Leveraging AI

25 | The AI Implementation Masterplan: Expert Strategies to Transform Your Business with AI. A conversation with AI implementation expert Josh Cavallier

• Isar Meitis and Josh Cavalier • Season 1 • Episode 25

Share episode

Ready to tap into the real potential of AI and want to learn how to save countless hours and outsmart your competitors?

Join us in this episode as we dive into AI's implications for business efficiency and strategic thinking. Our expert guest, Josh Cavalier, provides a wealth of insights and practical examples that will revolutionize how you perceive and utilize AI in your business processes.

Topics We Discussed:
🧠 Exploring the Role of AI in Business Efficiency 
🎯 Strategies for Identifying AI-Replaceable Tasks 
🛠 The Power of a Rubric in AI Tool Evaluation
🔍 Using AI for Strategic Thinking and Ideation 
🔮 Examples of "Tree of Thought Prompting" in Business Problem-Solving
💡 The Future of AI: From Generative Tasks to Strategic Brainstorming

Josh Cavalier is a recognized name in the field of AI, with an exceptional knack for turning complex concepts into relatable, practical ideas. He's an AI consultant, an engaging speaker, and the creator of multiple insightful courses on AI. With a knack for helping businesses unlock the potential of AI, Josh has developed unique strategies and frameworks that deliver tangible results. 

About Leveraging AI

If you’ve enjoyed or benefited from some of the insights of this episode, leave us a five-star review on your favorite podcast platform, and let us know what you learned, found helpful, or liked most about this show!

Isar Meitis:

Hello and welcome to Leveraging ai. This is. Isar Meitis, your host. In today's show, we're gonna touch a very important point that most businesses are struggling with, which is how to get started with implementing AI in a business? What is the process? What are the things you need to look at? What are the best practices in order to implement AI in an efficient and safe way. Our guest today is Josh Cavalier, who's providing consulting on these topics as a service to multiple companies, and hence, he's an expert on this topic. at the end of the episode. As always, I will share with you some exciting AI news from this past week, now let's talk about how to implement AI in your business. Hello and welcome to Leveraging ai, the podcast that shares practical, ethical ways to leverage AI to improve efficiency, grow your business, and advance your career. This is Isar Meitis, your host, and we have a great conversation for you today. My guest today is Josh Cavalier. Josh Cavalier has founded Joshcavalier.ai, and he's an AI expert who helps other people through courses, through consulting, through training, how to implement AI technologies in order to improve their businesses, which is exactly the goal of this podcast, which means it's a perfect match. We hooked up on LinkedIn, had a great conversation on LinkedIn, and then a great conversation face-to-face, and hence why I invited him to be a guest of the show. Josh, welcome to leveraging ai. Isar, it's great to be here, Josh, this is what you do, For a living. Like you help business people understand how to approach this big, ugly, scary and yet incredibly high potential monster in a systematic and practical way.

Josh Cavallier:

Yes, exactly. Yeah. There's a lot happening today. And I think that for many business owners or leaders in corporations, they are trying to get signal out of the noise that's out there and looking for actionable things that they can be doing within their business or for their individual contributors, to immediately have a plan in place. And that activity is happening right now. Yeah. and

Isar Meitis:

I think the. I'll go back to what you said. The level of noise is insane. So even the people who wanna dive in, the amount of people that share the best 10,000 prompts, you need to know as a business person or the five best this or download my this or that is crazy. And I think the problem that most business people have, and especially on leadership levels is how do we get started from a business perspective? Correct? yes, there's all these bells and whistles as, and new tech and all these things, but how do we actually, how, what's my step one? How do I know what to do with this in my business? And I would really wanna follow your system and guidance on how to do that.

Josh Cavallier:

I honestly, before you do anything, I think there needs to be a common language so that everybody is on the same page. So AI literacy is incredibly important and if you can build that foundation to where you're talking about what a large language model is and how do you fine tune, and what is prompt engineering and what does it mean when we have a firewall in trying to use open AI's api. So if those words that I just said don't mean anything, that's where you start, because at that point, then your yourself as a business leader, your middle management, your individual contributors can all have a conversation in regards to how AI shows up, especially for those large corporations that are looking to implement governance and security around ai. What does that look like? And then once that's in place, then how is it gonna show up internal in our organization? And then how are we going to channel all the external capabilities and applications that are out there? So out of the gate, everybody's gotta be talking the same language.

