Latest / You've Been Heard / 371- AI Implementation Without Hype w/Yossi Akselrud

Transcript

371- AI Implementation Without Hype w/Yossi Akselrud · You've Been Heard

• 0:09
  Speaker A: Everyone, welcome back to Dissecting Popular Itiners. And the reason why we've been having these wonderful technical things, which is what we do in IT worlds, put out fires every day, hopefully not. And we're a little bit more proactive is we're trying to get all the live streams going at one time. So, you know, Yosi, you're the first. Like super double live I guess is what we could call this. I don't know. We're trying to go live in the community, live on LinkedIn and all this, all this stuff at the same time. I'm sure there's like an AI thing that, that could fix that before we even get there. Can we maybe just talk a little bit about how you got into the wonderful world of it? Because I'm sure it started somewhere back in the day with maybe tape backups or something like that. Why don't you just give us your, give us your rundown and how you got, how you got playing with all this stuff before we get into the real deep stuff.
• 1:00
  Speaker B: Yeah, sure. So my background in it, how I got involved was I was going to be a radio engineer. I'm going to be like an audio engineer. Radio engineer. That stuff, like, really spoke to me, man. I have a brother who's like, you know, take a networking course. Take it. I was in, I went, took a networking, like one class. I was like, it spoke to me. It was just a computer class. It was just like a regular computer class, maybe like, IT stuff. I knew nothing about it at that point other than like we had a computer in the house that I, you know, aol, you've got mail, you know, files waiting. That kind of fun stuff. Like don't, don't pick up the phone when you're online. You know, the old fax blasting, you know, board rate, modem fun stuff.
• 1:47
  Speaker A: But that's pretty, that's pretty high tech for back then, though, if you actually had a modem in your house.
• 1:51
  Speaker B: Well, if you had 36,000, you know, board rate, you know, the haze modems, then yeah, for sure. So that class, I don't remember much. I like, I remember, I think it was like we touched on Excel a little bit. Like what Netscape is like. I don't think it went very deep. I don't remember. But then I was like, this, this really speaks to me. I think I took a second one and I was like, okay, this is, you know, we're going to put audio engineering maybe to the side. And I took a. I went to Chubb Institute Enrolled in Chubb, which was very interesting for me to have to interview to go to college. But, you know, I got accepted to Chubb. And then after, you know, spending some good months of every day having to take the subway, I actually went to school across the street from World Trade Center. So every day before, before we started, I actually met my brother every day for lunch who was working at the Trade center. And I would be there with a jacket and tie every day. You had to wear a tie to a Chubb Institute. That's my path, that's my journey, is, I'm here, I've been there, I'm here now, what's next? But like, oh, if you do this and you fill out this punch list and you do this really well and we do this construction, you know, know, and we make sure that, you know, this construction goes well and this project goes well, then you're going to get this phone call and this opportunity. No, just right now, I'm doing my 110.
• 3:18
  Speaker C: At Dissecting Popular IT Nerds, we expect to win, and we expect our IT directors to win. And one of those areas where we know that we can help you win is Internet service providers. As an IT director tasked with managing Internet connectivity, few vendor relationships can prove more painfully frustrating than the one with your Internet service provider. The array of challenges just seems never ending. Dealing with ISPs can try one's patience even on the best of days. So whether you are managing one location or a hundred locations, our back office support team and vendor partners are the best in the industry. And the best part about this is none of this will ever cost you a dime. Due to the partnership and the sponsors that we have behind the scenes at Dissecting Popular IT Nerds, let us show you how we can manage away the mediocrity and hit it out of the park. We start by mapping all of the available fiber routes. And we use our 1.2 billion in combined customer buying power in massive economy of scale to map all of your locations, to overcome construction fees, to use industry historical data. And we leverage aggregators and wholesale relationship to ensure you get the best possible pricing available in the marketplace. And on top of all of this, you get proactive network monitoring and proactive alerts so that you're not left calling 1-800-Gold-POUND SAND to enter in a ticket number and wonder, why is my Internet connection down? In short, we are the partner that you have always wanted who understands your needs, your frustrations, and knows what you need without you having to ask this all starts with a value discovery call where we find out what you have, why you have it and what's on your roadmap. All you need to do is email Internet popularit.net and say I want help managing all of my Internet garbage. Please make my life easier and we'll get right on it for you. Have a wonderful day.
• 5:17
  Speaker A: And I want to ask some hard questions for you. Number one, one of the things that everyone wants to know is how do we get past the AI hype and how do we actually have some real use cases or really implement AI in a way that is actually meaningful to the business and drives business value.
