Episode 56: AI with Dave Walters

Written By Pat Allen
 

This week we sit down and talk to the Director of Product Design at Slice Up, Dave Walters. We talk about what is AI, ChatGPTs’ massive explosion, some ethics around AI, the future of what AI is capable of and much more! 

Like us? Give us a review on Podchaser or Apple Podcasts to let us know!

Follow Breaking Down the Bytes!

 Linkedin |  Twitter | Facebook | Discord


Want to give feedback? Fill out our survey 

Email us! - breakingbytespod@gmail.com

Follow Pat and Kyle!

Twitter:

Pat | Kyle

  • Pat: 0:34

    Hey everybody. Welcome back to this week's edition of breaking down the bites. I'm your host, Pat, as always driving this bus. You can find me on Twitter at layer eight packet. That is the number eight. You can find Kyle on Twitter at Tana Danis 256. And you can find the show on Twitter at breaking bites pod. Alex, you're not on Twitter or any social media stuff. You want to get ahold of Alex, hit him up on the show and it'll get to the right place. So we're pretty active on Twitter. So come say hello. If you like the show, don't forget to subscribe on your streaming platform of choice. Most of you are come from the Apple podcast world, which is totally cool. So, hit that like, subscribe button. And that just makes us look better to the algorithms and the AI and the machine learning world that we that we live in. So, it's all good there. So, Alex, what's up, man? How are you? We're back for another week.

    Alex: 1:23

    Indeed we are. This is the episode that I've been waiting for. No pressure, Dave. But yeah, ever since we, well, ever since I got involved with this is the one that I've been hoping to find someone that can speak to AI and machine learning a little bit better than I can, but also didn't want to get somebody that's right in the PHD on it. So I think we found the right balance and I feel like people are just going to get to the point where they assume that we're just bringing back old evolve IP coworkers. Cause that's the only people we know, but I do think that everyone that we've brought in has served a niche, whether the Ryan talking about containers or in this case, Dave going to hopefully teach us something about AI

    Pat: 2:10

    And we had Alex, before you joined, we had Nikki Townsend

    Alex: 2:13

    and Nikki was on.

    Pat: 2:14

    So she was on. Quite a bit. So, if you're not familiar with Nikki Townsend, go look her up. She runs a nonprofit called tech aware. She's doing some really cool things in that space. So, shout out to Nikki cause I know she listens every now and then. So hopefully she listens to this one, but Kyle is not here this week. He's got other things going on, which is totally cool. That gives us room for, as Alex mentioned. A former coworker of Alex and I's at Evolve IP. We're just bringing back the band here, ladies and gentlemen. That's all we do. So we just recycle Evolve IP. People are like, Oh, you work at Evolve? You're good enough for this show. Rock and roll. Here we go. Dave Walters. What's up, man? How are you?

    Dave: 2:48

    Hey, thanks for having me guys. Good to be here.

    Pat: 2:50

    Yeah, man. It's awesome. This is a long time coming. We were looking for some AI folk and we had your coworker and former of all my P coworker as well with Alex and all of us, Ryan Young on a few weeks ago, talking containers and dockers and all that kind of cool stuff. So if you haven't checked out that episode right after this one. Go check that out too, because that was a really cool conversation. But Ryan is with you now at y'all both work at slice up. Yeah.

    Dave: 3:15

    Yeah, it's correct. Ryan actually stole me about six months ago to come work there at slice up.

    Pat: 3:20

    nice. Good deal. So Dave, why don't you give the people a little bit of quick rundown of who we are, where you came from and sort of what you're doing now, and then we'll take it from there. So the floor is yours, my man.

    Dave: 3:31

    Awesome. So I'm Dave Walters. I'm currently working as the director of product design at slice up and slice up is an AI ops platform for networks. I guess I got my start in machine learning back in undergrad. I double majored in computer science and mathematics, and there's really two classes there that. Grabs my heart. And that was algorithms and data structures and data mining. And then I went to evolve IP. I worked as a software engineer for eight years and didn't touch a lick of AI or ML.

    Alex: 4:01

    Yeah. And I thought that too, I was kind of, when I saw the, like what you've been working on in your LinkedIn profile, I was like, man, did I just really not have any idea what the guys at Evolve IFE were doing? Or did you manage to get into the field that everyone's kind of struggling to get into?

    Dave: 4:20

    Yeah. Well, after evolve I was looking for things were aligned with my interests. They were fun to do and I wanted to work in gaming. So I ended up at games 24 seven and their tagline is the science of gaming. And they're actually a real money gaming company. And the science of gaming was all about data science, actually. So we were looking to maximize user engagement working on optimization problems like automated AB testing and the multi armed bandit framework, if you're familiar, and ultimately getting people addicted to real money gaming was the goal

    Alex: 4:53

    So, sell your soul a little bit, but at least you get in the industry.

    Dave: 4:57

    Yep, exactly. And once I was done selling my soul, I saw another AI ML opportunity with slice up and I was happy to take it

    Alex: 5:05

    So, were you considering leaving at the point, at that time, or you just got this opportunity from, I don't want to say got it from Ryan, but were made aware of it by Ryan and then just knew right away you wanted to pursue it? You already, the writing was already in the law.

    Dave: 5:22

    a little bit of both. So I've been, you know, my background is a software engineer, but I've been working in the product and UX space for the last three years or so. But I love keeping my hands dirty with code. That's where I find a lot of enjoyment and I wanted to do more work on code. So I started doing some consulting and then Ryan said that slice up could use some of my skills. So I started writing some code for them and then working for them more and more. And I just really like the fit. With the team and I was really inspired by the stuff that we're working on. So I decided to make the switch full time and left games and started working on. So I said,

    Alex: 5:57

    So when you say you've been that slice up for six months, you've actually had some type of affiliation with them longer than that?

    Dave: 6:03

    yeah,

    Alex: 6:06

    Well, yeah, I didn't realize that. That's so in your current role today. Well, maybe I'll step back a little bit. You talked about it briefly, but since the topic is AI and machine learning and you admitted that you didn't do much of that or any of it at Evolve IP. What was your role at what was the name of the company you were at

    Dave: 6:28

    games 24 seven.

    Alex: 6:30

    Okay. And what was your role there and pretty much how did you land that role given that you hadn't done much of that before?

    Dave: 6:39

    Yeah. So I was a director of user experience there as well. And since I, most of the software that I wrote it evolve was on the front end. I got really involved in. Working with Scott Kinka more on the product side and like, Hey, how can we incorporate our users more into our product? Right? Because if you remember osmosis and other systems that we built, they were much, they're very much built by engineers for tech adjacent people, right? They weren't the best. They weren't the most user friendly. And as I got more familiar with the UX processes and design thinking and things like that, you know, I really saw an opportunity to move in that direction. And. Games 24 7 needed that for a new division they were starting, which was all about user engagement and retention marketing and things like that. So I was a really good fit for them for that. And through working in that role we had PhDs in data science, working on our team. And on our board of directors that I came in contact with frequently because we were strategizing that way, you know, from the predictive analytics optimization, data science lens for what we were doing and. That was super cool to me. It's something I've always been interested in. So,

    Alex: 8:00

    it would be interesting to have a conversation with them. So. With your role, where you kind of, you knew the end goal, but the data scientists were the ones that could help actually write the algorithms that got you to that end goal or what something could do.

    Dave: 8:19

    yeah, exactly. And the way that it ended up there is that a lot of times the data scientists that we had were, well, they were academics. We had one from one from Wharton and actually I think they're both Penn guys. But. It was at times much more of a lecture that we were receiving than a product that we were developing. So I ended up getting heavily involved in the productization of the data science that he was kind of. Teaching us, if that makes sense,

    Alex: 8:52

    Interesting. Okay, and so I guess when you started this role, you mentioned that gaming was. That got you into this role. So when you took this role, did you even realize that you're going to have like your first dip into AI and machine learning when you took this role, or that was just a happy surprise.

    Dave: 9:15

    bit of both originally it was just kind of needs based, you know, they needed someone to be front end team lead. And then we quickly realized that they needed someone to also teach them UX because again, a product made by engineers for engineers was not the most user friendly. And then. We also realized that we needed somebody who had strong product skills, and I happen to have an interest in background in mathematics and we have data scientists on the team. So my role is continually expanding every day, but it's a startup. So I'm happy to wear many hats and, you know, I love what we're doing. I think it's super interesting. So I'm more than happy to take that on.

