On this episode of The Geek in Review, hosts Marlene Gebauer and Greg Lambert explore innovations in legal search with Paulina Grnarova and Yannic Kilcher, co-founders of DeepJudge. This semantic search engine for legal documents leverages proprietary AI developed by experts with backgrounds from Google and academic AI research.

As PhDs from ETH Zurich, Grnarova and Kilcher recognized lawyers needed better access to institutional knowledge rather than constantly reinventing the wheel. DeepJudge moves beyond traditional keyword searches to a deeper integration of search and generative AI models like GPT-3. Partnerships provide financial support and key insights – advisors include execs from Recommind and Kira Systems while collaborations with law firms shape real-world product capabilities.

Discussing product development, Kilcher explains connecting search to language models allows generating summaries grounded in internal data without ethical or security risks of training individual models. Grnarova finds the core problem of connecting users to full knowledge translates universally across firms, though notes larger US firms devote more resources to knowledge management and data science teams.

When asked about the future of AI, Grnarova expresses excitement for AI and humans enhancing each other rather than replacing human roles. Kilcher predicts continued growth in model scale and capability, requiring innovations to sustain rapid progress. They aim to leverage academic research and industry experience to build AI that augments, not displaces, professionals.

DeepJudge stands out for its co-founder expertise and proprietary AI enabling semantic search to tap into institutional knowledge. Instead of reinventing the wheel, lawyers can find relevant precedents and background facts at their fingertips. As Kilcher states, competitive advantage lies in accumulated know-how – their technology surfaces this asset. The future of DeepJudge lies in combining search and generative models for greater insights.

Listen on mobile platforms:  ⁠Apple Podcasts⁠ |  ⁠Spotify⁠ | YouTube (NEW!)


Contact DeepJudge: info@deepjudge.ai⁠

Contact Us: 

Twitter: ⁠⁠⁠⁠⁠@gebauerm⁠⁠⁠⁠⁠, or ⁠⁠⁠⁠⁠@glambert⁠⁠⁠⁠⁠

Threads: @glambertpod or @gebauerm66

Email: geekinreviewpodcast@gmail.com

Music: ⁠⁠⁠⁠⁠Jerry David DeCicca⁠⁠⁠⁠


Marlene Gebauer 0:07
Welcome to The Geek in Review. The podcast focused on innovative and creative ideas in the legal profession. I’m Marlene Gebauer,

Greg Lambert 0:14
And I’m Greg Lambert. Marlene we did a little bit of traveling last week. I think I went to Chicago, you went to New York. I did. What else did you cover in New York?

Marlene Gebauer 0:22
Well, I actually went to New York. We were doing listening sessions with many of our attorney groups, which I think was was really excellent, got a lot of great feedback and also caught up with some of the folks at Lex fusion who were also having a meeting there. So that was quite nice, too. We got to see our friend Zack and a few other people there.

Greg Lambert 0:43
Nice nice. You also caught some music Did

Marlene Gebauer 0:45
I did I want to see here now concert in one of my favorite places Asbury Park and got to see lots and lots of bands right out there on the beach. So got to catch up with some friends and so all good. Oh, good. Good.

Greg Lambert 0:59
Oh, yeah. I did the legal value network. Meeting conference. It was awesome. And met up with with a lot of friends. Toby, Ryan and Zena are close friends as as well as had a good panel. We actually did the closing panel. I caught a little bit of music too, except it was kind of the other way around in that we we all went out for karaoke. And we we did live band karaoke. And that was really cool.

Marlene Gebauer 1:31
Is it really music though? Or is it just sounds

Greg Lambert 1:36
when you had enough drinks is music sound

Marlene Gebauer 1:38
cool? Sounds great. Yes, exactly. Right.

Greg Lambert 1:41
Before we get into this week’s guest, I did want to give a shout out to Karen Osterley. From PLI for launch here and for launching PLI is new the practicing law librarian podcast. And that’s practicing with an s the way that practicing law Institute fails. Now it’s a limited series available on pli.edu. And any of its available there or any of the podcast platforms. Great content so far that I’ve seen, including some good friends like Cornell, Winston, Elaine neck, and more. So it’s a great spot for information professionals to stay connected and learn from one another. So great job, Karen.

