What You Need to Know to Prepare For GPT-5

all right so today's video is a bit different from what I usually upload this is a 20-minute lecture that I held at the Google offices in Madrid so what I ultimately cover here is what I see as the future of products like chat GPT right now in other words what do I expect from GPT 5 and onward and not just that I also talk about what you can do about it today so you're prepared and you can get an advantage once these new and revolutionary tools come out and I start out on this journey by discussing the different features that were added to CET GPT over the course of the last 14 months I did this because the target audience was very mixed here it was people from all levels and only about 15 to 20% of the audience had Chad GPT plus so I wanted to catch them up on what happened and what is going to happen in my opinion and then we also talk about Gemini and the one use case that I found there right away that is just better than most other large language models right now I'm really excited about uploading this because this talk is a fantastic summary of everything that I covered on the YouTube channel over the last year and it's a fantastic way to catch up but also it's a great opportunity for some of you Advanced users to think about where this is going next and how you can prepare for that I also want to thank mindstone for organizing this talk and inviting me as a speaker they do a fantastic job at organizing these real life AI meetups across Europe and the US and with all that being said enjoy my little keynote on what I expect from the future of CET GPT 5 and what you can do about that

today I'm going to show you what you should be doing today if you actually want to get more out of these tools okay that's why we're going to arrive plus I'm actually going to address two pieces of news that are quite massive throughout the last week one of them is what open ey is looking at with which direction they're going in and secondly is actually the Google Gemini release and how to actually use that to your advantage yourself okay so before we do that we need to take a step back and talk about this product development that has been happening right under our eyes over the course of the last few years okay and I'm talking about chat GPT here so what I want to do is I want to briefly go through the history of how the product has developed ever since the release end of November 2022 okay because when you look at the different phases in isolation it can be interesting and it can be a good topic to talk about but rarely do people really connect the dots that have led us to the point where we are today and where we are heading because it does clearly in my eyes um but I do spend all my time kind of looking at this and talking about this and using this in my eyes and my team's eyes everybody agrees that the tra trajectory that this is in and the products that we have here like cat GPT Gemini and llama and all these tech companies what they're doing they're heading in a clear Direction and before revealing it I kind of want to talk about the different points that led up to today okay so first things first we had you know GPT 1 2 and three that were not accessible for a userfriendly interface like this most people in here including me probably weren't using that on a regular basis you had to have a very specific prompting style most of the results weren't that useful with gpt3 it kind of did cross a certain limit where it became very useful for certain use cases but only with the introduction of GPT 3. 5 and this interface that most of you probably know and have accounted and used and at the very least tried um only this really made it possible right having a little chat box where you can actually leverage and use your social skills that all of you have by the way even the introverted people here you can understand me right so you have some social skills you understand language and only by having this interface where you can actually talk to it like a human kind of that unlocked a lot okay so this was the very first release that really made a lot of people realize what kind of power Heights here okay before we move on I have one question for a show of hands how many people here have plus how many people have plus today subscribe to plus it's usually it's not most it's about like even at these events it's about 20% of the people which is interesting because that says a lot it says that you're not getting as much value out of it today as $20 at worth a month right so and I think that's fair I don't want to overstate the capabilities here okay I'm going to show you where this is heading as we said we're already heading there and then I'm going to show you what to do about it and how to kind of like get ahead of most people in terms of what you can do with this because then the upgrade came right went from 3. 