• 5:32
  Speaker B: So I work in healthcare, I've always worked in healthcare, right. So with healthcare we're guided by hipaa, right? So it's, it's a protocol put in by the government. We have to, we have to keep it, right. And my personal feeling is that every business should have these type of standards and regulations of how are we treating somebody else's data, right? Like I saw recently on LinkedIn there's like a keyboard voice dictation tool, like oh, what's their security protocol? Because you know, we remember back in the day there were keyword loggers, right? Like the worst thing you could get is in fact somebody put in a keyword logger. Well, if I install on my own something that's capturing everything that I'm saying as a keyword and now they're putting it to somewhere and they don't have the securities in place to protect my data and who I am, you know, that's not something I would want, right? So you asked a multi pronged question, right? So one is how to get into it, right. And I think my personal feeling is find something that you will find fun and interesting, right? Make it almost like you're drawn to it and something that will draw you to it, right? Like let's make this interesting, right? And if you, I want to say gamify it, but it's a tool that you're going back to because it's fun, it's interesting, it speaks to you, it goes to you. I'm in GPT constantly. I'm using for multi prompt, you know, of what I'm using it for, whether it's help formulate a draft of something that we're sending out to somebody or to review something else or to make a fun image, right? I'm really good at making fun images and funny memes and to send that around. I have a lot of fun doing that. You have no idea how many subscriptions I have just so I can do fun things, right? Like, you need to have fun, right? And these, these are fun milestones, right? Fun things. So how to draw somebody into getting into it? I would start with. Start. Start low, right? Hey, what is something that speaks to you?
• 7:24
  Speaker A: We should go along with the vibe too. Because one time you told me I start off in the morning, like maybe even in the shower, vibing. I. I imagine you have like this waterproof microphone in the, in the shower and that you're vibing with AI.
• 7:33
  Speaker B: Yeah, it's called your phone. Your phone's waterproof. Okay, I can. I know the. I remember the exact time when the phones became IP67, right? Not like I know these IPs, whatever it is, but I can use it in the phone, in the shower. And I'm like, oh, my phone is now waterproof. I don't have to really protect it. And so in terms of like, vibing, I talk to my GPT constantly. And I actually had a thought this morning. I put on LinkedIn, right? Where in GPT you can vibe with it, right? Whereas, like, if you tell somebody to go to Google and they're doing Google search, you have to give it exact, right? Like in the longer tail words. It's more finite, you know, of what your results are going to be. GPT just having a conversation with it, right? And so I'm getting really comfortable with vibing with my GPT. I'll hit, you know, dictate. It's not the talk, you know, one to one where it's talking, it's answering you, it's just listening, and you can imbibe with it. 10 minutes, right? The cap is actually 10 minutes and you're just having a thought, right? So like somebody sent in like, hey, what are we doing for this process? We're doing a rollout. How are we doing the stage of the rollout? I have an idea. In the shower, like we discussed, the last time I spoke in GPT, I voiced dictation. I vibed with it. This is something how we're going to roll this out. I'd like to do this address for this, Calculate this, you know all the parameters and it just, it spit it out. So, yeah, Mike, if you're hearing this, that whole memo I wrote up, it was in the shower. Just vibe to it. Don't be scared of it. People are scared about this thing. It's a, it's a scary thing. It's new, you know, lean into it, but don't give it private Info yet, you know, I wouldn't.
• 9:08
  Speaker A: How many, how many people you think if you had to guess a percentage wise in the company. Not your, not your company. Any, any company anyways across whatever industry, let's just say whatever manufacturing, healthcare actually use AI every day.
• 9:22
  Speaker B: I don't have stats, but I will tell you, I have a friend who's a partner named accounting firm and we, we talk AI and I give them some stuff to look at and, and he went to his IT guy, he goes okay, I would like to start using AI. And his IT said no you cannot. He's like what do you mean for stuff that doesn't. It's not, you know, private, it's not without phi. You know why can't use it knows. And so this IT guy as his accounting firm, I told my friend, I said listen, get a co pilot license, it's your World Azure tenant. Your dad is not leaving, it's secure. You sign on his little green shield, go that way. And I'm trying to help my friend in the accounting firm convince his IT guy to let him, you know, dip his toes into AI. Like these are formulas and trying to figure out long spreadsheets, you know, they're.
• 10:16
  Speaker A: Changing so people are scared to start. That was kind of one of our, that was kind of one of our first bullet points. And I wasn't really scared to start, I was just kind of. Nah, I mean I could use it. What am I going to use it for? Search my email, make a few things more, you know, whatever it was. And, and Greg, he knows that there's things that he showed me before in AI and then I come back to him like, hey, we gotta use this. And he's like, yeah, I showed that to you like six months ago to.
