Eureka moments with natural language processing

When is the last time you had a eureka moment? Chris had a chat with Nicholas Mohnacky, CEO and Cofounder of bundleIQ, where they use natural language processing algorithms like GPT-3 to connect your Google GSuite with other personal data sources to find deeper connections, go beyond the obvious, and create eureka moments. Sponsors: Fastly https://www.fastly.com/?utm_source=changelog&utm_medium=podcast&utm_campaign=changelog-sponsorship – Our bandwidth partner. Fastly powers fast, secure, and scalable digital experiences. Move beyond your content delivery network to their powerful edge cloud platform. Learn more at fastly.com https://www.fastly.com/?utm_source=changelog&utm_medium=podcast&utm_campaign=changelog-sponsorshipLaunchDarkly / TrajectoryConf https://trajectoryconf.com/?utm_source=changelog&utm_medium=podcast&utm_campaign=the-changelog – Software powers the world. LaunchDarkly empowers all teams to deliver and control their software. DevOps and feature management are reimagining how we build and release new products. On November 9th and 10th, LaunchDarkly is hosting Trajectory Conference 2021 — a two-day event for software innovators who want to break orbit, not systems. Trajectory is a fully-virtual conference that focuses on the technology, people, and processes that continuously deliver better user experiences and more powerful software. Register for free at trajectoryconf.com https://trajectoryconf.com/?utm_source=changelog&utm_medium=podcast&utm_campaign=the-changelogLinode https://linode.com/changelog – Get $100 in free credit to get started on Linode – Linode is our cloud of choice and the home of Changelog.com. Head to linode.com/changelog https://linode.com/changelog OR text CHANGELOG to 474747 to get instant access to that $100 in free credit. Featuring: Nicholas Mohnacky – GitHub https://github.com/mohnacky, LinkedIn https://www.linkedin.com/in/mohnacky, X https://x.com/mohnackyChris Benson – Website https://chrisbenson.com, GitHub https://github.com/chrisbenson, LinkedIn https://www.linkedin.com/in/chrisbenson, X https://x.com/chrisbensonShow Notes: bundleIQ https://www.bundleiq.combundleIQ | LinkedIn https://www.linkedin.com/company/bundleiqUpcoming Events: Register for upcoming webinars here https://practicalai.fm/webinars!

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  What we believe we can do better than any company in the world at Bundle IQ is we believe we can engineer eureka moments. So those aha moments that might come to you when they're least expected. That's kind of like our North Star as a company. And so when you think about research and analysis, whether it's on what models to use or other parts of your business, when you're going through this research process and trying to make sense of all of this information that's out there on the internet or in your inbox, that's like disparate information that kind of lives in multiple places. And so what we aim to do is basically be that like centrifuge that can kind of crunch all that data and make those unexpected connections across information using math.

• 0:45Speaker 01

  Big thanks to our partners, Leno, Fastly, and Launch Darkly. We love Leno. They keep it fast and simple. Check them out at leno.com/changelog. Our bandwidth is provided by Fastly. Learn more at fastly.com and get your feature flags powered by Launch Darkly. Get a demo at launchdarkly.com.

• 1:08Speaker 02

  Welcome to Practical AI, a weekly podcast that makes artificial intelligence practical, productive, and accessible to everyone. This is where conversations around AI, machine learning, and data science happen. Join the community and Slack with us around various topics of the show at change.com/community and follow us on Twitter. We are at Practical AI.

• 1:34Chris Benson?

  Welcome to another episode of Practical AI. I'm your host, Chris Benson. I am a technology strategist at Lockheed Martin. Normally, you would have Daniel Whitenack, my co-host, with me here today. Daniel is unfortunately on a plane right now and is not able to make this. So I'm going to go right into introducing our guest. I'm excited because there's some topics here that I'm really looking forward to diving into. I'd like to introduce Nicholas Mohagney, who is the CEO and co-founder of Bundle IQ. Welcome to Practical AI, Nick.

• 2:07Speaker 00

  Thanks for having me, Chris.

• 2:08Chris Benson?

  You're very welcome. I guess if you could start off a little bit, give me a little bit of background. We have some interesting topics, GPT-3 and others that I'm looking forward to. But if you could tell us a little bit about how you arrived into this space and a little bit about what you're all about to get us going.

• 2:28Speaker 00

  Yeah, definitely. Obviously, GPT-3 is, I would say, a big topic of discussion, but it's definitely making mainstream news. And that's definitely something that we've been excited here at the team to explore.

