Latest / The Adults In The AI Room / The Holy Grail of AI - Alex Babin - Earley AI Podcast with Seth Earley & Chris Featherstone - Episode # 030

Transcript

The Holy Grail of AI - Alex Babin - Earley AI Podcast with Seth Earley & Chris Featherstone - Episode # 030 · The Adults In The AI Room

• 0:01
  SPEAKER_00: This is the Early AI Podcast, hosted by Seth Early and Chris Featherstone.
• 0:06
  SPEAKER_00: Join us as we delve deep into the passions, expertise, and experiences of thought leaders and practitioners to talk about what's possible in artificial intelligence.
• 0:15
  SPEAKER_00: Early AI Podcast is sponsored by Early Information Science, your digital transformation journey with the design and deployment innovative technology solutions.
• 0:27
  SPEAKER_02: Good morning, good afternoon, good evening.
• 0:29
  SPEAKER_02: Depending upon your time zone and when you're listening to this, welcome to another edition of our podcast, Early AI.
• 0:36
  SPEAKER_02: My name is Seth Early.
• 0:38
  SPEAKER_02: And I'm Chris Featherstone.
• 0:40
  SPEAKER_02: And today's guest is a tech savvy individual.
• 0:44
  SPEAKER_02: He leads a growing, high-performing team.
• 0:47
  SPEAKER_02: He stays on top of both what's going on in the legal world and the AI world.
• 0:52
  SPEAKER_02: His goal is to help legal professionals do the work better and faster.
• 0:57
  SPEAKER_02: I'm really excited about today's podcast.
• 1:00
  SPEAKER_02: This is going to be tremendous.
• 1:01
  SPEAKER_02: We have CEO and co-founder at Zero, Alex Babin.
• 1:05
  SPEAKER_01: Welcome to the show, Alex.
• 1:07
  SPEAKER_01: Thank you.
• 1:08
  SPEAKER_01: Thank you, Stat.
• 1:09
  SPEAKER_01: Thank you, Chris.
• 1:10
  SPEAKER_01: Pleasure to be here.
• 1:12
  SPEAKER_01: You give me too much credit.
• 1:14
  SPEAKER_01: Like no one can be on top of uh everything in AI right now.
• 1:17
  SPEAKER_01: It's moving slowly.
• 1:20
  SPEAKER_02: You're on top of something.
• 1:21
  SPEAKER_02: We'll leave it at all.
• 1:24
  SPEAKER_02: You're a one-eyed man in the kingdom of the blind.
• 1:26
  SPEAKER_02: That's how I so uh I wanted to start with a question about what are some of the popular misconceptions that you are coming across and about AI and AI in the legal field.
• 1:42
  SPEAKER_02: What's operating under that that's just not correct?
• 1:45
  SPEAKER_01: Yeah.
• 1:45
  SPEAKER_01: Uh it's such a big can of worms you just touched.
• 1:48
  SPEAKER_01: So I even scared of opening it up.
• 1:51
  SPEAKER_01: There are so many misconceptions and how people understand AI and what they think about AI.
• 1:56
  SPEAKER_01: Um, and not just in legal industry that we serve, but also in uh Fortune 1000 and insurance and management consulting everywhere.
• 2:05
  SPEAKER_01: So we're not focusing on legal alone.
• 2:07
  SPEAKER_01: Uh legal is just one vertical um that we're working with.
• 2:10
  SPEAKER_01: And those misconceptions, they are very universal.
• 2:14
  SPEAKER_01: Number one, the biggest one.
• 2:17
  SPEAKER_01: The the price goes to like AI can work out of the box.
• 2:22
  SPEAKER_01: No, it can't.
• 2:23
  SPEAKER_01: And the reason that that is a biggest misconception right now uh is that open AI did an amazing job showing everyone what AI is capable of.
• 2:34
  SPEAKER_01: And we as humans, we have this kind of ability to extrapolate what we see and what we can touch to what we believe those things can do and how they can do things.
• 2:44
  SPEAKER_01: So if you go to the chat GPT and you can ask it to do something for you, provide an information and so on and so forth, extrapolate, like, oh, it can do everything for me.
• 2:54
  SPEAKER_01: And the second misconception is like, okay, if I believe that it can do anything or everything that I require, I just need to throw a bunch of my data in it, and it's gonna be El Dorado, Holy Grail, Narnia, you name it.
• 3:10
  SPEAKER_01: That's the second biggest misconception uh that we see today.
• 3:14
  SPEAKER_01: I would focus on those two because those are the biggest ones, uh, especially if we talk about enterprise.
• 3:21
  SPEAKER_02: And what do you recommend?
• 3:23
  SPEAKER_02: Like, what's the best practice to what should uh our our um get our our our listeners walk away with?
• 3:30
  SPEAKER_02: So, what did what what do you do?
• 3:32
  SPEAKER_02: What's the argument?
• 3:32
  SPEAKER_02: What's the simple argument for dealing with that?
• 3:37
  SPEAKER_01: Um the simplest argument is uh to start not from the end result from the very beginning, but what is the ROI that you are looking for?
• 3:49
  SPEAKER_01: Like, what is your return on investment?
• 3:51
  SPEAKER_01: What do you want to do?
• 3:52
  SPEAKER_01: And then basically uh reconstruct or reverse engineer all the steps required to getting there.
• 4:00
  SPEAKER_01: And in many cases, you'll be surprised.
• 4:03
  SPEAKER_01: But in many cases, our clients, our partners, they see like, wait a second, we don't need all those fancy new stuff, uh, those fancy new tools.
• 4:13
  SPEAKER_01: What we what we want to achieve can be achieved with a much less effort with existing solutions if applied the right way.
• 4:22
  SPEAKER_01: And most of those um conversations starts with the data and how the data is structured, its ontology, and all this uh other elements of data because topics near and dear to my heart.
• 4:37
  SPEAKER_01: I know we we we're dealing with that all the time uh because garbage in, garbage out.
• 4:41
  SPEAKER_01: Uh, and if you want to have something uh effective and working um to solve your problems, you have to have all the right ingredients and components inside.
• 4:50
  SPEAKER_01: Just throwing Chat GPT on top of it will not solve the problem.
• 4:54
  SPEAKER_02: So, and say more about that.
• 4:56
  SPEAKER_02: Say more about that.
• 4:57
  SPEAKER_02: Why, why, what are organizations?
• 4:59
  SPEAKER_02: And I know I have my approaches and my theories, and and again, generative AI is generative, it means it's creating stuff, it's not a retrieval mechanism, but we still need our knowledge and we still need to be able to deal with that knowledge and and put guardrails around the generative AI so that it's giving accurate answers.
• 5:19
  SPEAKER_02: So, what are your thoughts around that?
• 5:21
  SPEAKER_02: How is that dealt with?