Marlene Gebauer 2:28
Today we’re focusing on search and our guests are experts in this regard. I’m just nervous about it. While Nita good Narva, CEO and founder and Yannic Kilcher, CTO and Founder of DeepJudge a semantic search engine for legal documents Polina and Yannick, welcome to The Geek in Review.

Yannic Kilcher 2:45
Thanks for having us.

Paulina Grnarova 2:45
Hi, Marlene. Hi, Greg. Thanks a lot for having us. It’s great to be here.

Marlene Gebauer 2:49
So the first question I wanted to ask as as ex Googlers. How did you experience? How did your experience at Google shaped the development of deep judges AI models? What specific expertise did you gain there?

Paulina Grnarova 3:01
Yeah, I think our backgrounds play a big role into what models we’re using, but also the problems that we choose to solve. I think one thing that is kind of interesting about the DeepJudge team is that all three co founders actually have a pretty technical background. So all of us worked at Google for quite some time. And we actually worked on the Google Search and Google Assistant. In parallel, we also did a PhD in artificial intelligence at ETH Zurich, in Switzerland, where we are based. And we very much focused on semantic understanding of texts, and generative models. And we were super lucky that during this period, when we were both at at Google and at ETH, kind of experienced the wave of what deep learning can do when it comes to semantic understanding of text. And we experienced this as part of academia, but also as part of industry. And we recognize the inflection point. And we decided to take it to the legal domain where there’s obviously a lot of text and a lot of potential to optimize things. And it’s really the perfect domain, because a lot of that text is written by humans, for humans. And we kind of with the, with the knowledge of what we have been building before we analyze the market, and we decided to actually go towards the Holy Grail, which is connecting users to the full institutional knowledge inside the law firm. It was super interesting to us. Because you know, when you think about it, the two competitive power of a law firm is really the entire knowledge that they have accumulated over the years. And when you hire a big law firm or a lawyer from a big law firm, you really expect them to leverage all of that and build on top of all of the previous experiences. But it was really shocking for us to learn that what actually happens instead is that people would send PTI emails kind of inquiring across the entire firm. Has anyone ever done this before? Where this can be? You’re nodding here. Everybody says this when, you know, this decider of claws? Or have you ever had a case like this before, instead of really looking internally, so we decided to actually do this. And now do search again. But now in a more internal fashion, where we really connect the users, and we enable all their institutional knowledge to be accessible at their fingerprints.

Greg Lambert 5:27
Yeah, that that, unfortunately, is not a shock to any of us knowledge management, people that have been trying for 25 years to, to get them to clean up their information and being able to, you know, use best practices, get get, reuse, dock, reuse. The documents that you’ve already created stops starting over from scratch. So yeah, welcome to The Club.

Marlene Gebauer 5:52
So Yannic, what was your experience?

Yannic Kilcher 5:56
I think we just learned building large things are things that serve a large amount of people. Google needs to respond within 200 milliseconds. So there’s really no, no room for no diddling around and not making things that scale really well. And I mean, a law firm isn’t the whole world, but they do have usually terabytes of data. So being able to effectively handle that is something that we certainly learned in, in this industry, environment, much more than in academia.

Greg Lambert 6:29
And I just found out today that I think tomorrow is Google’s 25th anniversary of Huawei. And yeah, it’s hard. It’s hard to believe it’s only been 25 years, and it’s hard to believe it’s been 25 years. So it’s kind of this weird weirdness there. So, you know, we DeepJudge itself uses the proprietary AI technology to enable the semantic search for legal documents. And so, you know, we talked about knowledge management, that in how we’ve been working with things such as, you know, traditional keyword search, prior versions of semantic search, natural language search and a price or, you know, there’s been a hodgepodge of different types of searching available. So, either one of you, do you mind talking about, now, how the technology works with DeepJudge? And how it makes it superior than then all those traditional searches? Yeah, we

