So, I'm really excited to bring you this interview with the CEO of MindsDB, Jorge Torres. Jorge and I talk about open source, we talk about enterprise artificial intelligence, we talk about RAG, we talk about a whole bunch of really cool stuff. So, check out the

interview right now. So, um, we we got connected, and uh, I I saw MindsDB. I I do a lot of consulting, um, on on this side, and mostly for enterprise companies. And a lot of these companies are asking me, like, how how do I apply AI to my

business? Uh, obviously, ChatGPT just as kind of an assistant is is great. Um, but I think one of the most valuable use cases that I've seen is being able to, uh, essentially ask questions about your entire knowledge base as an enterprise company. And so that that's what MindsDB

does. But I want to ask you, how did you arrive at this problem? Like, what did you identify, uh, and then how did you look to approach solving it? Yeah, that that is a great question. Um, so maybe I can I can walk you back into what we

were inspired by and what we wanted to build when we started, and and how that fits into the story of today. Um, so, are you into science fiction?

Absolutely. I think I've seen every sci-fi movie there is. Yes! Yeah, so so we're big fans here as well,

And we felt that maybe the world would be great if humans had minds they could collaborate with. We felt that AI was only going to get better, like machine learning was going to transform more into applicable AI, and it was going

to continue to improve. We felt that maybe the consumer aspects of this, while interesting for us, were less important than the enterprise, which we all depend on. At the end, we all consume products from companies, we all work for companies, we all make our living from

enterprises. They could leverage this capability. So we felt at some point there's going to be enterprise AGI, and we wanted to be an important part of the solution. When we thought of enterprise AGI, we felt that if you think about the

definition of AGI, where you have three major milestones they have to accomplish: one is human-level cognition, the other one is the ability to learn, and the last one is having autonomy. Enterprise AGI seems to have an *our* interpretation of that,

which is: autonomy, at the end, has to be controlled autonomy. You don't want something that is fully autonomous just making all the decisions of a company, at least not in the foreseeable future. You want it to be a collaboration; you want it to have

constraints. So then you can focus a lot on: can you accomplish human-level cognition? Can you accomplish a system that can learn to solve problems? And to accomplish human-level cognition, you have to do a few things. You have to first have systems that can

understand their environment. And the environment of enterprises is all boiled down into data systems. Like everything that happens within the enterprise goes into some sort of information system or data store of some capacity. Like, if you think of every application, it has to store the data

somewhere. Um, and that's kind of like the perception layer for the enterprise. The second one is you need to be able to communicate with these things. Um, and the way that people traditionally communicate with this is with query languages. Um, it is a problem of

data being everywhere, so you can do it in a kind of partitioned way. So we thought, okay, to solve this problem, it has to be in a federated way. And to actually do human-level cognition, you have to do it in a way that the systems

learn how to communicate in natural language. This capability, we saw, was going to improve over time. So we felt that at some point, if we build a good connection tissue into the information systems of the organization, eventually they will merge into: Can you communicate to them in

plain English? And now we are in that stage where it's true. Now, the following thing is: Can you ask these things to perform a task, and they can kind of stay focused on that task and solve it within the domains of their

autonomy? And it felt that many of the basic level tasks that you want to do are discovery of data. You want to be able to find some information that exists somewhere within your data stores and use that information to make an action. And it is natural for humans

to start there and then transform some of those questions or follow-up questions into some form of analytics that you want to do. Uh, that's kind of like the basic thing that you can do with a system that has some human-level communication capabilities over the entire data that

you may have in the organization. And eventually, those systems will evolve to the point where you can say, 'Okay, well, now that I've seen this data, I want to automate some part of the process or automate some workflows.' And that's kind of like a lot

of the things that people do with agents today. Um, and nonetheless, all of those require having this layer that can talk with a user or a system, understand the requirement, and get the data. So that's kind of like where we are,

and to get to the journey, we first built a distributed federated query engine. Um, and then on top of that, we started building more of the cognitive capabilities that we felt were important to accomplish this task. Um, when a company approaches Myb to use you, um, what is the problem that

they're able to verbalize, or maybe they aren't, but they kind of have this issue or this thing that they need to accomplish? What are they telling you when they first come to you? And then what are they discovering after they discover more of what Myb is able to

do? Yeah, so many companies suffer from this natural consequence of existing for some time: the more a company exists, the more data they collect. Um, and this data sometimes is the same as not having data, and they also understand that this data