• 5:23
  SPEAKER_02: And I know it's a big problem.
• 5:24
  SPEAKER_01: Oh, yeah, it is.
• 5:25
  SPEAKER_01: It's such a clear solutions yet.
• 5:27
  SPEAKER_01: Yeah, it's such a big piece of conversation, can be the whole series around that, but uh let's try to summarize it into something.
• 5:33
  SPEAKER_01: So, first of all, generative AI is again pretty big concept, and a lot of people think about generative AI is like I asked uh a chat bot or chat GPT or uh any other solution.
• 5:44
  SPEAKER_01: I asked it something and it generates the uh content that never existed before, and I get it, right?
• 5:51
  SPEAKER_01: It's not correct.
• 5:53
  SPEAKER_01: So the content already is somewhere, right?
• 5:57
  SPEAKER_01: It's just like the combination, it's like uh chopping into pieces, putting it together.
• 6:01
  SPEAKER_01: There are so many uh components in there, but generative AI also works really well with the data and data retrieval as well.
• 6:08
  SPEAKER_01: So let's say you have a 50-page document, and how you and uh on a top level of your um ontology, you probably do what, like labeling and add metadata, what client it belongs to, what project it belongs to, but it's just scratching the surface.
• 6:23
  SPEAKER_01: What is it inside the document?
• 6:25
  SPEAKER_01: What are those uh clauses, due dates, components, like all of those elements inside?
• 6:32
  SPEAKER_01: Generative AI can actually generate the metadata by analyzing the document, extracting this important component, summarizing them into something searchable.
• 6:41
  SPEAKER_01: Of course, you can put it into um database later for search, but that also generates content, but it's not the way typically um people think about generative AI.
• 6:52
  SPEAKER_01: I ask it to uh create an image of a cat riding a bicycle on the moon, and here it is.
• 6:59
  SPEAKER_01: It's not everything that generative AI can do.
• 7:02
  SPEAKER_02: Right.
• 7:03
  SPEAKER_02: So you're saying that you can use generative AI to you can feed it your content and then leverage its ability to do things like entity extraction so you can have additional signals on on that uh piece of information.
• 7:17
  SPEAKER_02: Um, and then you can uh break it into components and you can understand what the elements are.
• 7:23
  SPEAKER_02: And then how how is that metadata used?
• 7:26
  SPEAKER_02: So, what what what's the next step to that?
• 7:28
  SPEAKER_01: That's a that's a very good question, Seth.
• 7:31
  SPEAKER_01: So that's one one like one piece of what General FAI can do for the data, right?
• 7:37
  SPEAKER_01: Like basically extracting the data.
• 7:39
  SPEAKER_01: Next layer of data or metadata that never existed but uh before, but is is being generated is the user-generated data, this feedback loop.
• 7:50
  SPEAKER_01: And actually, this piece is the most important one because let's say you have a value add product, um, an automation tool based on AI or whatever, whatever it is, and user interacts with the tool.
• 8:03
  SPEAKER_01: And during this interaction, new type of metadata being born, this kind of a reinforcement learning, and the machine uh now has more data to be fed back into the model, reinforced, and provide better and better results.
• 8:17
  SPEAKER_01: So it's kind of a flywheel.
• 8:18
  SPEAKER_01: Uh, and in this case, that's like how basically chat uh GPT was born out of uh GPT-3 because of the human in the loop, reinforcement learning with the human in the loop.
• 8:28
  SPEAKER_01: Otherwise, it's just a model, and we know what the difference between the GPT-3 and uh chat GPT, it's a conversational piece, and it required a lot of human work to interact with the model to get it to that level.
• 8:41
  SPEAKER_01: So that kind of a type of data, it's uh much smaller in volume, but the quality and the importance of that data is absolutely critical for the constant and like continuous improvements of uh AI solutions.
• 8:56
  SPEAKER_01: And this is the gold mine.
• 8:59
  SPEAKER_02: And the the thing that I have uh talked about in the past is being able to uh semantically deconstruct the utterance, right?
• 9:09
  SPEAKER_02: The thing people are asking for.
• 9:10
  SPEAKER_02: And certainly generative AI is very good at identifying intents.
• 9:14
  SPEAKER_02: And I would imagine it's going to be able to identify elements of those intents and entities in those intents, and then those can be used to boost the signal that uh you want to your particular product number, and then that metadata can be used in order to uh retrieve.
• 9:33
  SPEAKER_02: So we still have to have some retrieval mechanism, right?
• 9:36
  SPEAKER_02: We still have to have the ability to do that, and that's where I think the the balance is right now.
• 9:41
  SPEAKER_02: It's this is generative.
• 9:43
  SPEAKER_02: Now we also need to retrieve because we can't have it hallucinate, especially in uh a regulated industry, right?
• 9:50
  SPEAKER_01: Absolutely, yeah.
• 9:51
  SPEAKER_01: So um one way of uh, and that's another important topic you just touched on um hallucinations and the reliability of the output.
• 9:59
  SPEAKER_01: I would say it's probably the most critical component, right?
• 10:03
  SPEAKER_01: Because especially in regulated industries, in uh uh for large enterprises, being able to get the reliable output in one of most of the critical component.
• 10:16
  SPEAKER_01: Uh, and whenever I talk to people and they're like, okay, whatever, like it might have hallucinate.
• 10:22
  SPEAKER_01: I'm like, let me give you an example, folks.
• 10:24
  SPEAKER_01: Let's say you go to a bank and you put all your money in a bank, and you know that the my the bank is using AI models that can hallucinate.
• 10:35
  SPEAKER_01: So instead of your bank account, your money, there is a slight chance, just 10% chance of your money going into Seth's account or Chris's account.
• 10:46
  SPEAKER_01: Trust that bank, even if it's a slight possibility, no.
• 10:51
  SPEAKER_01: Um, and that's of course it's an exaggeration, but sometimes this reliability issue can uh can be critical for systems that are yes, not that kind of a dependent on this quality part.
• 11:04
  SPEAKER_01: I also use the um the the joke um about how important it is to actually have the source of data and verification because it reminds me of like uh when a uh two friends talking, and one asks, like, do you like Mozart, the music of Mozart?
• 11:21
  SPEAKER_01: It's like, no, I hate it.
• 11:23
  SPEAKER_01: Like, why?
• 11:24
  SPEAKER_01: Have you heard any of his masterpieces?
• 11:26
  SPEAKER_01: And the other, like, of course I did.
• 11:28
  SPEAKER_01: My roommate always sings in the in the showers, right?
• 11:31
  SPEAKER_01: So that's exactly what it is, it should be exactly the source of truth.
• 11:36
  SPEAKER_01: You can't sing something and say that this is the because a lot of that work is derivative, and derivative should be kind of a traced back to the original, right?
• 11:47
  SPEAKER_01: So it's pretty much Alex.