• 2:39Chris Benson?

  As part of this, could you tell us a little bit about what that is, just as you're, since you're already kind of referencing it right off the bat?

• 2:45Speaker 00

  Yeah, so there's a company called OpenAI, and I don't know how accurate this is, but I would like to say that it's sort of like the next space race to create artificial general intelligence. And so there's a company that was founded as a nonprofit initially, and I think they do have a for-profit arm as well. But basically to go out and try to build some intellectual property and try to do just that. So for people that maybe don't know what artificial general intelligence is, it's kind of trying to replicate the intelligence of human beings. You know, they've said that a fifth grader is more generally intelligent than the best AI out there. And that, you know, so, long story short, we got into the beta program and we've been exploring. And we're probably one of maybe only a few hundred apps that is live currently.

• 3:29Speaker 00

  Because you do have to go through an application process and get approved to go live with GPT-3. So they do have some throttles on the system. Some checks in place to make sure maybe it doesn't get into bad actors' hands, so to speak. I wouldn't say that was my foray into AI and bundle like you per se, but, you know, having done some maybe preliminary modeling and then moving on to GPT-3 has been a nice step in the right direction for us as a company.

• 3:55Chris Benson?

  Gotcha. And for those, in the past, we've had folks from OpenAI come on and talk a little bit about. But I don't think we've ever had a show focused on capabilities, especially with someone like yourself who's gone in and started using GPT-3. Could you describe what GPT-3 is for anybody that's not familiar with that as an algorithm? That like, what space is it in? Was it trying to do what kind of capabilities? And I understand that this is not your algorithm. It's one that your company is taking advantage of to great effect. But if you could just kind of give me your take on what GPT-3 is.

• 4:29Speaker 00

  Yeah, so essentially what they've done is they've ingested like ten percent of the internet. I think it's like one hundred and sixty something billion parameters worth of data, which is

• 4:39Chris Benson?

  That's a lot.

• 4:40Speaker 00

  mind boggling, right? So they've done that and by doing that, they've been able to create these really creative models for language. So you could, as an example, you could maybe say something like in the GPT three playground, say, create a science fiction script for automated robots inside of battleships. You know, like, and it would like create this story for you, you know? So you can kind of create those really fun prompts. So you could say, create a recipe for Halloween and it would come up with a recipe for Halloween.

• 5:10Chris Benson?

  That's pretty cool. So, I mean, is it essentially unbounded in terms of the types of questions that you would ask? Or would you think of it as average everyday question that you might ask Google? Which is kind of what it's sounding like a little bit, but then it goes and generates that output from scratch?

• 5:25Speaker 00

  Yeah, so I think on the simplest level, it's a great mirror to hold up to any sort of inquiry or prompt. So you can kind of start to get, you know, more context for maybe a very small amount of information. Finishing a sentence or creating a tweet or creating a paragraph. And there's some apps out there that have taken full advantage of that. Namely, there's an app called Jarvis that helps with like marketing copy. So let's say you're creating a LinkedIn ad and, you know, you want to target, you know, business professionals in the C suite. You know, you can kind of use GPT three to come up with like fifty potential scripts for that ad within like six seconds.

• 6:03Chris Benson?

  Wow. So how did you get? I know you got into the beta program and I remember when they were, when OpenAI was soliciting that. And I know that later on, Microsoft, I believe, has some sort of exclusive license to doing that. Are you still, are you still going through OpenAI's program? Or do you do this through Microsoft or?

• 6:22Speaker 00

  Yeah, we're not part of like any formal programming per se, but we are part of the beta and we're using the playground. And we've used a couple of their models to support the outputs, I guess, of like the results of the work that we do at Bundle IQ.

• 6:36Chris Benson?

  Yeah.

• 6:36Speaker 00

  And we've also been transitioning away from GPT three because a lot A lot of it is a black box. So just from like a technical standpoint, you know, if you're going to build out a company, it's obviously not good to just use.

• 6:48Chris Benson?

  There's a risk there. Yeah.

• 6:49Speaker 00

  Third party software. Yeah, exactly. So if they cut us off and, you know, we can't create the results for our end user that we want to. So we, we used it mostly for training purposes and to see what the capabilities might be. But we've, we've since even, I think since you and I connected initially a, a month or so ago, we've already started to transition away from it.

• 7:09Chris Benson?