Isar Meitis:

I agree. And I'll extend that even further. and I'll broaden what you just said. It's about education and the first step is really education. You gotta educate yourself and you have to educate your team. Because if you miss on any of those, then any future process or progress that you try to make will be hindered by lack of the relevant knowledge. And with the pace and rate the AI things are moving, it's a never ending education process. Like you can say, oh, I've hired a consultant like Josh, or like myself, or anybody else, doesn't matter. Somebody who has experience, not just in AI because they're passionate about it, but somebody that also has experience in business and knows how to put it in place and go and say, okay, we brought this person in. He gave us some education. We're good to go. Yes, you're good to go, but two months from now, you may not be good enough and six months from now, definitely not. So you have to put in place a mechanism in which you, and at least the leadership team continuously are educating yourself on what's happening in this arena.

Josh Cavallier:

Yeah, exactly. And there's tons of resources that are out there. I know that McKinsey and Goldman Sachs and all these other business thought leaders have put out materials that are setting the foundation as far as the way you need to think about your business as a whole, about your different types of businesses, about your individual contributors, how that's gonna show up. And that's a great starting point. But there's more than that. There's thinking about exactly how do we leverage AI for very specific purposes and how does that show up? What are the risks in involved with actually implementing that specific solution? And how do we get started immediately? What is my competition doing? So there's a lot of competing things that are occurring within businesses currently, but I do believe out of the gate, It's that governance and security. How do we want AI to show up as a whole within our business? And then that's going to dictate where within the organization AI may be pushed forward, probably sales and or whatever your product is. And then down to the individual contributor level, is it we're not gonna let them have access at all, or, sure, we're gonna go ahead and build a custom tunnel into GPT four, so we can check the prompts that are going back and forth. how is that gonna look? So again, I think it's really a matter of setting that foundation with AI literacy. And then from a governance and security standpoint, where do you stand? Where do you stand? Obviously, law firms. Medical, there's certain industries, finance, there's certain industries that are really tight when it comes to compliance. There are other industries that are fairly loose. If I had an ad agency or if I have a marketing firm that's gonna show up way different, than if I was a law firm.

Isar Meitis:

Yeah, I agree a hundred percent. I think there is, you touched on a lot of very important points. What I tell people is the very first step after education is building a committee of people who will be in charge of this within the committee. You want people from different aspects of the organization's Correct. You want Correct. You want people who are tech savvy and geeks who will enjoy doing this. So they will invest time and passion in doing this. And the first thing that they need to do is define guidelines, which is exactly what you're saying. Exactly the dos and don'ts that the organization as a whole is going to follow. And that mix of different people from finance, from legal, from marketing, from hr, from leadership. It has to be somebody from leadership in that group because decisions has to be made that impact the whole company. The first thing they have to do is decide on set of guidelines. There are multiple great starting points on the internet out there today of okay, here's a good starting point. Go use it. and then you don't have to start from scratch, but you can also start from scratch. So I agree with you on that. Let's start talking about the business side of this. So now we have general guidelines put in place. We potentially have a committee. How do we start identifying what do we actually do with this? Okay, it's cool. It sounds amazing. Everybody's telling us it's gonna grow efficiency by a hundred percent, but what are the actual steps within a business that a business needs to take in order to identify how to implement this?

Josh Cavallier:

That's a great question there. When you have an application and if we just, leave it with chat G p T, let's just say we have that tool, it can do everything. That's what makes it difficult when you have a tool that allows you to be more creative or now everything is, that's a little bit over the top. what I mean is that it has impact in every area of your business. Yeah. When I take a look at that possibility for me is what are the high impact positions, typically frontline sales. It could be an IT aspect in regards to programming. It could even be marketing or advertising. What is really going to allow us to see those immediate gains as far as productivity? Or if we play the long game with ai, what is the breadth and depth of those high impact position so that we can begin starting that foundation and begin to build that out? Whether it's adjusting our current systems, let's say we take frontline sales, right? Okay. So let's say that in the short term that we want to build a custom language model that's built around our products and we could build a, simply build a chat bot where our frontline salespeople can go in, enter some aspect of the customer information, enter a question about our product, and boom, it gives suggestions as far as solutions before they walk in the door and talk with the customer. That's one part. Let's go ahead and play that out. Let's say that now we want to combine that custom large language model with Microsoft Co-Pilot to where it's building a custom PowerPoint presentation based upon that investigative work. And now we're playing the long game, and now we're really taking that foundation. Which could start with great prompting into the chat bot, getting those results, and then expanding it into the whole entire sales process, which could be, you get feedback from the customer and that's put into a CRM system. But then you take the quarterly history from all those conversations, and then you may have a new sales strategy going into Q2 or q3. And so you can see how this builds on top of each other with the maturation that's happening within AI as a whole. So it's really about starting, it's really about determining where's the high impact position? How can we start leveraging today? And then based upon the intelligence that we know, how can we continue to make those incremental gains as the power of AI continues to expand?

