Transcript 0:00 [upbeat music] Welcome back to the Middle Tech podcast. This is Evan Knowles. I am doing a solo recording this week, just like Logan did on the previous week, on a topic that I have been deep in the weeds of. 0:14 You know, last week, Logan went into, uh, nuclear energy and all the learnings that he's been, uh, pulling off the internet and doing, um, finding his own research. 0:25 Um, I have been doing the same in the artificial intelligence field, and, um, I've been building an artificial intelligence company that, you know, I'll talk about, uh, here on this podcast later. 0:36 But I wanted to do a quick episode to walk everybody through, you know, what's happening in AI right now, what is AI, what are the origins of it, uh, what are the components of it so you know how it works behind the scenes, and how it's being applied and used in everyday real life, how it's gonna affect society, and key concepts that I think everybody should know so they can apply it to their, you know, everyday lives. 0:58 Um, I think this is an important episode, uh, that needs to be recorded. 1:01 Uh, everybody in Kentucky really needs to make sure that you understand, you know, what's happening with AI and can become a part of it versus get, uh, left behind, 'cause it's moving very fast. 1:12 Before highlighting our sponsors, we'd just like to state that the views and content shared on this platform do not necessarily reflect those of our show sponsors. 1:20 Middle Tech is presented by KY Innovation, the Kentucky Cabinet for Economic Development's Office of Entrepreneurship. 1:26 KY Innovation exists to support and develop Kentucky's startup ecosystem, and we are proud to be supported by an organization whose mission aligns so closely with ours. 1:35 If you're a founder building in Kentucky, you need to check out the resources that KY Innovation has to offer. You can find more information by clicking the link in our show notes or going to kyinnovation.com. 1:46 Middle Tech is sponsored by Bolt Marketing. Take your website to the next level with a website that's built to work. 1:52 At Bolt Marketing, they're revolutionizing websites for small businesses that are affordable, customizable, and hassle-free. 1:58 Whether you have a construction company, a boutique clothing store, or you own a hot yoga studio, they have options for you. 2:04 Click the link in our show notes to explore their marketing options that can transform your marketing and grow your business. 2:09 And as a personal note, Bolt Marketing built our website, and they were awesome to work with throughout the entire process. We highly recommend working with them. 2:16 Middle Tech is also supported by Hunsicker, an impact venture studio based in Lexington, Kentucky. 2:22 Hunsicker is creating companies out of the ordinary by founding and funding the next generation of purpose-driven startups. 2:28 They partner with founders to create indispensable solutions to the world's most fundamental problems. 2:33 On a quarterly basis, they host an invite-only intimate gathering called Fireside Convos, where attendees representing a wide network of stakeholders in the regional entrepreneurial ecosystem discuss timely topics, including the role of AI for startups out of Kentucky and the challenges of identifying and attracting talent. 2:50 To be considered as a guest, visit hunsicker.co/fireside to apply. So I wanted to start with, you know, what is AI, and where did it originate from? 3:00 So the term artificial intelligence, uh, itself was coined by John McCarthy, and the discussions actually dated before him by a man named Alan Turing, who introduced the idea of an ur-universal, uh, machine capable of performing any computation. 3:18 So the Turing test is probably well-known at this point as if you're wondering if a computer is, uh, a super intelligence or using artificial intelligence, and it can trick you into you believing it's a human, that is basically the Turing test, where you cannot tell the difference between a computer and a human in the way that you interact with it through natural language. 3:42 And, you know, many would say that we're there today. But all that to say the origin dates back to about 1950, and it's been developing, you know, ever since. 3:52 And the official definition of artificial intelligence is, uh, really a machine that can perform tasks that's typically, uh, required of human intelligence. 4:03 So activities such as understanding language, recognizing patterns, making decisions, uh, learning from experience, artificial intelligence is, uh, good at those things. So really, it is exactly what it sounds like. 4:15 It's, it's computers that are representing and showing intelligence, uh, so it's artificial. So what types of artificial, uh, intelligence are there? I wanted to walk through... You know, there are several types. 4:29 Uh, one is, you know, narrow AI or weak AI, and really you can think of this as the Netflix or Amazon recommendation engines. They're really good at one narrow task, and they really can't be applied to many other things. 4:45 Next, you've kind of got, uh, general intelligence, so AGI. This is what, you know, many businesses, especially OpenAI, are racing towards. 4:53 Uh, so this is really where the artificial intelligence has a general ability to understand, learn, and apply knowledge across a very general set of tasks in a wide-ranging applications. 