  Okay, understood. From a business standpoint, that's a very interesting perspective because clearly if you build a business around something that you don't, you know, control directly, especially as beta, let's talk about that for a moment. Like, how are you thinking about that as the CEO of your company? And you're looking at this really cool technology, but it has a definite risk in terms of that dependency. How are you thinking about it? You know, if you're looking at other alternatives, other algorithms in the natural language processing space? That help you get to the same destination that your company is going. How do you look at that landscape? How do you evaluate it? And what are the options looking like to you?

• 7:48Chris Benson?

  Not trying to get you to reveal your specific business plan, but if you could give us a little perspective on how you see the world in that way. I imagine that's something a lot of people would wonder.

• 7:59Speaker 00

  Yeah, definitely. And it's a good question. You know, it's a question that obviously a lot of investors have asked, like, okay, if you're using this, then what? You know, what are your dependencies? What are your vulnerabilities? That sort of thing? What is your technical moat? Your technical moat really look like if you're sort of using this as a, a black box solution. So what we've had to do is basically build out our own ensembles. And in kind of the AI world, you, an ensemble, you think of it as like, you know, like an orchestra, like you have the woodwinds and the, and the horns and the, you know, the brass and the strings, right? So those different types of instruments in the ensemble of like maybe an AI model that you'll be passing data through looks like, you know, maybe filtering, maybe summarization, maybe, you know, some other component?

• 8:44Speaker 00

  That can help you sort of have this like beautiful symphony, you know, harmonious like melody coming. Out on the other side, we basically looked at GPT three as kind of the benchmark. Said, okay, if this is the best and they're saying this is the best, then like, how close can we really get by doing it on our own? And that was kind of what we sort of used GPT three for initially.

• 9:04Chris Benson?

  It kind of gives you something to target, but you're kind of inferring another point, I think, and correct me if I'm wrong, but GPT, GPT three is so big and is so expansive. If you look at the inputs about how much of the internet, you know, was used to develop the model and any business, not just yours, though, has a narrower scope, you know, of trying, you have a specific business plan, you're trying to do stuff. Would it be fair to say that recognizing that GPT three is so expansive that you don't have to produce the same thing to satisfy your business need, that you can have a narrower, smaller scope and still be as good or better

• 9:45Chris Benson?

  than you might have been with GPT three, if you can get the right model going, if you get the right algorithm developed, is that, is that a fair way of looking at that?

• 9:52Speaker 00

  Yeah, you did a great job at reading between the lines there. So the answer is yes. So Chris's data is not 10% of the internet. We basically build the model, each individual model and ensemble is like attached to Chris's data.

• 10:04Chris Benson?

  Okay.

• 10:05Speaker 00

  And the people that Chris works with, it's definitely much narrower and more explicit in nature. It's not, it's not as creative. It doesn't need to be.

• 10:12Chris Benson?

  Gotcha. You're able to kind of personalize based on your user in that way.

• 10:16Speaker 00

  That's correct.

• 10:17Chris Benson?

  That user is, has a certain amount of data. It's not everywhere. God, I hope my data is not everywhere on the internet. I know it is, but at least you can do that. And this is just a stab in the dark curiosity. Our last episode that we released as I recorded this was on federated learning. Is there any interest or intention of using federated learning to build up better models, but yet letting certain data, you know, that's in private, you know, you have a user and they don't want their data out, but you can train models on that and infer back to the aggregated model. Has there, has there been any thought of doing that? And I'm mainly just curious because I've been talking about federated learning a lot the last few days.

• 10:53Speaker 00

  It's a good question and it's one that we've discussed, but it's not a bridge that we've crossed yet.

• 10:58Chris Benson?

  Gotcha.

• 10:58Speaker 00

  So I'd be purely speculating on like where we might go with that, but I think there's, there's definitely value in federated learning and sort of extrapolating context and meaning, maybe based on industries or categories or et cetera.

• 11:13Chris Benson?

  Gotcha. So it sounds like you're in a bit of a transition right now as you're looking at the world outside of GPT-3. Can you describe a little bit, aside from your choices and your decisions for your business, but just in a broader sense, you know, as you're looking at that, what does the state of this particular branch of natural language processing look like to you and your team at this point? You're a business that's already in flight, you're already moving and you already have users that you're supporting and you're out there doing what your business does. And this landscape is changing around on you on a day-to-day basis. How does that look and how do you, how do you evaluate all the options? Whether there are options that are interesting to your specific use case at your business or not?

• 11:58Chris Benson?

  How do you do that? How do you stay up with it?