    Alex: 9:52

    Okay. And. Where are you saying that's what your current role right now at slice up is Or how is that different than your previous role?

    Dave: 10:02

    My previous role was mostly focused on the user experience where now I'd say my role is more focused on the product holistically and the product is.

    Alex: 10:15

    Okay, and maybe we should even started with this and I'm sorry, I don't want to offend anyone. I'm not too familiar with the company. So, when you say it's AI machine learning base for networks maybe you can elaborate that on a little bit more, like what actually is the product and

    Dave: 10:31

    Yeah, so we'll forgive you for not being super familiar since we're a startup, but basically we're

    Pat: 10:37

    Thanks, Dave

    Dave: 10:40

    we're developing an AI ops platform that uses AI and ML to detect anomalies and correlate network events from various data sources. To provide high fidelity context around network issues. So if you think of the noise that you generally have with like the 30 different monitoring tools, you use like net cool and Nagios and all the server and container logs, you have to dig through to figure out what's going on slice up. Takes all of that and processes it through our AI pipeline, and it puts the puzzle together so that people don't have to, and instead of digging through all of that noise, they can just focus on fixing the issue. So our platform process this and spits out, you know, here's the issue here, the relevant metrics. Here are the logs associated with this event. And then we've also built an auto remediation feature that lets you say, Hey, if you see these conditions happening again, run this set of instructions, which comes via an Ansible script and then fix it before it becomes a problem. It's super cool stuff. I'm not a networking guy. So a lot of that stuff mystifies me still, but

    Alex: 11:46

    Yeah, the whole idea is self healing networks. Yeah, it's a term that's been thrown a lot around a lot last few years, but now it might seems like it's a real thing. Well, we are networking people, so I am somewhat interested in this, maybe more than. Some people are, but what's the like ingest method. So if I want to use slice up, am I sending you just any type of log SNMP net flow data, or what can I send you?

    Dave: 12:14

    Yeah. So we were taking in syslog from devices. We're taking in application logs. From whatever application servers your apps are running on we're working on exchange server data data from load balancers like the F5 networks, load balancer applications. We take an SNMP data. It's just, it's as much data as we can from whatever source and then. Using that to show you, you know, latency path changes with delay associated giving you warning about possible over utilization on interfaces or on T cam modules on your Cisco devices. Yeah, lots of use cases for different things. I think the most interesting is the log parsing that we do on some of our customer networks. As soon as we've deployed the system and let it run, It's Picked up things that they weren't even aware were happening, right? Because logs are the number one thing that tend to be white noise, right? They're going and maybe somebody will look at them after something happens, but here, you know, duplex mismatch came up right away. And we could say, Hey, we're seeing latency here. Here are logs that are spitting out saying, this is why easy fix, right?

    Alex: 13:24

    Yeah. And I kind of think that the, what's the word I'm looking for, this this type of work, this type of business model, it seems like it's getting a bit crowded. Who do you think is kind of like the main player that, I don't want to say it's your rival or your biggest. Contender for business, but is there 1 that pops up right away that you think of that somewhat competing with

    Dave: 13:55

    It's hard to say. Cause like you're saying now everybody's starting to get into it, but I mean, solar winds is big and they have a lot of these types of capabilities.

    Alex: 14:01

    And maybe this is a difficult 1 to answer. Is there a feature that you see for my solar winds or for someone else that you kind of feel like. Slice up could do better to mimic that. Maybe when you're a little jealous about, or one that you feel slice up does a lot better

    Dave: 14:24

    no, I'm not sure. I haven't gotten to, you know, we have so many things on our plate right now for the upcoming use cases that we've promised to our customers and our roadmap is so packed that. A lot of the research that I probably should be doing on competitors, you know, I just haven't gotten to so yeah I'm not sure, really.

    Alex: 14:45

    Understood. Well, I asked a lot of questions there, so I'll let you take the floor pat, see if there's where you wanna

    Pat: 14:54

    ah, just a couple that I had in mind. So obviously with AI sort of taking the world by storm, at least at least a little bit here in the last I don't know, say probably eight months to a year, somewhere in there. You know, when it comes to AI what do you, like, what do you think you like with, in relation to the IT field, I know that's a broad thing. What do you think of AI when it comes to IT, in the IT terms, whether that's, you know, again, with logs and sort of, you know, smart logs, if you will, or at a help desk level, trying to, you know, automate the boring things of that nature, or or, you know, almost like a predictive Being proactive instead of reactive sort of thing. Like, what do you think of AI in that space? Or what do you think that's going to kind of turn into, or what, where is it as current state, I guess? I know there's,

    Alex: 15:43

    well, that's a loaded

    Pat: 15:44

    unpack. So yeah, I know.

    Alex: 15:45

    may. How about like, wt, I think AI is just like a topic that could mean so many different things now. So like when you think of ai, like how would you even go. To begin to define what AI can do for IT or what it's already doing.

    Dave: 16:05

    Yeah. The way I like to think about AI is I think of it as the development of computer systems can perform tasks. It used to be humans only, you know, things like problem solving and learning and pattern or object recognition, understanding language, perception, and things like that, and I think that the opportunity. That it's bringing to it is it's going to turn. It's going to allow every individual contributor to become a manager. Right? Because. I think that it takes away a lot of the things that we don't want to do anyway, To talk about it in non it terms, everybody's using chat GPT to write things, right? It's great for that. It's a generative AI. But we still need people to edit it, right? It hallucinates it might come to wrong conclusions that you weren't hoping for. Right. But it speeds up generation process, right? Like, you can brainstorm infinitely faster now. And that's true with coding as well. Right? So, you know, I was a software engineer for a long time and then eventually a team lead and then a manager. So, As I progressed, I wrote less and less code, and then it became more about reviewing pull requests and checking for optimization opportunities and making sure that code wasn't blatantly wrong. And that's what all software engineers are going to become, right? Because we won't need to write the same boilerplate code that's always been written. Chet GPT understands that very well, and it can produce that. But what we will need to know is how this code relates to other parts of the system and the impact that it can have and things of that nature. And I think that's really exciting. And you know, that it goes to the help desk level two, like with what slice up can do, you know, there's no need for you to dig through all of your customers logs and old tickets and see to have, they had this problem before we have AI that can do that for you. And it'll say this customer has had this problem 13 times. Just tell him to turn it off and on again.

    Alex: 18:16

    Okay. Yeah, and I think everyone's fear is that it's going to eliminate some jobs. And I think what you're saying is that. you still likely have these tiers of jobs. It's just even tier one guys are going to be doing more than they've ever done before. It's just because they have a tool that can assist them with it. Just like, you know, 15 years ago or maybe not 15 years ago, they had 20 when did Google come around before search engines were really popular.

    Pat: 18:42

    God, we're

    Alex: 18:42

    you were capable.

    Pat: 18:43

    God.

    Alex: 18:45

    Yeah, I know. And now they've been around a while. It's just like, I'm thinking, man, I'm almost 40. So yeah, this has been around a while. Yeah, and do you see maybe a situation where we don't have the concept of tier one people anymore, but it's not like they're losing their jobs, they're just becoming more in line with what a mid tier person could do nowadays, is what they're

    Pat: 19:11

    I think it's a shift.

    Dave: 19:13

    Yeah, I hundred percent agree with that. I think that it's. It's going to elevate everyone in the same way that instant information via your smartphone did, right. Or the calculator, right? We don't have people who manually compute things anymore. We barely need lawyers anymore. Like there are things that,

    Alex: 19:32

    Right, yeah, I mean, before the

    Pat: 19:34

    goes our

    Alex: 19:35

    of the calculator,

    Pat: 19:36

    Sorry about that.