• 11:50
  SPEAKER_04: Let me ask you something because we have a number of of um you know managed services out there with predetermined models.
• 11:57
  SPEAKER_04: We've got now Gen I Gen AI, you know, that's generating you know content.
• 12:01
  SPEAKER_04: We've got all this this stuff that's out there.
• 12:04
  SPEAKER_04: At what point do you see?
• 12:05
  SPEAKER_04: Because I I feel like there's organizations that will pivot over, you know, with the pendulum swing over to hey, we need to do everything generative and we need to look at this from the the perspective of what are we doing, what are we doing?
• 12:18
  SPEAKER_04: And I feel like it's almost a um a little bit of paranoia moving you know them over, whereas, hey, listen, I've got my my models, I know exactly what the outcomes are, I've trained them, I've got moderation put on top of them, I've got governance put on top of them already, and I'm generating my output, right?
• 12:37
  SPEAKER_04: So where do you what's your what's your um your perspective and your recommendation centered around that we we still will use all these models and we have this generative piece, and it's just a matter of functionally putting it together to drive outcomes, right?
• 12:52
  SPEAKER_01: That's uh it's an amazing question, Chris.
• 12:54
  SPEAKER_01: So I call it sometimes I call it uh self-supporting hysteria because uh everyone is running this way, and then like everyone else, like, oh, I should be running this way.
• 13:04
  SPEAKER_01: Um and literally generative AI and AI in general um is the topic number one for everyone, including CEOs of the large organizations.
• 13:14
  SPEAKER_01: I've been having those conversations with Fortune 1000 companies, where uh even like six months ago, I was finding hard time sometimes to explain what large language model is, even to the technical folks.
• 13:28
  SPEAKER_01: Now, every CEO and their mom knows what uh large language model is, right?
• 13:34
  SPEAKER_01: Um, and in this case, it's again very, very important to trace it back to the original intent.
• 13:41
  SPEAKER_01: Why you need this?
• 13:42
  SPEAKER_01: What do you want to achieve with that?
• 13:44
  SPEAKER_01: And there is so many components that are there, they're they're not there yet.
• 13:49
  SPEAKER_01: Let's say ethical walls.
• 13:51
  SPEAKER_01: A lot of folks think like, okay, we'll just set up a model and it'll uh source the data in and it will give us the superpowers, and then they realize that the uh data proliferation happens inside the organization, and some sales manager in Nebraska gets all the financial statements of the Fortune 1000.
• 14:09
  SPEAKER_01: I'm exaggerating, of course, making things up, but it happens recently.
• 14:13
  SPEAKER_01: Samsung employees were using Chat GPT and putting the source code in of the chip design uh that they were working on asking Chat GPT to fix the errors in the code.
• 14:24
  SPEAKER_01: It's got exposed, like how crazy, more crazy that so it will going will is going to be going down a bit.
• 14:33
  SPEAKER_01: Yeah, uh, there's going to be normalization of this um kind of AI strategies for enterprises, but they will have to have components to set up AI the right way, and that's what actually we do.
• 14:44
  SPEAKER_01: We provide our clients with the infrastructure layer uh and orchestration layer to actually work with AI, both outside models and internal models, our models, and all the components required to do it effectively, safely, and without shooting themselves in both feet.
• 15:03
  SPEAKER_02: So let me ask a follow-on question.
• 15:06
  SPEAKER_02: When you said, you know, everybody in their mother knows what a large language model is, and I think they know at a certain level, but maybe you could just kind of dig in a little bit and say, what are the things that people need to be aware of of large language models?
• 15:18
  SPEAKER_02: What is, you know, what is your definition of a large language model?
• 15:21
  SPEAKER_02: And then what's important to understand about a large language model?
• 15:25
  SPEAKER_02: And you know, my premise on this is, you know, uh uh they're generalized, so they cover a broad swath of topics and and content, and you know, they're trained on enormous amounts of information, but they may not be specific enough for an industry, maybe a highly technical industry, because it won't necessarily understand or interpret highly technical terms or engineering terms.
• 15:46
  SPEAKER_02: And again, it's getting better all the time, so it's it's training on more and more content, but then there's company-specific IP.
• 15:53
  SPEAKER_02: You just mentioned Samsung, right?
• 15:55
  SPEAKER_02: That's where the differentiation is.
• 15:58
  SPEAKER_02: If we say everybody's going to use a standardized language model, that's great for efficiency, but it's not going to give you any competitive advantage.
• 16:06
  SPEAKER_02: We need differentiation for competitive advantage, right?
• 16:09
  SPEAKER_02: That's where we're doing this based on our knowledge assets, based on our data, our content, of customers, and so on.
• 16:16
  SPEAKER_02: So talk a little bit about what large language models are, and then this idea of uh specification or specificity for industry and company.
• 16:26
  SPEAKER_01: Yeah.
• 16:26
  SPEAKER_01: Um, again, such a big topic.
• 16:29
  SPEAKER_01: I'll start with a very basic thing.
• 16:30
  SPEAKER_01: So large language model is the type of a neural network that is being trained on an enormous amount of uh data, like big corpuses of data.
• 16:38
  SPEAKER_01: And uh one of the critical components in um uh in um LLM and how they've been developing is a transformer.
• 16:46
  SPEAKER_01: So uh it started with a paper published by uh Google folks, uh, which surprisingly they didn't take advantage of.
• 16:53
  SPEAKER_01: Uh, it was probably 2017, if I'm not mistaken, called Attention is all you need.
• 16:58
  SPEAKER_01: And transformers actually allowed to get those large language models to the next level because they were they existed before.
• 17:05
  SPEAKER_01: We've been uh working with BERT model uh for that, uh, and we're training it on the data.
• 17:11
  SPEAKER_01: Uh, but to get to that level of sophistication that we see it right now, the attention mechanism that not just predicting next token in the in the sequence, but also understand the importance of the elements inside that particular sequence of tokens, that would make a big difference.
• 17:29
  SPEAKER_01: And when that happened, apply to the uh large uh corpus of data, and I salute open AI folks who actually made a bet.
• 17:39
  SPEAKER_01: They said, Well, now we have all those components.
• 17:42
  SPEAKER_01: What if we throw in an enormous amount of data like internet?
• 17:46
  SPEAKER_01: And they did.
• 17:48
  SPEAKER_01: And I know that there uh they didn't expect those results at the very beginning themselves.
• 17:53
  SPEAKER_01: It wasn't basically an experiment, but the results, the results were there, and that's how this whole thing started.
• 18:00
  SPEAKER_01: So, large language model in a nutshell, uh, is a neural network that can can uh operate with a large amount of data, um, and if trained well enough with a specific mechanism, can provide uh specific results.
• 18:16
  SPEAKER_01: So the second piece here is large language model, like large, is a very broad and big concept concept.