• 12:00Speaker 00

  I think it's, software in general is sort of always evolving. It's a living and breathing kind of organism unto itself. And so we're constantly iterating through the process of optimizing the outcome. And some of that includes, you know, maybe changing the infrastructure because some of our assumptions on how we were doing something one way with like tokenization of keywords and extraction and that sort of thing. And creating all these cues and like, you know, moving data in certain directions without, I guess, boring you with the details. Like what we've learned is that by instituting new models that maybe turn natural language into math, you don't have to do as much of the language part of it versus doing the more the math part of it.

• 12:43Speaker 00

  So like vectorizing data, for instance, you know, and then in building machine learning on, on an individual like customer basis. It's much more scalable than maybe trying to manage, you know, an entire organization's knowledge base. But just kind of maybe zooming out a little bit, so just for the audience, you know, what we believe we can do better than any company in the world at Bundle IQ is we believe we can engineer eureka moments. So those like aha moments that might come to you when they're least expected. And so, you know, that's kind of like our North Star as a company. And so when you think about like, you know, you do a lot of research and you know, you're constantly doing analysis. Whether it's on, you know, what models to use or, you know, other parts of your business

• 13:28Speaker 00

  or people that you bring on to your podcast. When you're going through kind of this research process and trying to make sense of all of this information that's out there on the internet or in your inbox or a note that, you know, Daniel shared with you or whatever it is, that's like disparate information that just kind of lives in multiple places. And so what we aim to do is basically be that like centrifuge that can kind of, you know, like crunch all that data and like make those unexpected connections across information using map. That's kind of like the ultimate, you know, vision and, and we've been able to achieve that thankfully by using, working with our users and using their data sets initially from their notes and then now moving into their emails and now moving into their Google

• 14:13Speaker 00

  Docs and other forms of information like PDFs and, you know, SharePoint files and things like that.

• 14:29Speaker 02

  Change log plus plus is the best way for you to directly support practical AI. Join today and unlock access to a private feed that makes the ads disappear, gets you closer to the metal and help sustain our production of practical AI into the future. Simply follow the change log plus plus link in your show notes or point your favorite web browser to changelog.com slash plus plus. Once again, that's changelog.com slash plus plus.

• 15:14Chris Benson?

  So that sounds really cool in terms of how you are trying to find or, you know, locate or maybe even to some degree construct those your recon moments for people from that. Before we dive into the, fully into the kind of the logistics of what that means to the user, where did you get the idea from? I'm just curious, like when you were, you're in the space and you're looking and you're seeing these tools, you know, we talked about G P D three, you've talked about other, developing other algorithms based on your users' data. What was the inspiration for going down this path to begin with?

• 15:47Speaker 00

  Yeah, so we're going to zoom way, way, way, way out.

• 15:49Chris Benson?

  Okay.

• 15:49Speaker 00

  And we'll talk about internet 2.0. And basically, you know, as a society, we've done a really, really good job at getting a lot of information online. And so we've sort of encapsulated the information age with a ton of data and databases. And now we've like moved from the information age into the experience age. And now we're trying to figure out how do we get our time back from sorting through all of this information? And that's like the real problem is like, you know, information overload is not only is it real, it's like palatable. And we have like chronic stress as a result of it.

• 16:24Chris Benson?

  Indeed, we do.

• 16:27Speaker 00

  So thinking about like, you know, the whole elephant equation is like one bite at a time. It's like, okay, let's, let's like zoom in to now Chris and like just his, what does he want to do in life? What is gonna like help you solve problems and what are the problems that you have? And it's like, you know, humans are really, really great at better than machines at three things. Creativity, innovation and imagination. And so like for us, we, it's like, how can we stay in those lanes? And let the machines like come alongside the machines to like do all the heavy lifting on the information management side. So that, that was kind of my thought process and it.

• 17:04Chris Benson?

  So let's take that motivation and a little bit of that history. And what have you gone and kind of built out and kind of pulling some of the, the strings together? Cause we opened up the conversation, kind of talking about, you know, the algorithms and GPT three and what it can do. You've talked about these different points that are the touch points in the user's life. You know, like email and, and you know, we have all these different messaging platforms that we're all using and stuff like that. And you'd also talked just now about keeping us in that area that we're strongest at, which is innovation, creativity and imagination. So how are you pulling these, all these little pieces together to create something at bundle IQ that addresses that?