    Alex: 19:38

    okay. So I can, I guess I touched on it a little bit, but it bleeds into this the next question I'll ask. And it sounds like you're not, but... Are you one of those people right now that are in any way concerned with like it's rapid adoption say in the last six months or it's just seems like chat GPT. I know it's been around for a while. I didn't do any demos with it and chat GPT version three. It's just when four came out that I got really involved with it. But to me, it just seems like how. How did they make such a monumental like increase in just usability performance and it's stay under wraps. I just, it blew my mind how much better chat GPT 4 was and that scared me thinking like, Oh man, if it got that much smarter, has it really just, you know, have they just been sitting and letting it learn and it's gotten that much better? But yeah, I'll just go back to my question. Are you at all concerned at how quickly this is improving? Are you all for

    Dave: 20:46

    I'm all for it. I'm a hundred percent in I love to see the innovation and it's been around for a while. I think that the reason that chat GPT in particular has captured the imagination of. So many is that even though this is a tech, I mean, transformers themselves have been around for six years. Google wrote a paper on them in 2017. I think chat GPT was the first time that users could interact directly with a model, right? Like if you look at the chat GPT experience, we've been using AI on our products for years now, you're. Your recommended feed on YouTube, your Netflix recommendations, the Google search results are the result of neural networks at this point. But they've, that AI experience has always been wrapped inside of a product with a very specific outcome. Chat GPT was the first time people were just given an AI model and. Told do what you will with it, right? Like, we've had auto complete, which has been using, you know, different forms of machine learning and things like that. And chat GPT is the fanciest auto complete that 1 could use. And I think that it was a great move by open AI to release it. Google Bard has been ready for a long time. They wanted to release it as a product that was ready for users to consume. Where open AI was like, we don't know what this is. And we think that's a good thing. Let's see what happens.

    Alex: 22:20

    So like the world's biggest beta ever,

    Dave: 22:24

    Yeah, absolutely. And one of the reasons they were able to make such fast improvements GPT 4, I think is even kind of an old model, but. It's more recent tunings have been based on RLHF, right, which is reinforcement learning from human feedback. And that's taking your thumbs up and your thumbs down and your reasons for giving those thumbs up and thumbs down during the beta of chat GPT to really improve those responses improve its reasoning capabilities. Fix it's bugs in code, things like that. So doing that really big beta has just moved us ahead really quickly, you know, hitting a hundred million users in a month or two months or whatever

    Alex: 23:06

    Yeah, something like that. It was mind boggling. And you just threw an acronym out there that seemed like a really good one. Say that one

    Dave: 23:13

    RLHF it's reinforcement learning from human feedback. And that's basically just letting people. Tell the machine when it's right and wrong so that the machine can apply those learnings in the future by adjusting the weights in the model

    Alex: 23:29

    And I think that's, would have been the most impressive things that I've done with chat GPT. Cause as you said, in its simplest form, you could just take that as a thumbs up or a thumbs down and Netflix uses to suggest movie titles to you. Or whereas chat GPT, and it just blows my mind away is having these conversations with it, where I'm iterating over things that I've already asked it, like I've used it to write code as someone who doesn't code, but is technically minded. And I've written, I shouldn't even say that I've written, I've had it write code for me by just speaking to it as a normal human would. And when I realized those things are. Not working correctly and the way how quickly it responds back and understands. Oh, I see what you're getting now. I can just iterate over what it's done in the previously. It's it's just mind boggling and that's a constructive way to use it. I've used it to just play around to and have fun with it. Like I have 1 chat right now that I use it. For everything, any random thing I want to talk to it. But I asked it to now continue speaking to me as if it were a Texas cowboy. So now every time I ask it something, it says something like how y'all doing? I reckon this is what we should do here. And I just kept it going because now it's, you know, from my feedback, it knows that's the way I like how it responds to me. So it's given me very constructive. Text based on subsequent questions, but it still remembers that I've asked it to talk to me like that. Which I I think that's mind boggling. I just, it's so much fun. And,

    Pat: 25:15

    of all, can I get it to, can I get it to talk to me in like a weird alien output? I would love to be like, hello, take me to your leader. It's a dumb shit like that. That would be awesome. I'm in.

    Alex: 25:27

    and it's almost like, I hate, I don't want this to just be a chat GPT conversation and we will get off the topic, but I, one thing that. So many people have asked me, so I might as well bring it up here and just get some feedback from you two. Is what is the coolest thing that you've gotten ChatGP to do or the thing that just Impressed you the most and just see if there's anything out there people haven't seen yet And there's a few that I have that but I'll leave you guys to say yours first

    Pat: 25:58

    Dave, you're up.

    Dave: 26:00

    Okay, I so I don't use Chet GPT directly a ton but one of the things that I think is super impressive that it can do is translating languages and here's why. So when people talk about Chet GPT being a fancy autocomplete that's true, and it's also. A bit false because of the reasoning that it can do. And if you think of translating a sentence from English to a gendered language like French, what Chachapiti can do that requires reasoning blows my mind. So think of the sentence, the book will not fit in my book bag because it is too big. And the sentence, the book will not fit in my book bag because it is too small. There's only one word difference in that sentence, right? Big and small. But when I say. Because it is too big and it won't fit, I'm obviously referring to the book, and when I say it is too small, I'm obviously referring to the book bag. We can recognize that because we're people and we recognize those nuances, but for a machine, there's a one word difference in those two sentences. But when it translates to French, book, I believe, is a masculine word, where the book bag is a feminine word, and it translates the word gender correctly in each instance. which used to be something that would require a human, and that's why Google Translate used to kind of suck. But now that it uses neural networks, it's learned more than just patterns in language. It's learned a bit of reasoning and small things like that blow my mind. It also blows my mind when it screws up. Really simple things like that too. But on the impressive side

    Alex: 27:54

    Right. And I think that's the part that gets scary is because it's going from just like a variable translator to it understands intent and well, maybe it doesn't understand intent, but that's what it seems like to the end user. It understands what I'm really driving at. right.

    Dave: 28:10

    Yeah. The inference that it can do is spectacular.

    Pat: 28:15

    I'm just a nerd and I only use it for nerdy work things, which I probably should open that up a little bit and use it for other things to not related to work. But I just, I had to just write a Python script for me because I'm an idiot when it comes to writing any sort of code. So I had to write a Python script to change some. Config on Cisco iOS stuff. So, it, it did a decent job of, you know, putting it all out there, commented a bunch of stuff too. So it was nice and clean, which was great. And then I just basically had to put in my variables of what, you know, what I was trying to connect to things that interest. So I had to tweak it a little bit. And it was probably more of my lack of in depth like Python language that kind of. But it did work pretty well and I just had to do variables and plug and play. And it was pretty cool. But the other thing I've played around with too, is I wanted to give me a list of it podcast topics and most of them we actually already touched. So I was pretty proud of myself on that one. I was like, Ooh, yeah, I could. I'm as smart as chat GPT. This is awesome. So it's all good. So I was like, that's kind of cool. But yeah, I don't really use it for a whole lot outside of like work things that I'm really kind of stuck at, but I probably should open, open that box a little bit wider and just start using it for dumb stuff too.

    Alex: 29:31

    Oh,

    Dave: 29:31

    work things, kind of jumping back to the, how is it going to change it topic? One of the things that I think AI is going to enable us to do is it'll be the death of specialties in a way, right? Like right now you have Python programmers and Java programmers and go programmers. With the assistance of AI, I've been, and GitHub copilot, I've been writing in every language. Right. We have go in our stack. We have Python in our stack. We have JavaScript and TypeScript and everything under the sun in our stack. And before it'd be like, Oh man, I don't really know Python. I'm going to have to spend three weeks figuring out the syntax and, you know, the data structures and things like that. Where now it's like, Hey, I can just write the equivalent JavaScript, put it in chat, GPT, tell it my intent, what I want, look at the output. It's 95% of their most of the time, if not 100 and then knowing programming, you can fix the bugs, right? Because. Data structures, function calls, et cetera.

    Alex: 30:30

    Yeah. It'll just, I guess we'll just become like pseudo code bas, if you just understand the structure of the code, you understand how coding works in general, then an AI can write it in any language, I guess.

    Dave: 30:45

    Yep.

    Alex: 30:46

    yeah. That is interesting.