5:06 And this is, you know, competitive with human intelligence. Uh, so some examples of this would be everybody has experienced, uh, ChatGPT. 5:14 Um, it's not quite AGI yet, but you can see how it's starting to get closer and closer to that, and OpenAI's mission statement is to explicitly create AGI or artificial general intelligence. Someone is at the front door. 5:27 There's my Alexa. Uh, that's an example of narrow artificial intelligence. You know, it's really good at what it's programmed to do and really nothing beyond that. 5:36 And I think we've, we've all experienced the comparison between ChatGPT and Alexa. Alexa is pretty dumb compared to ChatGPT. 5:45 And then we've got, uh, the next tier, which is, you know, what some refer to as the singularity or super intelligence, and this is the point where computers are much, much smarter and outperforming any human mind. 6:01 Uh, in any field. And so you could take the aggregated intelligence of the smartest people to ever live, and they're still not capable of what super intelligence is. 6:12 And so those are kind of three ways to think about, you know, the stages of artificial intelligence. 6:17 And then one term that I wanted to make sure, uh, was also understood was a sub-field within AI that's known as, uh, NLP or natural language processing. 6:27 And this is really the core of what has been developed with ChatGPT, is we've basically been able to train computers on how to use natural language as an interface. 6:38 And I wanted to make sure that now that you understand kind of what is AI, what are the origins, and what types are they-- are there out there, why is this important? 6:48 So it's important because, you know, many believe this is the greatest invention to ever be created. You know, it's up there with electricity, it's up there with the internet, the wheel, fire. 6:59 Um, I think as time goes on, we'll realize that the pace of change and the impact that AI will have will be greater than, you know, any of those that I mentioned because of how fast it's gonna change our, our minds and society as a whole. 7:12 And so there's a few ways that, you know, many believe that artificial intelligence is gonna change society, and really this is believed to happen, especially when we reach AGI or artificial general intelligence, as I just described a little bit ago. 7:28 And here are some of the ways. So first, uh, this is kind of the obvious one that's probably at the top of everybody's mind, which is economic transformation. 7:37 Uh, when these computers get to a point that they are representing and showing that they have the capability of AGI, you know, a very wide range of jobs will be displaced. You know, manual labor, uh, knowledge work. 7:53 You know, we used to think that computers couldn't be creative. We're sh- we're showing signs, especially with ChatGPT and its writing abilities, that these are creative machines. 8:03 And so there are many, many jobs in our economy that are gonna be rendered, um, really distinct, uh, when AGI becomes a thing. 8:12 And, and even happening now, I'll get to a story here in just a bit where, you know, 700 jobs were replaced by a single instance and application within a large organization. 8:23 And so economic transformation is coming, and new industries and new types of jobs and new opportunities will pop up. Um, but it's gonna be a bit uncomfortable. 8:32 But the thing that is very clear is that productivity is gonna increase come-- and compound very quickly. 8:38 Uh, we're gonna be able to produce goods and knowledge and intelligence and labor at really a rate we've never experienced. 8:47 It's really kind of always been a dream of what artificial intelligence is gonna allow us to realize. 8:53 Next is, I think, something that's super important that, you know, I've, I've thought about a lot, and I think many people feel and, and want to experience every single day, and really it's, uh, healthcare advances. 9:07 You know, with AGI and the amazing amount of healthcare data we have in the United States and within the hospital systems, you can apply these models to all of that data and figure out new genetic makeups, lifestyles. 9:22 We can analyze medical histories, uh, and we can create new medications that can cure diseases using all of the data that we've collected so far. 9:32 And I think this is where, you know, the impact on our physical bodies is really gonna speed up, and, you know, we could start to see the, [chuckles] the average age, you know, increase, uh, pretty substantially if we're able to figure out, you know, how to, uh, affect our genes in positive ways based on what we're eating and our lifestyles. 9:52 You know, that's really, uh, in a lot of ways today, uh, there's not enough science on that front. 9:59 And when we're able to run all this data through models that traditionally a human would have to go through, uh, we'll be able to do a lot of amazing things. 10:07 And so hospital systems are introducing this, um, at a large scale. Uh, scientific and technological progress. 10:14 So science is really all about, you know, testing hypotheses and, and seeing what we can invent on the other side of those hypotheses. 