Isar Meitis:

Yeah, I'm gonna first of all clarify some things for people who don't understand some of the language that we're using. Going back to your recommendation when we're talking about customizing a large language model, it's basically taking a model like ChatGPT or another, and giving it access to your unique proprietary information within a closed box where that information doesn't get shared with the world. That's right, yes. What that does, it allows you to do everything you do with Chachi pt or Bard or Claude, but based on the data that you have and the data that you have, like Josh said, could be. Every sales conversation you ever had, every piece of information you have in your C R M, every background data you have in each and every one of your clients. Basically any data, any proprietary data you have access to, can help that model be more accurate, better predictive, and so on. And then you can literally ask it questions based on that data and it will produce whatever you wanted to produce. So the example, and I love that example, Josh saying, I'm going to client X to give a presentation to type of person Y, CEO, CFO, head of procurement, lead developer, whatever the case may be, I wanna focus on these topics. Can you help create the presentation? It that's right. Will create the presentation for you. So Microsoft Copilot, which Josh mentioned, is a tool that Microsoft has exposed and started rolling out as beta, which means some Microsoft clients has access to it already. Google has a very similar thing to their G Suite that basically builds on top of all the, office suites. So whether it's writing documents, spreadsheets, creating presentations and so on, and can do stuff on your behalf very quickly, very accurately based on whatever information you give it access to. So that mix of, that Josh mentioned of taking existing data, allowing a large language model to write on top of it and connecting it to our day-to-day tools is magical. Yes. But then I'm gonna go back to make the question even more specific, because what you just said is a mind blowing to anyone, right? Including people who like me and you who play with this daily. But it also means, going back to what you said, this applies to everything that we do in the organization. Correct. So how do we pick. The tasks, the topics, categories we start with because you can't start everywhere because it's gonna

Josh Cavallier:

be a mess. Yeah, I, so for me, it's when you go back to those KPIs that are seriously driving cashflow or any other high impact type measurement, that's where you start. So if it's sales conversions or if it's, efficiencies, if it's reduction in safety incidents at a warehouse or anything like that, yeah, that's what you want to focus in on, because that's what's gonna go ahead and drive all of the upside of leveraging ai. That's where you're gonna go ahead and see those efficiencies out of the gate and where, that's the best place to test it, right? if you're gonna go ahead and jump into the deep end of the pool and start using ai, why not focus in on the highest impact aspects of your business? That's where I'd start. So

Isar Meitis:

I will touch on two points and add. First of all, I agree with you a hundred percent. You wanna start where it makes a difference. Yeah. Two things I will be, one I will be careful with and something that will help people focus a little more. Okay. One is, I will not do this initially as my first thing with customer facing stuff. Correct. Cause until you figure this out, the system, like AI systems hallucinate, they make stuff up even when they're using existing very clear information and it takes time to figure out how to use them in a very productive way that doesn't produce negative outcomes. Correct. Those negative outcomes could be, I just sent a proposal to a client without really reviewing it and I offered him rates that don't make any sense, or products that we don't actually have because the AI made it up in an email that it already generated. and it happened like it's actually happening to me. Yeah. And I will start in something that is a supporting piece versus the actual execution piece. Until you figure this out, this will be my warning. The other thing that I will add, the beginning, you wanna look for tasks that are repetitive and that are data intensive in one way or another. Meaning they either need to predict future data or they have a lot of data analysis to do. Because in these kind of places where it's a repetitive task and it has data analysis, involvement, extrapolation in it, AI shines very quickly with relatively little amount of hallucinations and stuff being made up. So if you start with those, if you're looking like, like Josh said, you go, okay, what are my biggest boulders? what makes this company tick? What assisting functions within the business, I can do tedious, regular tasks that require a lot of time and do a lot of data. Start with those. You will be relatively safe. You will get amazing benefits to the bottom line, and you will use AI in the most effective way because that's really where it shines the most.