5 to four and as many of you know that kind of use this regularly massive leap in terms of quality right like literally for writing and for Creative tasks and reasoning um I personally like the creative capabilities the most massive leap a real Improvement okay in quality consistency less hallucinations better results big leap okay then we went ahead and added multiple features here over time so as you might know when we switch over to four we have capabilities now like uploading files including images you can give it image and it's going to recognize the image now what's the point of that it recognizes your images that's the point what's the point of the chat box that I addressed before it takes your language what's the point of the code interpreter that's built in the data analysis tool it takes your data all of this is about taking stuff from your life the expression that you have through your words the documents the Excel sheets the files the images everything that is in your life on your phone on your computer and you're able to interact with that for now that interaction has to be initiated by yourself through so-called prompt right as we just talked about in Lang chain The Prompt is one part of the pipeline right but a very important part because that is the one where you speak stuff into existence right with the prompt you tell it that hey get me this result now that can take many forms we won't get into that but all of these different input methods like adding data adding images are just different ways to prompt now I personally am a big fan of the image input because it just makes it so simple to kind of communicate right it's just a few weeks ago I had this accounting use case and from the European Central Bank I had like this table of conversion rates between euro and dollar and yeah just transferring that myself would have took like I don't know 15 20 minutes annoying task right doing it for 12 months so all I did is I just took a screenshot of it all threw it into GPT and I was like turn this into a CSV it did it just copied the CSV and used it so simple to prompt right way simpler than having to like explain o take every field one by one and then add you know one column is going to be conversion rate second this it would take a while um so all of this is making it simpler for you to use it okay and here we get to kind of like the end of this like product road map feature right there's few more things that were added one of them was custom instructions and I think this is by most people the number one overlooked feature in here okay gpts we'll talk about that in a second are just the evolution of this but this is really where you personal it so as I was talking about when you upload your own data when you prompt it in your own way for your own use cases for your own life you make it do what you want what you need right like what's the what's a good prompt I always say this when I teach chat GPT like what's a good prompt well it's one that gets you a result that you're happy with that's it right it doesn't need to be over complicated if you're happy with the result it's good custom instructions help you really dial in who you are and what you need okay so this is the next level of personalization because this is persistent the cross time this personalizes the product it's kind of a baby step towards fine-tuning a model which we won't talk about here but this really tells it I mean look in my case you know I teach AI I have a YouTube channel all that and it knows about it so if I just flip this little switch every single answer here is going to be customized to my own needs beautiful thing it also works in 3. 5 so this is what I'm trying to get to having a set of instructions like this that tells it just a little bit about yourself defining these few variables like profession specific interests current projects that you're working on really helps and the cool thing about that it doesn't just help now it helps like in this product it helps in Gemini it helps in Lama it helps in anything that's going to come out okay because custom instructions really personalize it to your own needs okay and what are gpts you might have heard about them they're kind of a combination of all that okay I forgot to mention one thing which is plugins and that is where we enter brand new territory it's not you speaking stuff into existence that you want to make happen Okay plugins are apps like canva giving out what they do and making it simpler for you so when you tell it hey create me a PowerPoint presentation canva can go ahead and like arrange the things for you know it's primitive I'm not saying it's great right now but it's the worst it's ever going to be so when we bring all that together that's where we get gpts you get all the capabilities of GPT 4 you get image inputs you get um code interpreter with data analysis you get custom instructions you get um plugins in the form of actions and you get to attach your files as a knowledge base you get

to attach PDS databases and all of a sudden you created a specialized worker and this is the way I want you guys to think about this because really what this is an assistant but as soon as you enter kind of GPT territory it's more of a worker it's more of an employee so to say that is doing a specialized thing okay so this is a general purpose model that does a lot okay a GPT does one thing ideally you set it up to do one thing you give it the documents that it needs to do the thing you maybe give it an action to perform that one action that one thing you give it the instructions to teach it about the context of the task it has and then it can do that can it do it very well to be honest not really in most cases people will agree that today uh when using these models there's a lot of exceptions with use cases and stuff but today in most use cases it's an assistant right it doesn't it's not a finished product it's not a polished thing that you copy paste and use but for rewriting for data analysis many more examples can maybe even get the whole task done right like I'm not a data analyst with the release of the feature today I am I know how to ask it what can I do with this data and then it just gives me a list and then I know how to ask one or two more times on that list of questions then I ask it those questions and I can do stuff that you know one year ago I wasn't able to do um no background in that but all of a sudden pulling all the YouTube analytics and asking and getting to insights that I wouldn't have arrived at a year ago simple as that

