Stanford Webinar - Identifying AI Opportunities: Strategies for Market Success
I'm excited to introduce our speaker aditia uh so just before we get into the webinar I wanted to just say a little bit about adtia um adtia chapali is a machine learning engineer and product lead he currently works at Microsoft where he helps build gen products that are used by millions glob globally he's also an adviser to several silicon valy startups and Fortune 500 companies aditia lectures on product management and AI both at Stanford School of Engineering and Center for professional development Adida is here as part of a new course as well that just launched on July 29th and if you want to learn more about the course there are links to the course listed here within the platform so without further Ado I'm going to turn it over to to aditia to start our webinar awesome thank you very much Jess uh I'm super excited to be here and the title for today's webinar is identifying AI opportunities like a strategy for Market success but really what I want to do is bust a lot of myths with you that people seem to have about Ai and gen Ai and one of the most common set of questions that we get around gen is how do I personally succeed in gen and AI you know how do I get in and if I'm in there how do I succeed as a business or technical professional and we've done a lot of research on that topic as well and so I'm super excited to share those bits of questions with you and obviously there's a Q&A section where you can post questions and we'll stop very regularly uh to get some of those questions as well as we'll have points throughout this presentation where I'll ask you questions just uh to understand if you are understanding the process so far Okay cool so uh a little bit about the information in this webinar it's taken from the course that Jess mentioned that me and a Stanford professional teach to get the information in this course we spoke to 50 plus Executives uh we did a user research survey 300 plus users and we've put a lot of thought into making sure that these answers are correct that they're valid and sort of backed by a lot of data so let's dive into it so the one of the first things I want to talk about is this curve where I want to talk about a phenomenon with you specifically you can post your guesses in the chat about what you think this phenomenon might be and so this is the I'm going to give you the answer the first version of this answer right now this is the internet so initially people thought it's a bubble where they said you know is it a bubble or is it going to create real value there's a lot of discussion and then people started saying oh we have other priorities we've built an internal Tool uh you know to compensate or to deal with this uh work and then people said oh you know companies started getting disrupted new ones started getting created and then we're here right now in this new process and exactly this is Gen this is basically gen is following the exact same curve process as the internet literally what bit by bit people initially thought it was a bubble and then they had some discussion around it now we're past that with J we know it's clearly going to create a lot of value but a lot of companies now are dealing with it by just building internal tools and not necessarily building user facing products and those people are losing those same people who did that with the internet have lost essentially and the people who took advantage of it disrupted those previous companies and what's more astounding is the internet went on to create a thousand times more value Than People estimated in 1999 so at the peak bubble people thought that you know this is as big as it's going to get and now almost let's say 20 25 years later we have created a thousand times more value than that and so we expect that same curve to happen with ji and so we can use a lot of those same lessons specifically we're still in the early days of the internet so because it's a new platform nobody knows who's going to win and the tech is improving extremely rapidly such that users themselves don't even have solid preferences yet so the stuff we're going to tell you is very early research but we have a strong sense that it's correct but it's super evolving and the most exciting part about all this is the game is yours to capture so I would almost say we're gonna tell you some secrets in this webinar because this data has been let say locked up for people in our course or or you know webinars or executive workshops that we've done and now we're releasing it and people have used these secrets or these myths to create millions of dollars of value and so we're super excited to share that with you today but that's the premise of this webinar so the first thing we're going to dive into is the actual Tech behind gen at this point people are very familiar with what gen can do in general but there's still a few things people get wrong number one note that geni or generative AI is a marketing term it's not an actual technical term generative AI is actually composed or refers to various different types of models like generative adversarial networks or Transformer models or variational Auto encoders basically what that means is if you're a non-technical person like you're a business professional don't go around saying you work in gen we see this happen all the time uh with business professionals especially nobody's a Nobody calls themselves a gen engineer they call themselves an ml engineer er a machine learning engineer so you shouldn't call yourself a gen business professional necessarily you should be calling yourself a machine learning professional or some sort of technical professional in some sense or a ml business professional definitely definitely not a generative AI business professional unless you know you're doing it for marketing sake if you're going to an engineering team just don't use the word gen use the word machine learning okay there's various types of gen applications and I will be using the words and terms gen going forward because obviously we all understand what it means the first type is a collaborative type of generative a application so this is every idea is Amplified and every possibility is explored while it while you work so you have an assistant that works along beside you's personalized gender of AI which everything is created for you so everything will become a market of one and then there's proactive generator doesn't action for you those are the T three types and why those are the three types is because if