• 18:26
  SPEAKER_01: So it's how long is a piece of a rope?
• 18:30
  SPEAKER_01: So for specific tasks, 100 million parameters is a pretty big model for other tasks, even 100 billion parameters would not be enough.
• 18:39
  SPEAKER_01: And I believe, and that's what we see with our clients, that it's always gonna be a combination that's there's not gonna be a one large language model that rule them all.
• 18:50
  SPEAKER_01: Never.
• 18:51
  SPEAKER_01: They're gonna be a combination of large language models, task-specific models, external models, internal models, and all the things between.
• 19:00
  SPEAKER_01: It's gonna be an ecosystem, it's already an ecosystem.
• 19:03
  SPEAKER_01: LLMs have become a commodity faster than any other technology I've ever seen.
• 19:11
  SPEAKER_01: Like four months in, and that's it.
• 19:13
  SPEAKER_01: Everyone has the uh large language model or claim they have.
• 19:17
  SPEAKER_01: So um, in this case, there's some models that can be trained on specific corpus of data, might be a small subset of data, but it's very, very high quality data with a lot of ontology uh efforts being put on top of it, making sure it's absolutely crystal clear and the high quality.
• 19:36
  SPEAKER_01: And that model will provide absolutely amazing results, even results, even uh GPT-3 model or cohere will not be able to match.
• 19:45
  SPEAKER_02: So I was actually going to ask that question about ontology.
• 19:49
  SPEAKER_02: So tell me how, of course, we should define ontology for our people who may not be as familiar, but it's but my definition is it's a knowledge scaffolding, it has conceptual relationships.
• 20:01
  SPEAKER_02: It's a it's imagine if you had all the taxonomies in an organization for different entities and elements, and then how all the relationships between those different entities and elements, including non-preferred terms and the thesaurus structure.
• 20:13
  SPEAKER_02: So, what it does is it helps you understand how things are conceptually related, it becomes that knowledge scaffolding for the organization, and it can contextualize uh information and knowledge.
• 20:25
  SPEAKER_02: And um, some of the ontologies that I've seen for um uh life sciences, you know, are too big.
• 20:31
  SPEAKER_02: The language model is too big.
• 20:33
  SPEAKER_02: So talk because it's it has things that are not relevant to the particular client, right?
• 20:38
  SPEAKER_02: So they have to be fine-tuned.
• 20:40
  SPEAKER_02: So talk about the relationship of ontology to language model.
• 20:43
  SPEAKER_01: That's that's my one of my favorite topics, actually.
• 20:45
  SPEAKER_01: Thank you for bringing this up.
• 20:46
  SPEAKER_01: So actually, I wanted to um to get it down to the level uh of a bit more simplification, right?
• 20:53
  SPEAKER_01: So it's important to start not from the ontology, but actually taxonomy.
• 21:00
  SPEAKER_01: And uh because those two things are, and I'm trying to kind of uh bring it up as uh as I was explaining it to my mom, for example, because I like this level of uh abstraction and explanation, because sometimes uh people might be like, uh, what are you just talking about?
• 21:15
  SPEAKER_01: Um, and I'm trying to bring it to the level that anyone can understand.
• 21:18
  SPEAKER_01: So in this case, for example, taxonomy is something that has a structure, and we're talking about the data, of course, like a catalog.
• 21:26
  SPEAKER_01: There is a uh relationship between the data, like if we're talking about, let's say, consumer goods, and I'm in manufacturer, I have uh men's clothes, I have uh women's clothes, and then pants, shirts, socks, and so on, so forth under each category.
• 21:40
  SPEAKER_01: That's a very simplistic our brain works this way.
• 21:44
  SPEAKER_01: We catalog things.
• 21:46
  SPEAKER_01: This is a taxonomy, it's a very simple but straightforward um way of doing things.
• 21:52
  SPEAKER_01: And we have our module, we have our own um our own model, it's much smaller than uh big models, but it's designed just.
• 22:00
  SPEAKER_01: For that classification and labeling of the data.
• 22:02
  SPEAKER_01: So let's say your email email comes into the inbox and it's not classified, it doesn't have its place in the taxonomy of your data.
• 22:11
  SPEAKER_01: But suddenly the model says, wait a second, this is a client A, project B, stage C, boom, you have a you have a taxonomy, and it it pieces of these pieces of the data have its place in the taxonomy of your data.
• 22:22
  SPEAKER_01: So we are matching the unstructured data to the um uh taxonomy of of your organization, right?
• 22:29
  SPEAKER_01: Because if you don't have that classification label in your taxonomy, you are probably, for example, you're not classifying by matter and uh project or client, like no system would do that.
• 22:40
  SPEAKER_01: So, anyways, this is a simple way of structuring the data, but then we go into um ontology, and this is a much more important concept here because if taxonomy answers the question what and how, ontology answers the question why, and this is where we get into the uh realm of models, large language models, because interconnectivity between the data and having this um meaning and context captured, and also labeled and basically this metadata around the data um surface, then you can apply uh models on top of it and get maximum power out of it.
• 23:24
  SPEAKER_01: So, I'll give you another example, okay, related to the email.
• 23:27
  SPEAKER_01: Let's say email comes into your inbox, taxonomy first, of course.
• 23:31
  SPEAKER_01: What is a client?
• 23:32
  SPEAKER_01: What is the project?
• 23:33
  SPEAKER_01: What is the stage?
• 23:34
  SPEAKER_01: All those surface level um metadata uh types.
• 23:39
  SPEAKER_01: But what is inside the email?
• 23:41
  SPEAKER_01: Is it a due date?
• 23:42
  SPEAKER_01: Is a client asking for a report?
• 23:45
  SPEAKER_01: What is this report?
• 23:46
  SPEAKER_01: Have I been doing those reports before?
• 23:48
  SPEAKER_01: Where the data for that report leaves, is it in Salesforce?
• 23:52
  SPEAKER_01: Is it in my document management system, content management system?
• 23:55
  SPEAKER_01: There is so many, it's such a kind of a door to Narnia.
• 23:58
  SPEAKER_01: You open it up and it just goes beyond belief.
• 24:02
  SPEAKER_01: So in this case, this is the next level.
• 24:05
  SPEAKER_01: And a lot of organizations, I would say most of them, uh, as far as I know, don't go beyond advanced taxonomy.
• 24:13
  SPEAKER_01: The holy grail, the Narnia of AI and large language model lives in the realm of the ontology.
• 24:21
  SPEAKER_02: Yep.
• 24:21
  SPEAKER_02: That's great.
• 24:22
  SPEAKER_02: So that's great.
• 24:23
  SPEAKER_02: And and and and when you were talking about enterprises having taxonomies, many of them don't have enterprise taxonomies, or they have 50 or 60 taxonomies, those do need to be reconciled, reconciled at the right level of granularity, and you do need to build those relationships to them because that is the contextualization of the data and the content.