Yannic Kilcher 7:29
avoid calling it semantic search by now, because it’s kind of an overused term. Now, what do you want to call it? Now, if we just call it search, search on semantic search, nowadays, kind of means vector search. And that has its advantages, especially with the newer generation of, of language models, you can get really great embeddings for text, and then you can search by sort of concepts inside that text. But it’s only it’s useful for some stuff to nuts. Why, you know, if you if you go into developing this, the first demo, you’ll see is like, you’re looking at getting search Amazon product reviews. And that’s great, because Amazon product reviews, they’re like, it’s broken, it’s great, or I never got it or something like this. And so this is it, this is all semantics, right? But if you want to find like a name of a person, you really want that name, you don’t want a semantically similar name, you want your that name, or some time period or so you need to essentially take what’s good about yo tried and true enterprise search, and, and all of that, and you need to augment it with these newer technologies, rather than doing one or the other or something of that. So that’s how the best search engines work in this what we do.

Greg Lambert 8:49
So what would be some examples of if I’m, if I’m sitting down and using DeepJudge? To look at my data? Well, what would I need to know differently? It’s kind of, it’s kind of like we’ve talked a lot this past year about the difference between searching and prompting. And so how, how do I go about digging through the information and using using something like DeepJudge? You know, more properly.

Yannic Kilcher 9:20
That’s really interesting, because we feel people have gotten so used to Google, that if you show them a search engine that work there, they’re sort of they’re like, oh, yeah, of course. So it’s more like the search that exists right now, inside of, of law firms really doesn’t work. Really, you have to either be an A super expert at combining really big operator chains. Or you just have to scroll through hundreds of search results to find one that you want. So a a search engine where you just type in what you want to find and it actually gives that to you. That’s what your experience or your experience is Very much like Google. But internally. So that means we take a step back, and we try to understand what are you actually looking for. And in that you don’t exactly have to hit, you know, the exact words that you type, it’s not that important that they exactly match that first of all allows you to be more precise, meaning that all of the kind of crap doesn’t get returned. And second of all, it means that you’re also going to find stuff that someone else may be used some different words to describe. So it’s not exactly prompting, it’s more like you type what you want to find. And we take care of not only finding it, but also mean there is on top of that, there’s a lot of crap in and of itself. And by crap, we mean stuff that you maybe don’t want to find, like tons of Red Line documents, tons of, you know, like emails where it’s just, Hey, you want to go for lunch, okay. And you want to save all of this, and you want to have it searchable. But when you search for something you ideally want to search engines that understands, okay, this is a redline, that’s not important right? Now, these 10 documents are actually the same document. So I’ll only show it to you wants to all of this kind of things is things that we would handle.

Greg Lambert 11:16
So all the hard work kind of done in the index and on the back end,

Yannic Kilcher 11:21
correct? Yes. Yeah, there is a obviously, we’re deeply ill impressed by people who want to organize their data. And we know that especially knowledge management, that is a big topic, how do we organize our knowledge, and that certainly helped. But usually that’s done for the knowledge collection, which is super cool. But then you still have all the the actual matter documents, and all the actual communication and everything that goes in there. And there’s just, it’s just a Herculean task to keep that organized. So you’d rather have a good search engine.

Marlene Gebauer 11:59
So you have an impressive group of advisers, including former executives from recommened and Kira systems, how have they helped deep judges product development and go to market strategy? Yeah, I

Paulina Grnarova 12:11
think we we benefit quite a lot from having them on board. So we have young pussycat who was the founder and CTO for a combined up until the acquisition by open text, and he has built a search engine for law firms himself. So leveraging his knowledge is really really valuable to us. And we work very closely with him when developing the product. We also have Steve Urbanski, who was the previous chief strategy officer at Cura, who is helping us really greatly with go to market deciding which problems to tackle, how to present them and bring them to the world. And now, obviously, with the with the US expansion. Yeah, actually, that’s, I

Marlene Gebauer 12:53
think how we were connected was through steep

Paulina Grnarova 12:55

Greg Lambert 12:57
Well, in addition to the advisory board that you have, I know that you also have collaborated with a number of organizations. And so Pauline, could you share some of the examples of some of these successful projects, or partnerships that, that you’ve used, or that have used DeepJudge? As solutions?