you project all of these going forward you can see a world where gen surrounds you in everything you do so now you open up Tik Tok or Instagram and you see all these creators in the future that could potentially be all done with generative AI collaborative you potentially have a lot of team members that work with you in the future that could be all generative AI assistants and proactive something is done for you you have to go out and do a list of chores that's all done for you and so there's a bunch of enabling Technologies behind this which we won't dive into like text image audio video but just know that when people talk about generative AI now people think about oh I'm gonna do I'm doing like a writing chatbot or something like this but the most significant sort of advancements are sort of five 10 years ahead of us um cool so the first paradigm shift I want to talk you about with the tech associated with Gen is previously you needed a lot of data and infrastructure and algorithms to build your own model so this is something like a Facebook or a Tik Tok they had all these data centers and all these data scientists to make an amazing recommendation algorithm now those models come fully functional out of the box and people don't realize how drastic of a change this is not only is generative AI really amazing and really smart open Ai and all these people have actually made it public for everybody that's the equivalent of Facebook saying you can use a recommendation algorithm just give us your content and we'll rank it for you and that's amazing so that's a second change that's happened and because of that or uh you know in part of this we see a lot of more opportunities and sort of investment opportunities open up and we're going to talk really quickly about how to access these models and then we're going to talk about those investment opportunities in a second and how you can potentially make a lot of money and how people have potentially used this webinar to sort of really Advance their goals and you know either make money or get into their own careers the first one is to call a model Creator so the first one is you can call something like an open AI or a deep mind with an API the second one is you can host your own open source model and you know there's plenty of Open Source models plenty that arrival the more close sourced versions like llama 3.1 and the amazing part about this is that these resources are democratized like never before you know if you take one thing away from the session take away that most people aren't recognizing the massive paradigm shift that free intelligence is providing to people completely free intelligence that you can host by yourself and that opens up a ton of new investment opportunities which which we're going to talk about in a second but we're going to take a quick pause does anybody have any questions so far uh about the technical section you can post them in the chat and we'll wait a few seconds if you do so ad Dida one question that did come up um in general that we have is do you need to know the technical details generative AI to be a good leader that's awesome yeah so um this is an amazing question which we're actually GNA cover in depth in a few sections so I'm not going to talk about right now because um we are going to talk about it in more detail in the personal section specifically in how you can succeed as a professional in really succeeding in general AI or in this new post AI World um Okay cool so um someone else I don't entirely understand the question of can you add more details what for staging open source model means could you describe that in a little bit if you if you describe that question a little bit more I'm happy to answer it um okay cool uh and for yeang and Yousef we'll get to those two questions in a bit as well and just you can Mark those questions if I don't answer them later for youf and Yan and Allan them at the end cool um let's talk really quickly about some types of gen companies and specifically we want to talk about where the most money is and where people are going to make the most money and so um there's three types of major companies that take advantage of geni the first one was we called takers they just take the model and they put a user interface in front of it this is the best place to start the second one is a model customizer which is a customer that takes a model adds their own data into it and then customizes the model this is the best long-term place to be and then there's a final model Creator who actually goes out and create the models and we're gonna talk in a second about how everybody's wrong about where most of the money is in these cases but first we're going to layer these companies in a second and so um give me a second okay cool the first one is you have to to understand where those companies make a lot where you can make a lot of money with those companies or who's going to make a lot of money we first have to understand the technical stack of bit and the industry stack so just know at the bottom there's compute Hardware these are comput these are companies like Nvidia that make the chips Cloud platforms that sit on top like Google and you know Amazon and and Azure and then on top of those there's people who create the models using Azure and all those data services and then people who take those and then there's people who create those models and make them open source and then people who customize those open source models or sort of take them and put them in front of users that's important to note because I'm going to ask you a question in a second and you can post your answers in the Q&A um of who do you think you should invest your money into and we're going to talk about how most people get this question wrong but we still want your answers so what company would you invest $1 million into mistal AI which is on the left is a model Creator so they make models and they've recently raised a lot of money and they've made Cutting Edge models and they're doing really well an amazing tomson reuter's West law is a legal database as in it holds a bunch of legal cases and who won what and how they won and their results and Associated information if you had a million and they're super old company if you had a million dollars what company would you invest into and we're gonna we're going to wait a few minutes and and get people's questions and get people's answers in the chat okay so maybe I gave it away too much because most people are saying Westlaw and that is correct Westlaw is the one you should invest