    Pat: 30:48

    interesting. The one thing I want to jump in with, and this is not chat GPT related. It's more AI centric. I was I was listening to there's a PhD guy. Oh Miku, I think it's Miku Kaku. I think his name is. He's a professor out in, out west somewhere. And like, I've seen him on like, like history channel, like shows and things like that, because he's like super smart, like, he does all like the space stuff, and he's like this crazy, like, he's just, he's brilliant. His mind works in ways that mine's not even possible. But he was on Joe Rogan. Last week he was on Joe and Joe likes to get in all that crazy, like alien stuff and you know, all that weird stuff which I like too. so yeah. So like, I was listening, I was half listening as I was working last week and just some of that conversation just fascinates me, but they got on the topic of AI. And there was almost like a, there's almost like an ethics conversation that. There's a piece of that comes with AI as well. Cause then they were talking like, okay, AI is great. It's the world's at this point, you know, they compared it to the world's best Googler or Google on steroids, you know, sort of thing. Right. But then they're also like peeking behind the covers of like, okay, like how does it learn that? And like. It wasn't smart enough to like understand nuance and understand, like, basically just regurgitating what it can find on Google. So that means, you know, is there an inherent, like an inherent, like bias to it, or like, it's not smart enough to kind of. Do like things without sort of nuance built in. I just kind of interested on the take of like, yeah, AI is spitting it back to you. It's making your life easier, but there is a behind the covers aspect of that to be like, is it true information? Is it just spitting misinformation? Cause it doesn't know the difference between actual truth and like a misinformation, crazy hot take, right? That kind of thing. So I'm curious on that. It was like. It doesn't get better as we kind of grow with this thing. Does it get better in a couple of versions? And it's like, then it comes to the thing of like, okay, it's really getting smart. Like just because we can, does it mean we should, you know, that sort of thing? I can't, that's where my ethics sort of come in to be like, yeah, okay. You know,

    Dave: 33:03

    Yeah. Yeah. And that's the concept of responsible AI. It's something that's really talked about right now and it's important. And that is one of the reasons why, even though Google was ahead in the AI race, let's say for the longest time, the reason they fell behind open AI is because they didn't want to release their model for those reasons.

    Alex: 33:23

    Right.

    Dave: 33:24

    And aI doesn't know the difference between right and wrong. Right. So,

    Alex: 33:28

    right. And I think that was even in it. I was just gonna just chime in there that I think that's also chat. GPT had some media concerns where they had to address Yeah. Ethnic related issues where it would say things that were somewhat questionable, but Yeah, that's a

    Pat: 33:45

    Yeah. I'm interested to see where that goes.

    Alex: 33:47

    because, yeah, it's since it doesn't understand that like I'm trying to like the worst case scenario is like hate speech or something like that. It's just like, well, then what dictates what hate speech is? And then now you're starting to get to some really tough questions that. Who has answers to that, and do you need a governing body that is saying what these AI models can learn? And I don't know what the answer is. I mean, I guess when you're dealing with computer networks the idea of ethics is not quite the same as when you're working on something like

    Pat: 34:18

    Right.

    Dave: 34:20

    yeah. I think, and I think that's where kind of the humans as the editor layer, you know, is going to continue to be important. I think that where we really get in trouble with AI is, I don't know if you guys have. I've been paying attention to like the auto GPT and the LLM agents that have been super popular very recently. And what that is basically people taking large language models like jet GPT and hooking them up to the internet and also giving them. Execution authority to do things. So instead of just having a conversation with you, it'll have a conversation and determine the next steps for your desired outcome. So if your desired outcome is, I want to write an email to Pat about the next podcast. Instead of just giving you the content of the email, it will give, it will generate the content of the email. It will open. Whatever it needs to, it will insert the body of the email. It will hit send. It will watch all that. So I think that once we have AI is determining their sub tasks with no morality and no ethics, that's when things start to get a little dark for me. Because who's to say that the best way to do a task is what the AI is going to come up with. It might be the most efficient. It might not be the most ethical. Right. And then you have companies like Palantir who are. Using AI to make weapons, and that is their business model and their goal, and you think about just because we can, should we? So think about a self healing minefield, for example, right? Like, what if we have the mines talk to each other, and once one detonates, they reconfigure themselves, so there's no blind spots in the minefield. Like,

    Alex: 36:02

    Oh,

    Dave: 36:03

    an effective, but not a good use of AI, right?

    Alex: 36:06

    And then it decides that we could increase our affectability if we eliminate this neighborhood nearby and we can have more mines. This is a good thing.

    Pat: 36:16

    it. That's it. I just don't want to be like. In a world where like Skynet 2. 0 is taken off and then like all of a sudden they just start talking it in a language that nobody understands you're like, I'm just going to unplug this real quick. We just pulled a, you know, you know,

    Alex: 36:32

    And I think what Dave was touching on, I think big groups like open AI or Google they'll always have that editor in charge. That's going to try to, you know, that team that's going to try, but if they. Create like these just open plugins that anyone can use and they can just bring in that technology, then who's going to govern. Anyone making programs that utilize it and just saying, Hey, for simplicity sake, I don't want to spend you know, however much money on just employees to review this, just I'll accept whatever it tells me

    Dave: 37:06

    hmm.

    Alex: 37:08

    it'll just eventually get to the point where it, and I don't know how you can stop it. I mean, short of you don't. Allow plugins, but that's not going to happen. It seems like everyone's using it right now.

    Dave: 37:20

    Yeah. And, you know, governments want to go the regulation route, but, you know, people that want to do bad things don't follow the rules. So how effective is that going to be? And the other problem that we're encountering very rapidly is that the progress of open source with these large language models has been exponential. As well, right? Like we talked about how fast PT four got better compared to three. Well, once the. Llama model got leaked by Meta, the open source LLM community has really taken off and there are open source models now that perform to about 94% of what GPT 4 is doing. And that's good enough, right? To do some damage. So

    Alex: 38:06

    For sure.

    Pat: 38:09

    that's interesting. You got anything else, Alex on that?

    Alex: 38:12

    Oh man. It's a scary topic. So I'm all for AI machine learning and chat GPT and everything it's doing. And I don't want to hear any more negatives about it.

    Dave: 38:21

    there you go. Yep. Hey, I'm right there with you. It's there. There's a lot of positive. That's going to come for sure.

    Pat: 38:27

    I was going to say as far as like, so chat GPT is the juggernaut in the room when it comes to AI and things like HR, are there any others that people should be taking? Note of like, any equivalence to chat GPT, whether that's text based or is there other flavors of AI for other things? I guess. I don't know if that's the best way to kind of put it, but is, are there other AI things that outside of the chat GPT realm that people could take a flyer on?

    Dave: 38:54

    For sure. So just that, since we weren't talking chat, I'll talk about some LLMs or maybe even chats. So hugging face has it. Yeah. Hugging Face has a chat. They're an open source community and library full of large language models. Hugging face.co/chat I think is their chat url which is using open assistant, which utilizes some of the leaked meta model. There is Hey Pie, which I think is using una, which is a large language model that's part of the open source community. Of course there's Google Bard. Yeah, there's lots of things. Popping up, as I said, like open source is catching up very rapidly.

    Pat: 39:32

    usually does.

    Dave: 39:34

    yeah, especially, you know, if a big company's secret sauce gets out, that's,

    Pat: 39:39

    Right.

    Dave: 39:39

    it's always going to speed things up a bit. Yeah, exactly. But other parts of AI and machine learning that I'm super interested in are things like object recognition. And the way that's going to change the game. One of the things that people might not know about the AI and model training process is it takes a ton of compute to train these models. Like chat, GPT, you know, there's thousands, maybe tens of thousands of GPUs because the model has a trillion parameters and it takes all this energy to create the model, but running the model takes far less

    Alex: 40:16

    Oh,

    Dave: 40:16

    and we will. Get to a point where we can run models and people have already run models on a pixel six, right? Because the pixel phones have a dedicated tensor chip for running AI. We'll have AI in our refrigerators with object detection that can just tell you exactly what you have in your fridge while you're at the grocery store. So you will never have to look again or worry about things like that. We have AI in our cars. There, there's just so many applications of computer vision and things like that, that are going to change our lives for the better. They make me really excited about AI.

    Alex: 40:51

    Yeah. Even something like, like, because you talked about computer vision and cause we, we focused on language models, like chat GPT, but Google lens has been around for a while. And I remember the first time playing around with Google lens. It like blew my mind. Like, how is it, how's no one else talking about this? Like, do you have any idea what this is accomplishing? This is amazing.

    Dave: 41:10

    Yeah, I use it for. that I see that I don't know what it is, if there's a weird bug, Google Lens. If there's a cute dog, Google Lens. Like,

    Alex: 41:20

    And by bug, you mean like an insect, not like code. Okay. I

    Dave: 41:23

    No!

    Alex: 41:24

    to say I never went that far.

    Pat: 41:27

    yeah That's awesome

    Dave: 41:29

    Hey, these days it might work for that too.