• 24:44
  SPEAKER_04: Let me ask you two something, and then I want your example, Alex, because the other piece of this is you know, Seth, you brought up a good point.
• 24:50
  SPEAKER_04: There may be a bunch of variations.
• 24:53
  SPEAKER_04: What happens?
• 24:54
  SPEAKER_04: Because nobody gets it right the first time.
• 24:56
  SPEAKER_04: How do you embrace evolution of the taxonomy over time because of changing needs?
• 25:03
  SPEAKER_02: Absolutely.
• 25:04
  SPEAKER_02: Well, we always like to say that a taxonomy is a living, breathing thing.
• 25:08
  SPEAKER_02: When you're done with you're done with your taxonomy, when you're done with sales, you know, right, or manufacturing because there'll be new products, it'll be new opportunities, there'll be new topics, there'll be new issues, right?
• 25:18
  SPEAKER_02: And so we always have to have that change management and we have to be able to build domain models so that they're broad enough to be able to include extensions of the taxonomy.
• 25:29
  SPEAKER_02: The other thing is not every application cares about the detailed granularity, right?
• 25:33
  SPEAKER_02: I like to say there are two people, two types of people in this world, lumpers and splitters, right?
• 25:37
  SPEAKER_02: Or two types of taxonomists, lumpers and splitters, people that group things together and people that divide them into finer points.
• 25:44
  SPEAKER_02: And so the idea is that you're going to have some situations that require lumping and some situations that require splitting.
• 25:50
  SPEAKER_02: And the example I used to use is you know, a telecommunications firm had a routes to market.
• 25:55
  SPEAKER_02: On the website, they would only have about six categories of industries.
• 25:58
  SPEAKER_02: In the CRM, they'd have about 50 categories because there were different routes to market, different conferences, different, you know, titles, different is much deeper granularity.
• 26:08
  SPEAKER_02: So when you have those, those might be called industry, it might be the same vocabulary, but then you have to add additional metadata on the terminology to tell it what the right context is.
• 26:19
  SPEAKER_02: And that's something that has to be done with ontologies.
• 26:22
  SPEAKER_01: Yeah, I absolutely agree with you.
• 26:23
  SPEAKER_01: It's living breathing organism.
• 26:25
  SPEAKER_01: And if if it stops breathing and uh moving, it's dead.
• 26:30
  SPEAKER_01: So in this case, more data comes in, data creates through with the help of AI creates more data, and it's kind of a self-supporting flywheel.
• 26:38
  SPEAKER_01: Because if it's not providing kind of more data, more results, it's just stacked.
• 26:43
  SPEAKER_01: And in this case, it's not AI, it's uh rule-based something.
• 26:46
  SPEAKER_01: So um, I was going to bring an example of uh policy enforcement, for example, where two different ontologies uh that are not typically connected together should be connected to provide extra results.
• 27:01
  SPEAKER_01: So let's say you have something, the the the set of rules or guidelines uh or policies that you need to enforce inside your organization, and you have basically ontology of it, like what should be done, what can't be done, how it can be done, and so on so forth.
• 27:16
  SPEAKER_01: Typically, it's like 50 pages documents.
• 27:17
  SPEAKER_01: Uh, no one reads literally, I've never seen a person except for those who create those, um, read reading those documents, but they see it somewhere, and then there are other people who actually should be doing things according to those uh guidelines or the policies, and they have their own ontologies of the data they work with and so on and so forth, and those two worlds never or rarely overlap.
• 27:40
  SPEAKER_01: Of course, it depends on the on the industry if it's medical or very highly regulated, of course, it does overlap, but less regulated the industry becomes, you see those worlds go apart.
• 27:51
  SPEAKER_01: Uh, but AI can actually match those and bring those because understanding the ontology on one side, the uh the way it should be, and understanding the ontology of the data, and then matching those saying, wait a second, this is this should not go there, or you can't expose that here, and this information should go this way.
• 28:10
  SPEAKER_01: So, in this case, I think the AI um can be a stitching mechanism between those ontologies, it's not yet there, it's it's getting there.
• 28:22
  SPEAKER_01: One of our module modules of our engine actually does that.
• 28:25
  SPEAKER_01: Um, but that's where where to go.
• 28:28
  SPEAKER_01: Because if you if we talk about the uh let's say fortune 1000 organization, there would be thousands and thousands of different ontologies related to different data sets, different uh lines of business, different departments, you name it.
• 28:44
  SPEAKER_01: Right.
• 28:44
  SPEAKER_01: And those are not interconnected, and it's like basically trying to um uh boil the ocean.
• 28:50
  SPEAKER_01: If you start with fun, it's it's across the organization.
• 28:55
  SPEAKER_01: That's where AI can help.
• 28:56
  SPEAKER_04: But that's that's to me, that's why it's a science and not you know something solved.
• 29:01
  SPEAKER_04: Because to your point, you know, I may have a master taxonomy and an ontology to go with it that maps to a catalog.
• 29:09
  SPEAKER_04: But the beauty of that, what we're talking about, is that catalog may be multiple catalogs, and I may have to conjoin those catalogs in order to get, like you said, additional and or you know, better results.
• 29:22
  SPEAKER_04: And so we get into this notion of all the data governance for system people and process.
• 29:27
  SPEAKER_04: Then we actually you know filter that up to give access access to you know folks that are looking for it, especially the data science teams, and now they're doing feature development on top of that and adding new features on top of that and almost generating completely new, you know, born taxonomies and catalogs and everything else at the same time while they're testing everything.
• 29:48
  SPEAKER_01: You're right, you're right.
• 29:49
  SPEAKER_01: But uh, Chris, let's take it level higher.
• 29:53
  SPEAKER_01: So that's a foundation.
• 29:54
  SPEAKER_01: Without that, nothing would work.
• 29:57
  SPEAKER_01: But what actually the organizations need?
• 29:59
  SPEAKER_01: They need end-to-end solutions, yeah.
• 30:01
  SPEAKER_01: They need solutions to the problems.
• 30:03
  SPEAKER_01: So without having the right data, right ontology on top of taxonomy, it would be possible, but actually that's not what they're looking for.
• 30:10
  SPEAKER_01: They they're looking for solutions that will solve the problems, and that's actually another uh thing that doesn't exist on the market right now, or at least an enterprise scale.
• 30:19
  SPEAKER_01: There is so many fragmented solutions that work here and there.
• 30:24
  SPEAKER_01: Um, but there is no way for enterprises to say, hey, this is this is the layer I'm going to be applying on top of my data, this is the level of APIs, integrations, protocols that will help me to get those um modules or those solutions to work on top of my data, providing me with the end um end-to-end solution.
• 30:43
  SPEAKER_01: Um, and do it easily instead of like two years development.