10:23 And that's really what computers, and especially artificial intelligence, is really good at. 10:27 You can define, uh, basically barriers within a test and feed a bunch of data into it and see what comes out on the other side, and this is traditionally done by, by humans. 10:38 And we'll be able to do this at a scale that, you know, again, has never been seen before. 10:43 So the acceleration of research and figuring out new experiments to run will be amazing to see, and it'll lead to all kinds of innovation and new technologies that, you know, again, are, are part of our dreams and thoughts I don't think have really been, uh, around for very long. 11:01 Uh, you know, now we're talking about, you know, what does the metaverse look like when it combines with artificial intelligence, and, and we just don't know. 11:09 But you can imagine simulating the world and creating worlds, uh, just using your imagination in real time. That's kind of where this is heading. 11:17 And so that's an innovation that, you know, we wouldn't have thought of without, I think, artificial intelligence showing the promise that it is today. 11:24 Ethical and societal considerations and, and social considerations, um, is another category of, of how we're gonna be affected by this. 11:32 You know, if, if, uh, we have instant access to knowledge at any time, uh, what does that mean for how we interact with each other? Um, and what does that mean for how we interact with humanity as a whole? 11:44 With AGI as a thing, there will be robots walking around that we can conversate with, and they'll be just as good at speaking to us as another human being. 11:54 And I'm not sure what that means from an ethical and societal standpoint, um, but we have to create new frameworks within our mind on how to deal with that. And that might sound crazy, but that exists today. 12:05 Companies like Tesla and Figure and OpenAI are building, you know, humanoid robots that are capable of speaking to us, seeing the same things we see, and being able to navigate physical space just like we can. 12:19 And so that introduces all kinds of new, um, ethical considerations. Privacy concerns fall underneath ethical considerations as well. 12:27 You know, all these models have scraped all the data on the internet, and I'll get into this in a bit, but that level of, um, safeguards and privacy violations is something that, uh, we should be concerned about. 12:40 You know, what data do we own, and what data do the companies building these models own? That's a very important question around privacy. Security and governance, this is one that is not talked about enough. 12:52 Um, this is gonna require international collaboration. 12:55 When we have a supercomputer in the United States and China has a supercomputer, we're gonna have to coordinate with China on what that means, and we're not sure yet what that looks like. 13:08 But there will be new international conflicts and races, just like there was when the nuclear bomb was created, to reach AGI before another country does, and that's gonna create a whole new level of global governance that's gonna have to be created. 13:25 Um, and so that, that's something that, you know, we really need to look to our politicians to talk more about, uh, because our national security is, without question, at risk with, uh, artificial intelligence. 13:37 The last thing that I think many people are worried about is really this existential risk of, you know, what does this mean for humans? Are we gonna be extinct? And nobody really knows that answer. 13:50 Ultimately, we are the ones creating this, this technology. But where it goes after we create AGI, you know, we really just have no way of predicting. 14:00 Um, and when we create a computer that's smarter than us, which is going to happen very soon, um, I'm, I'm-- Again, I'm not sure, and I don't think anybody is really sure, uh, what that means. 14:11 You know, we'll have to have kill switches, and we'll have to have a lot of security gu- safeguards in place. 14:16 Um, but, you know, it's up in the air, and I think it's up to all of us collectively to have an important conversation around how we wanna introduce safety measures into this, this technology. 14:25 So I mentioned earlier that there was a company that cut, uh, a large portion of their staff because of this. And so I really wanted to get into some practical app- applications that, um, you know, coming out of 14:40 what is artificial intelligence, what are the origins of it, um, what are the different types of artificial intelligence, and how should we think about it as a society? 14:47 You know, I really wanted to lay out how are people using it today in very tangible ways. 14:52 Um, and so one personal use case that, that I've done is I've been taking pictures of, uh, food labels, and I prompt the model to tell me what each of the ingredients is, uh, what are healthy doses, and how does it affect my body? 15:07 And all of that can be done with just taking a picture of the back of whatever your food is. Um, and so that, that's a great application, uh, that you can do with ChatGPT. 15:16 Uh, and then again, I alluded to this big company that cut jobs. The name of that company is, is Klarna. It's a European company. 15:24 Uh, and what they had done was they created, uh, basically a model that helped them with customer service and customer support. 