• 17:48Speaker 00

  Yeah, I guess this in the simplest form, we're basically vectorizing then information and attaching that to, to you, the user. So whatever you're inputting into the system in the form of emailing people or writing notes or saving a document to the cloud or whatever, that is tied to you. And then anything that you are. Shared like on a folder or a drive or a channel that you're shared in that is also tied to you. And then basically just storing that and building the mechanics around just that. That's like the simplest form.

• 18:23Chris Benson?

  What do you do with it? Like taking that raw data? If I'm your user, you're taking all that Chris generated or received stuff, things that I'm touching in my day to day. And you mentioned the eureka moment, which kind of grabbed me a little bit because I'm looking for some eureka moments. Quite honestly, how do we find that? What are you taking all of that input data about me and the thing in my activities in my day to day? How are you trying to find what is the kind of thing that you're going to pull out of that? To give me that that thing that you described.

• 18:53Speaker 00

  Right. So I'll give you a user story that just came to me about a week ago. So you're a manager, you sit down to email and you had just interviewed maybe five or six candidates. And you need to reply to four of those and with a rejection letter. So you know that like maybe three months ago, four months ago, you had just written an email. That was the perfect email for this rejection letter. But the problem is it's not titled rejection letter. So how do you find that? And you're like, I don't know, it was Susan. I don't even remember her name. Like, I don't know her email address and that's where it's like, again, connecting like information that's already out there.

• 19:33Speaker 00

  So imagine now you're writing this email and within the first few sentences, this nudge comes up and says, hey, Chris, you might want to check this out. And it's the email that you wrote to Susan, you know, six months ago. And it was, and you get to copy and paste that content into that email in that moment. That was a, you know, a pretty cool one that I was like, oh, wow, that's cool.

• 19:55Chris Benson?

  So it's kind of you're using those capabilities in the NLP model that you're using. I know you started with GPT three, you're looking at other options going forward at this point, but you're using that capability to kind of, I'm a little afraid to use the word, but kind of create some context around that person.

• 20:14Speaker 00

  Absolutely.

• 20:15Chris Benson?

  In terms of what they're doing, is that fair?

• 20:16Speaker 00

  So it's recognizing what you're saying and it's like, oh, you might be talking about. These things and then it's looking into like the corpus of content or data from your emails and notes and all these different places. And then it's it's starting to recommend the most relevant suggestions that might be useful in that moment.

• 20:34Chris Benson?

  Okay, in the use case you talked about, it kind of feels like personal assistant.

• 20:38Speaker 00

  Yeah.

• 20:39Chris Benson?

  In terms of kind of helping me be better at what I am than than I would be by myself, which is kind of ironic. Is that kind of the focus there is is trying to to help people with their own personal assistant that's trained on them?

• 20:51Speaker 00

  Yeah, it's meant to, we don't use the word assistant.

• 20:53Chris Benson?

  Okay.

• 20:54Speaker 00

  It's a little bit hierarchical and we're just like, you know, it's not really an assistant. And so we were like, what if it was Robin? It was like your sidekick.

• 21:01Chris Benson?

  Okay.

• 21:02Speaker 00

  You know, so like it's your AI sidekick that sort of works alongside you and it's like, it's got your back, you know, it's like helping.

• 21:08Chris Benson?

  That makes me Batman, though. I win that analogy, so.

• 21:11Speaker 00

  Exactly. You're Batman and this is Robin. Yeah.

• 21:14Chris Benson?

  Oh dude, I'm feeling much better about this already, so.

• 21:18Speaker 00

  Yeah, I mean, you don't want to like belittle Robin, right?

• 21:20Chris Benson?

  No, Robin's cool. We like Robin.

• 21:22Speaker 00

  Yeah.

• 21:22Chris Benson?

  Robin's cool. Okay, so I get to be Batman. I got my Robin coming along and Robin's kind of kind of helping me get the things done that I need doing a bit better. And is that really defined that? I know we've been doing a lot of NLP the last few years, but it's still exploding outward. We're still having so much, you know, work in that that area. Are there other aspects of it or does that kind of, is it really NLP focused? Do you have other algorithmic things that have to be considered? Is there any kind of reinforcement learning, anything like that, or is this doing it? I'm not trying to to push you in other directions. I'm just wondering, is that large area of NLP really cover it fully in terms of what you're trying to achieve?

• 22:00Speaker 00

  Yeah.

• 22:00Chris Benson?

  Are there any gaps? Might be a better question in terms of like, what do you wish was there or something that you guys are exploring even outside the NLP space?