    Alex: 41:31

    Yeah, and it's and phones do it too. I guess Android has a built in type of lens functionality too. But the idea that you can search your pictures, cause like most people, you probably have like a thousand pictures on your phone. And I just used that the other day where I knew that I had my insurance card somewhere on my phone. I needed it. And I just put insurance card and sure enough, it came up with eight images. And one of them was my insurance card

    Dave: 41:55

    Oh yeah, I love that feature. That tells me the other day, I didn't have my license cause I lost it. And we are going out to Pete's dueling piano bars down on sixth street here in Austin, Texas. And I didn't have my license with me. And, you know, it's an embarrassing moment where like, Oh, well, we can't let you in and then a bouncer says, do you have a picture of your license by chance? I was like, Oh, I know I do it. Just typed it in driver's license came right up picture from like two years ago,

    Alex: 42:22

    Yeah. And I think some people just like, especially if you're not in the IT, we'll just kind of glance over that. That's a cool feature, but how like, just amazing of a technology. It is what you're thinking about, especially if you're interested at all with AI and machine learning, what that thing just accomplished. And it's almost like, you know, it's just an add on feature that people just kind of ignore it, but that is an amazing tech, amazing.

    Dave: 42:49

    Yeah. And that goes back to what we were saying about how AI is all around us and in so many of the things we use, but it's always been invisible, you know, until we've interacted directly with

    Pat: 42:59

    I guess the question is so we kind of covered on what you're looking forward to the most some of the object recognition, things of that nature. Do you have any predictions for the next Year, two years, five years to kind of where we're living. I know the fridge thing is a big one. You're starting to see that now, like, yeah, with things in the fridge. And she's going to hate me for saying this publicly, but I wish my wife would take a flyer on that. Cause she always buys like she, she buys stuff. And then we have like two of them already at home. And I'm like, what are you doing? Like, this is just,

    Alex: 43:28

    You got six balls of mustard and no one in the house likes mustard

    Pat: 43:31

    Like, I'm like, honey, we don't need three taco kits. I mean, I love tacos. Don't get me wrong, but like, why don't we have three of them? And she's like, I don't know. And like, and she buys the kids, which is great. She buys she buys all the vegetables and all these fruits and stuff. And my oldest likes the likes cucumbers. And so they make them in the, like they're little tiny cucumbers. So she can actually, you know, she can handle them and not the big, you know, not whatever like, but they come in like a cellophane plastic, you know, sealed thing. There's like four boxes of these things in my fridge right now. I'm like, she's not going to eat that many cucumbers. What are you doing? Like, but I totally see that you know, coming into play with with the household items, the stoves.

    Alex: 44:11

    have extra boxes of cucumbers. Two year prediction, Skynet.

    Pat: 44:16

    That's right. That's right. That's how we roll. That's it. That's it. But no, that was, it's interesting to see like the smart home is obviously it's been taken off the last couple of years. AI has a big part of that. You know, internet of things, right. That's a big part of it. AI is tying into that again with the object stuff and things of that nature. And I always, from. I always remember this. There was a video out a couple of years ago. This is probably going back to our evolve days. So that's probably a good seven, eight years now. It was from Corning glass and they had a video. It was like a smart home of the future sort of thing. And everything was touched. So like. The stove, there's no buttons on the stove, you would just touch it, and, you know, it would be, you know, electric heat, or, you know, whatever, that kind of thing, or, like, the the one girl came down and, you know, touched the fridge, and, like, she was calling grandma, but she was using the fridge surface as, like, the video, and, like, it was just crazy, I was just like, what sort of Jetson's voodoo is this? Like, this is insanity, and now it's

    Alex: 45:15

    a Black Friday deal

    Pat: 45:17

    Yeah, no, it's just like, I can go get one if I really want to drop a couple grand on it, but I can totally get one. I'm like, what a time to be alive. This is insanity. I love it. But yeah, I think AI is totally kind of driving that. You know, innovation that driving that you know, home of the future and, you know, everything sort of around it. Cause like you said, you had to have an app to front that before they'd been using it for a while, but you had to have an app that interface with you. Now you're just given the raw engine and they just said, go. And you're like, Whoa, nobody's ever had this power before. This is kind of cool. And here we are, you know, so it's really interesting to me to see. But I guess you have any predictions for five years down the road? Like anything that you can kind of carve out or is it too early or

    Dave: 46:02

    it. I'd say so for five years down the road, I think we'll see the biggest change in. Medicine and education.

    Pat: 46:09

    okay.

    Dave: 46:11

    talking about object detection and pattern recognition, we've already seen that AI can now identify things. At a level as good as or better than a radiologist, for example, because you think of the best radiologist in the world, they'll see maybe what 10, 000 x rays or MRIs in their career where AI has seen billions, right? That's its job and it will find my new details and change things like that. You know, we've mapped the human genome, but we're still trying to understand it and the way different things interact. AI is terrific at things like that. I don't know how much you guys watch the. Clinical trial space, but Moderna's doing some really cool things with personalized cancer vaccines. They're mRNA based and they're making a lot of progress with that and AI is helping hugely with that. And you know, those vaccines are personalized to the person that is using them. And able to see the variability in the individual human and still generalize solutions to fix those problems. There's incredible applications there. And then the education space, if you guys have seen what Khan Academy has been doing, they've completely changed the game using AI. Every student has their own tutor that's AI based. And every teacher has their own teaching assistant that is an AI. And the way that kids can learn with this. Is so different in ways that are incredible. Like we always think of students using chat GPT to write their papers. Let's say,

    Pat: 47:44

    Yeah.

    Dave: 47:45

    but now we have an AI. They can take on the persona of George Washington and the student can interact directly with a version of George Washington to ask him questions and. It's such a more immersive way of

    Pat: 48:02

    Yeah,

    Dave: 48:03

    And I think that's going to be so beneficial. Incredible. I mean, teachers spend so much of their time lesson planning. Great thing for AI to do grading things. Great thing for AI to do, right? Like we have overworked underappreciated teachers. We have help coming, you know, so

    Pat: 48:24

    my wife is one of them, second grade, she is burnt, she's burnt to a crisp, let's put it that way, so, yeah, I get it, I totally get it, but that's interesting, cause also, going back to the Joe Rogan and Miku Kaku. I hope I'm saying his name right. Dude is brilliant. Anyway. They talked about, you know, the AI and everyone that's living now, like having a digital space or a digital footprint that will live on literally forever. And so our great grandkids will be able to literally have a conversation with us like we are talking in real time. Like, it was just, I was just like, wait, what? I had to rewind that part of the podcast. I was like, hold on, let me see, let me hear that again. But yeah, every, everybody that is living today has a digital footprint that is going to live forever. And you know, our great grandkids will be able to virtually meet us and talk like we're talking now and get a feel for who we are and things of that nature. And I just found that so fascinating and AI is totally driving that whole. Sector of things and it's just like it just it blows your mind. It absolutely blows your mind.

    Alex: 49:34

    Yeah and that technology, I guess, is really here today. I mean, I know they kind of, with these language models today, if you fed it enough. History of your speech. I mean, like I've actually talked about this with some of the people at work. Just spitballing ideas. Like, I've used Slack and Teams for I don't know how many years now. If I were to just feed it every conversation I've had in the last three years, this thing could probably mimic me as if it were me. And that's alarming that the technology is there for that. Today, I think.

    Dave: 50:08

    And if you think of the deepfake technology, you know, deepfake being its bad usage, but the good usages of it, there's enough video of you to recreate you on in any circumstance. That they want, right? But that'll be good when our loved ones want to talk to us in 200 years, you know, you can talk to great granddad in his 20s

    Pat: 50:27

    Yeah,

    Dave: 50:27

    realistically, which is cool. You can talk to any president from, let's say the 80s on where we have sufficient footage and recordings of them, like.

    Alex: 50:37

    oh man, maybe you could even have a debate with a younger version of yourself. Like, I'd love to debate myself at 25 versus now and about,

    Pat: 50:48

    I just wanted to be known I was an idiot at 25 versus

    Alex: 50:51

    this is, you married me. I can't believe this because I got

    Pat: 50:54

    my god

    Dave: 50:55

    Yeah, I don't think I'd be able to finish a debate with 25 year old me. I would walk out frustration.