in and we're g to talk about why in a second uh and the myth is that model creators will make more money actually it seems like it's 50-50 Westra mistol but oh actually it seems like most people are saying mistol and it's actually wrong uh the myth is that model creators will make all of the money and that's wrong in fact the reality is models are becoming customized and so these model creators are actually sort of competing against commoditized competing models and they're sort of they're not struggling to make money but they don't there's not a lot lot of money to be made there in fact um you know we're GNA really talk about how this democratization is making it such that these gen models are providing these really accessible Foundation models to app developers such that these app developers can unlock 10 times more businesses without running any significant investment in their own data science centers and and because of that we're going to talk about how this affects the profit of each layer and so we're going to talk about compute Hardware the hardware folks like Nvidia are super defensible and they're going to make a ton of money with that cloud platforms will probably provide the same sort of profitability profiles as they've previously provided with Cloud so generally so quite high but not insane as compared to actually model app users so these takers and these customizers we were talking about earlier specifically these takers and customizers were finding or actually making the most amount of money and we're gonna spend a few seconds explaining why so um number one the biggest myth is that you these rappers which are these apps that take chat GPT and L large language models and put a user interface in front of them don't make a lot of money so an example for this could be perplexity or uh there's various ones like Julius and things like this that customize chat GPD for data scientists or for researchers or whoever in fact these apps are making a ton of money and this is the this is why chat GPT or these llms are only used regularly by by let's say maximum around a 100 million people a a month that means there's still 8.9 billion people in the world who don't use chat GPT or these llms regularly which also means that making chat GPT or these llms accessible to those 8.9 billion people is extremely lucrative like extremely lucrative opportunity and that's what these rappers are doing and they're making a ton of money doing that that's number one number two people have pointed out that data is the real Oil we're going to talk about in a second how that's not entirely true data is quite significant but what's even better is another type of competitive advantage that large companies have which I won't give away just yet because we're going to ask it is a question in a few seconds but if there's any questions on this we'll answer this at the end of this section but I'm going to talk enough that's enough about the profitability section I'm going to now ask you a question which we just talked which I just said I would ask which is what is the biggest Moe in generative AI so that could be a newer or better model which includes taking a model and making it better with customized data having more gpus having better user experience or having better distribution which means you know accessing a large number of users at once so post your answers quickly in the chat and then we'll talk about um who's right okay nice a lot of good answers coming through okay the split seems seems to be mostly between user experience and distri tion with some people answering other P other places oh sorry somebody asked what is a mode a mode is a competitive Advantage a mode is something that you know allows you to differentiate yourself with with your competitors okay so the answer seem to be split between user experience and distribution mostly with some people saying gpus the people who said distribution are correct so congrats to those folks why that is is because you know people talk about poen data being the new oil that used to be true when you had to make your own ml models and your own AI models it's not it was necessary that you needed a lot of data because you essentially constructed these models from scratch but now that's not true anymore you don't need you don't necessarily need better models or better data because these models are so good out of the box so that's number one data is not necessarily the biggest killer user experience is really valuable if I had to give the second best answer here it would be for sure user experience but it's still not the best because even if you have a subpar user experience if you have access to a large number of users that's actually also fine because our research has found that people will take even a slightly worse user experience if the results help them it doesn't really matter if the user experience has to be amazing as long as the results are somewhat helpful and that's what distribution offers here and so distribution is really the key to winning here and we're going to talk about how that affects the types of companies that are going to win and so when people think about hey making these you know if you're a startup or if you're a large company let's say your startup you know previously potentially the world the way to win would be hey how do I like um you know gain the best model or gain the best data or make the amazing user experience that people always come back and now sort of the game is how do I get this into the hands of everybody as quickly as possible in this new market this rest of this 8.9 billion people so that they can start to use my product that's the that's the new people are way of winning Okay cool so let's gonna let's Bust The Myth versus reality then the previously like we just talked about the biggest advantages used to be talent and infrastructure and model stuff that big tech companies used to to have because you needed that to make your own model now the biggest advantages are distribution like we talked about user experience with a second and then data and so what that means is if we look at the Moes they've been flipped completely with Gen so now the strongest Moe is distribution and the weakest most is algorithm which is amazing so you can see here distribution user experience and then data for the people who guess user experience and dat data and so uh the paradigm shift is you previously needed a lot of money a lot of models and thousands of employees to win and now all you need is just really good uiux some unique data and 10 