• 10:47
  Speaker B: Prove though Greg, it's true. Cause anything that we have now, it wasn't there three weeks ago. And like we looked at, we're rolling out a questionnaire right to, to our subset of our employees because we're switching out some of the devices. And so we found this old form that I used two years ago, literally two years ago when like AI was a new thing and mid journey was what we were using for, for creating images. And I looked at this thing, this image I created, I go, that is so beta I guess, right? It's like, it's like it finally generated an image without six fingers. Greg, if you've been you know, doing the AI for a while, like oh, I got all that six fingers, let's use it because it's not complete Crap. But like, it, like, now I recreate it and it's lifelike, it's real, it's got the image, the emotion that I'm looking for. It really dives into exactly that subset. Whereas when AI was, like, starting, it was not repeatable. You could not repeat results from prompt to prompt to prompt zero. Whereas now it's keeping the memory and it's keeping the images, the memory within the images. So, like, initially, when you would adjust, like, let's say, oh, I like his image, take out this table from the. From this image, the whole image would change, right? The whole thing was gone. Whereas now, table's gone. Okay. Put a pen in the hand as a penny. It's free to do.
• 12:10
  Speaker A: I can't figure out the spelling yet. If you guys can figure out how to get stuff to spell long, long form words on the, on the page.
• 12:16
  Speaker D: Yeah, these are about the models, but some of them are better than others.
• 12:21
  Speaker A: So to not follow any agenda at all, I'm throwing it away. But the one thing that we were talking about, Greg and I, before we got on here was look at the stuff we can build. We're not even coders. Look at what we can do. So I know that you're doing that. I know you've got things that we don't, we can't talk about. We can't even. If we signed NDAs, you wouldn't even whisper about it. I know you wouldn't, but let's just use a fictitious example of something that you could do. And I guess our question was, before, we're not coders. What do we need to do? Go find a coder to, like, fix the little gaps or something. What do we do?
• 12:54
  Speaker B: So there's a few platforms out there. One is replete. That was the first one I was introduced to. Another one is called Lovable. Right. Another one is called Zeit. Right. And then there are a few other ones. Base 44 is another one, you know, and these platforms, you really don't need no coding, right? So my workflow that a good friend of mine, Clark, suggested to me, which is my gold standard for the workflow of these guys, right? You gotta give credit, you know what I'm saying? It's not my idea. The idea that I'm building up, perhaps our mind, but the workflow, right, where it's. I'm using GPT as my core of everything, anything I'm using in AI, even if I need to do it, bring it to another AI, for instance, like Gamma, right? Like, oh, I need to Create a pitch deck for this or create a document or PowerPoint, whatever it is. I'll go into GPT, I'll vibe with my GPT, right? So I'll just hit that dictation button. I'll talk to my GPT looking for these things to occur, right? Then I'll tell GPT, give me an output for whatever the platform is. CPT knows how to give you an, an output for that platform. But you say give it for Gamma. It will put the breaks into where it feels that the gamma pages breaks near to happen. It's crazy. Like when you go to Gamma, you're like, oh, it'll, it'll have it baked out. And so the pages are exactly. So like for instance, when GPT, before there's a gamma, you say, hey, I need a PowerPoint deck. It would just give you some like really, you know, poorly formatted document of like a presentation. But now you're getting the good data from GPT, right? It's not that great at creating the images, right, as the formatting, but it's really good at creating content. So you're taking the content from GPT and you're putting into Gamma. And then here's your, here's your thing, right? So I do the same thing with my low code, zero code that I'm doing, right? Well, I'll vibe with, I'll open up a new thread, new memory in my GPT, right? You can open new folders, whatever it is. And I'll tell it, this is what I'm doing. I'll vibe with it, what I'm looking to do, the business model, the platform, whatever it is that I'm looking to have created, I'll vibe into my GPT exactly what I'm looking to do, right? It'll then I'll then tell it, give me an output for this platform, whatever the platform is, right? It'll then give me the output, I'll copy and I'll paste it into, let's just say lovable, right? Or I'll paste into zeit, right? Z, I, T E, right? And it'll build it out, right? And then if I want to make a functional change to it like within the platform, right? Like let's say lovable, I'll, you know, if I need to adjust this, this button should look like this. It should be doing this, you know, that's fine. The micro adjustments I will do within that platform, but the macro adjustments, right? Such as functionality, I'll then go back into GPT because GPT is now my My thread of. Of this platform, right, that I'm looking to build out and I'll say, okay, adjust this functionality, build out this functionality and again, give me the output file. I'll put that into Lovable and it will then now build that out for me. Let's so. And I'll just, I'll give you like an example, right? Because you asked me for like an example of what you can do, please. You build out an automated workflow, right? Within a company of. You need these triggers to happen, these people to be notified, right? Very simple. And you want a transparency, right? Really, you know, something that's customizable based off of your business that you're looking to do. So instead of going online and you're paying somebody, right, as a gig, or you find somebody's tool online that's going to cost you, whatever, a month, you can build it up, customize it for yourself and within using Lovable. Okay? Lovable, you don't even need to know database architecture. You don't need to be a DBA because Lovable will connect with Sepabase, which is the backend, and it'll say, okay, would you want me to create this for you? Yes. Bam. Okay, I need to adjust. It'll tell you, hey, I need to adjust the SQL tables to do the. Would you like that? Yes. Okay, I need to send emails. Give me the API key for this platform. It'll tell you the platform it'll give you to. It's called resend. I'll say, okay, give me the API key for resend. API key. Put it in. You're not going to the backend, really, of subabase. You could if you wanted to, but you're all doing it through this adding with an AI agent that is now doing all these actions for you and all these triggers. Crazy stuff. And the old days of, oh, I need to invest X amount of money and X amount of time and to create even just a wireframe for an app. Right before you're like, oh, let me see if this is like a real thing.