15:32 And the AI model handles about two-thirds of all their customer support chats, and this equated to 700 full-time human agents. 15:43 And it was able to cut all 700 of those human agents' jobs, and it did so with a drop in productivity from 11 minutes to two minutes in response times. 15:58 So you now have, uh, humans have become the less productive part of a business, and again, they can't compete with, again, 11 minutes it took a human to two minutes it took the artificial intelligence productivity-wise to respond to these queries, these customer support queries. 16:17 Um, and they also saw a 25% drop in repeat queries, inquiries, and that's crazy too because these models are way more accurate than humans. 16:27 And so you're starting to see how this affects businesses, but more importantly, people, and it goes straight to the bottom line of these businesses. 16:35 That business, just by introducing this one application, increased profits by $40 million in a single year, and it's not that big of a company. So imagine what happens at scale across massive, massive companies. 16:49 And so it's really important to really consider as a business, if you're listening to this as a business owner, where as quickly as you can, can you introduce AI in a responsible manner? 16:59 And then how do you communicate to your employees what this means for them? A couple examples here in Kentucky of founders that are introducing AI, um, is a company called EZOS. 17:10 So Chase Minnifield is using artificial intelligence to analyze, uh, leases, and so they build technology and an application for property managers, and they're using artificial intelligence to read hundreds and hundreds of leases and extract insights and compare them to rent rolls. 17:31 This took people, humans, weeks, and they're doing it in a matter of hours or days with their technology. So massive increase in productivity again there. 17:41 And then there's another company my friend Danny Perry has started called Brevity, and he's using it with video captions and cutting video clips from long-form video and using them for social media clips. 17:55 So he's able to build a model that can analyze a long-form video, like a YouTube video, analyze which clips from that video are best suited for social media It will automatically cut those and add captions, and then you can customize it right inside of Adobe. 18:12 And it did all of this using several different models, but it cut the time it takes to make social media clips down substantially. 18:21 And so you have people that traditionally were watching an entire YouTube video trying to find the right clips and then adding captions manually, and now they're able to do it in just a matter of seconds. 18:33 So those are just a few tangible examples of how people are using this technology to increase productivity substantially. Now, I'm wanting to get into the weeds just a bit. 18:44 So I mentioned, you know, the whole background on AI, how are people applying it, how should we think about it, and how it's gonna impact us, but I really think it's important for everybody to understand the different components of what makes up artificial intelligence and some terms that you might be hearing, uh, as it relates to AI and, and just not sure what they are. 19:03 Um, so I wanted to go through several of those. Number one is, uh, large language models, so LLMs. 19:11 And an LLM is officially, um, essentially we've given these models the entire corpus of text that we have access to. We've given them the internet. The entire internet has been scraped. 19:27 Every book, article, and website has been put into these models, and the computer learns patterns and structures and nuances of, you know, our human language and converts that into data that it can take action on. 19:41 And it can complete sentences, answer questions, write an entire document. Um, and all of this is also, you know, mixed up in the term generative AI. 19:52 And so it's using its deep understanding of language, again, large language models, to generate new text or new articles, new images. 20:04 Uh, and it's doing all of this, uh, using, again, all of the data that essentially we have, uh, on the internet and in books and everything. 20:12 And so we've shoved all this knowledge in these models, and we can interface with them. And we interface with them, as many of you has ex- have experienced, in something like ChatGPT or Microsoft Copilot. 20:24 And whenever you type something to one of those models, you're giving it a prompt. 20:28 And so this is the next word I want everybody to understand, is a prompt is a set of instructions that you give the model to generate some kind of output that you want. 20:37 And so the clearer those instructions, the more details you give it, the better that you're gonna get a response. 20:43 And so, for instance, if you wanna write, let's say, a story, you could say, "Write a story about a cat who goes on an adventure." And that is the prompt. And of course, you can give it way more details. 20:53 You can be much more instructional, but that is an example of a prompt, and you would get an output of a story about a cat. And you're putting that text into what's known as the next term, which is a context window. 21:07 A context window is essentially all the information that the model takes into account. 