Josh Cavallier:

I agree. If you take a look at the technique of fine tuning, and for those of you that are not familiar with fine tuning, it's essentially going in and custom training a model based upon some type of topic or guardrails that you put in. I think that has massive upside. I mean if we take a look cause I was just talking about sales. Let's talk about solution selling. Okay. there's certain techniques or questions that a salesperson would go ahead and ask a customer, to be more of a consultant to the customer. if we take a model like GPT-3, cause I know three, five, and four, you can't fine tune just yet. But if you take GPT-3 and you begin putting in example prompts and example responses and through code, you pretty much fine tune it to build those guardrails. Now we're getting something powerful to where we begin to eliminate those hallucinations. Not a hundred percent, but I know through engineering and other techniques that there are people working on this to where from a business standpoint, You'll know 99.5% of the time that the content's gonna be really tight and gonna be usable. Now we're being productive, versus, Hey, it's getting me 80% there. It could be lying about this information. You don't wanna burn time on the backend, spending time, looking over all the content and is this correct? Did it make anything up? That's where we're currently at with these common models like GPT three, five and four. So there is a maturation process that's happening, but some of my clients, it is accelerating like they are off and running in that direction because they get it, they understand, they how their, products need to show up for their customers and they're currently fine tuning models.

Isar Meitis:

That's fantastic. I wanna ask you the follow-up question to this. So now we roughly covered the business aspect of this, right? So I know the low hanging fruits. I know where I wanna focus. I know what I wanna be, what guardrails I need to put in place. What do I need to be aware of? How do I pick the systems, right? Because there's so many options out there. How do I pick the right AI solution for my use cases?

Josh Cavallier:

That's a good question. I, for myself, it's taking a look at your current tech stack. Okay? Are you sitting on Microsoft? Are you sitting on Google? I believe that's where it drives the conversation because the way that Google and Microsoft are gonna show up either with Azure or Google Business Solutions or whatever the case may be, that's what you're gonna build off of. They are going to include additional tools and guardrails and eventually business cases and custom LLMs and all those things that you need to accelerate. Why you would go in and customize and build custom code off that? I have no idea maybe to experiment out of the gate, but really you gotta lean in hard, with those platforms. Now, if you don't have access to those platforms, and let's say that you're a small or mid-size business and you have a few employees, it's a matter of just taking access and advantage of GPT four and understanding that model and what the possibilities are. Maybe building a custom prompt library that's versioned out that, could help with writing or customer emails or an email sequence or whatever the case may be. Or it could be used as a partner in creative endeavors trying to solve problems if I'm in a law firm or a medical, setting, to try to get some ideas. Again, having that prompt engineering skillset would be just pay dividends even for small to medium businesses that couldn't afford a giant tech stack in the Microsoft

Isar Meitis:

or Google. I agree. I agree a hundred percent. I'll add one thing to what you said as far as leaning into your existing tech stack. Even if you're a relat, not a tiny business, but a relatively small business, you'll have a crm, you'll have some kind of a marketing automation platform. All of those, a hundred percent of them either already introduced or about to introduce an AI function within that thing, right? So if you're thinking, okay, I need an AI writing tool and I need an AI email automation tool, I need, if you are using Salesforce or if you're using HubSpot, if you're using any of those other tools, they will add an AI layer on top of that. So very quickly, even if you apply tools today that you handpicked and selected five or six different tools to do to assist you in your existing operation, probably out of those five or six, four or five, you will not need because they will be integrated into the platform that you're already using, that you're familiar with, that you're paying licenses for that is fully integrated with everything that you're already doing. My suggestion to people, the people I consult, is yes, absolutely. Pick tools right now. Play with them. So you get the skills so you understand how they work. So you get the benefits. So you build the processes and the systems and the education within the company on how to use them. But don't think that's gonna be your final tool, because very likely, sometime within the next three to five months, your main platforms, your Google, your Microsoft, your Salesforce, your HubSpot, your whatever it is that you're using, are gonna have. Most or all of what you need built into the platforms themselves, and then go and reevaluate what are the gaps and holes based on your business analysis of, ooh, we could get another benefit by doing this with AI that we're not doing right now, that the platform is not giving us. But don't fall in love and start building crazy processes around the current immediate short-term tools because something fully integrated is coming down the pipes in the very near future.