could be a competitive advantage if it could be used and leveraged in ways that they can't necessarily do today. So that is kind of like the common denominator. I feel like what happened in 2022 and 2023-24 paved the

way for most of the decision-makers that were not necessarily technical to understand, "Oh wow, this capability is probably ready for me to actually leverage." Like, no one before 2024, even 2023, would think that the CEO will wake up and think, "Oh, what's going

on with my AI?" But that's true for every single organization today. So where most organizations are at this moment is, "Okay, AI is something I can go into ChatGPT, log in, and I can play with. How can I do that for me? How can I

do that for not just me, but for every single person in my organization, for every single team?" Now, today, a few years later, since GPT 3.5 onwards, a lot of people are making this decision based on a framework of risk.

So they have, of course, many considerations that they have to take into account. One is, "Do I build systems, do I make systems that are to improve the performance of my teams internally?" That has very little risk. "Or do I make systems that are customer-facing,

and if something goes wrong, it will happen?" So most organizations are a little bit hesitant to take the leap on that last part. But for the problems that are about optimizing workflows that exist today that could be highly optimized if people had better

information, uh, every time they make a decision, then that's that's very appealing for people. And the reason why this is very appealing is because they understand right now that there are inefficiencies, inefficiencies that, in theory, they could be solving, um, and

many of these inefficiencies have to do with, "I know that the answer to my question is somewhere in there." Um, and if I could find it, then I could perform my task without having to reinvent the wheel every single time that I do it. And the alternative to not having

AI is just that you don't find the information. Uh, and we all experience this, even at, like, if I were to look into my email, for example, and ask her a question like, "Hey, what are the conversations that I have with Matthew over the past six

months that have to do with, like, the interview?" I can't, like, Gmail just doesn't give me that. Like, it is impossible to find information, and you're talking about, like, uh, Gmail, which is like a data— now, now you can imagine how it is for

like all, every single organization that doesn't have this capability. So all of a sudden, the possibility of being able to bubble this information, and bubble it from the point of view of, uh, it can be natural for users to ask these questions and get this data, it's very appealing.

And that that's where, where most of the market is today for, for us. Yeah. Yeah, I see. Um, I'm going to oversimplify it greatly, but like the chat with your doc's knowledge, you know, context, signal, whatever you want to call it, that was really the most

Impressive first real-world use case that I saw with AI. And so, Minds essentially does that, but on steroids, supercharged for large organizations, huge data sets. Um, where do you see real-world AI, actual applicable AI for the enterprise going in the mid and the long...

run? So, if we have "chat with your docs" now, and it doesn't need to be specific to Minds, but what should company leaders start to think about with regards to solutions that might be coming in the next two to three years, and then, you know, five to ten years to help them just be

more productive, be more efficient, whatever that metric they're aiming for is? Yeah. And I think that many of these intuitions are guided by how people work today with other humans. Um, but then when, when there's, like, a person, they're limited by how much data they

can process at a given time. Um, so one good example of this is, you, you clearly start, like, okay, well, "chat with your documents." Um, most of the companies, organizations, or most of the company's data is not just in documents. It lives on... Let's say you have Salesforce. So you

have, like, some data in Salesforce, you have some data on your emails, you have some data on your system of record, your, your PC. Like, if you think about it, they have so many systems. So being able to do that across the board is like the next

um, kind of type of solution that people want. And we, we enable, uh, to a great extent, that. But once you enable this capability, it is phenomenal to, to see how... if you enable people with the capability to find the data they want, it is natural for them to then ask...

questions that involve transforming that data into information that they need for the problem they need to ask. So this is now, the data doesn't exist in that shape, but you're asking the systems to transform it in a way that is useful for your problem. So this is kind

of like crossing the line into, from search and discovery, to analytics. Traditionally, analytics was solved with, like, BI tools and all this tooling that was invented as a solution for kind of like bubbling this information, but still has many problems. Like, in order for you to create a dashboard, you

need to kind of be versatile in SQL. And even if you build like many dashboards, it becomes also a problem of finding the dashboard that has the information that you actually want. And then if you optimize for a few dashboards, there's no way that you can cut all the

information that you need in one single way. So, conversational systems change the game dramatically because it starts with like, first, let me find the data I want, and then let me transform this data in a way that is useful for my use case. And that's specific to the question that

each individual may have at the time. So the next evolution that we're seeing today, and we'll continue to see, is the merger between search and analytics from the conversational point of view. And if you keep on playing this line of thought, usually what happens