21:12 So it's the prompt, and it's anything else that you're passing along with that prompt, like, uh, old chats and other communication that you've had, uh, farther up in the thread. 21:24 All of that is basically considered the context or the memory of the model. And when you pass all of that memory to the model, it uses all of that to generate the outcome. 21:33 And so if you've noticed, when you talk with ChatGPT, it knows what you said higher up in the thread and considers that in the current output. And that's all, uh, living within the context windows, and there's... 21:45 Today, there's a limited amount of context that you can pass it. Uh, we're increasing that pretty quickly. 21:51 We've gone from 4,000, what's called tokens, which is basically the, uh, individual small components within the context window. It could be parts of a word or a single word could make up a token. 22:04 But we've gone from 4,000 tokens to 100,000 tokens to a million, and now there are some models that can do 2 million, uh, token context windows. 22:14 And that just allows you to put more information in it, which allows you to introduce new use cases. So it could literally write an entire book right now. 22:22 It could generate books of hundreds of pages, and you could feed it books of hundreds of pages. That's what the context window is for. 22:30 Next, you've got, uh, we're starting to get really in the weeds here, uh, but vector databases. And a vector database is a new type of database. It's been around for a bit, but it's really becoming effective today. 22:42 Um, it's a new type of database that stores and manages data in the form of vectors and, like, uh, coordinates on a map is the way to think about this. 22:51 Um, so rather than really structured data like names, emails, phone numbers, addresses, uh, they all have, you know, very clear relationships. 23:01 With a vector database, you can feed it a ton of text, and it maps that text to coordinates. Again, there's basically an infinite amount of coordinates here. 23:09 It can map that text to coordinates, and next time you go search for those words, it finds-- Using the current words you're passing it, it finds the most similar words using the coordinates. 23:20 So it basically generates coordinates, and it searches coordinates. And you no longer have to have super structured, uh, databases anymore, and this really allows for new types of search and user experiences. 23:31 So an example of this is Amazon has been using vector databases for a long time to suggest shopping items that when you create a profile and you go onto Amazon, it takes all of your data, and it puts it into a vector database. 23:44 And the next time that you log in and it has new information about you, it can suggest really well items to you. And that's just not possible with really structured databases. 23:54 So that's a really powerful tool that, uh, many AI companies, um, are leveraging today. It's pretty much essential. 24:01 Next, uh, that ties into vector databases is this concept of retrieval-augmented generation or RAG, as many people are calling it. 24:10 So, as I mentioned earlier, you have a context window, you have the prompt, and oftentimes you would struggle to put most relevant information that might be living within your organization in that prompt. 24:22 And so an example of this is if you want To put a file in a window, in a con- uh, context window and in your prompt, you could basically just copy and paste that file and that prompt. 24:33 But if you want the model to reference a bunch of files, then it would go against that vector database I just described and pull relevant data into the prompt using the vectors, and then you can use that to basically search across a ton of data within your organization and chat with that data. 24:53 Um, and so it's just basically a way to inject prompts with relevant information that traditionally you would have to search for, uh, within all your files. 25:05 And so this is a really powerful tool that allows AI to create all kinds of cool user experiences and search over all, all of your files. 25:13 And really, it's a way for the model to essentially have memory of all your files and understanding of all your files. Then using all of this technology that I just described, agents, AI agents can be introduced. 25:26 And so an AI agent can take action on behalf of humans to achieve a goal. 25:31 And so essentially, it's a set of tools like vector databases, uh, retrieval augmented generation, uh, and large language models, and all of this stuff works together where you can pass many prompts and chain-- basically chain them together to autonomously create, uh, some kind of output or achieve a goal. 25:50 Um, and so you can use it to analyze large documents and extract information from them. You can use it to complete tasks that are multi-step. 26:00 And AI agents are really kind of the next big revolution and next big, I think, in- innovation that needs to be introduced and is being introduced by many companies. 26:09 Um, I'm developing AI agents right now within my product. And so it's really kind of where things are heading, and it's a culmination of all of those technologies that I went into the, the weeds for. 26:19 Whether you, you know, wanted to get into that much detail or not, I think it's important. 