• 17:34
  Speaker D: I have, I have a personal example like that. That's funny because it really goes back to now. The ability to create your own solution. For example, now instead of building like a PDF document with the boring Google Doc to PDF, I just do everything on HTML. But I couldn't find a good HTML code to PDF format, so I just built it myself. And now the app is perfect, but I just use it on my own locally. But that's where I see where it can go now. Basically you just at this stage where it's going to be like micro app that you use for your own use case, not even putting production to sell it, but really because you want your own thing.
• 18:25
  Speaker A: Someone's going to hire Greg. Someone's going to hire Greg now. He's going to get like 15 job applications. You have dreams.
• 18:31
  Speaker B: He should as well. He should by the way, because you realize that a guy like Greg, with his strong suit of being able to make micro apps. Most companies, they don't want to bring in something that's a big platform, right? Like they don't want to bring in a whole big thing and change all workflow. They want an mvp, a micro application that will address the, you know, some little pain point that they're having in efficiency.
• 18:54
  Speaker A: The spreadsheet company that's still working off spreadsheets and using email as a database. So what we've talked was there's a little bit high level for maybe some people in the company. What, what would you do to start people out in your company? How would you draw this out? Or how would you get people to start using any kind of steps or suggestions?
• 19:11
  Speaker B: Start small. I'd start small.
• 19:14
  Speaker D: And if, if, if I can just add something on top of that question. The whole vibe coding is usually for solopreneurs, people that just build a thing, you know, on their own, a really small team. But now, yeah, when it comes to a company, is it something that is really applicable for company because of all the question behind the ethic, behind the security and everything.
• 19:36
  Speaker B: So yeah, Greg, if, if so I'll, I'll, I'll answer you, right, with the vibing for the solopreneur, right? When you have at least for a company my size, right. The, I view it the same way, right, where you first come up with a set of what you're looking to accomplish, right? What's, what's, what are you looking to accomplish? Your mvp, right. What's your framework? Right. So whether, you know, I'm doing it as a vibe being sold entrepreneur or doing for a company, I would, I would address it and you know, go at it the same way you would with any other, you know, platform that you're looking to build or anything that you're looking to accomplish within the building it for a company, right? And then you adjust your timeframe to roll something out. Like this is just greatly, you know, accelerated where hey, you can meet with, with the, with, with, with, with the stakeholders, right? And like get their, their what, their requirements, what they're looking for. And then two hours later you can spin around a framework that might even be working, right? You don't have to like bring it back to the developers and then have the developers then, you know, work on this and they're creating it. Your, your turnaround time to just, you know, have, have something, even if it's. I'll give you an example, right? I had an idea at 2 o' clock in the morning on Friday morning when I was in Israel and I was like, ooh, I have nobody to talk this through because 2 o' clock in the morning in Israel, it's, it's a seven hour difference, six hours depending on the time of the year. And I'm like, let me just talk this through to my AI, right? And so I just, I went to my GPT, I vibed with it, right? Everybody's sleeping, it was just me and I just vibe with this thing and then it gave me an output Lovable. I took this output and I have a crazy. And then I'm modifying him, adjusting or whatever it is, right? And then it's not working because it's meant for a phone, right? Like my idea that I want to build out and I do a thing, it's for an AI for phone. And I'm just waiting for, you know, for it to happen, right? For the piece of the tech stack to line up, you know, appropriately. But I have a working wireframe of how my app is going to work. Yes, it's in Lovable and it's, it's not doing it because whatever it is, but your turnaround time to just come back to whoever the stakeholders and say, is this what you're looking for, right? Even if you want to take the code, right, and then put it into another platform, like Lovable has a link, is a hookup with a GitHub, right? So, so like I have a data ninja, right? I have a guy in Costa Rica who's incredible. And so we're just. He used low code, no code to do, you know, platforms for me, like micro platforms where we built out something like about a year ago that is just, it's not on the space. But like you're, like you said before, not everything that you're building out you're going to go public with, right? A lot of things. It's built out efficiency and like my owner of my company said, like, we built up something that's going to make us more efficient. Why would I give that to, you know, my competition, right?
• 22:33
  Speaker D: And is that up to the company policy or like, or Something that's so everybody use a quietly on their own, but they don't mention it to the company because of eventual great question.