employees and those three things will enable you to get good distribution okay uh oh well between these two companies the answer is already highlighted here but between these two companies IBM and JP Morgan if you had to guess which company was better positioned to succeed in geni it would be JP Morgan which is kind of interesting because IBM has a lot of talent it has a lot of money it has way more gpus it has way it has all these things but JP Morgan has a wider distribution to actually more end users and then it can potentially build a flywheel with the data and the user experience in that iterative model so JP Morgan is the better position to win in generative Ai and in in this case you're not that I'm giving you any stock advice or investing advice but most of the market doesn't understand this and so most of the market thinks oh IBM is going to win so whenever they make a generative AI announcement or whenever generative AI comes out IBM stock goes up but in fact it should be JP Morgan stock that's going up in fact if anything JP Morgan stock or its equivalents are undervalued compared to their potential for generative AI again this is not any investment advice I'm just telling you that the M this is a mistake a lot of people make Okay cool so that means just to recap that the key AI winners used to be big tech companies because they had a lot of uh money a lot of models these like very specialized employees now the key generative AI winners are actually non-te companies and that's they themselves don't realize this and the mistake that they make is they keep on building internal tools and things that don't really take advantage of this new world and hopefully if you're part of those companies you'll start realizing that you are actually better positioned to succeed than these tech companies because the real values in integrating AI to enhance exist existing systems because you already have a lot of distribution and you can pump generative AI experiences through those so they have a lot of data and they have a large user base and so of these three compan of these four companies let me explain the last question before we go to qu before we I take some questions on who would you think is be best positioned to succeed in generative AI chat base is a company that takes tppt and puts it on a website that's it that's all it does and you can ask questions about the website through that chat bot coher makes models they make amazing models like some of the best in the Enterprise world but they make models Riz GPT allows you to test uh essentially allows you to improve how you talk to other people in some cases it's for dating in some cases it's just to have a conversation with a chat bot uh and perplexity allows you to search and get better results with generative AR so which of these companies you think is best position to succeed and people are asking where's the chat button I believe that's just your I think the Q&A box is the chat button so just post your answers in there of which four uh which of the four companies you think is best position to succeed okay so the answers mostly seem to be split between chat-based perplexity and Riz GPT [Music] um and you know what and most most people are saying perplexity uh so the real value is uh nobody said cohere actually a people are now saying cohere the only wrong answer is coher the rest are all well positioned to succeed because coher and actually we made this slide a while back and now coher is actually laying off employees they're cutting back a lot of deals so you're actually seeing some of these cracks happen because and we're not commenting on a specific company in this case we're just commenting on the general Trend and I this is again not any investment advice but the rest of these companies have these ility to put really good user experiences in front of these models and are getting a lot of distribution in this case let me explain really quickly why all of these could work well Riz GPT got sold for I think millions of dollars chat GPT chat base sorry he's making millions in ARR and perplexities has I think 100 million users or something so all of them are doing quite well I I can't really compare if any of them are doing better than the other I think more people are familiar with perplexity because it's a b Toc product okay so we're gonna pause the end of this section and we're going to take some questions before we go to personal tips uh cool great so first question do you think the current models are good enough I agree you don't need large data but models are not specialized to a specific task and nowadays big companies need to specialize models for a specific task for example due to EU act privacy issues ETC Okay cool so um this okay so this is a good question so number one this is the mistake a lot of people make so they often think that they need to do a lot of like they need to update a lot of their models a lot with their own data to really differentiate themselves and research from our course s basically highlights that that's not necessarily the case if you implement even the most basic gen functionality into your uh app or website or experience somewhat seamlessly that itself has a lot of value let me provide an example there's a company called CarMax that in which you can sell and buy cars and all they did is they didn't do a chat bot that's customized to help you buy cars or take a model and helping you determine between cars or anything like that all they did was they sumarized car reviews that's all they did and it is one of the most helpful generative AI features that's been launched and it's one of the most helpful features that they themselves have launched and it there's no customization there needed and in fact what we found is a lot of companies get stuck in this customization refine tuning process when it's not really necessary the best companies we found succeed when they distribute just a model out of the box find a user scenario that somewhat works and then based on those signals they then start to fine-tune it afterwards and in terms of fine-tuning models being better than General Foundation models uh that is technically true but what we've also found from our research is um in most cases even if you find tuna model with a lot of data the next iteration the next Frontier iteration of the models often beats fine-tuning even the fine tune models for your specific use case and except in really specific Parts where you have really data like Medic in the medical field or legal field um but yeah Okay cool so next question