• 22:08Speaker 00

  Yeah, so there's a lot of discussion internally about supervised versus unsupervised learning. And everything to date has been unsupervised. You're not telling the model or telling the, you know, IQ to do anything. And so we're trying to sort of figure out, does it make sense for Chris to say like, atta boy, like high five, you know?

• 22:28Chris Benson?

  Ah, get the feedback, essentially.

• 22:29Speaker 00

  Yeah, exactly. So we're trying to sort of sort through that now as far as like machine learning is concerned.

• 22:36Chris Benson?

  Yeah, so it sounds sort of like a reinforcement thing where you have a reward that you're offering back to kind of steer it a little bit. And that would make a lot of sense.

• 22:44Speaker 00

  Yeah, you know, there's like sort of knowledge graphs that exist in the world. Obviously, we're graphing the knowledge, so that's why I bring forth the phrase knowledge graph that exists. But, you know, they're all brute force. So it's you having to make those connections and then having to go back and sort of roam through those graphs and figure out what the connections might be. And I think the challenge with that is ultimately is scale. So let's say you get 10,000 or 200,000 documents in, like to make those connections. Like we, our minds just aren't really capable of doing that, like to continue making those connections. So unsupervised is the route that we chose to take.

• 23:26Speaker 00

  And we're, um, yeah, and I think that's ultimately where Eureka moments can live because I'll give you another example. So let's say this is like an investor or an investment advisor. Let's, so let's say I'm the advisor, you're my client and you're like, I have $2 million I want to spend with you. And I want to spend it on like battery technology and like innovative real estate mo- business models for millennials. And like et cetera, et cetera. And so like you're telling me this and I'm like, how am I going to filter that? Like, you know what I mean? Like these models don't have what you just told me.

• 23:57Chris Benson?

  Right.

• 23:58Speaker 00

  So there's a way to use bundle to like support you in that. So let's, let's pretend that my company uses bundle and you know, bundles graphed 2,000 investment models. And I just start typing in battery technology and real estate and whatever. And I just click analyze and like semantically, I'm like, where? Oh, oh, wow. I didn't even know that this like ESG like value, whatever is like a, you know, moderate risk. And they have like, you know, elements on the periodic table, like, you know, whatever lithium and et cetera. So anyway, you know, that's another way to kind of create a Eureka moment. If you're like, wow, no idea. And I don't even know how I would search for that otherwise, other than just spending a lot of time trying to figure it out.

• 24:40Chris Benson?

  I like the two use cases. And the first one, it's a business use case, but it's also very personal in that you're busy trying to, to draw on your own creativity. And then the second one, you're trying to pull different data points that aren't necessarily. Just obviously connected together in such a way to create the value that you're looking for. I'm curious as you, and I'm starting to see how that goes out and there's both a personal and a business side of it. Can you talk a little bit about what data points you're integrating at this point? I know you mentioned email and stuff. What all are y'all pulling together currently that you're publicly talking about? That you're able to, and what should users that want to get into this? Know about the data that they're sharing and privacy, what are you know, kind of some of the disclaimer stuff? Can you share some of that?

• 25:24Speaker 00

  Yeah, absolutely. So first and foremost, we have a have a workspace where you can take notes. That's a really easy way to kind of prime the system.

• 25:33Chris Benson?

  Okay.

• 25:33Speaker 00

  Is you can take notes and bundle IQ workspace and we have an AI that sits in the editor. And sort of, you know, as you're robbing, as you're writing saying, hey, you might want to check this out, check this out. So there's notes and then we also are connecting with the G suite. So we've got connected Gmail so far. So the next one is Google Docs. Those are, you know, long form information that we graph out. And then we have a new one, which is a custom integration that we've created called books. So you can upload, imagine this, right? We've been testing it. We just uploaded all seven Harry Potter books.

• 26:07Chris Benson?

  Okay, that's good. You just won me over.

• 26:10Speaker 00

  Yeah.

• 26:11Chris Benson?

  I like Harry Potter.

• 26:12Speaker 00

  It took like six minutes, right? So we graphed all seven books in six minutes. So you can literally like as you're taking notes or asking, you can ask questions and like query against all seven books. So let's say now maybe you're an academic, right? And you or a student getting your MBA and like you have to do all these research papers. So you just dump in a bunch of, you know, PDF white papers. And like, it just changes your life because like you could just write your article and like, it's literally bringing you stuff from these white papers as you're writing.

• 26:42Chris Benson?