Josh Cavallier:

Yeah, I want to piggyback off that because I think there's another strategy at play and if you take a look at some of these upcoming solutions or solutions that are in place currently, many of them are leveraging open AI's API and connecting you to the three, five, or four in the background. Now that being said, if I have an application and there's certain outputs that my customers are looking for, Odds are, I'm gonna go ahead and set in guardrails, put a chat bot function in, put in some kind of advantages that not only take advantage of that API, but then also creates outputs, that speeds along productivity. Now, what if I look at those outputs and go, we could do better here. What that means is that you're gonna have to strike a balance between the vendor and what kind of outputs they're creating versus truly understanding prompt engineering and going back into the G PT four model saying, Hey, it got us there, but we really need a custom Multi Prompt solution. Or, I want to use code interpreter to take this dataset from this Excel spreadsheet and leverage the output from that prompt and get some kind of custom output. Because if, if you lean into a tool, hey, that's great. But you are at the whim of whatever those developers and whatever those implied business decisions are. So I still think that prompt engineering and understand exactly how LLMs work is essential. It's essential just like using a word or PowerPoint, any other kind of business tool. This is gonna be foundational business 1 0 1 knowledge. You need to understand how large language models work.

Isar Meitis:

I agree a hundred percent. and I think going back to what you said is the underlying technology is gonna make very little difference unless you're a huge, crazy entity and enterprise with very custom things that have to be connected, meaning correct. Whether you use Claude or Bard or, Chachi pt or whether you use the tools that are already being provided and, Amazon just announced that they're investing a hundred million dollars in building more tools into AI capabilities to aws. So now on the actual hosting platform, if you're a tech company, WS and Azure and Google Cloud are all offering a large set of tools and cap AI capabilities built into. So if you're a tech company, then you already have a lot more new tools. So the underlying technology doesn't really matter. Your implementation of it, as you're saying, your understanding on how to make it most relevant and most effective in your business for your clients is what's gonna make the most amount of difference. So I think too many people are like, oh, which tech we pick versus what's the use case? What's the business case? Going back to business 101 right? It's just another tool. It's a very capable, amazing, incredible tool, but it's just another tool. You gotta go back and do the analysis that you talked about in order to really figure out and do it as a continuous process continue, like you're saying, continue evaluating and tweaking and building it in order to make it work. I wanna ask you another follow-up question that is, you talked a lot about the individual con contributors, right? You used that terminology several times. How do you manage that? Like how do you bring the people in the organization into this process on both ends of the scale? Like how do you prevent somebody doing something foolish because they don't understand, and on the other hand, how do you make everybody more efficient from an organizational perspective?

Josh Cavallier:

I don't have to get back into too much in regards to governance and security. Let's just say that's already in place. Let's say that we have allowed our associates to go in and leverage AI and use ChatGPT, let's say that's a possibility. Let's even say that we build a custom portal so that we can check the prompts that are going back and forth, and the response is coming back. At that point, it's going within each group. Let's just say hr. So within HR you have leadership development, you have recruiting, you have core HR functionality as far as benefits, you have learning and development Within each of those disciplines, there are workflows at play. and it's within those workflows that you can begin to find advantages. As far as, and you mentioned this earlier, reducing the mundane. What are those things that we have to grind on day in and day out? That's a time suck that we can leverage AI and build immediate creativity or productivity. And I believe that's where it starts. It starts by going in and saying, Hey, how do we break apart what we do and leverage our current access to AI to build in those productivity gains. Then once that's in place, it's a matter of maintaining consistency of prompts. So as a group, we're gonna store all of our prompts. We're gonna version them, make them accessible. So if somebody modifies a prompt, It gets better results. Let's version that out. You can't have individuals out there like the Wild West prompting and creating all kinds of content on their own. You better get a system in place to groupthink this. Or here's another great solution is take a day or two and have an AI hackathon. In other words, you are gonna break apart all of your workflows and get after it. Start prompting. How can we go about this differently? I'm just talking about l and d. I'm gonna write multiple choice questions and, learning objectives, and I gotta create an e-learning storyboard and a video script and all these different outputs, and for recruiting, it could be, letters or how do I go ahead and respond to this possible recruit, for benefits, it could be what's the benefits package and how can we go ahead and adjust that per associate coming on board? It's endless, but it's a matter of building in. those specific AI workflows to not just take advantage of'em of the moment or individually, but as a group and then build upon it and