to a lot of people is like, okay, well, I found the data. I don't want to do this all the time. Like, I don't want to be asking every single morning what is this, you know? Like, you start wanting the system to automate things for you. So you

want to start setting those instructions in a more conversational way. And I think that that's an endeavor that some organizations are thinking about today. But the way that we see it is that if you enable one capability, that itself unlocks a demand for another capability because

that's the way that we humans have been doing things for a while. Um, and all the way to, you know, like how can you automate some of the tasks that people do so they don't have to do them all the time? Um, then it translates into how do you

optimize the decisions that are made on those tasks? Like how can you, for example, introduce things that before were the Holy Grail of machine learning, etc.? We're like, okay, before you make this decision, forecast the possible outcomes and make sure that you make a decision

based on, you know, some likelihood that is based on the data that exists, not on some intuition. So you said something interesting. Um, companies have all of this data, and one part of it is retrieving the data, and then it's manipulating the data as you see

fit. Um, I want to ask you something because this is something I've been thinking about a lot recently, I should say. Um, like there's this notion of just-in-time UI, and it kind of feels like MindDB is doing that today and then

also very well positioned to do that even more so in the future. So when, either as a consumer or an enterprise company, you're thinking about what needs to be delivered to that person in that moment, and so you have systems of record like Salesforce, like HubSpot,

and again, a super oversimplification, but essentially it's a UI layer on top of a database, right? And so if you think about it like that, in the future, if you can just ask an agent through Minds to pull the data and present it to me or you in

a specific way, where do the Salesforces of the world or the HubSpots of the world, what do they do? Where do they lie in this stack? And I'm going to ask you a subsequent question after that, but let's start there. Like what, this notion of

just-in-time UI, where do all these other system of record interfaces belong in that stack in the future? Yeah, that's a great question. And I think that the way that this will likely evolve, in my opinion, is this just-in-time UI will still require some widgets

for you to interact with whatever information you need at the moment, uh, in the most articulate way. So many times, like question, question there, question back is not the best way for you to provide information. Sometimes you want to, like, click a few things and then that's it. So many of these systems will

evolve into, 'Hey, you have a question.' Then the answer may be you get a table and you get to explore the table like it's Google Sheets or Excel. Um, and some of the things, you're like creating a ticket, and you may not want to make all the elements of this

conversational, even though you may have, like, a co-pilot next to it that will help you populate most of the information that you need. But what will happen is that these kinds of systems that today we understand as, like, the SaaS applications

will transform into applications, and then there will be some centralized platforms that allow you to kind of pull these widgets, ideally in a way that is, um, kind of like one decentralized standard that you can ensure every player gets to use.

in a similar way. Kind of like the way that it happened on the web. HTTP made that possible. Um, so ultimately, I think that it will evolve in that direction, and therefore each of the players gets to truly understand the domain, uh, and how this

interface continues to be very smart, but also not tedious. And what you can do is reduce the number of steps. And for them, the goal is that they are continuing to be the state of record. Like when it comes to customers, they're going to be highly

incentivized to retain that data and also enrich it with data that may be useful for you. Um, and what will be the responsibility for systems outside of that is to ensure that you can deal with the reality that data will continue to be federated. And if you

want to build a solution that is holistic, you will have to be able to talk to all the systems and guarantee that you interact with them through, again, like a very simple UI that provides complexity as the user demands it but continues to be a

single entry point. Like, go. The truth is that even today, if you look at ChatGPT, people start asking questions, all of a sudden start creating demands for, like, 'Okay, search can be something similar to this.' And it just changed the

way that people through search yet, uh, right now they start figuring out what is the best way to show the results in a way that are actionable. Um, so that's that's the way that I that I think it will evolve. It will be more more like widgets that will get brought

into, you made by those providers of, the holders of the state of record, or that information. Um, and then you will continue to to to interact with those systems and therefore there will be highly incentivized to provide the best experiences. But now there will be

conversation-centric. Okay. Yeah, I'm I'm I keep going back and forth on whether I believe there is a place for those types of companies in the distant future, uh, you know, 10 10 years from now. Uh, I, if everything can be just predicted, what

you need in that moment proactively, it's it's hard for me to really envision where an interface company would would be. But I want to actually change topics. I I I want to, one of the reasons I was actually really excited to talk to you

today, and and really glad that we met to begin with, is because you have, uh, an awesome open-source version and kind of are very, you've contributed a lot to the open-source side of things. So I want to ask you first, first tell me, tell me