  That sounds good. I like that. It sounds like you're really into some pretty cool use cases for these modern algorithms in NLP. In terms of things you can do, you're really focused on creativity. I like that and being able to kind of enhance. Do you feel like you're still just scratching the surface on what might be possible here? Or what are your thoughts on, do you have any? Thing that you're able to share on any things that you'd like to do next, you know, building on where you're already at. It seems like a pretty cool, a pretty cool space to be in.

• 27:13Speaker 00

  Yeah, it's been a lot of fun. And I think the challenge always is like, what do you focus on next? Right? Like, you have to do something really, really well, solve a really well-defined problem first, and then you sort of land and expand from there, right? We started with notes and then email, because people spend so much time in email. And we sort of give people this sidekick to support their knowledge work in email. But that is really like a mile wide and an inch deep, you know? So where do we go next? And we're still trying to figure that out. Like, I would love to go in the medical space. I think that'd be really cool. But there's so many barriers to entry on that front. So like, you're a doctor and, you know, maybe you see a thousand patients a year and you're like, you know, you see a rash on somebody's left arm and like something you've never seen before,

• 27:58Speaker 00

  but you get to describe it and then you're tapped into some, you know, the broader data set that could potentially pull up, you know, anonymized people's notes, but like, I wouldn't know that it was Chris that had the other rash, but start to make those connections, right? That would be pretty cool. But where we went recently is the climate crisis. When IPCC launched their global climate crisis report that came out recently, it was like 3,000 pages long. You and I are not reading this report. Who's reading this report?

• 28:28Chris Benson?

  No one.

• 28:30Speaker 00

  So like, how do we support climate literacy, you know, for policymakers? You know, I'm a millennial, like, I care about the environment. I'm like, you know, I'm constantly thinking about like, social good. I know, like, you're a plant eater, right? Like, you're socially conscious.

• 28:44Chris Benson?

  Yeah, I'm a vegan, so anyone that's listened to me for a while sadly probably knows that about me.

• 28:49Speaker 00

  So how can we maybe support this massive data set that doesn't have a whole lot of history behind it? There's certainly a lot of dissonance between it. You've got these, like, attorneys and Congress and municipalities are going to be writing legislation and writing policies to like, solve these problems that they don't really even understand. And they, it's no different than the infrastructure bill that's 1,300 pages or however many pages I think that one is. So I think we want to kind of go deep on that. We've got a meeting with a large international law firm on Friday to talk about that. They're pretty excited about it. They work with a lot of, a lot of cities, you know, and climate policy is something that's at the forefront. You know, we got to figure this out. There's a ticking clock against it, so.

• 29:31Chris Benson?

  I like that aspect. You know, we're talking about the business, but you clearly have a motivation that is very personal there in terms of trying to address a big problem. With kind of a new set of tools in the large that people are still just learning and understanding. With these capabilities, as you have been exploring new grounds with a new set of tools and really kind of trailblazing use cases. Whether we're talking GPT three or other similar capabilities. Have there been any moments where you have been kind of going, I found a gap or I've found something. Have these algorithms really solved everything in the in the use cases? Or do you struggle at times with saying, I'm missing a tool.

• 30:15Chris Benson?

  I have a great toolbox with some good tools here, but there are, you know, I wish I had something that did X. Have you had any moments like that where you were trying to solve a well-defined problem? To use your words a moment ago, and you said, I need this thing that's missing from my toolbox right now. Has that come up?

• 30:31Speaker 00

  Yeah, I think in some ways, like so speaking specifically to the model side of it, the AI side of it. I think there's a lot of room for improvement, but I don't know that the technology is there yet. And I think there's just going to be a lot of R and D that needs to be had. Part of it's like adoption, right? So like, you know, up until just a couple of years ago, like we didn't have these AI assistants. Like conversion AI and like these, these tools didn't really exist. And now as we're moving into this like human-centered AI, software as a service enterprise. Companies, they've got big budgets carved out for AI, but they don't even know how to spend it. Most of them.

• 31:12Chris Benson?

  I've noticed that.

• 31:13Speaker 00

  They don't know what to do with it. So there's kind of a gap in time and maybe usability where there's going to be so much just learning, right? So we're, I mean, we are so far ahead. Of where the world is already and kind of this knowledge sidekick that we created. But I think there's just a lot of unknowns. I mean, yeah, I wish we could get a lot, maybe a little bit more horsepower out of what we're doing, but it's already pretty damn good. So we just gotta basically find more use cases and make more revenues so that we can buy ourselves more time to do more things, you know?