• 30:48
  SPEAKER_01: And actually, we see this as the probably the biggest problem because it's a wild, wide west right now.
• 30:56
  SPEAKER_01: Data sources, uh uh lack of uh instruments and infrastructure and protocols and concepts, all of that together.
• 31:06
  SPEAKER_01: It's just such a big mess.
• 31:07
  SPEAKER_01: We're trying to bring in structure into it.
• 31:09
  SPEAKER_01: We have our own um orchestration and uh infrastructure layer, Hercules, that sits inside the security perimeter of an organization, prepares the data, uh helps to get the ontology in place, uh, interconnects data, to personalizes it.
• 31:25
  SPEAKER_01: And then organization can use external models, internal models, their own models, whatever it is, to get end-to-end solutions.
• 31:32
  SPEAKER_01: And they can actually build those end-to-end solutions quickly.
• 31:36
  SPEAKER_01: We call them SAMs, skill the AI modules.
• 31:38
  SPEAKER_01: And each SAM can provide a small specific subset of the value.
• 31:44
  SPEAKER_01: But if you multiply those modules by number of um uh use cases you apply to your organizations can immediately get to the next level, and this is I think this is the future of enterprise and how AI is going to be applied uh in enterprise.
• 31:59
  SPEAKER_01: Yeah, all time value.
• 32:01
  SPEAKER_01: Right.
• 32:01
  SPEAKER_02: What is the that's it's a really interesting word?
• 32:03
  SPEAKER_01: What does the S stand for again?
• 32:05
  SPEAKER_01: Skilled AI module.
• 32:06
  SPEAKER_01: The main component here is skill.
• 32:09
  SPEAKER_01: Skill is a is a small uh subset of like basically a function that uh this module can tap into, and there is a library of those skills, like for example, summarization is a skill, data extraction or entity extraction is a skill, and once you have those many, many, many skills, and they might be that's what we talked about, there might be a small model that provides high-level accuracy for a data entity extraction, and if a SAM that is more content and your and your reference data, right?
• 32:39
  SPEAKER_01: Yeah, so in this case, for example, there is a skill day module that needs to do um invoice processing and then connect it to internal system and put it in there so it can tap into multiple SAN as multiple skills.
• 32:53
  SPEAKER_01: One is entity extraction, another is API level connection to the system, summarization, and so on and so forth.
• 33:00
  SPEAKER_01: So this provides basic basically Lego blocks that build on top.
• 33:06
  SPEAKER_02: I wanted to ask a quick question about um the business and about your role.
• 33:11
  SPEAKER_02: And you know, what is keeping you up at night in this field?
• 33:14
  SPEAKER_02: I mean, it's moving very quickly.
• 33:16
  SPEAKER_02: Uh, there's a lot of unknowns, there's a lot of solutions that need to happen, the competition is moving quickly.
• 33:21
  SPEAKER_02: What are you concerned about?
• 33:23
  SPEAKER_01: Um, well, of course, the main concern for everyone on the market right now is where things go.
• 33:29
  SPEAKER_01: The comforting thing, though, no one knows.
• 33:32
  SPEAKER_01: So we'll all in the same boat together will figure it out.
• 33:36
  SPEAKER_01: Uh, I would not say we predicted what is going to be happening, but we knew that uh enterprises will require AI uh to be part of the business processes.
• 33:48
  SPEAKER_01: And we started about seven years ago now, uh, over just over seven years ago.
• 33:54
  SPEAKER_01: We knew that it is coming.
• 33:55
  SPEAKER_01: We've been building our building blocks.
• 33:58
  SPEAKER_01: It's like in a gold rush when uh most of the money is made by guys who sell shovels and picks and gins.
• 34:05
  SPEAKER_01: We were building shovel, picks, and gins factory for ourselves.
• 34:10
  SPEAKER_01: Uh, we didn't know where this gold rush would be coming, and like at least what form it will take.
• 34:15
  SPEAKER_01: And uh when in November uh OpenAI released Chat GPT, I was telling my co-founder saying, This is it, it's the best ever marketing done uh for us without us spending a dollar.
• 34:29
  SPEAKER_01: So in this case, um we we realized early on that everyone would need infrastructure, everyone would need end-to-end solutions, and if we if we work hard enough and long enough to create it preemptively, when the gold rush starts and everyone is rushing for AI, they will need those components no matter what.
• 34:52
  SPEAKER_01: And we're gonna be first supplier in the town to get those components in.
• 34:59
  SPEAKER_01: So we didn't know where yeah, we didn't know where their gold would be coming from.
• 35:03
  SPEAKER_01: Eventually, there was the large dynamic models and uh availability of AI and how it was uh popularized um across all industries, and now we have these components and focusing on Fortune 1000, right?
• 35:19
  SPEAKER_02: Right, and there still has to be human judgment in here, right?
• 35:22
  SPEAKER_02: There still have to be services to help organizations understand those processes, understand where they want to have an intervention, understand what data is required, you know, fine-tuning taxonomies and ontologies, you know, choosing the specific language models, maybe building their own language model, you know, based on the content they they have.
• 35:41
  SPEAKER_02: So there's still a lot of work that needs to be done to connect these pieces together and to be able to have those solutions.
• 35:47
  SPEAKER_02: So uh I certainly don't see my company going out of business.
• 35:51
  SPEAKER_01: Oh no, you guys are going to be not booming because actually that's what I started with.
• 35:56
  SPEAKER_01: The biggest misconception is that AI can work outside the box.
• 35:59
  SPEAKER_01: Right.
• 36:00
  SPEAKER_01: We can provide uh all the components and the building blocks, but we're not serviceable companies, serviceable companies.
• 36:05
  SPEAKER_01: We are uh technology company, and guys like you and uh Bain and Deloitte and others they provide actually the value add services helping organizations get to the ROI faster, right?
• 36:20
  SPEAKER_01: Right.
• 36:20
  SPEAKER_01: Like I see your goal and our goal is to get clients to ROI as fast as possible without mistakes in the in the process.
• 36:28
  SPEAKER_02: Totally agree, totally agree.
• 36:30
  SPEAKER_02: Accelerate that time to value, totally agree.
• 36:32
  SPEAKER_02: Let me ask you a quick uh question.
• 36:34
  SPEAKER_02: Um, what kinds of problems have your team solved recently?
• 36:38
  SPEAKER_02: So, can you give us some examples of you know either a client story or something you've done recently uh is notable?
• 36:45
  SPEAKER_01: Absolutely.
• 36:46
  SPEAKER_01: This example is the financial analyst um uh process where uh private equity notice is being processed.
• 36:56
  SPEAKER_01: When the private equity notice, like capital calls and distribution notices coming in, and the company, one of the business divisions, uh, is processing those notices, providing the uh fund management um services to the clients.