about the open-source MindsDB. What's the difference between that and and your enterprise version? And then I want to talk a little bit about the business of having, you know, you you're a for-profit company, but you also have this open-source arm. So talk to me a little bit

about what is the open source version right now? Yeah, um, so have you have you heard of the, like, the old Rick G, uh, principles of systems complexity? Probably no one has heard about this, but, um, no, it has a law that says that it is impossible to make a very complicated

system from scratch unless you start from a very simple system. Um, and and this law is that if you try to, like, engineer a system, like, end-to-end that is very complex, um, you're likely going to end up with a system that doesn't solve the problem, and you're going to

spend the rest of your life kind of patching that system to get to to the complexity that you want. Um, if you think of something like AGI or enterprise AGI, etc., um, it it will be almost one-to-one, in our opinion, um, a mistake if you try to build

a system that has all the capabilities that you think a system like that should have. And therefore, you want to start with what is the simplest part of a problem that you can solve and then build upon that. In our mind, the the simplest but most important part of the

problem to start with was being able to talk to all kinds of data. And and we felt that that was the fundamental part of of our thesis: was data in enterprises lives in so many different places, and if you cannot tap into this data, you all, what you have is generic AI. You don't

have, like, AI that is that is for that enterprise. So we built, um, this federated query engine, um, with the principles of, it has to be AI-centric. It will allow anything that is machine learning to talk to this data. And given that we wanted to

understanding we were going in the right direction, we decided to open-source it. And for us, open source is this beautiful way of creating software in which you can even start by putting a journey with like, 'This is my idea.' And you will find individuals that would be like,

'Ah, this is interesting. Why don't you do it this way?' Or, 'I tried it and it actually sucks.' And they don't have any hesitation, and you don't owe them anything. They don't owe anything back to you. So when they tell you something is...

...because, uh, there's no, like, common interest in trying to make something that sounds cool work. Um, and we couldn't exist without that. So today, if you ask me, I don't know how people build software that is not open source. It's just like so hard for me to think about.

You build something and then you convince someone that doesn't get to... to... to reward in any other way than just kind of like, 'I'm going to try your product.' When it's open source, at least they feel like they belong because they really do belong to the solution. But

when you're building on a... on a vacuum, I don't... I don't really know how is it that you find people to try your product when it isn't perfect. It... it really means that you understand exactly, uh, the market very, very well and this solution is kind

of already predetermined... Um, what if I can interrupt for one second? Why... why do you... why do you think, and maybe, maybe this is just my perception, but why do you think AI companies in particular have been so, um, open and and willing to have an open-source version of their products?

Is it AI in particular? Have I just not noticed other industries open-sourcing as frequently as AI? What are your thoughts there? I think since AI is incomplete, it's very humbling to kind of say, 'Hey, this is my approach, try it.'

I mean, I don't know if this is the best solution, but it is *one* solution. And I think that is what is appealing to the AI community. It's like, 'Man, this solution will continue to improve, and you can try it. Even more, you can

give feedback, and if you want to, you can even help me fix it.' And that's beautiful. Other industries have problems that now are a little bit more robust, but at times they weren't. For example, when people were creating databases, people were like,

'Well, there's the Oracle way, but there might be other ways.' For example, like Guy from MongoDB, they're like, 'Maybe for the web, it's better to store information in JSON.' And I don't really know if this is going to be the

best solution, but why don't you try it?' And I think that led to what is... When the problem is already well-defined, then you actually want a solution that just kind of works. For example, when Snowflake came into play, they were like, 'Data warehousing? I mean, you

can try to go the open-source way, but what people want is to be able to search this information really fast.' So, as long as the solution is faster than what you can find, then you can do it. So I think that when the market is already perfected, then it lends itself to

less open source, but when you're figuring things out, um, then then it's it's a good alternative. Another one is where you have a player that is the only player in the market for a very, very important solution, and then you feel that unless there is

competition, uh, it's not going to land well. Operating systems were one of those. Like, you you had like only closed-source operating systems, and the guys from, uh, the BSD, like FreeBSD and, you know, then Linux, etc., they were like, "Well, there needs to be an alternative, uh, because"

you cannot operate a computer with a, an operating system." So when that happens, um, then also there's an open-source movement, which you saw with, like, OpenAI and and what the guys from Facebook did, which was beautiful. And M, etc. Otherwise, uh, it would be too scary to leave OpenAI