Paulina Grnarova 13:18
Yes, certainly. Um, as a startup, we, it takes a village to build a business. So we leverage a lot of partnerships. And I think they they can roughly fall into three different categories. The first one are kind of a lot of financial support. So the ecosystem of startups in Switzerland is very lively. So we’ve been lucky to receive a lot of awards, a lot of grants partner up with a lot of institutions. And that is great for both kind of traction, but also as financial support. The second type of partnership is more strategic type. So we are part of the eth AI center, we’re in fact, the first pin have to come out of the eths Center, which is one of the largest hubs for AI research worldwide. And through that, we leverage a lot of the talent and kind of information flow. Another strategic partnership we have is with Microsoft, for example. We are part of their exclusive data service program for startups, which means that we get early access to all of their models, we get access to cloud architects cloud computing, so all of it, all of which is really beneficial to us. And probably the most valuable partnership type that we have is with customers themselves. So when we decided to build knowledge search, we build it very closely together with the law firms that we are working with. And I think that was super important because we learned a lot of different features and things that we added on top, some of which Yannic mentioned before. So for instance, we learned that obviously as a search, it’s super important to bring the relevant result at the top but then also what is super important is how do you order And I send presents the results themselves. So this comes in the form of recognizing what is a red line or recognizing what is an email. So maybe you can very easily remove that, or organizing duplicates in your duplicates version, because you don’t want to see the same result on the first three pages just because it’s slightly different. And another big one that we actually learned to our partnerships with law firms is the ability to recognize whether a result is relevant or not very quickly, not just by the content, but also by by all the metadata surrounding it. And while there’s a lot of metadata already in the document management system in place, and other places, we learned that even though when you upload a document, and you need to select what type of document it is, almost no one really does it, it’s always the default type that is docked. And that’s not really helpful. So what we do is we have our own classifiers, that add a lot of metadata to the collection. So for instance, if you have even just very few labels by the very good souls that actually select the correct document type, we can extend that to the entire collection, or we can classify out of the box what type of document it is it is, is it a contract? Is it something else? If it is a contract, what type of contract it is, we have a classifier of over I think 400 different classes that tells you the type of contract this document belongs in. And this is very good for narrowing down the results and being really specific when you’re searching for something.

Greg Lambert 16:29
I’m just curious, do you also does it learn as people use the system so that if the certain searches, returned certain documents, and then within that results, there’s kind of favorites that people tend to always go to does that? Does that change the search results? As time goes by,

Yannic Kilcher 16:49
as time goes by, but it’s if you’re a public search engine, like Google, or Bing or so you got, I don’t know how many million impressions a day, in a in a law firm, even in a big law firm, you maybe get a few 100 Or a few 1000 a day. So it takes a while to accumulate enough data to be able to robustly learn from it. So we we are our strategy is to mix metrics of quantitative assessment such as that, like, what do users click on? What do they actually look at? When are they happy? When are they unhappy? And more qualitative measures where we actually go and sit with the people and ask them about, you know, what they’re looking for why and what they’re looking to achieve, and then improve by that

Marlene Gebauer 17:37
can they actually kind of see how it’s working, like behind the scenes in terms of like, how it’s how it classifies the way it classifies are the results that it’s given.

Yannic Kilcher 17:47
Um, some of it is certainly certainly visible. We expose obviously, all the metadata that we infer about documents, a lot of a lot of the internals of search engines are also hidden, meaning just the consultation after relevances, and so on, they would be not too meaningful. Or people would start to read wrong things into the numbers where they’re like we had, we had too many numbers at one point, we had, like the relevance as a number, and then people will be like, No, this result. That’s not 85 That’s like 99. And that’s really important, whether it’s higher or lower than something else. So we’re now trying to sort of give more visual indication of relevance. But yeah, so we try to expose as much as it as is useful

Marlene Gebauer 18:39
Now. Yannic and Paulina, you know, your team’s expertise in in, you know, AI research is, is quite impressive how you know, how, you know, what is your manner in terms of staying in the forefront of AI advancements? And how does this expertise benefit deep judges users,

Yannic Kilcher 18:58
I think we obviously keep very tight with our upbringings, let’s say from academia, and from the big industry labs. So we still are very connected with people in both places. So I think we keep up to date naturally through diffusion in there, there is quite big buzzers in this field. And it’s hard to, I mean, it’s easy to get overwhelmed. But if something good, big is happening, it’s also hard to miss it. So we make we make big efforts to stay up to date on anything that’s relevant to us so we can get it to the to the customers in the fastest manner possible. Think the current research environment is really special in that there has never been such a small gap between what’s state of the art and what you can actually put into productive use. We try to make the maximum benefit out of that. So whenever there is anything new, we immediately try it out. And we immediately look for a way if it is actually working some research results to bring it into into the product or apply it.