    Pat: 51:01

    Yeah, my wife totally settled down. I married up. Let's just put it out there right now. She, she

    Alex: 51:06

    so hard. Hehehehe

    Pat: 51:07

    Oh, yeah, so hard. What are you doing wife? You're killing me. Dude, you could have had so much more. You settled for me? Anyway, another discussion.

    Alex: 51:19

    Alright, and then We talked about realistic, 1 5 years. Do you see any, is there anything far fetched that may, I don't know, people, most people don't even think about that you could see as a possibility in our lifetime.

    Pat: 51:36

    We're all cyborgs just using Google Glass until the end of time.

    Alex: 51:40

    How about true, are we gonna see True artificial intelligence, like sentient.

    Dave: 51:49

    AGI, artificial general intelligence,

    Alex: 51:51

    we going to get AGI in our lifetime?

    Dave: 51:57

    you know,

    Alex: 51:57

    years.

    Dave: 51:58

    three years ago, I would have said, no, it's complete science fiction. Today within our lifetime, I think, yes,

    Alex: 52:06

    Wow, there it

    Pat: 52:07

    So wait, you mean to tell me...

    Dave: 52:08

    I think it might be different

    Pat: 52:10

    I can have a robot bring me a sandwich? Like, what? Hehehehehehehehehehehehehehehehehehehehehehehehe Hehehehehehehehehehehehehehehehehehe Hehehehehehe

    Alex: 52:13

    No, you're gonna have a robot says it would rather not bring you a sandwich.

    Dave: 52:18

    exactly.

    Pat: 52:19

    He's like, wait, just, did I just get talked to? Yeah, did I just get talked to by a robot? What the hell?

    Alex: 52:25

    that's AGI. Get it yourself, you lazy sOB. That's

    Dave: 52:28

    seriously,

    Pat: 52:29

    lazy ass.

    Dave: 52:30

    it's like we have determined you do not need a sandwich. Like,

    Pat: 52:32

    That's right. Lose a pound, you fat...

    Alex: 52:35

    get you this sandwich.

    Dave: 52:38

    Yeah.

    Pat: 52:39

    ten reasons why the sandwich is not healthy for you. Just sit back down.

    Alex: 52:43

    Pros and cons, yeah.

    Pat: 52:45

    That's it. That's it. Do you know how many calories are in this salami sandwich? Sit down.

    Alex: 52:51

    That's interesting. Well, that's probably the easy question to ask. Well, that's interesting because I think I would have been the same boat if you would have asked me even a year ago, I'd say that's, no that's science fiction. It's a cute idea. We'll have something

    Pat: 53:02

    That's some Star Trek

    Alex: 53:03

    will mimic it pretty well, but it won't be what we Are defining as AGI, but now I'm, I don't know.

    Dave: 53:12

    As you think about all of the data that we have, if we. Just gave it access to everything

    Alex: 53:17

    Yeah.

    Dave: 53:19

    seen, written, observed.

    Alex: 53:21

    Oh my gosh, because then you're going to start getting into some really ethical debates on like, well, what is human then if not just a collection of everything you've learned and what makes that, what makes us sentient and that not, especially if you have to try to throw out religion, you can't talk about religion in this sense. So, you know, what does differentiate AGI from humanity?

    Dave: 53:45

    Mm hmm.

    Alex: 53:46

    topics. Probably something that you can spend a lot more time than that. Yeah. 10 minutes on a podcast talking about so anything other than AGI, maybe

    Dave: 53:56

    Yeah, just on the,

    Alex: 53:57

    We're not thinking of.

    Dave: 53:58

    well, on the biomedical front, I think that we might be one of the last generations to die.

    Alex: 54:06

    Ooh, immortality.

    Pat: 54:08

    Wow.

    Dave: 54:10

    There's been recent studies and things that say that aging doesn't seem to be a process that needs to happen.

    Pat: 54:19

    Wow. Mind blown.

    Alex: 54:22

    and

    Pat: 54:23

    Well, then the question is, do you want to live forever? That's the other side of the it's like, ah, you know,

    Alex: 54:28

    then there'll be, there's a finite amount of resources and. At that point, do we just say, Hey, we're capping at

    Dave: 54:35

    reproducing? Yeah.

    Alex: 54:37

    there you go. If you want to live past this, you can't have kids or something like that. Otherwise you stop getting the treatment you need or something. Yeah, that's interesting. Wow. That's really far fetched. I shouldn't say far fetched. That's really out

    Dave: 54:51

    hmm. Well, it used to be, right?

    Pat: 54:54

    Yeah.

    Dave: 54:55

    But yeah, if you look at lobsters and telomeres and, you know, all those things. It's, there's

    Alex: 55:00

    the Greenland shark, too, I think is technically immortal.

    Dave: 55:04

    Yeah,

    Alex: 55:05

    Some people say they've seen some that are 400 plus years and it just doesn't, it's cells don't die.

    Dave: 55:12

    we have examples and I mean, what's machine learning besides, through examples. So we have examples of this and

    Alex: 55:22

    Yeah,

    Dave: 55:23

    mapped our whole genome. We have CRISPR technology. We have AI now.

    Alex: 55:28

    So we'll live forever, but we'll start growing fins and gills.

    Pat: 55:33

    that's it.

    Dave: 55:33

    Just think Gen Z is the immortal. Gen Z is the immortal generation,

    Pat: 55:40

    yeah, exactly. Yeah. Don't leave me that.

    Alex: 55:42

    oh man. We picked this generation to be immortal.

    Pat: 55:45

    Oh, Jesus.

    Dave: 55:48

    right? Millennials, we miss everything.

    Pat: 55:50

    Oh, our whole life's gonna be one big TikTok reel. Oh, my God. This is terrible.

    Dave: 55:57

    with Tide Pods, ended with immortality. Geez,

    Pat: 56:00

    right. Oh, Jesus. Yeah. Some gens have all the luck. Son of a...

    Alex: 56:05

    yeah, they won't create anything, so they'll just use the millennial technology to, yeah, just, they'll have a two billion TikTok subs, but,

    Pat: 56:15

    it. Talk about failing up. Jesus. We have a whole generation that fails up. God damn it. Oh, that's funny. On the business side of things, how do you think, like, either as an individual, so we sort of talked about that or business can prepare for the increasing invasion, if you will of AI, right? So is there like a, Crawl, walk, run sort of thing, or is AI just be like, you know, I'm going to start using AI tomorrow and it's going to transform my business from 1 million a year to, you know, 10, 10 million a year, whatever it is. So like, is there a model to sort of, get that introduced and use that to the most efficient, or is it just jump in with both feet into the pool in the deep end and figure it out later?

    Alex: 57:01

    That's how I'm picturing it now. If you're not a huge company, like a meta or something that can do 10 billion to fund research in AI, and you're just a smaller company, it seems like a daunting idea, like how do you get in front of this, and I just feel like you're gonna have a board of directors that just say, you know, like, AI, am I right? We gotta do it this year. I don't know what that means.

    Pat: 57:29

    Yeah, you have more directors that can barely spell AI.

    Alex: 57:32

    Yeah, spell check on that. Yeah, so I mean, what do they even do? Is there anything... In your opinion, that these companies that don't have these massive budgets can do to prepare for this?

    Dave: 57:49

    I don't know. I think, yeah, I think you have to dive right in. I think that for most companies that are non tech related, it's going to just change the nature of the work that people do, right? Personal assistants lives are going to change. Copywriters lives are going to change any type of content creator. but these things. Are also becoming so much more accessible in the tech world. If you look up lang chain, it's a really nice framework for working with large language models, and you can use the models just through APIs and connect them to your databases and connect them to. whatever else you need to connect them to. I think that it's going to transform the way that we interact with products. So for example, going back to the networking example, we're used to troubleshooting things by looking at different graphs and looking at metrics and trying to figure things out. I think that we're going to see a lot more interaction with network directly, right? What if you could just ask your network, what's wrong? Why do we have latency right now? And it will do that work for you. Just like we talked about with the education example, if you could just ask Hayden from the Holden from the catcher in the rye, you know, what were you thinking when you did this? And then you can get an answer back, you know, from that point of view. I

    Alex: 59:13

    Okay, interesting.

    Dave: 59:15

    Yeah.

    Alex: 59:16

    It's

    Pat: 59:17

    It's all a good answer here. no, yeah, there's no writer in. It is.