for for in the in the past, not so much currently, or or currently they were like the closed player. And and some people thought, "This is such an incredible capability. It just kind of sucks that there's not a, a way for for the general public to see what's going on"

and to see where it's going to land and to kind of get a say on how the systems are built, to participate on it." And luckily, uh, the market kind of like, again, showed that when that happens, the open-source community just kind of like somewhere

and is going to say, "Well, let me build it and kind of give it away so I can leverage many more things from the open source. Like, let me figure out how this is going to play." And for Facebook, it's playing out pretty well. They're building

all this tooling around the models, and now they're using it for Meta, so then their ecosystem becomes the standard. Yeah, I I I like it. Um, I really, obviously, I've said it a million times on the channel, I really appreciate everything Meta is doing. Obviously, I also

appreciate what Minds is doing because, you know, open source, the ability to download something, play around with it, have a bunch of eyeballs on it, it just hardens the security, it evolves the product much more quickly. I I'm a big fan, and it's interesting, I'll I'll share

something I don't think I've mentioned on the channel. Um, I I would, I didn't really understand open source for a very long time. I was, you know, I I had formerly created a SaaS company. Um, I worked for closed-source companies. It was just never a thing we talked about,

and then AI, as I said, it was kind of just the standard, it seemed, and I really liked it. Then I started understanding it better and and trying to figure out, okay, in what ways does open source benefit everybody? And and I have a much better understanding now, so I I find that

Real World AI Use Cases, Open-Source, Hallucinations (MindsDB CEO Interview)

Artificial Intelligence (AI), Open Source, Generative Art, AI Art, Futurism, ChatGPT, Large Language Models (LLM), Machine Learning, Technology, Coding, Tutorials, AI News, and more

** Exclusive Pinecone Partner **

Matthew Berman

Interview with Jorge Torres, CEO of MindsDB, discussing open source, enterprise AI, and RAG.

Discussion on enterprise companies seeking ways to apply AI to their businesses, particularly using AI to query their knowledge base.

Jorge explains MindsDB's origin and inspiration, drawing parallels with the concept of collaborative AI minds from science fiction.

Focus on enterprise AI due to its potential impact on businesses and the economy.

Explanation of Enterprise AGI and its milestones: controlled autonomy, human-level cognition, and learning.

MindsDB's approach to achieving human-level cognition in enterprise AI by connecting to data systems, using query languages, and natural language communication.

The importance of data systems as the perception layer for enterprises and the role of query languages in communication.

The evolution of communication with AI systems from structured queries to natural language and the focus on data discovery and analytics.

Discussion on the challenges faced by companies with growing data sets and how MindsDB helps them leverage this data as a competitive advantage.

The shift in perspective towards AI adoption among decision-makers, driven by the rise of tools like ChatGPT.

Companies are now exploring how to implement AI solutions across their organizations and teams, focusing on optimizing workflows and improving decision-making.

Addressing inefficiencies by enabling access to information within existing data stores, similar to the concept of 'chat with your docs'.

Discussion on the future of real-world AI applications in the enterprise, including just-in-time UI and the role of traditional systems of record.

Moving beyond document-based AI to integrating with various data systems like Salesforce and email platforms.

The merging of search and analytics through conversational interfaces, allowing users to transform data into actionable information.

The potential impact of AI-driven just-in-time UI on the future of SaaS applications and the role of centralized platforms.

Discussion on MindsDB's open-source approach, the benefits of open source for AI development, and the business model around it.

The importance of starting with a simple system and iteratively building complexity, as applied to the development of MindsDB's federated query engine.

The value of community feedback and collaboration in open-source development, allowing for rapid iteration and improvement.

Why AI companies are particularly inclined towards open source, due to the evolving nature of the field and the need for community involvement.

Discussion on the business model of open-source companies and how MindsDB balances its open-source offerings with enterprise services.

The importance of building a strong community and achieving product-community fit before developing enterprise offerings.

MindsDB's enterprise offerings focus on security, observability, and scalability, addressing the specific needs and requirements of enterprise clients.

The focus on delivering value to a wide audience through open source while also providing valuable enterprise services for those who need them.

Addressing the challenge of AI hallucinations in enterprise settings and the strategies MindsDB employs to mitigate risks.

The importance of considering the risk of AI mistakes versus the existing alternatives and the role of human accountability in AI-driven processes.

Techniques for reducing hallucinations, including retrieving evidence and ensuring the information provided to the LLM is sufficient and well-structured.