• 37:10
  SPEAKER_01: And those are thousands and thousands of documents, different documents uh from different uh sources coming in uh each week.
• 37:19
  SPEAKER_01: And there are 80 people basically opening up the PDF, looking for uh information, extracting it, and putting somewhere.
• 37:29
  SPEAKER_01: But here is the deal typically IDP can can deal with that pretty easily, right?
• 37:33
  SPEAKER_01: Like you open up the document, extract stuff, but there is a second component was never done before because there is a business logic after you extract the documents, they all interconnected, and that's where actually the the new approach with the large language model comes in, kicks in because you need to understand the logic, you don't just take the data and put it somewhere, right?
• 37:55
  SPEAKER_01: You understand where it comes came from, what the uh amount is, you do the math, you do the you apply business logic, and only then you provide it somewhere, and that logical block reasoning block was missing until now.
• 38:11
  SPEAKER_01: So, our example not just takes the data out beautifully with zero short learning, it also applies business logic, it mimics the cognitive process of that particular financial analyst.
• 38:23
  SPEAKER_02: So now they don't have to kind of do the snapking math and then provide some and the error rate just goes right, and so that sounds like that process needs to be defined, and you need to extract those rules or you need to extract that judgment.
• 38:37
  SPEAKER_01: You need to understand those rules, like it's those, those are sometimes not rules.
• 38:41
  SPEAKER_01: Sometimes people actually can't even explain how they do it, right?
• 38:44
  SPEAKER_01: Like, that's the way I do it, and another person might be doing it differently.
• 38:48
  SPEAKER_01: So, this reasoning module needs to really understand what is the end goal, what is the user wants to achieve, and reconstruct back like basically uh do their uh reconstruction of the whole process knowing how to replicate it.
• 39:02
  SPEAKER_01: And uh it was it was not possible before, now it is, and those examples they exist everywhere.
• 39:09
  SPEAKER_01: You you look at any organization in the world, there are thousands of processes like this where the processes of knowledge workers should be augmented, and that's where the power AI of AI comes in.
• 39:23
  SPEAKER_01: But uh it should be possible without the uh the large language models and uh the infrastructure layer that we provide.
• 39:30
  SPEAKER_04: But you mentioned something in the beginning too, Alex, where I think you said the orchestration of let's say these SAMS is super important to string them together in the right way to get the outputs, right?
• 39:40
  SPEAKER_04: And I'm assuming your technology stack does something like that.
• 39:43
  SPEAKER_04: Is that correct?
• 39:44
  SPEAKER_01: Yeah, okay, because like knowledge workers, like real humans, do not work in isolation.
• 39:51
  SPEAKER_01: Yeah, the same do the modules.
• 39:53
  SPEAKER_01: Modules need to talk, they need to send data, they need to share.
• 39:57
  SPEAKER_01: And uh that's where it's important because otherwise it's just gonna be isolated items uh without providing each other with important information.
• 40:06
  SPEAKER_01: Those are protocols, much more complicated process here.
• 40:10
  SPEAKER_04: I mean, it's super it's super easy to understand, let's say an ASR model talking to an NLP model, right?
• 40:15
  SPEAKER_04: But I think to your point, it's when you pull in, you know, like you say, a classification or a you know, or something like that in there so that you can actually get to you know resolution quicker or whatever the outcome is, but you could string however many of these together, I'm assuming, and then you know generate these, you know, these outputs.
• 40:35
  SPEAKER_01: So that's yeah, that's that's where we need it.
• 40:38
  SPEAKER_01: Chain of tasks, chain of events, chain of uh data points that are combined together.
• 40:43
  SPEAKER_01: Um, and that's where the real power is, and that's where the interpretes go.
• 40:48
  SPEAKER_01: That's great.
• 40:48
  SPEAKER_01: That's interesting.
• 40:49
  SPEAKER_02: Let me ask you this question: what are your what what are the biggest challenges for you today?
• 40:55
  SPEAKER_02: And then what excites you about the future?
• 40:57
  SPEAKER_01: Uh, sometimes those are the same things, right?
• 41:00
  SPEAKER_01: That's not future exclusive.
• 41:02
  SPEAKER_01: The things that are exciting us also makes us like uh sometimes uh be very careful.
• 41:09
  SPEAKER_01: I would say um the ethics of AI, right?
• 41:15
  SPEAKER_01: We are very laser focused on the AI that is a hundred percent reliable and accurate.
• 41:22
  SPEAKER_01: That's our nature.
• 41:23
  SPEAKER_01: We work with enterprises, our AI, our models need to be traceable, referenceable, and uh things like this.
• 41:31
  SPEAKER_01: Um, the advancements of AI, though, on the other side, go the way it's like literally it's an arm race.
• 41:39
  SPEAKER_01: It's good for us because better models exist outside, better for us.
• 41:43
  SPEAKER_01: We'll just provide our clients access to those models in a safe and secure way, but they are advancing so fast, it's a it's an arm race, bigger, faster, more powerful.
• 41:54
  SPEAKER_01: There is no controlled um or governance layer that kind of helps everyone understand what's going on.
• 42:04
  SPEAKER_01: So, uh, and because we do not understand what's going on, by we, I mean the market, everyone.
• 42:10
  SPEAKER_01: We don't understand what's going on and where it goes, it's less predictable, less controllable.
• 42:15
  SPEAKER_01: So I would say this is this is a shaky ground, the shaky um situation, but we'll collectively I believe we'll figure it out.
• 42:24
  SPEAKER_04: It feels like a bit like the transaction arms race of the 90s with databases.
• 42:29
  SPEAKER_04: Do you know what I mean?
• 42:30
  SPEAKER_04: Which, if you take it, take a step back, it's like, what in the hell is the real business value of that?
• 42:34
  SPEAKER_04: I can do four billion you know, um transactions in a single second for my business that has long-running transactions.
• 42:42
  SPEAKER_04: It doesn't make any sense.
• 42:44
  SPEAKER_02: Yeah, yeah, yeah.
• 42:45
  SPEAKER_02: You know, let me switch uh gears just a little bit.
• 42:48
  SPEAKER_02: Um, tell us about who you are, you know, where are you from?
• 42:53
  SPEAKER_02: How did you get to this space?
• 42:54
  SPEAKER_02: How did you get into AI?
• 42:56
  SPEAKER_02: Just give us a little uh rundown 47 hours.
• 43:00
  SPEAKER_04: Alex's origin story.
• 43:02
  SPEAKER_04: Yeah.
• 43:02
  SPEAKER_01: Um I've been into um AI for about, well, as we call it right now, my uh previous company was actually doing interactive video, and we were doing the uh trying to understand the elements on the video, uh automatically tracing it and then applying interactivity to it.
• 43:19
  SPEAKER_01: So it was pretty interesting um concept.
• 43:22
  SPEAKER_01: Uh no tools and no processing power existed at that time.