    Dave: 59:21

    I think there's gonna be a slow change in the nature of work, but as far as. AI into your business. I think that's going to come from within. It's just going to be people using AI to change your business. Just like we, the way that we use Google in our roles now and things.

    Pat: 59:38

    Yeah.

    Alex: 59:39

    Just organically happen. You don't have to have this five year master plan. It's going to happen at least to some degree.

    Dave: 59:45

    Yeah. People are going to you.

    Alex: 59:47

    right for sure. And then the tools that you consume will eventually start using it. So maybe you'll use it without realizing it.

    Dave: 59:54

    Yeah. Like GitHub co pilot, maybe all, or, you know, that becomes the normal part of an IDE where it's coding with you instead of you typing in every character.

    Alex: 1:00:04

    for sure.

    Pat: 1:00:05

    Cool. Interesting.

    Alex: 1:00:07

    All right. Final topic.

    Pat: 1:00:09

    I actually have one. I guess it just kind of came to me, but do you, so now AI is sort of open and it's, you know, it's, you know, quote unquote free for everyone, that sort of thing. Do we get to a point where it's just consuming that much data? Open AI is just pouring tons and tons of money into it. Is there ever going to be like a paid version of an AI or like AI as a service? Cause everything else on today's planet is a service, right? Infrastructure and PAS and IAS, PAS, all that kind of crazy stuff. Is. Does it eventually hit because of what it can do or what these places that are kind of driving forward, is it put behind a paywall at some point or because it is open source? Does it stay open like a Linux for however many years Linux has been around a billion? It seems like, like, is there a paywall coming or is there, do you think these places have a way to monetize that? Because it's so hot right now, right? Everybody's trying to make a buck.

    Dave: 1:01:09

    I'd say yes and no. I think there's definitely a paywall for specific applications of AI because. The products that integrate it the best for whatever it is, whatever outcome they're looking for are always going to be the best user experience that people are going to pay for. But I think that something like a chat GPT or an AGI is going to become more like a utility, like the internet. Maybe you'll need to pay for some sort of access to it in general. But not a specific version of it. And I think that'll be because of the progress of open source and things where it's kind of not tenable to gatekeep it any longer.

    Pat: 1:01:47

    interesting.

    Alex: 1:01:48

    Yeah, and you also mentioned something earlier when it came to cost too, because in my head I was thinking like, how on earth are they not charging for a tool like this? I didn't realize that it's the learning that is the costly part. Because whenever it just, we go back to chat GPT just cause it's hot and everyone knows it. But it comes right out and tell you that it's learned it's a, I forget how they word it, but you know, it's only learned up to. Sometime in 2011 or

    Dave: 1:02:17

    Yeah. It's training cutoff.

    Alex: 1:02:19

    right. Yeah. And then now that makes a little bit more sense that it's not actively training, at least not the model that we're interacting with. So that makes sense how a hundred million people can use this. And this thing is not like, how is this thing responding as quickly as it. That blew my mind when I was, when I've been messing with it. Just like, if there's so many people using this, how on earth is this keeping up? That's interesting to know that it's significantly less compute. So, I mean, there could be language models that people just, you get the base trained up to X year and it's even in that capacity, it's so helpful. And you would just need just so minimal of the resources to run it. So I guess maybe that's kind of like future is how long is it going to take for us to get a model open to the public that is real time, updating and training itself. Maybe that's further down the road than people think because of how intensive that would be. Interesting.

    Dave: 1:03:25

    The interesting thing that we're seeing now is that. Originally, we thought that you needed to always retrain the model once you made sufficient progress, right? And the training process costs so much to do, but what we're seeing in the open source community is that there are a lot of things that you can do to supplement the model that you already have.

    Alex: 1:03:49

    Yeah. I guess most people, at least I did, I just assumed that it was always kind of iterating over and supplemental, but I guess in the past it's been a forklift, you know, idea to retrain it.

    Dave: 1:04:05

    Yeah. And you can completely retrain on the new data that you've learned, but you can also use different fine tuning techniques like low rank. Adaptation of large language models, which was part of the stable diffusion thing, if you're paying attention to that. So instead of, you know, training billions of parameters, which is ridiculously expensive, you can freeze the pre trained model weights and then put in new trainable layers. In the transformer GPTP generative pre trained transformer which just drastically reduces the GPU and memory requirements.

    Alex: 1:04:43

    Interesting. Yeah. It would be nice to really understand how these are structured. I guess in my head I was trying to relate it back to like just basic coding one on one And I was thinking like, do they have like 2021 training, and that's kind of like a module that I can call back and it doesn't have to do a lot of stuff with it. And it's just, someone's got this crazy thing written out in some ID somewhere. Um, yeah, that's interesting. Yeah. Probably should start learning more about that. And because I would assume, even though open AI is kind of the forefront of the news right now. I assume they're not doing anything radically different than what people have done in the past. I mean, I assume the same type of structure that they're using to train the models are similar to any other language models out there. And I don't know if that's open. Well, I guess it is open to the world now, even by mistake. So. Yeah, I mean, now that we have that code, were they doing anything that was just so radically different than anything else has been done in the past? Or are they just, you know, they fine tuned it better than anybody else?

    Dave: 1:05:52

    I think it really came down to scale. I mean, the, so transform, like the transformer architecture has been out since 2017, but even going back to the eighties. The idea of a neural network, you know, has been around since then. The problem was that back then neural networks sucked compared to symbolic AI, which was just very specific instructions to get the outcome that you wanted that kind of appeared like it was intelligent, but neural networks and the idea of having. Something that learns performed terribly. Let's say it had 30 to 40% error rate and it just wasn't tenable. But the problem back then was that we didn't have the compute or the data for it to actually learn. You know, they were trying to build neural networks with a couple hundred examples of things and it would do an okay job, but it was never going to become something that people wanted to use where now we have billions of data points and. Tons of compute to train on you know, these things have become simply incredible. So,

    Alex: 1:06:58

    Well, I guess your opinion, your thoughts is right now, the, to see the next evolution in these AI models or AI in general, is this really just an idea of just We probably already have the algorithms in place and really, we just need more data points and compute. Is that what's the bottleneck? Or do we still need data scientists to make all these different algorithms and next iteration of algorithms?

    Dave: 1:07:32

    I think there's definitely plenty of room for improvement and the data sciences will keep working at that. There's new papers coming out every week about different ways to improve on, you know, what's already been done. So that's going to keep moving. And now with all the interest that's in it from. The academic side and the accessibility of it now as well. yeah, I think we'll have large scale architectural changes that will continue to move us forward. But as we've seen, like we're already being able to run some of these sophisticated models, which we know we can run lighter weight versions and then build layers on top of, you know, on smaller devices, smaller chip sets, things like that. So I think the combination of downsizing the requirements to run, increasing the compute that we have and the researchers working on the models, that's. That's why those 10 year predictions don't seem so far fetched anymore.

    Alex: 1:08:29

    Wow. Amazing. All right.

    Pat: 1:08:33

    Yeah, wait till Quantum Computing gets here, then it's all, then it's it's game on.

    Alex: 1:08:40

    that's one of the classes that I signed up for. Pat Cisco. I was quantum networking. So, yeah, I'll let you know how that goes. That's a fun one. And it's with a PhD in mathematics. That's who's given the lecture. So hoping I can understand it. If I don't, I'll just put it in chat GPT and tell it to explain it to me in terms I understand.

    Pat: 1:09:03

    right. Dumb it down. That's

    Dave: 1:09:05

    Explain like I'm five.

    Pat: 1:09:08

    right. So as we sort of... Wrap up here. The last kind of topic and we're right around the hour. So, which is perfect. This has been awesome conversation. But as far as like landing a job in the AI space or transitioning to it, like, do you have a roadmap of kind of how, like what it takes to get there? Do you have to have a software background such as yourself and then kind of move that way? Cause we all know it's a software world going forward, right? That's no secret. So, you know, do you have to start there and kind of move, move with it or can you jump right in and depending on how you learn and things of that nature, like, can you get there without being a software dev? It's like, what does that look like? Or what's your opinion on that?