Jess yes so do tech companies like Google Apple that have large distribution networks win as well would they win as well yes so this is again again a key question so this they would win so number one they obviously have a large amount of distribution I work at Microsoft I understand that you know we have access to a billion users you know we when we build products we can immediately publish them to a large audience but this is the advantage that non-te companies have all these tech companies can't train to talk when we talk about you know really specific data all these uh these banking legal healthc care companies and all these places have all this data and access to a set of users that we these tech companies don't have access to necessarily these tech companies don't have banking products or medical products or things like this and so these companies have access to different users who use their products for a different use case and th you can almost think of them as a different set of users have access to a lot more unique data and a different user experience and a different UI so you know these tech companies are positioned to win but these non-te companies are positioned to win like never before um and that's what I would say and in fact if they execute really well like if JP Moran really executes really well they could potentially even perform better in generator VII than a lot of these standard tech companies because gener works best when there's other context involved like banking context or medical context or whatever Jess so um how can um startup companies leverage this and how can they find the right opportunities in the space because we I know you were talking a lot about bigger companies so uh the questions around like startup companies and leverage leveraging yeah so um this is a great question because startups are also extremely well positioned to win in this space again you know people say hey like you know if distribution is how you win then the large companies have everything it takes to win but the real issue here is these large companies aren't moving on this opportunity some tech companies are obviously but especially these non-tech companies they're moving super slowly like extremely slowly and so what you can do as a startup is Target these companies these industries like law Finance health care that have these sort of unique modes or these unique sets of data and build really good experiences or build sort of uh experiences that either help those professionals or help those end companies a company that's gaining a lot of hype is a company called Harvey which does this for in the legal profession where they help lawyers do a sort of customized chat GPT for Law and you know there's plenty of versions of these in healthc care and all these places basically the advantage for startups is there's tons of opportunities for these non tech companies they're not moving at all they're super slow and so as a startup you can quickly move and take advantage of it okay like for example let me let me tell you right now Westlaw has amazing dat data amazing access to a large number of users is doing nothing with g at least nothing substantial Harvey's coming in complete startup and they're sort of completely disrupting the space and they're started by some small 20 year olds like some young 20 year olds Okay cool so we'll move on any other questions we'll answer at the end uh to because we want we have a lot to get through okay cool we're going to talk about personal tips and this is a bit of a jump because so far we've talked about you know where do you succeed as a company where is the most money but I want to take a bit of an aside because this is the most common question we get from professionals how do I succeed what do I need to do and especially we get this from non-technical professionals so we get this from more business-oriented professionals and so I want to talk through you know how we've seen people succeed and the curriculum what we're going to put out has helped people increase their compensation or their success significantly you know I think speak to one person who doubled their income to $500,000 or more than double their income significant increased their income to $500,000 because they sort of went down and took this curriculum quite seriously so you know that let's dive in the first change is non-technical people used to not be able to build because you had to learn how to code and do all the system architecture and all these things now non-technical people can build and I'll explain why and there's different sort of there's a beginner to advanc level of building but let me explain how your job as a non-technical person can change previously let's say you're a PM and you don't necessarily have to be a PM you could be any sort of business professional you can start to act as a data scientist way more than before because now what a data scientist does or an applied scientist is they don't make new models what they do is they give chat GP these llms new data and ask it to you know be fine-tuned essentially and it's as simple as uploading an Excel file and or they tune prompts so they they basically fiddle with the English language to make sure that they get the results that they want and it's an reasonably easy skill to contribute and if you pick it up and you can start going to these engineering teams or going to these technical teams or start doing it yourself you can start saying hey I want to build this app where I have this idea and look chat PT does it super well let me give you an example we worked with a banker who or we didn't necessarily work with this Banker but we worked with this bank and the banker in some Regional Branch put in a customers information to for a loan approval into chat GPT it was a company approved instance and chat GPD provided a great answer as to why this person should get the approval and they took that version and they they they provided the approval to the customer for the loan and then you basically use chat GPT for the rationale and it was so helpful that the company uh sort of approved this use case for everybody to use and now this Banker is actually leading this initiative across the bank on launching generative AI so their sort of career rent from being a regional sort of rank representative to now leading this companywide technical initiative and it's all because they were just fiddling with prompts and there's such a demand from these companies that even some small efforts really are appreciated you know we talk to leaders