Marlene Gebauer 20:09
And that’s something I guess it’s sort of disseminated or or, you know, it is the team is encouraged to go seek that that knowledge out as well.

Yannic Kilcher 20:17
Yeah, we have our internal internal channels are going with people discussing research and, and finding things.

Paulina Grnarova 20:26
It also helps that that Yannic is kind of at the center of all of it with his YouTube channel. So, he actually has a YouTube channel discussing working

Marlene Gebauer 20:34
on YouTube channel.

Greg Lambert 20:38
Yanaka get revved up, because we’re going to talk about the YouTube channel here here in a minute. So, but Pauline, I’ve read, that you’re growing the team internationally that you’ve hired, Christina Godfried, and you’ve got another key hire that you’re going to be announcing soon in the US. So how are you finding the reception in law firms in departments? Is there a difference between the US and other countries that you’re finding?

Paulina Grnarova 21:12
That’s a great question, it’s really interesting that law firms started reaching out to us. And we figured that the same problem translates pretty much everywhere. Similarly, as you were nodding, everybody is kind of telling us, we’ve been waiting for a tool that really works for kind of when it comes to search and knowledge management for yours. So in terms of the problems, we are not finding any differences, really, I think there are differences in the in the law firms themselves, obviously, the largest law firms in the US are significantly larger, larger than the ones we have seen in Switzerland. And it comes with with some otter differences in terms of the structure. So for instance, there are specific people and even teams that are dealing with with knowledge management, and even people responsible for search. I think another interesting aspect is that a lot of the large US firms, they would have also data scientists themselves. They’re playing around and building different models, which is also super interesting for us, because then we get to kind of work together and build other things on top of the search and on top of the data that we expose to them. Yeah, so, so pretty excited about the US expansion.

Greg Lambert 22:25
I wonder, are we using our data scientists correctly? How are you? Are you finding that that that law firms that have data scientist are putting them to use and in a good way? I guess, let me let me ask it that way? Yeah,

Paulina Grnarova 22:41
I was gonna say, We’ve definitely seen some some great examples of people playing around with the models, which helps them kind of understand the limitations of the models and the things that you can do with the models. And I think that that knowledge is pretty helpful also, for them to know what they can build internally themselves and what they should look for in a vendor. So yeah,

Marlene Gebauer 23:05
yeah. Could you provide insights into any recent developments or updates to deep judges products that the users can look forward to?

Yannic Kilcher 23:13
I’m sure, we have lots of updates and things that we’re working on. Currently, I think the obviously the whole world is buzzing about generative AI. And even the, you know, the business world also took notice since last year or so. And in the search engine, as such, we do use language models, but not the type of generative ChatGPT models, because we have to process in terabytes of data. And we have to search really quickly through all of that. So having a model that goes line by line is not not going to cut it for us. But what we are doing is what’s called generative search, where you connect the search engine to a generative model. And so we’re ourselves not unlike an LLM company, although we can train such models, but we connect these models to the internal data. So the has some advantages. Everyone’s now looking to use generative models in, in legal. But if you just use them as such, they’ll they’ll have hallucinations. They’ll just tell you something, even if they don’t hallucinate, you have no clue where it’s coming from, right. And some one route is to start to like train a large language model on internal data, where you say, Hey, we have all this great data, can we train something on it? And there’s a bit of a misconception of what that will do? What it will do at least what we know from research is it’s going to adopt your style of writing like a bit of your your language, Nick’s and your, your way of phrasing, sentences and so on. It’s still going to have hallucinations because you just can’t get around the fact that it’s a statistical model and And there’s other problems like, if there’s a new document tomorrow, what are you going to do? retrain it? Like? Yeah, Bloomberg spent millions on their model, it’s already out of date. Right? So and, and even other problems like how do you know that it’s not generating something from a matter that where you are where there’s an ethical wall between you and that matter at the law firm is not going to train an individual LLM for each user. So all of these kinds of problems, you can elegantly solve by actually connecting a language, a generative language model to a search engine nowadays, that’s broadly known under the term of retrieval augment to generation, we go further than that we have a much deeper integration with the search engine and generative models. So think we’re pretty excited about that. And we can work with any generative model like once we we train on ourselves, or commercial ones, or even the ones that law firms train, if they really want to have their style, they’ll train one. And it’ll still be super useful. Because with the search engine, you, you can see where all the information is coming from. And it’s coming really from your own documents. And that goes back to our core philosophy that we mentioned at the beginning, we believe the internal knowledge is the competitive edge that really a firm has, across, you know, a firm versus 100 individual lawyers, the combined power and the internal knowledge is the competitive advantage. So just makes sense to not only connect humans to that, but also to connect generative AI to that,