• 31:45Chris Benson?

  So you're kind of getting back to, you know, that imagination, that creativity, that innovation, doing what humans do well. So I, I want to ask you, as we finish up here, you've kind of talked a little bit about where we're at and kind of what the things look like a little bit going forward. Can you just, Daniel and I do this all the time, where we will try to make a prediction on how we think the world might look, or maybe how we want the world to look going forward. But beyond just the immediate future, where do you think this is going? What kinds of use cases might you be able to address? What kinds of tools do you think might evolve out of where we're at today? Where is this space going? As someone who is constantly thinking about the future and about how to bring these amazing new capabilities into productive use?

• 32:30Chris Benson?

  Where's the world going with this at this point?

• 32:32Speaker 00

  So I'm definitely not like the technical guy per se, like I've never written a line of code in my life.

• 32:37Chris Benson?

  I'm not sure you have to be technical on this one. It's, this is really like, how is the world changing going forward from where you're sitting?

• 32:43Speaker 00

  And trust me, I hear you and I'm gonna, I'm gonna get there. I'm gonna start with by saying like, I went to a liberal arts school. Like I didn't go to MIT, didn't go to Harvard or, you know, any of these tech schools, Berkeley. And I'm like such a humanities, like history, like I want to go to the opera, the symphony and like that, you know? And ultimately what I think might happen is because there's so many knowledge workers, I think humans, like individual knowledge workers will be able to own their IP. And I think that will bridge the wealth disparity, like the gap in disparity. If Chris and Nick, like if you owned your IP and you're able to own all of the knowledge that you put forth in terms of like moving these enterprises forward,

• 33:27Speaker 00

  and you were able to attribute that as like your unique fingerprint, so to speak. I think Bundle IQ has The propensity to basically like say that Chris is worth this because of all of the knowledge he's brought forth. And like you'll have your own fingerprint, your own ID. And maybe you'll be paid royalties on that. I don't know how you would renegotiate your salary, but like essentially you would get paid dividends on like what you've been able to contribute within an enterprise.

• 33:56Chris Benson?

  So by having the technology that we're using in terms of going forward gives me the ability to maintain IP that I create. And then you kind of just need a business model for how that works in terms of finding the value or other people finding the value from you and stuff going forward.

• 34:14Speaker 00

  Yeah.

• 34:14Chris Benson?

  It's a beautiful idea in terms of it's very liberating in terms of the, you know, what it might imply going forward.

• 34:20Speaker 00

  Well, think about it. If we've bundled all of your knowledge, then the company, the organizations or multiple companies of your contract that you've contributed to like there's transparency now on what you've been able to bring to the table. And you're able to basically have like a fingerprint. Like if you write an email to me, I would know that it that Chris wrote the email because of all the, all the emails that Chris has ever written.

• 34:44Chris Benson?

  That makes sense.

• 34:45Speaker 00

  If you wrote a marketing document, I would know that Chris wrote the marketing document at like a very high percentage of accuracy.

• 34:51Chris Benson?

  Understood.

• 34:52Speaker 00

  So that's what I mean by like vectorizing and bundling your, your IQ and your knowledge.

• 34:56Chris Benson?

  Gotcha.

• 34:57Speaker 00

  Potentially like you're a walking, talking digital wallet of Chris's like intellectual property that you could contribute somewhere else or just get paid on for everything that you've always contributed, already contributed, rather.

• 35:09Chris Benson?

  That is pretty inspiring right there. On that note, I'm just gonna say, yeah, I can't think of anything that tops that. Thank you very much, Nick, for coming on the show. It's been a fascinating conversation. You've definitely given me something to think about here in terms of what that would mean. Very cool endeavors. Thanks a lot.

• 35:25Speaker 00

  Thank you. Thanks for having me and give Daniel my regards.

• 35:28Chris Benson?

  Will do. Thank you.

• 35:29Speaker 00

  Cheers.

• 35:33Speaker 02

  That's our show. Thanks for listening. For more like this, check out our master feed. It is all Changelog podcasts in one easy to consume place. Let your podcast app snag everything we produce and then pick and choose which ones to listen to. Subscribe today at changelog.com/master or just search for Changelog Master in your podcast app of choice. You'll find it. Special thanks to Brake Master Cylinder for providing our music and to our longtime sponsors, Facily, Long Starkly, and Linode. That's all for this week. We'll talk to you again next time.