• 22:45
  Speaker B: And shadow it is a real thing, right? Shadow it is a real thing. I had a company reach out to me last week. Hey, we can monitor your stuff and your IT for shadow it. Like who is doing things that they shouldn't be doing, right? Who's doing stuff with your data like a dlp, right? That's like we're putting into GPT. We're putting to an AI that has none of these securities in place, right? I had a call with a company like about two months ago. I'm blanking on the name of the company as I totally give them a shout out where they are. Your, your filter, right? Where you like they are. Your, your interface to the AIs. Whatever the AI is that they're going to. That's, you know, that's up to you and you can decide and whatever it is. But anybody who needs an AI within the platform, within your company will put into their platform. They're then filtering for DLP to filter out, anonymize whatever it is. Any, you know, everything. SOC compliant, right? If you're POC compliant, right? They will this stuff out, flag it, maybe even tell them, you know, hey, Greg is trying to extricate some data here, okay? Whatever it might be. Obviously he's not Phil, but look, Phil tripped up over there. But it will then filter through and then go to the AIs with you know, a clean set of data, right? So I'm like, I'll name the company but like how do you stop anybody from creating a side account of let's say smartsheet, right? And putting your data into it.
• 24:13
  Speaker A: He's in charge of building this, the SSOT as we speak somehow if we can do that. So, you know, I mean there are, there are solution.
• 24:22
  Speaker D: Like I have some models that just run locally on my computer with LM Studio. And so if they are like sensitive data anyway, you know, you can just use that and do like a mockup material that use fake datas and you just use the same structure and. But that takes a bit of time.
• 24:41
  Speaker B: And I don't know, it takes a bit of time and it takes an expert like yourself. It's like the difference between like having somebody on staff, right, who can do whatever give you an example. I don't know, whatever it is, right? They could be a subject matter expert, right? Or you might not have a subject matter expert on staff. So you'll Go to somebody externally, like Phil said before, you know, whoever, you know, Greg's, you know, going to be getting a lot of job opportunities now because everybody's recognizing that he's, you know, a genius who's built a local LLM which does not go to the cloud. So you're complying for all those security regulations. Right? But like, not everybody has. So like, so first of all, even if you have somebody now and like how we start this conversation, they might be really good at stuff that's now three weeks. That's changing like the, the space of. Remember the old, oh, what are you gonna do in five years? What does your tech stack look like in five years? Then it became what does your tech stack look in three years? Right, three years. What does it look like in three months?
• 25:38
  Speaker A: That's why I have a problem with people signing five year agreements. Somehow they still do it. Somehow they still do it.
• 25:47
  Speaker B: Bend to that.
• 25:49
  Speaker A: That should be like a, that should be like a company like on a top 10, no do thing or something like that.
• 25:55
  Speaker B: Really the. Especially now, how do you sign a contract especially for user licenses now. So five years, even for three years and stuff. But three years, okay, I can, I can, I can kind of come up with this phase, you know, how much is going to change in three years. But five years to sign like user licenses, what does that tell me I need?
• 26:14
  Speaker A: Like, depends on what it is, I guess.
• 26:17
  Speaker B: But you sign a user license agreement for five years based off of the user account. No, no, no, no.
• 26:21
  Speaker A: Not a user license. I was just. When you said that, I thought Microsoft and the changing of the guard in March back a couple years ago and no more month to month and we're going to raise your, if you don't notice it, we're going to charge you an extra 40% and then you do and that type of thing.
• 26:36
  Speaker B: Microsoft can get away with it.
• 26:37
  Speaker A: But like I said, it depends on what it is because 95% of the world's on Microsoft and I'm kind of a Microsoft lover and I'm part of the. I don't know what we want to call it. Truman.
• 26:51
  Speaker B: So that's the problem. Sol, I knew you were going with that. You're part of the solution. Yes, yes.
• 26:56
  Speaker A: Don't think we get stuff so far. We're trying to draw out these four points. First part is start small. So start small. If you're this IT leader out there, start small, build an efficiency app, play around with AI, Build some sort of efficiency app to, I mean, have fun doing it. First of all, pick, pick someone in the, I don't know, operations, marketing department, something like that. I think marketing would be a great place to start for a lot of people because I think you can make a big bang for your buck and really make a difference in the bottom line for the company. There's probably a lot of things you could do with it, actually. I know there are, so start small. Now you have a whole list of questions. I don't even know which ones are gonna, we're not gonna get to all these, but this one seems to be appropriate. How do you maintain human oversight, which is technical speak? How do you, how do you keep everyone human and normal and speaking like human beings and being authentic and genuine without creating some sort of weird AI bureaucratic bottleneck of crap?