all the time who say you know our employees don't do enough of this and we want to see more and so another thing is if you're a business professional one of the things you have to go ask is you know hey can is this even feasible you know if I want to build an app how long would it take you know let's like you know how many Engineers would we need and you have to consider all of those estimates now you don't need to do any of that now you can go to chat GP and get the answer yourself you can just say hey like can you do this if so like you know what do you need all these things and it provides the answer for you we'll talk about this in a little bit more technical detail in a second or in a few slides and uh yeah and we so we'll talk about it in a little bit more detail but if you're in a tech adjacent role how you communicate business requirements completely Chang changes now you can come come in with examples of prompts and output at content you can be the leader in how generative AI limits or expands your scenario so you can say hey you know what like you can be the leader in your organization for what gender of AI can or can't do and how feasible it is to get it to do a certain piece of the app then you're GNA research data and privacy requirements beforeand we're going to talk about it in a few seconds how more specifically about how you can do these things specific specifically to advance your career so what can a PM do and in this case a PM is a non-technical person Bally a business professional we went out and asked 50 Executives and product leaders uh basically Business Leaders about how in this new world a product manager or a business professional specifically can expand their respect and legitimacy and grow the organization the first thing is they need to understand the tech in depth and we're going to talk about it in a second what they mean by that and how much Tech you need to understand the second thing is understanding product Vision the third thing is can or does a PR and a PR is a poll request so if you can start to code even a little bit even with prompting that's a earns a lot of legitimacy and then the fifth one is gathering and defending requ or the fourth one is gathering defending requirements we did this survey three years ago ago or we've been doing this survey consistently this year understanding that Tech in depth was by far the biggest job so how important it's it used to not be super important to understand the tech now it's extremely important leaders are really searching for business professionals who understand a lot more of the tech uh and you know that's becoming increasingly important and so and it's in fact the most important thing so how do the best ml PM so the ml business professionals grow and succeed obviously the first advice was join a fast growing company the second advice is become technical then it's become a domain expert and then it's build more side projects okay so basically there's three paths to get into generative AI get technical get Niche domain experience and ideally you do both of the three the most common question then get is you know which one should I do should I go out and get domain experience like learn about a specific industry or should I get technical and most often the advice we give for our business professionals and the advice we see works is get technical and let me spend a few minutes explaining why that is most business professionals already understand a little bit about their domain a little bit about the industry like even if you're an accountant let's say you understand how accounting works you don't necessarily understand how a specific industry works but you understand how accounting works if you're in manufacturing obviously you understand how manufacturing works and so what we find uh is that a lot of people opt for getting more deeper domain expertise and number one when we see get deeper domain expertise here it's still with a mix of generative AI so you still have to say hey I I I know manufacturing really well but I know manufacturing the context of AI so I can tell you really specifically where AI is the most helpful what use cases have been tried before have not been tried where has AI been successful where has AI not been successful where are some Primal from from initial customers I can get to try out some initial use cases for generative AI in in a specific industry and because a lot of people opt for this domain expertise route it's actually a really competitive because some people have been in this industry for 30 40 years and they know so much they know everybody in this space it's really hard to get competitive you have to know an extreme level of knowledge on the technical side a lot of business professionals shy away from it they say hey I don't know how to code I don't really understand that world and we find that if those people try to get technical and they get more technical they sort of become unicorns because they are one of the few in the world who can bridge the gap between the technical side and the industry side and I'm going to explain in a second what get technical means because there's there's beginner intermediate and advanced but the path that is more valuable we often find is get technical the ideal is you have whatever 10 20 years of domain expertise and you get super technical then you're a unicorn you know these are the types of people that we've seen make $500,000 a million dollars in a few instances because startups and Industry leaders and companies pay so much money for this advice and I can explain what examples of advice we mean by that but okay let's go talk about how to get technical so there's three stages like I said beginner intermediate and advanced if you're in this session You' already probably passed the beginner section of how to get technical because in specifically related to generative AI you just have to understand what are the rough tools what are the rough implications what can it roughly do at a high level and that's pretty much it the intermediate advances really where most of the money starts to come in uh for a lot of these people especially at the advanced level so at an intermediate level what we mean by how to get technical is you get extremely good at prompting and this takes about four to six weeks we've found and this is like number one the basic level is you know understanding how to use chat PT at a pretty deep Lev level intermediate is using system level prompting and multi- prompt examples so really understanding how to coach it to get a set of