Greg Lambert 26:38
as you as you’ve been implementing these within firms and working with, with the people inside the firms, has there been any kind of surprise or something that that you’ve learned that the product can do that you may not have realized until you actually put it into action? So

Yannic Kilcher 27:00
we’re noticing that some some people nowadays, whenever they see a text field, they they immediately think is ChatGPT, and they start to write, you know, we’ve we visited a search engine, but then they start to write, please give me a document that has, right, and then we’re like, you

Greg Lambert 27:19
are a third year associate.

Yannic Kilcher 27:22
But then there are others. There are others who so the the generative search we’re currently in, this is in more of an experimental phase, right? And we’re, we’re really getting into the workflows of people. And we’ve noticed some people have actually become super billet well versed at prompt engineering, like you’d be astonished of, you know, how well people immediately sit in front of it and tell it right, exactly this though, right? You are this, this is the background and so on. And we’re like, wow, it’s been what, half a year since these tools have been available. And people have already adopted, you know, very sophisticated techniques around them. So that must mean great. There is there is something that people get out of these productively, which is very cool. Very cool to see.

Greg Lambert 28:11
So Paulina, you have in you mentioned both of you have a PhD? I believe it? Is it in AI or machine learning or both that? How do you define it?

Paulina Grnarova 28:22
I guess, I don’t know, machine learning?

Marlene Gebauer 28:26
Well, for one, or what does it say in the certificate?

Greg Lambert 28:31
To says PhD?

Paulina Grnarova 28:33
In general, in artificial intelligence, and then yeah, okay, different than

Greg Lambert 28:37
kale. So how has your experience, the academic experience influenced how you’ve developed the product and the direction that you’re taking? DeepJudge?

Paulina Grnarova 28:50
Yeah, as I said, at the beginning, I think our backgrounds really shaped our journey, including what type of problems we pick to solve, and I think we we picked a really difficult one, just because we can solve it technically, I think, you know, it’s, it’s a tough problem, because you are dealing with hundreds of millions of data, if not more, you have to be really fast and really performant when it comes to ordering up the results. But on top of that, you’re dealing with a lot of security aspects of it. All right, you’re you cannot really send the data to any third parties have to respect ethical walls and all of that. So it’s a combining all of this is it’s a really difficult problem. And I think we we decided to tackle it because of a little bit because of our backgrounds. Yeah, yeah.

Marlene Gebauer 29:37
And you have a PhD as well in AI. Yes. Correct. Yeah. You and you also have a very strong presence on YouTube. In AI research. How does your role as a content creator complement your work at DeepJudge? And do these aspects intersect in any way?

Yannic Kilcher 29:57
Oh, well, they both Rob mice. Sleep?

Marlene Gebauer 30:01
Well as part as podcasters. We know what you’re talking about. It’s a labor of love. Right?

