Sure, and also no, in this sense. I was talking earlier today with my colleagues about this. Was there a real business model or value proposition or something interesting behind Pets.com?

 

Gyi Tsakalakis & Conrad Saam (56:36)

That’s the same example I used.

 

Cain Elliott (56:39)

No, there was not. Are major AI providers, or what’s going on in legal tech right now, having serious business models that are bringing in revenue that’s real economics? They are. So that’s

 

Gyi Tsakalakis & Conrad Saam (56:51)

Yeah, absolutely. That’s a great point.

 

Cain Elliott (56:53)

not just speculative stuff. Now, all the insider deals and all the other things going on at these big shops,

 

I would attribute that not to something special about AI, but to things humans do and bring on, how

 

Gyi Tsakalakis & Conrad Saam (57:05)

Yeah.

 

Cain Elliott (57:06)

people like to speculate, right?

 

Speculation. Yes. I think that’s their own. I would say, look. I think I would—

 

Gyi Tsakalakis & Conrad Saam (57:17)

Here’s the thing.

 

Cain Elliott (57:19)

Building

 

things with it, my team is— I can see outcomes of what clients are doing with this tech. Will there be a longer adoption curve for getting productivity out of it? A lot of domains, of people figuring out the best ways to use it within their organizations? For sure. But that’s where I kind of tell everybody, you know, AI is a weird thing in the sense that we don’t have a history as a species of making tools that make better tools themselves.

 

So that’s a little different. Like,

 

the hammers don’t produce better hammers. It’s a little odd, right? So that’s the kind of part where, when people say they know for sure what’s gonna happen, it’s like, no, no, no, this is my job and has been for my adult life, and I can tell you— like, you think you know what’s happening five years out from now? Like, okay, sure. That’s good. I hope that, you know, you enjoy that, and none of that will come to fruition or be very valuable for people. What I would say for sure is that the productivity gains

 

that can happen are pretty massive. What I think is going to be most useful for firms especially is pick narrow areas of domains you want to go after first, and then go after how you utilize it to get better outcomes in that, before you, like, redo an entire stack, a way of thinking about your practice, on the basis of this. And, like, talk with myself and other people who want to share what’s going on, have more conversations about it. Is there a— I don’t know. People are just driven

 

to do that where there’s a lot of value to be had. But on the other hand, if you invested in a lot of the companies before the big burst in 2000, you still would have done really well. And nobody’s sitting out saying, like, you know, the internet, what use was that? As an example, all of the stuff we talked about playing with AI, it wouldn’t be possible without internet sets of data. So that was one of the things— back in ’86, when we talked about the first sets of these models, the neural nets, they weren’t going to work because chips weren’t fast enough and there wasn’t enough data

 

out there. Now there’s both. The internet provided that as well. So I think it’s, you know, iterative, building on itself, and if there is some kind of, you know, incremental downturn or something, I don’t think it stops the train from the direction it’s going in.

 

Gyi Tsakalakis & Conrad Saam (59:39)

Any final questions? No. As usual, mind-blowing conversation with Dr. Cain. Thank you so much. If people want to have their minds blown as well, how do they learn more from Dr. Cain?

 

Cain Elliott (59:52)

Hit me up on Twitter at CainElliott, two L’s, two T’s. You know, you can DM me on there. Emails I really tend to only answer for business purposes, and who likes to spend more time in email than they have to? And otherwise, Filevine puts up a lot of material, I go to different events, check out Filevine.com and the stuff we’re building. Thank you guys for coming up.

 

Gyi Tsakalakis & Conrad Saam (1:00:15)

Dr. Cain Elliott, thank

 

you so much, that was amazing.

 

Gyi Tsakalakis (1:00:18)

And with that, dear listeners, another episode of Lunch Hour Legal Marketing comes to a close. If you are in DC or Boston or San Diego, we would love to see you in real life. If this is all brand new to you, please do like and subscribe and stay a part of the conversation. Our show depends on your input for topics, questions, and awesome reviews. Thank you.

 

Gyi and Conrad saying farewell from Lunch Hour Legal Marketing

Continue Reading