results that are really valuable and you you and regular and consistently what you do with Advanced is you get achieve consistent results with multi-channel prompting we'll explain what that means in a second but uh think using things like Json formatting and Checker llms let me give you an example like let's say the llm like Chach PT you're say you're using doesn't give you the right result the first time you say hey can you give me a a a workout uh regime and it gives you a like a sort of wrong workout regime you have another llm check the response with the specific things you're looking for like saying hey I told them it had back pain I had back pain is it recommending squats to me because if so it's a bad workout program so you know redo it so you have a checker llm set up that checks that for LM and knowing how to understand Json formatting and structured outputs from opening and all these places and doing things like Chain of Thought which is like prompting llms to talk about how they get their reasoning right and you know we won't talk about it in more detail but that's the advanced stage and if you really want to understand how these companies are making so much money this is where most of it is and especially in these rapper ones and so this is where if you also learn this you you would also make you would also start to access those sort of Worlds the second one that's important to understand is data boundary and Associated systems architecture so you know really quickly and we won't spend too much time on this because we have so many other things to cover this is just understanding how you can call gender of AI in your own company or personal boundaries for example we work with so many companies who think that they can't use gender of AI or they can't use these closed models from open Ai and all these places because Oh They'll break the EU act or they'll break these privacy acts and that's just not true all of these Cloud companies if they work on cloud run these models Within These companies data boundaries they just have to call it within their data boundary API but it's astounding how many companies don't know this like you know I think we let's say if we work like 99% of the people we work with don't know this which is you know astounding and so they go through all of this infrastructure sure to host their own model but they just need to know this so if you and the few business professionals who know this provide so much value because they're saving millions and thus they're rewarded with an Associated compensation package so just knowing data boundary and Associated systems architecture how to call generative AI in your own beta boundaries understanding data flow diagrams understanding Storage storage systems is really helpful Advanced is really where we get to those really big numbers so there's two paths you can take if you're a big or or worker or we find this also works with technical adjacent workers dive into systems architecture really deeply so what you want to be able to do is you need to practice communicating these technical Concepts to non-technical and Technical people really deeply and what that means is you need to understand like what is you know eup what is eup like what is personal information what are things like caching what are things like stuff like this so you can dive into systems architecture you don't necessarily have to code and where to put in generative AI super valuable these people are paid an incredible amount of money for solo workers or non- Tech adjacent folks maybe you're not in the discussions with those systems Engineers all these people these people make a lot of money by using low code or no code tools with gender VII capabilities to either automate parts of their jobs or start to build so some of these people we've seen learn to build apps without ever actually learning how to code they don't even understand what chat GPT outputs they just prompt chat GPT or these llms to give them code or give them an app and copy paste that code into you know an app or some sort of browser and have it work and so these people also make a lot of money because you know we've seen for example a nurse who made a patient intake tool not by learning how to code by just prompting chat gbt to make that app and then sort of that app made a significant amount of money okay so we're going to take a quick pause on that personal section uh just any questions yeah so we'll do one question here because I know we're getting closer toward the time of the top of the hour so we'll do one question and then move forward so we have time a little bit at the end for some more Q&A so one question is I see we're getting a lot of questions from people with specific problems they want to solve with generative AI so what would you recommend as some good first steps and how should they evaluate the opportunity and decide about the best next steps for their pain points and ideas you know what this is great because it works well into our later sections of the most common mistakes people make and you know how you should go about evaluating what is really a good idea or bad idea so this works well so let's Dive Right In um so the SE second section should should answer this you know what really quickly we're going to cover how generative AI models business models are different and this is the first thing people sort of forget we're going to go through it really quickly um before we talk about how to evaluate gender of idas because number one most people the thing people get wrong is they think of gender of AI like a typical subscription model where they say hey there's a typical subscription price there's a total gen that it's a typical subscription price and typically software would always operate within that sort of subscription price like if you charge $ eight a user you typically wouldn't incur a cost every time somebody uses that new product but with generative AI that cost goes up with usage and so that means and it often looks like this where generative AI goes up exponentially as usage goes up so often a lot of SAS or softwares or service business models don't really end up working out so that's why you'll see companies potentially like perplexity that actually lose a lot of money because even though they charge users for their usage especially if you go to the Premium plan they quickly start to break that cost period pretty soon so just remember that as you start to investigate these ideas know that a lot of the dollar per seit uh or dollar per person or dollar per month business