26:23 But all of that to say, when you combine all that stuff, you know, there's really new, exciting ways to interface with AI, and particularly AI agents. 26:32 So I mentioned, you know, I'm building a company right now, and I'm building a company called Devise. 26:37 And what we're doing, I've gotten a ton of great personal experience with this, is we're building a tool that helps streamline, uh, document-heavy workflows. 26:47 And so we're targeting first the request for proposal workflow, where you're able to feed very large documents into, um, our platform, and we use AI agents to deconstruct those documents and create a workflow to quickly respond to something like an RFP or an application or a grant. 27:07 So we're really speeding that up and making it easy. And the way that I really kind of arrived at this was I, I had been using large language models well before ChatGPT launched. 27:18 And so I kind of saw what was coming, and I really used them very heavily in my last company. 27:22 And when we exited that company, I knew I wanted to build something in the AI space, and my co-founder and I had something in mind, and so we started building it. 27:30 And just a few of the lessons that we've, we've learned along the way is, you know, these models are evolving extremely, extremely quickly. 27:38 And while they're evolving and becoming much more performant, they're becoming much more cheaper. And so that's with any, you know, scalable technology that when it becomes something that is a household 27:52 topic, you want to see it improve rapidly and get cheaper rapidly. And examples of this, you know, there's plenty of them throughout history, whether it's cars, electricity, um, airplanes. 28:05 You know, it always starts out as a big luxury, and then it gets taken out to the masses, and that's exactly what's happening with AI right now. 28:13 And so that, that's something I've noticed as I've been building this company, um, how easy it is to develop a product with artificial intelligence, um, and, and just build any product now. 28:25 The cost has gotten so much cheaper. Next is I mentioned RAG earlier, so retrieval augmented generation. That's where the magic is happening within AI right now. 28:35 So the models really are gonna be probably, um, limited to two or three companies will, will win the proprietary model race, and then there'll be a ton of open source models. 28:47 And it's gonna be extremely competitive, and only the people with the most money to train these models, um, are gonna be able to introduce, you know, new innovation on that front. 28:58 So, uh, I would advise people not to try to build, you know, a, a brand new model. I think, I think many will find that's just not, not plausible. 29:08 But try to figure out how to use retrieval augmented generation to build a product. 29:13 Um, that, that again is where the innovation is happening for you to be able to take a ton of data within a company and build workflows on top of that data using AI. That's, that's where you need to spend your time. 29:24 Um, next is, uh, collaboration. So ultimately what we've built is a collaboration platform, and collaboration is a really hard problem to solve with AI. 29:34 Today, most people are really familiar with using ChatGPT or Microsoft Copilot, and if you've noticed, it's a one-on-one chat. 29:43 Your team cannot be in that chat with you, and that's a problem, because if you're working on a project with your team and your whole team that's working on that project wants to leverage AI, there's really not many ways to do that today, and our company is really introducing some pretty creative ways to do that on projects, especially again, very document-heavy projects. 30:02 Next is, I alluded to it earlier, but you have to move extremely fast or you're gonna be overtaken by better-funded businesses. 30:12 There has been a ton of money flowing into artificial intelligence, and if your team is not capable of moving really fast and you're not, uh, willing or if you're willing to raise venture capital, you're gonna, you're gonna have to, frankly. 30:28 Um, this is something that if you wanna reach scale, there's gonna be competitors. 30:32 If your idea is something that, you know, has a lot of legs, you have to be prepared to raise venture capital, in my opinion, in order to move really fast just because of how AI is evolving right now. 30:42 Um, so that-- those are some things that, you know, I've witnessed and experienced firsthand. Uh, so just wanted to share some, some personal experience. 30:51 Now, this is something that we're gonna continue to talk about, um, within Middletech. Uh, we're super excited to bring you more of this. 31:00 Um, I wanna tease something that, you know, we're bringing to many of the hubs around Kentucky. Um, and one thing we're gonna be doing is, uh, in-person educational events for AI. 31:11 So we will be hosting some really awesome events on ways, just like with this conversation, people can use AI in their everyday lives and what they need to know about it in order to be a part of the movement. 31:24 And so we'll be doing that, uh, in several cities across Kentucky in partnership with, uh, KY Innovation and the hubs like Awesome Inc. Uh, so we're looking forward to bringing that to you. 31:35 Thanks for listening and look for more of this in the future. Thanks. 31:38 [upbeat music]