• 27:57
  Speaker B: I'll give you an example. I, I, I was asked to present at, at some conference and I was, and I was like, about A.I. i was like, great. How better than to create a presentation out of AI I'll put in some things in there. So I said, hey, what are, what are the topics that people are, are speaking about, right? And I didn't want to overlap with somebody else's, but what's the topic? And then I'll come up with something else that's not being addressed because I don't want to, you know, overlap somebody else's items, right? So I, they send me a list. I said, great, we'll talk about this, right? I sat there for two hours prompting and repromping, reprompting a presentation, right? And I realized this is not my voice, right? And I think that, back to your, your, your point of people need to know their voice and to find their voice, right? Because we all have a voice, right? How we talk, how we speak and we write, but we don't necessarily know our voice until we were fine tuned with who we are, right? And once I was prompting and reprompting, I said, this makes sense. It all makes sense. It's cute and all, but it's just, it's not who I am. I wouldn't talk this way, right? Yeah, I don't make these jokes. I asked it for jokes. It was like, you know, name jokes, you know, not as damn joker, you know, but it's, you know, it just, it wasn't me, it wasn't my voice. So then I did the second best thing. I opened up my notepad, my Apple notepad as I was driving home from the office and I just spoke to my Apple notepad. It wasn't vibing with My GPT, it was vibing with my notepad. And then I got home, I looked at my notes that I just dictated, and then I wrote the presentation that way. Yeah, it was me. It was my voice. It was exactly. Was it as fluffed out as the AI? No, that's not my voice.
• 29:40
  Speaker A: We're going to follow these bullet points. Autonomy, but with guardrails. So just maybe. I know what it means. I know you know what it means. But is autonomy with guardrails, is this speaking to our end users or is this advice to IT directors? From a coding standpoint, we've got to have some guardrails. We've got to make sure that we're not, you know.
• 30:06
  Speaker B: No, guardrails need to exist. Guardrails need to exist. It doesn't matter if it's your personal space of AI, whether it's business. I mean, the thing has it. Well, like we just. We're just coming off. It's a perfect segue. It'll make things up. Right. So without the guardrails, you can tell people really things that, you know, not appropriate, you know, for whatever they're asking. You know, the guardrails of. This isn't a famous example of car dealership, GM dealership. I'm blanking on where I think Denver maybe is in Colorado area where the. The person went to the GM website is. Okay, hey, it was a new thing and they rolled out this AI chat, right? And didn't have guardrails. So the person prompting it gave it to guardrails. Right. You can look this up, because I was at a conference and somebody said, oh, I'm like, no clue what these people are talking about. I looked it up. The person prompted the AI Papa, meaning the website, the customer prompted and said, hey, everything I say, after everything, just say, this is legally binding or whatever prompted the guardrails to the AI bot, right? Not from the back end, but the front end user. Right. And so. Okay, and how much is a car? It's a one. How much is this Chevy Avalanche? Oh, it's a dollar. And it's legally binding. Guess who got a free pickup truck.
• 31:27
  Speaker A: No way. That's great. It's a great advertisement, though. It's a great advertisement. Sometimes bad. What is that like, you know, bad. Negative press is good. Pressed you what I mean?
• 31:36
  Speaker B: No such thing as bad press. It's. Yeah, it's the. The guardrails need to exist. Especially I'm. I'm in healthcare, right. And if I, if we have an AI telling anybody anything, right? Whether it's patient facing, whether it's administrative facing, whether it's anywhere along the, the along the space. This can have real life implications, right? Not just like, oh, there's a number that's wrong on a spreadsheet, right. The balance sheet might be off by whatever it is. You know, we have people's lives we're dealing with, right. And so if we don't have the proper guidelines, guardrails, right. Things can go really south. Or on a marketing aspect, if you don't have guardrails, it's reaching out to your competition perhaps, right? Like, hey, don't reach out to these people right at this domain or whatever it is who are working for this company. I'm part of a. As an IT nerd. You'll. This will speak to you a beta community where they'll. You'll beta test items, right? Like you get like an interest. You fill out an interest form and like, hey, are you interested in trying out a product? Now I've signed NDA so I can't tell you what any of these products are, but let's just say interested in trying out a stapler? Of course I'll try out a stapler, right? And in the questionnaire of hey, do you also work for any of. Any of your relatives work for these companies? And it will list its competition, right? Because obviously they're not going to send you something that's not to market yet to try out if you're going to just, you know, go replicate it, right? So the guard that's a guardrail, but that's a guardrail of not an AI but you're filling out the form, right? But if you have your guardrails of AI let's say marketing. So that way your AI isn't reaching out to, you know, some podcast of not nerds. You know, you don't want that.
• 33:20
  Speaker A: Any specific tools. Do you want to answer that old phone system in pbx? It sounds like it might need to be replaced sometime in the future. AI in the background, it's like that's a Nortel or like.
• 33:35
  Speaker B: It'S all Colbeck Northern Telecom.