    Dave: 1:09:48

    I think you can. It depends. I mean, it's a broad field that needs a diverse set of skills. So I think you can get there from anywhere. I think coding absolutely helps. There are no data scientists that don't know how to program Python and are taught as part of what you learn in data science, because working with these large data sets is not something that humans can just do, right? Like, we need computer assistance. So you'll absolutely learn that. But programming itself is becoming more accessible with AI. So I don't think. Did anyone that doesn't have those skills should let that kind of gate keep them. They're also great courses and materials to review online. If you want to learn more about machine learning. Cassie Kazakov has a six hour talk, which is also broken up into more consumable chunks on YouTube called making friends with machine learning. I definitely recommend checking that out. If you want to dive a little deeper into the concepts and they're not. They're technical concepts, of course, but you don't have to be a data scientist to understand what she's talking about. there are data analytics courses on Coursera offered by Google that are really good. You can get your certificates there. I think that as AI continues to. Revolutionize the workforce, kind of like I talked about before the death of specialties, I think certificates will actually become more of a common thing and a more useful thing because knowing a lot about a lot is going to become more important than knowing a lot about something specific, if that makes sense.

    Pat: 1:11:19

    Yeah. Interesting. Alex, you got anything else?

    Alex: 1:11:25

    Well, I'll just touch on the learning a little bit because when you are looking to get into AI and machine learning, do you go straight to courses on AI and machine learning? Is there like a prerequisite technology that makes sense? Or like you said, you mentioned making friends with machine learning. Is that just. who just graduated high school has never worked in I. T. Can they digest the content like that?

    Pat: 1:11:52

    The wheels are turning. I see smoke.

    Dave: 1:11:57

    I think so. I think it's approachable enough where you would have a solid understanding coming out of it. To be effective at learning on the job when I think back to, you know, the things that we learned in college. Anyway, how much of it transferred directly into what we needed to do once we started working 10, 15% of it. Right? So. I think that going through putting in the work to, you know, get a certification for whatever that's worth would be enough to lay a foundation for the things that you're going to learn. Once you get into it from my experience, you know, being a math major and being someone who's always been interested in this stuff and who's taken those certificates and things, it makes the conversations that you do have with data scientists go a lot better. It helps your understanding of. Thanks. The direction that you want to take things or why certain solutions may or may not work, why you can't just shove chat GPT into everything, no matter how much your CTO wants to

    Pat: 1:12:59

    Ha. Ha. Ha.

    Dave: 1:13:00

    things like that.

    Alex: 1:13:02

    All right. And last question I have that is you mentioned certifications a couple of times now. Is there. Is there like a go to AI ML cert that has some weight in the industry? I mean, as far as networking goes, people understand the CCNA and people understand like AWS certified, whatever, anything in machine learning that you can think of or know of.

    Dave: 1:13:29

    Not that I can think of, cause I'm pretty new to exploring kind of the world of learning with it. Historically it's been people that have come out of a degree program in it because it's new and those people are just getting snatched up. But I think that. As the need for AI people grows, it's going to expand out into, you know, certificate programs and things like that, especially because it's so new, right? Like there are tons of smart people like on this call where it just happened too late for us. And we've been in careers,

    Pat: 1:14:00

    Yeah, Alex and I are both looking over our shoulders like, wait, there's

    Dave: 1:14:04

    well, Pat, your wife came down the stairs for a second. So I

    Pat: 1:14:06

    Yeah, she's, yeah, she peeked around the corner and said, she's the real smart one.

    Dave: 1:14:10

    Yeah, but I mean, there's tons of talent that could get involved where like, You know, we think like, ah, you know, I missed the boat on that one. And I just don't think that's the case. I think there's going to be a great enough need that if you're inspired enough to go after it, you know, there's going to be resources out there and knowing the foundations and being able to speak intelligently and pick up things, you know, once you get in there, all you're going to need.

    Alex: 1:14:33

    well, I think we have a new business model, Pat. So. no one's got a great cert and everyone wants to do it. So cuz we, we saw this with the adoption of the cloud infrastructure too, where there were businesses that did nothing other than get you to aws. I can see a correlation now between people, like train people on AI and get us to start using ai and like

    Pat: 1:14:55

    That's it.

    Alex: 1:14:55

    that he started an AI revolution in his company.

    Dave: 1:14:59

    Yeah. I mean, if you look at like the prompt engineering role that we're seeing now, where people are getting hired to just tickle the right answers at a chat GPT, it's like, Is that a. Is that a skill that you learned somewhere? Absolutely not.

    Pat: 1:15:12

    It's almost like the social media marketers back in the day when social media exploded, like the people, businesses were literally hiring people to specifically post on social media about, you know, they're about the business. It's like, man, literally a whole, like a whole industry was born out of that, like, it's just a whole career path was born out of social media. It's like, it just exploded. So I could definitely see AI

    Alex: 1:15:37

    And that's a. And that's the thing. It's still here. I mean, I know Disney has a vP of digital marketing, which is nothing more than they, they run the Twitter and the Facebooks and of our company. Yeah, it's a thing. So maybe we'll

    Pat: 1:15:52

    it's crazy. It's absolutely a thing.

    Alex: 1:15:55

    initiatives or something. I don't know.

    Pat: 1:15:57

    That's it. I'm gonna quit my job tomorrow I'll start brainstorming Alex you and I'll we'll we'll talk

    Alex: 1:16:05

    the smart thing to do. Let's just quit our jobs. That, you know, that really adds some fuel to the fire.

    Pat: 1:16:11

    that's it Start from scratch literally

    Alex: 1:16:14

    Alright, well I think that's a good way to stop since Pat and I have a lot of work to do apparently. Starting our new

    Pat: 1:16:20

    that's right. I'm going to study AI as soon as I get off this call. I'm out. That's it Before you know it this whole podcast of the AI you'll just be hearing robots talk about for

    Dave: 1:16:32

    talking to each other.

    Pat: 1:16:33

    That's it. It'll just be digital faces of Alex and I. We won't really be here. We'll just phone it in every week. That's it. Write me a podcast, slave. I love it. I love it. Oh, Dave, man, this has been awesome. I really appreciate you coming and hanging and talking all this kind of crazy stuff. And really it's good to see where the AI stuff goes. So again, thanks for coming and spending a couple of minutes with us and dropping some hot knowledge. It was it was really fun, really great conversation. So now I really appreciate you being here, man.

    Dave: 1:17:03

    Yeah, absolutely. Thanks guys for having me.

    Pat: 1:17:05

    Yeah, man. We'll definitely have you back when another version of AI comes out or a chat GPT, and we'll talk about that too. So pros and cons.

    Dave: 1:17:12

    Yep, we can complain about our robot overlords.

    Alex: 1:17:15

    Yeah,

    Pat: 1:17:15

    right. That's right. Skynet's not here yet, but damn

    Alex: 1:17:19

    yeah, 200 years down the road, we can have a follow up conversation.

    Pat: 1:17:23

    That's right. Damn Gen Z ers. That's it.

    Dave: 1:17:27

    The last generation.

    Pat: 1:17:29

    That's right. There's no more letters after Z. What are you going to call them next? I don't know. Type it in chat, GBT, it'll tell you. Anyway, thanks everybody for joining this week on breaking down the bites. Make sure you visit our website, breakingbytespod.io. io so you can get to, you can subscribe to the show on your platform of choice Apple podcasts, Spotify, Google podcasts Stitcher pretty much anywhere that has a podcast platform. We are there, or there's an RSS feed there as well. So just if you need just a plain RSS so you never miss the show. Throw us a rating on the Apple podcast. That's where most of our listeners come to, or at least that's what our statistics tell us. That would be great. That fools with the algorithms and the AI and all that crazy stuff that Apple uses over there to get podcasts to people's ear holes. So, the more we get ratings and reviews and that sort of thing, it It can only help. So, or if you simply tell a friend, right, that works just as well in today's crazy tech AI driven world, you know, talking to actually human to human talking is, you know, still works pretty well. So it's still a thing. Yes. We're not there yet. So still a thing. Tell a friend and certainly get them to tell a friend or whatever, so that'd socials are in our. So LinkedIn, Twitter Facebook, there's a discord server out there. The invite is in show notes as well to come hang with us. So we still need a little bit of help on the discord thing. So if anybody's got some discord wizardry, let me know and we'll we'll talk. So that'd be cool. Again, thanks man. Dave has been awesome, Alex. We will see you next week and that's it. Bye everybody.

 
Pat Allen
Previous

Episode 57: Why You Shouldn’t Get into Tech
Next

Episode 55: SD-WAN Deep Dive