Yannic Kilcher 30:07
Yeah, it’s it’s been, it’s been a, it’s been initially a hobby of mine to make few videos about research topics, I never expected your people to listen to someone explain very heavy mathy research for 45 minutes at a time. But, you know, here we are, apparently, apparently people like, or some people do, and it gives me so I think generally being connected to people who are interested and fascinated by research is just a great thing, overall soul. And I think that benefits also the company to a great degree, meaning that we’re quite well connected in all of these networks, the research network, the industry, network, and so on. And yeah, the YouTube channel also forces me a little bit to stay up to date on everything that’s happening in the machine learning world. And

Greg Lambert 31:01
yeah, and I just looked it up, just to make sure I get the numbers, right. 227,000 subscribers on your YouTube channel, so not too shabby. Although

Yannic Kilcher 31:13
the community has been in large part, like, extremely amazing. For all the things people say about internet comments, I think this is this has been one of the most uplifting group of people I’ve I’ve ever encountered.

Greg Lambert 31:26
Sao Paulo, we’ve gotten to the part of our interview where we ask all of our guests that are crystal ball question. So I’ll start with you and then Yannic, I’ll get asked you as well. What do you think are going to be some changes or challenges in the industry? Especially the future of AI development over the next two to five years? Paulina, you want to start us off?

Paulina Grnarova 31:54
That’s a that’s a difficult question. I think, you know, I’m excited about the combination of Ai plus humans, I think a lot of the talk these days is AI versus humans, and what one would replace the other. And I think I really think in the future that a lot of humans will be supplemented by an AI that can understand a lot more than just kind of text or visuals, but pretty much a lot of things and supplements the work of a lot of knowledge workers in in a lot of different industries,

Greg Lambert 32:27
and Yannic what do you what do you see in your crystal ball and in the next, you know, we say two to five years, I really, we should probably change this from the two to five months. And

Yannic Kilcher 32:37
I think we’ll get really set up by chat interfaces that don’t really make sense. Because everyone’s not putting them everywhere,

Marlene Gebauer 32:47
are already fed up with chat interfaces that seven two and 815. I said any target uh, you have

Yannic Kilcher 32:54
the trend of the trend of larger and larger models continuing probably for the foreseeable future. Because it’s, it’s such a, it’s like, if you have a good movie, you make a sequel, because you know, it’s going to make money. For now we have a pretty good idea of where it will land if we just make these models bigger and bigger and bigger and throw more money and resources at them. So that’s the current plan in the, in the mid term, long term future someone will have to come up with some sort of a smart new invention to keep that going. But that’s not not right now.

Marlene Gebauer 33:31
Very well, I like to ask remains

to be seen. So, Paulina Grnarova, CEO and founder and Yannic Kilcher, CTO and founder of DeepJudge. Thank you both so much for taking for thank you both so much for talking to us today.

Yannic Kilcher 33:46
Thank you very much.

Marlene Gebauer 33:47
Thank you as well. And of course, thanks to all of you, our listeners for taking the time to listen to The Geek in Review podcast. If you enjoy the show, share it with a colleague, we’d love to hear from you. So reach out to us on social media, I can be found at LinkedIn at gave our M on Twitter and at M gave our 66 on threads

Greg Lambert 34:04
and I can be reached on LinkedIn or Ed glambertpod on Twitter, or glambertpod on threads. So Alena, and Yannick. If someone wanted to learn more about DeepJudge DeepJudge or reach out, where can you be found online?

Paulina Grnarova 34:21
Our websites deepjudge.ai We’re also on LinkedIn. The company profile is DeepJudge and me myself I can be found on LinkedIn ads. bottino goodnight, over

Marlene Gebauer 34:31
so much better than I said it. To you to

Yannic Kilcher 34:37
type attention is all you need and to YouTube.

Marlene Gebauer 34:41
And as always, the music you hear is from Jerry David DeCicca Thank you, Jerry.

Greg Lambert 34:46
Thanks, Jerry. Marlene, do you notice I got rid of the telephone?

Marlene Gebauer 34:49
I did. The telephone was gone. I was gonna have something I felt like there was something missing and we just jumped into Jerry. There so Okay,

Greg Lambert 34:57
there we go. Alright, Jerry. Thanks, Jerry. All right, Jana. You can follow him. Thank

Marlene Gebauer 35:00
you again. Thank you, by

Unknown Speaker 35:08
the way Hey, hey welcome back on your store back devils back at the devils back

Transcribed by https://otter.ai