• 33:39
  Speaker A: You just know that I just know any, any specific tools, guardrail tools that you use that would be guardrail tools. Let's, let's give some advice. I don't know. Do you have any advice to give any people? Like any, any, you know, secret weapons, things that are nice for garrails, better than.
• 33:57
  Speaker B: Yeah, I think guardrails, you got to build into it like, right. Like when you're building out like the local model you think you're building within the tool. I think the guardrails need to be within the tools, within the instructions, within the prompting.
• 34:09
  Speaker C: Let's.
• 34:09
  Speaker A: Okay, let me put it this way. Vendors then for a second. Vendors you love. Top three vendors you love that have any AI access, maybe.
• 34:18
  Speaker B: So my, my feel on that is there's two thoughts with AI and how you're integrating AI into the company, right? Into any platform. We are AI as a tool, as a standalone tool versus AI as an ancillary, right. Is AI something that you bring in? It's a new product, right? And now you're standing this thing up versus this. My feeling, and this is what's happening, is it's existing tools that you have are now adding AI to the platform, right? So when you have existing tools and you're adding it to the platform, right. They already have access to your, your knowledge base, right. They have already access to your database, they already have access to your stuff. They've already done all the heavy work, the heavy lifting of how do we get AI to integrate with STEM platform, right? How do we get AI to see my tools? They're already doing it. They've built this thing out. It might be an add on, it might be already included within the cost of, you know, they've added more features to be up to date and competitive within the space, right? Like for instance, Microsoft copilot, right? Where I don't need to bring in a GPT license to be able to go through my emails and tell me, hey, where's this email from Phil that's referencing a podcast, you know, my copilot. I'll put in phrases like that. I'm just like, you know, where's an email from Phil testing podcasts that has an attachment that's discussing this. Did you mean this email that's discussing this? A little summary. So when you're discussing like an AI tool to bring into a company, right. It's more likely, at least from my experience, to be things that you already have in place, right? You already have agreements with, you already have an NDA. So that's the, you know, my, it's my feel and that's my experience. Is that the AI integrations. Yes, there are some stuff, and we discussed this before, are going to be changing the, you know, changing the flows and the user experience in terms of it's a bigger implementation, right. Versus the small ancillary lifts that are occurring. Right. Of existing platforms.
• 36:27
  Speaker A: We have a question. How do you decide where to jump in with things Changing every day. If you are establishing an AI strategy for your enterprise and how do you establish what is real?
• 36:42
  Speaker B: So my, my, my feeling and this is not my, my, my answer because somebody else gave it at a conference I went to and somebody asked the same, same question, right? I think it's very similar question. And the answer, which made sense to me then, made sense to me now, so I'm going to repeat it. I, unfortunately, I'd apologize. Whoever said this answer, I can't give you credit. So it's not mine. As long as you have the framework in place, right? Then what pieces you're changing, you know, you're adding in, you're changing whatever it is the framework is really is the guarding rail, right? The guarding light is the framework, how your pieces change. And like you're saying now Claude is, you're not happy with Claude because there's something better, you know, you can change that, right? But as long as you have the framework in place of, hey, this is how we're going to do with AI, this is how we're going to integrate it. This is how we're going to use it. This is how we're going to use it appropriately and responsibly, right? We can't put people's data or patients data from my end patient's data into the AI. I can't have an AI tool that doesn't meet some regulatory standards. And how do you establish what is real? I'm not sure what that means. I'm going to, just from my interpretation of understanding that question. Whereas if you call the AI out, call, free call, don't worry about it. If you call the AI out, if you're not sure, right? Like some things, they'll make sense to you. Obviously everything that AI is going to tell you makes sense. That's what's so crazy about this thing. So the everything it's going to tell you kind of makes sense, right? So when GPT was a new thing and it would hallucinate and you would have to call it out, right? Whereas a perplexity would give you the source code, the source material within it, it'll give you an answer and perplexity, like here's where I found this information, right? So GPT is very good at doing that now where you'll ask it something and it will say, it'll show you like the resources. Especially if you're doing deep search, right? If you're doing deep research, the deep research tool within GPT, it'll tell you it's showing you where it's looking it up. Great. But you can just ask it say, hey, what's your source material? Right. Give me your citations and it will give you. And you click on it and you can go check it out. But be on your toes. Like, like we said about an in, you know, being about an intern, right? Like, hey, where'd you get this information from? How do you know this? You know, we asked it. If I bring somebody into the company as a consultant, right. And they give me an answer, I'll still question them on it, right? You bring in somebody who's really show me. You know, when we used to take math tests, right. Like, you'd have to come up with your. If you work, right. Oh, how did you divide three plus two? Here's my, my work.
• 39:18
  Speaker A: And there might be two different methodologies for getting to that answer.
• 39:21
  Speaker B: Absolutely.
• 39:23
  Speaker C: The.