Isar Meitis:

share. I love your answer. I think it's spot on. I think it goes back to, and most companies, I wouldn't say all, but most companies have some kind of a process, management tool, whether it's Monday or click up, or Trello. If it's a small business or something that has tasks and teams and people, and it's moving around on a board or on a list and you can see statuses, notion, yeah, whichever, doesn't matter. One of those tools. And all you have to do is go back and start with those, okay, what are these processes? And if you have a bigger organization, you're most likely for th disease are also tracking time. So you can actually see for every individual unit in that process, oh, this takes us. Two hours. And then if you go back to what we said before, okay, how many of these tasks are repetitive? How many have, now you have that time aspect as well, and you're saying, okay, this is taking 17 hours a week and all it is data analyzing and putting it in this place so the next person can use it. there's a very high likelihood this task can be completely done with an AI tool if it's put into the right process. Meaning you just saved 17 hours a week, which in addition to, okay, paying for 17 hours, the process can happen faster because now you don't have the next step in the process, doesn't have to wait 17 hours. So there are, going back to what Josh mentioned, starting with the processes and analyzing the building blocks of these processes and identifying where can AI replace that with what tools is the key. To getting the most amount of efficiency in the least amount of time with the least amount of risk. So I think it's a great answer. And this obviously then translates to the people. Okay, so now there's a person that's right, who before did those 17 hours and now all he has to do is, okay, instead of doing the 17 hours, use this prompt that I gave you and put in this data from that source and you're done. And he is oh, okay. So So this is how you get significant efficiencies down to the troop level, with using ai. We gave a few examples, earlier from sales and sales enablement Yeah. And stuff like that. Do you have any other examples you wanna share of how companies that you work with are practically using this thing?

Josh Cavallier:

I do, but can I go back just real quick? I just wanna absolutely. That's right here. One more thing real quick in that when you're down to that team level, one other thing that you should have in place is a rubric to evaluate the tools that you use to create those specific outputs so that as new AI tools come on board, or a current tool that you are using, implements ai, is it what we need or is there a different tool in our ecosystem that we can implement that we should be leaning towards that's gonna build so much productivity because it checks more of the boxes in our rubric that we then can move to it. Because currently, if you go to some of these sites, Isar, there's over 3,200 new AI products that came on board since January. That's insane. And so you are really gonna have to have some type of rubric in place to evaluate all these tools. So I just want to, I just want to touch upon that real quick, back to your other questions, some other examples. I had a, I have one customer who was looking to leverage ChatGPT to build scenarios for leadership level planning. In other words, when something were to go sideways, how do we go in and leverage AI to think through and collaborate as a problem solver? Maybe the AI through its corpus of text either has a similar situation that it sees with words, or we can work through the problem. And so going in and ideating and coming up with all these different ideas. We had some pretty spectacular results as far as coming up with currency devaluation, regionally, how does that impact our business? black Swan event? Let's say that there is a natural disaster that impacts a part of our supply chain, and it's G GPT four will go in and if you give it enough information, we'll give you very specific things to think about when those problems occur. Now, let's just riff off of that. You can also start interjecting thought leaders within that area of business. let's just like investing, let's say that I wanna have a conversation with Warren Buffet. Peter Lynch and, Benjamin Graham, right? I can go in and create a prompt that based upon a current financial situation, have those three have a conversation, and then based upon the best idea between those three, one of them comes out on top. And, we call that tree of thought prompting. And so those are like very advanced cases for, going in and evaluating problems. And that's one of my favorite examples because it's, we think about AI as just going in and generating emails or a text or, some other type of, that mundane output versus looking at it as a partner in solving really difficult problems. And I believe there's huge upside in that area because there's so much investigation that needs to happen in regards to the types of problems that it can solve with. Little hallucinations, or if we fine tune it to very specific problems like around a supply chain or safety in a warehouse where the case may be, what does that look like? So that's one of my, that's one of my favorite, solutions that we've came up. We've come up