• 43:25
  SPEAKER_01: So um if you guys uh have seen recently, uh Facebook released the um uh segment uh segment all model, which does amazingly well what I was trying to achieve uh uh 12 years ago.
• 43:40
  SPEAKER_01: Um so and I was like super excited to see that it's uh being progressing now in the videos, every every on the photo, every element is being segmented.
• 43:49
  SPEAKER_01: Um, and that's where I got this AI bug.
• 43:52
  SPEAKER_01: And uh then we stuck my co founder and I we started the company and we wanted to build the cognitive automation.
• 44:00
  SPEAKER_01: Uh, for knowledge workers, and that's what we still do, just more tools, more powerful elements.
• 44:05
  SPEAKER_01: But at that point, we were focusing on summarization of the uh information, classification of the basically spend a lot of time building those building blocks that now pays off, and now everyone needs what we were trying to explain to people why it's cool.
• 44:20
  SPEAKER_01: Like five years ago, no one could understand, or at least two people understand.
• 44:24
  SPEAKER_01: Um, and uh, I'm I live in Silicon Valley, that's where our headquarters is.
• 44:30
  SPEAKER_01: Uh, father of two beautiful children and the husband of a beautiful wife.
• 44:33
  SPEAKER_01: I have a how old are your kids?
• 44:35
  SPEAKER_01: How old are your kids?
• 44:36
  SPEAKER_01: 16 and 10.
• 44:37
  SPEAKER_01: Oh, okay.
• 44:38
  SPEAKER_01: And uh, I have a crazy hobby.
• 44:40
  SPEAKER_01: I don't know anyone here in the valley who does the same thing that I do.
• 44:43
  SPEAKER_01: I do Japanese blacksmithing, Japanese swords, Japanese blacksmithing.
• 44:48
  SPEAKER_02: Wow, swords.
• 44:49
  SPEAKER_01: Yep, that is fascinating.
• 44:51
  SPEAKER_01: How long have you been doing that?
• 44:52
  SPEAKER_01: About five years now.
• 44:54
  SPEAKER_01: Wow, and I'm still pretty, pretty bad at it.
• 44:58
  SPEAKER_01: Are you what it's I'm getting better?
• 45:00
  SPEAKER_01: You do kendo, you do kendo as well.
• 45:02
  SPEAKER_01: Yeah, no, I don't do kendo, but uh yeah, but yeah, Kendo, I don't practice uh kendo sword maker, yep.
• 45:09
  SPEAKER_01: Uh to me it's just more of a meditation rather than uh craft.
• 45:13
  SPEAKER_01: Yes, um, but anyways, um we're here headquarters in Campbell, Silicon Valley, uh right between uh San Jose and um uh Palo Alto.
• 45:24
  SPEAKER_02: And uh and and and let me ask you another kind of reflective question.
• 45:31
  SPEAKER_02: If you were able to go back in time and give yourself some advice uh as you were getting out of college, what would that advice be?
• 45:44
  SPEAKER_01: Interesting.
• 45:45
  SPEAKER_01: Well, uh this go back in time, uh back to the future type of uh advice are they can actually backfire because what we know right now might not be that helpful if we go back, right?
• 46:00
  SPEAKER_02: But just anything, anything about your history or anything?
• 46:03
  SPEAKER_01: I would say I would say very simplistically, if I go back to 1999, I would tell myself buy Apple stock.
• 46:11
  SPEAKER_01: Okay, yeah, fair enough.
• 46:13
  SPEAKER_01: Forget about everything else, you'll figure it out.
• 46:15
  SPEAKER_01: Just buy Apple stock as much as you can.
• 46:18
  SPEAKER_02: So um so tell me where people can find you.
• 46:22
  SPEAKER_01: Um zero system zero systems.com, uh that's our website, or at LinkedIn, um uh zero systems.
• 46:30
  SPEAKER_01: So Alex Bavin on LinkedIn as well.
• 46:33
  SPEAKER_02: Great, great.
• 46:34
  SPEAKER_02: And uh, you know, I think this has been wonderful.
• 46:37
  SPEAKER_02: I'm really excited.
• 46:38
  SPEAKER_02: I love when people tie together the things that I'm passionate about, such as ontologies and knowledge, and knowledge work, and taxonomies, and large language models and AI and cognitive systems.
• 46:50
  SPEAKER_02: So this has been really wonderful.
• 46:52
  SPEAKER_02: Thank you, Alex, so much for being here.
• 46:54
  SPEAKER_02: Thank you, Seth.
• 46:55
  SPEAKER_04: I'm just I'm just more excited that I now know a sword maker.
• 47:00
  SPEAKER_04: Right?
• 47:01
  SPEAKER_04: I mean, AI said, let's uh let's dig in.
• 47:03
  SPEAKER_04: Maybe we should do another another uh hour just on the sword making, but no, that's super you know, you know how my wife puts it.
• 47:09
  SPEAKER_01: Uh she says, Well, when AI will take over the world, finally, that's my wife's word.
• 47:14
  SPEAKER_01: Um so when AI will take over the world and we go back to Stone Age, your hobby is gonna be a very important profession.
• 47:22
  SPEAKER_04: Exactly, exactly.
• 47:23
  SPEAKER_04: That's yeah, I actually fabricate, so I agree with you 100%.
• 47:27
  SPEAKER_04: So yeah, it's awesome.
• 47:29
  SPEAKER_02: I want to also thank uh our audience for listening.
• 47:33
  SPEAKER_02: And if you've enjoyed this podcast, if you learned something, certainly share it, tell somebody about it.
• 47:39
  SPEAKER_02: And again, thank you, Alex.
• 47:41
  SPEAKER_02: I really appreciate your being here.
• 47:44
  SPEAKER_02: It's been a pleasure.
• 47:45
  SPEAKER_04: Yep.
• 47:45
  SPEAKER_04: Thank you.
• 47:46
  SPEAKER_04: All feedback is well.
• 47:47
  SPEAKER_04: Thanks, Alex.
• 47:48
  SPEAKER_04: Appreciate the time.
• 47:49
  SPEAKER_02: Okay, this has been another episode of Early AI, and we will see you next time.
• 47:56
  SPEAKER_04: Thanks, Ed.
• 47:57
  SPEAKER_04: Thanks, Alex.
• 47:58
  SPEAKER_04: Take care.
• 48:00
  SPEAKER_00: Thank you for joining us on another deep dive into AI innovation.
• 48:04
  SPEAKER_00: Join us next time when we introduce another expert and discuss how to maximize AI in your world.
• 48:10
  SPEAKER_00: Early AI Podcast is sponsored by Early Information Science, CMS Wire, and Marketing AI Institute.
• 48:16
  SPEAKER_00: To learn more about us, visit Early.com.com.
• 48:23
  SPEAKER_00: See you next time.