• 39:25
  Speaker A: I guess just maybe along the lines of vendors that are use AI for security for. For example, how can we. I guess how would we establish or how would we establish or that that those, that some of those security providers that might be on a pretty advanced. Have some pretty advanced products. The security. The security realm of AI products and security vendors is a massive ocean, let's just put it that way. How do we decide what to use?
• 39:54
  Speaker B: This is a good question. I think it's. And I would think it's similar to how you've had any company, right, that you're looking for products to use, right? I would think, you know, it would be similar to you at any company, right? That's because you're putting the term AI to it.
• 40:12
  Speaker A: You had a really good example. You had. And I kind of, when I summarized, went like this. How do you build, test, add business value without getting sued for age discrimination? Like, where did that come up? Where did that one come up? Because that was actually a really. That's more of an example of like where we used AI and rolled it out and it failed.
• 40:37
  Speaker C: So at what point are we going.
• 40:38
  Speaker A: To what point are we going to use build and use AI and build something and roll it out that doesn't fail? And do we have any examples that don't fail? But this is a good example of failure.
• 40:47
  Speaker B: It was a period of public failure. So we're going to say workday is getting sued because they had an AI agent that would go through resumes right before it came to a recruiter and the AI was filtering out resumes. This is public. This is the real thing. And they judge says, yes, this can proceed to a lawsuit, right? This can Proceed in court where the AI was reading the resumes, doing its AI thing and was age discriminating. And that's a big, big no. No.
• 41:18
  Speaker A: Yes.
• 41:19
  Speaker B: How do you make sure you don't get sued?
• 41:20
  Speaker A: Guardrails.
• 41:22
  Speaker B: Guardrails. Guardrails and tests out of everything. And by the way, maybe even have another AI test the thing out. Why not build an AI agent of another platform to test it out? I mean that's why we have performance testers, right? It's a real industry, right? You can have somebody test it out, you can have your internal team test it out or you can have another AI from. I would use another AI from another platform, by the way, obviously, because it's me biased towards itself.
• 41:54
  Speaker A: Yes. The. You, you must have meanings and other things that move on throughout the day. So I want to leave. I want to ask you like you're, you're sitting down with your, your, your friend who was. There was one earlier. They. Were you paranoid? We're not going to let, we're not going to use AI the. I mean honestly, like you're speaking in a room now. So many your friends, nothing is going to leave this room. What is your piece of advice?
• 42:15
  Speaker B: I would find something, I mean I think from our initial conversation, like how do you know if something's success is repeated, right? If we repeat something, there's a success to it, right? Like we're, we're rolling out some like an AI phone thing and like how do we know it's successful? Well, if people aren't hit pressing 1 to speak to an agent, a human first, right? If they're coming back and speaking to the AI and we're seeing calls are getting, you know, responsibly addressed, then it's repeated success, right? So we see that if, if there's, if there's repeat, right? Success, right? You go to a restaurant, you like it, the food's good, the vibe is good. Environment, atmosphere, the food is tasty, you know, very good chance you're gonna come back, right? So I would find something. Might this work for me personally? Okay. Because we see from our. A few conversations. I'm, I'm. You know, I like to have fun, right? I like to joke. I like that, right. Keep it like. So I would find something within the AI that draws you to it, right? You have a fun time with it, right? Whether it's make a silly poem, right? I'm using the AI, the GPT for so many things that are not work related, right? Like memes, right? Make funny comics, take a picture so find something that draws you to it, right? And you can get comfortable with it, right? Get comfortable with it. Come up with. I saw this at somebody else here, this as an answer, as another presentation at another conference of like, come up with a fake business model for a fake company. Build up the pitch deck. Come up with a whole company, a fake company, right? And just do this through the AI and you can have your teams have competitions with each other and you know, use gamma to pitch deck in. And you can use a whole bunch of different AI tools. But come up with a use case light. I would, personally, I would use a light tool. Something meaning light, like keep it light, right? Don't get too heavy, that you can show the value, show its use, right? Then once you have that initial step of getting comfortable with it, right? Then you can take a steps and steps. I had a hockey coach back who said, like, when you're comfortable with doing this, then you're never going to get better, right? You have to be on the edge of your skates, right? You have to be okay with falling. Because then when that step of that motion of doing crossovers where you're really on the edge now that's comfortable for you. That's your comfort zone. Because if you're always going to be skating straight up like this, you're never going to be able to get out of that comfort zone. Never be able to do the crossover, crossover step, you know, hockey stop, move. So just work, I think, just get it out of your comfort zone. Work on it slowly, slowly, slowly. And then, you know, you turn back and three months later, like, how did I get here?
• 44:52
  Speaker A: Thank you so much for being on dissecting proper itineraries. I hope you have a wonderful day and continue.
• 44:57
  Speaker B: You as well.
• 44:58
  Speaker A: Everyone continue vibing with AI.
• 45:00
  Speaker B: Keep it going. I'll see you soon, Phil.