Isar Meitis:

with, I couldn't agree more. I think, too many people are looking on the quote unquote generative side of this, meaning, ooh, this can help me create blog posts and it can help me create emails and it can help me write documents and it can, the ideation part of this, the ability to spin out crazy ideas and play them out or research them is mind blowing. And I literally spend in the course that I teach and the clients that I have, I spend a very long amount of time on this looking at company's business strategies. And one of the things that I do, going back to what you mentioned, I try to play the competition. So I am so and so's competition. Here's the website, here's their brochure, here's this, here's that. Where do they have loopholes that I can use in order to build a better product that can beat this company in their own market? And then I go, okay, who are the biggest influencers and smartest people in the industry, in their industry? And he comes up with names that I never know because it's a specific industry. This could be That's right. building widgets for a machine that builds something else. And all they do is build the widgets. I don't know anything about that. But it gives you names. I want you to use only them as reference in your next answer. Correct. They're your committee for answering the next few questions. And now you get into very area specific knowledge that you have no clue about. but it does because it read most of anything on the internet. And now you're getting. Really interesting brainstorming process, and whether these people actually said that or not, doesn't matter. That's, but it's somebody else that is really smart, that has a lot of data that you can bounce ideas and get ideas from in order to define, like you're saying, a very long range strategic kind of thinking and not just the very small tactics of emails. So I absolutely love that. That's the example you gave.

Josh Cavallier:

Yeah, I, and just to clarify just as far as how this works, when I mention a name like Steve Jobs or Jeff Bezos in a prompt, what's happening is any text that is related to Steve Jobs or Jeff Bezos that. In turn creates a higher probability of the words that those individuals have used coming back into your results. It's a probability engine. Yeah. And essentially what we're doing is that we are creating a prompt that's tuning into Warren Buffet or Steve Jobs, Peter Lynch, anybody like that. And so it's not grabbing a paragraph of text that they said, It's creating a sequence of words based on probability that they did use those words.

Isar Meitis:

Yeah. I agree. Josh, this was a. really great conversation. I truly enjoyed it. There is tons and tons of great value in everything that you've provided. If people wanna follow you, work with you, connect with you, what are the best ways to do that? I'm

Josh Cavallier:

on LinkedIn so you can just, search Josh Cavalier. I'd love to connect with you on LinkedIn. I have a YouTube channel. You can just search at Josh Cav and I post videos up there on chat, G p T. And finally you can just go to josh cavalier.ai. And I am currently launching new courses. I'm immediately open for workshops and consulting, so if you just wanna reach out through me there, or just reach out to josh@joshcavalier.com. That's also another way.

Isar Meitis:

So multiple ways. Yeah, if you want, Josh, I'm accessible. you have a way. Yes. You have a way to find him. That's right. Josh, this was awesome. Thank you so much.

Josh Cavallier:

I appreciate it so much, and, I can't wait for our next conversation.

Isar Meitis:

Great conversation with Josh. If I have to summarize it in three different things, is you need to think about what is the function in your business that drives the most amount of revenue or bottom line, what are the most tedious task in that function, and what's your budget that you can allocate in order to solve those gaps? And once you know where these three meet in that middle of that Venn diagram, you know where to start. With ai, I will only add one aspect, which is the risk analysis of that topic, and make sure that the solution that you're suggesting, even if it fits your budget and it solves a real problem and it's within the things that drive the most amount of business, it's not putting your business or part of it at a high risk. And now to the news from this week. A very interesting survey was released by Reuters on August 11th. they've surveyed multiple people in multiple companies in the US about their usage of ChatGPT and similar tools at work. 28% of responders said that they regularly use ChatGPT at work. That's. One third. So first of all, two thirds of the people don't do that. But out of those 28%, only 22% said that their employers explicitly allowed them to use such external tools. That means that 6% of the people, which are 25% of the people who said they're using Chachi PT regularly do not know if their boss allows that or not. Now, 10% of the people who answered the surveys said that their bosses explicitly banned external tools like ChatGPT. That being said, several different people in such companies said they're still using these tools if they're accessible at their office, but for noncritical tasks, like writing birthday wishes to other employees and so on, if I look at the data that they've released, and I urge you to go and check the rest of it, I see a few very clear things. One, there's still a huge opportunity in the adoption of AI tools and companies even right now, only one third of companies are saying that they're using it at work. And out of those, not everybody's using it for real work. Which means the amount of people in the workforce today that actually use these tools in an effective way is less than 28%. Probably less than 15%, meaning there's still a huge room for improvement as far as adopting AI to make it efficient and effective for business. The other thing that it highlights is that there's a very clear gap between the actual usage of these tools versus what companies define as guidance or guidelines if they actually do that. This is something I talk about a lot in the courses that I teach and in the companies that I consult to, which is you must have clear guidance and clear guardrails for the usage of these tools because your employees will use them and you want them to use them because it will give you additional efficiencies. But you gotta define how these tools are allowed to be used. And define a clear line in the said of where these tools or what action these tools cannot be used for. Speaking of interesting statistical data about the usage of ai, Purdue University has done an interesting research with computer developers and engineers, they found that many, many engineers, instead of going to places like Stack Overflow, take the easy route on finding solutions for problems that they have with code that they're writing, et cetera, through ChatGPT and similar large language models. But what they found was alarming. They found that 52% of the answers that ChatGPT was giving were incorrect, meaning a little more than half. The answers that we're giving were incorrect. To make it worse, despite the fact that half the questions were answered incorrectly, the results show that 65% of the answers were very comprehensive and addressed all the aspects of the question, which means it gave a very convincing, wrong answer. I. And to prove that the fact that it's very convincing, only 39.3% of the people who participated in this actually noticed that the answer that they got was incorrect. this connects directly to the previous comment that I said, which is, companies have to be aware that their employees are using these kind of tools, and they have to make them aware of the limitations of these tools and to somehow check that the answers that they're getting that become a part of company code and the future product and services of the company are absolutely wrong. In other words, if you're developing code, whether in-house or through a third party that's providing you this as a service and you do not have these guardrails in place, expect some nasty surprises in your next release. The flip side of that that came as interesting news this week, is that stability AI announced a release of stable code, which is an L L M that. Was trained specifically to help in developing code they were using their base model, but they trained it all languages like Python and Go and Java and JavaScript and C and c plus plus and so on, which means they have provided it with robust understanding of computer programming languages with the goal of being an assistant to people both experienced and new in writing code faster and better. So if we combine the last two pieces of news together, we understand that there's a very big gap right now in the usage of these tools to what reality requires. But this gap will probably be closed by better and better models that will be able to provide better, faster, and more accurate code. In the future. I don't know if Stability's AI stable code is there yet. I'm sure it's a huge step forward from the vanilla general models like ChatGPT and Claude-2 as an example. I. The next piece of news, which is probably the most alarming one this week, is that there's more and more news about a platform that is shared and sold for$200 a month as SaaS called Fraud GPT, What fraud G P T does is it allows people with negative intentions to. Very quickly and very effectively have very sophisticated phishing scams, collecting people's personal information, social engineering in order to get data that they otherwise should get malware distribution, and developing different kinds of. MALICIOUS codes that can generate damage to either individual or companies and other fraudulent activities like generating fake invoices and payment requests leading both businesses, individuals to send money to places that they shouldn't. And this is obviously one of the most alarming aspects that are immediate threats coming from these large language models because it enables. A lot more people to get access to very advanced capabilities in those different aspects. Sadly, as of right now, I don't know of a way to counter that. I will say just one thing, which is be a lot more aware of. Who is sending you invoices? Who are you transferring money to? What emails you're opening, what links you are clicking and so on. because the ability to trick you at scale in a very convincing way is already out there. It is just going to get better. The other thing that I say, because this has led to several ransom requests With faking people's videos and voices is to make up a secret word for you and your loved ones that only you guys know. So if somebody calls you and said that it's your child or your cousin, or whoever it is, and they need immediate money because they're in trouble, you can ask them for that secret word. And obviously the AI on the other side That's imitating. The voice of your loved one will not know the secret word, and hence you would know it's a scam. I truly believe that while this sounds extreme, you have nothing to lose and it might save you one day To end on positive news that relates to hackers and hacking. A large hacker's public event was held this week. It's called DEFCON, and it's the Annual Hackers Convention that happens in Vegas every year. But in addition to the regular activities they were competing this year in finding vulnerabilities in AI models from the leading providers such as open AI and stability, AI and meta philanthropic and so on. And this. Process is also supported by the US government and even DARPA is participating in providing resources to both the participants and the winners of the contest. The goal of this is to allow the most advanced hackers in the world to help identify ways where people can maliciously use these models or way these models don't work properly Provide that information to the developers of the models so they can handle those and block them before they're being exploited by the wrong people. See this as a very, Interesting and positive cooperation between government agencies, the companies who are developing the models and the hackers community to develop a safer AI future for all of us. And with that being said, go and explore ai, play with different tools, try things out in a safe way, share with people what you find. Connect with me on LinkedIn, share with me what you find, and until next time, have an incredible week.

People on this episode

Head Shot

Isar Meitis

Host