models are breaking with G of AI okay so we're going to talk about now answering the question prioritizing generative AI opportunities how should you think about uh how which generative AI product or project to pick to pick and we're going to give you some some quick tidbits so number one let me give you three of the most common options people come to us with build an internal facing tool because they want to build up the expertise and capabilities while helping increase efficiency internally a user facing feature or buying a gen chatbot software or some other software in white lab it most people build the internal facing tool and they're wrong to do it what you should be be building always is a user facing feature almost never an internal tool you know it's just not recommended and why that and we're going to talk really quickly about how common this problem is you know we did an analysis of the people who've implemented generative AI product products and almost everybody either does a chatbot or just internal tool 95% of people do an uh chatbot or internal tool so that's 60 plus 35% and when we ask people you know how useful is your internal tool 75% said not useful 25% said somewhat useful so basically 20 95% of people don't find it useful so if you're building an internal chatbot don't do it or internal tool don't do it it doesn't help you build the skills that you need to build the skills that you need let's just walk through this really quickly what you should do is keep whatever feature you're doing in private preview avoid chat Bots or free form interactions because it allows people to put in inappropriate information and and potentially unethical pieces and you get into a lot of legal trouble if you give the wrong answer always do in product features and just test regularly and what does allows you to do is it allows you to build the capabilities you need to build great user facing features and the research from our course highlights that companies who focus on user facing featur so that's number one um but the basically don't build an internal tool if you're picking a gender to a opportunity almost anything or a chat box also doesn't really work super well though both of those are pretty tough to iterate and work with but if you focus on spec almost any other problem it works out well and we're going to quickly talk about how to implement those really quickly and then we'll take some questions um the thing that most people get wrong is when they go out and create these generative AI products they create products for creation so they'll say hey a product that helps people write emails and essays within just a few prompts is what people love versus a product that helps people read documents and emails easily and summarizes key Concepts but research from our course shows that users love the second option and they hate the first option so when you're building a tool or a generative AI feature prioritize content creation or sorry do not prioritize content content creation that's not recommended what you should prioritize is content consumption which means making it easier for people to consume content and understand things better so and why we think that people often do the content creation side is because the first wave of generative AI products was this sort of naive back and forth talking to chat GPT you know can I get an answer can I get this could I get you know could you answer this question for me a request response setup but users are getting tired of that in fact what people want now is they want to take a lot of information and just take the key insights out of it through gener generative AI in fact research from our course shows that content consumption is on average useful to very useful on a scale of one to four it's three and a half on how useful it was content creation was just a one or two so people love content consumption for gener of AI they hate content creation so let me give you some examples like instead of creating a legal document or helping having lawyers create legal documents you should analyze historical case outcomes to advise on legal strategy so instead of creating personalized documents or or proposals for potential clients you should leverage market trends and sort of refine sales pitches and you know instead of generating product specs or generating product emails aggregate user feed back and tell product people like what they should be building next we won't go through too many of these examples but there's they're here for you you know some of the best generative AI examples we've seen companies that work really well are companies that listen to sales calls or customer service calls and tell people how they can perform better which is really exciting versus for example creating a customer service agent okay so um we won't walk through more we won't walk through all these examples the final tip we're the second final tip is don't build a chatbot instead build AI features natively into a product users hate building it using or working with a chatbot you know most people aren't willing to use a chatbot almost 99% of people are not willing or only somewhat willing to use a chatbot and if they do 80% find it not useful and this is the modern version of gen chatbot versus if you build a gen feature into a product 90% are pretty excited to use that product and almost 55% find it useful so this is you know 10 times more useful for users Okay cool so we definitely answer that question in a little Speedy format because we want to leave time for some questions at the end so Jess yeah so I think because we are a minute away from ending at the top of the hour um unfortunately we don't have the time right now for some more questions but um is there any tips or takeaways that you want to share before we end the webinar with folks uh nope I think that's that's if if you want any if you want any of your questions answer just message me on LinkedIn and I'm happy to answer them there just one onone great well thank you so much for joining us everyone thank you aditia um and just real quick as a reminder our new course Adida has launched and if you're interested in learning more please refer to the link here um on the platform and feel free also to reach out to our team via email and we also have some other generative AI courses that are launching and we can help you choose from some of those courses as well so as mentioned the recording of the webinar will be sent to everyone within the week thank you so much everyone have a wonderful rest of your day
2024-08-31 21:00