okay so if pts bring it all together then what is important here okay now that I gave you this little like overview of kind of the product and the different things that are in here what is important what does all of this have in common and how can you actually use this to get better results than most other people the answer to that is learn more about yourself and how to communicate that info about yourself that matters to these language models okay all the tech companies that you interact with smartphones computers they have a lot of data on you matter of fact in certain ways they know you better than you know yourself you know your transaction histories your browsing histories you could name what you browsed two weeks ago on a Tuesday at 4 p. m. right they know exactly so I would encourage you to actually do a little bit of self-exploration and find out what matters about yourself so when you use this Tech you know how to communicate that because you can only communicate stuff that you understand right everybody that teaches knows this it's really like the only way to communicate something properly is understanding it from multiple angles and and communicating it concisely so if you manage to do that you will arrive at something like this and I'm here to help you I'm actually going to give you a preset and this framework um that you can use okay I came up with this I gave this out for free it's on my YouTube channel there's guides and everything but basically this is just a set of 12 variables profession role current projects specific interests values and principles and a few more and it might seem obvious right but if you're like putting this together would you consider adding your learning style your personal background maybe not but trust me it is helpful in certain use cases right this is crafted to be super Universal so you can drop some of these you can add some others but I'm just advocating for the fact that you should have a profile like this because whatever you do today with llama GPT Gemini Palm whatever you're using whatever you're going to be doing in a year from today this is going to help you and this is universal so whatever chatbot You're Building whatever use case you're looking at this stuff helps okay so what did we do well we went ahead and we first tried to craft a system that like kind of helps you generate this for yourself turns out to be a little tricky and especially for non-technical technical people not that easy if you have like some skills with writing it becomes way easier fun fact is that philosophers are really good at this real philosophers are excellent prompt Engineers that's kind of like one funny thing that I found they're just like good at understanding things at like multiple levels and that turns out to be really helpful here but for anybody you know who's not a philosopher and anybody who's not a professional writer this little framework um works really well but if you don't want to write this then what I did I went ahead and I created this little preset that I'm going to give you a q QR code tool in a second here so basically you know if you ever check out one of the over 100 YouTube videos that I have there you're going to find that I reference this all the time because it's what I base a lot of my learning on top of and it's just the set of custom instructions okay and these have been customized to these different roles okay so our best attempt that kind of like finding what works for you is just giving you like different ones right different starting points and then you can go ahead and mix and match these um now you can delete some you can add some but here's the one part and this is what chat GPT should know about you right it's his first field and then there's the second part is how should it response this is all about the formatting of the response so you know if you prefer it to answer in Spanish then you have a point in here in the second part which is language so this one for the recruitment specialist is set up to be clear language is favored use HR specific terms there you go so you could change that right you could just go ahead and say always answer in Spanish and then guess what if you're going to do that I just have to watch out because otherwise everybody's going to receive a Spanish recruitment manager if I leave it like this so I'm going to reset it okay and if you add this in here okay always answer in Spanish guess what I'm going to say right me an essay about penguins There You Go Spanish answer so it's a little things like this right and you can see how a few words just change everything so you can just go ahead and copy paste this into your notes and have it with you whenever you use a language model um you're going to have this at your disposal whenever you feel like the answer is not personalized enough to you which matter of fact in like 80% of the use cases for most people that's the case you could have better answers that are hyper personalized to you this is what a lot of companies find by the way too this is just a little bit tangential but I do want to say that a lot of people that I talk to a lot of entrepreneurs the number one use case they're looking at is hyper personalization right so like hyper personalizing messaging right the number one scares that you see Sam Alman talking about in interviews is actually people going ahead um and manipulating people for political campaigns with AI right because you can have robocalls that have like analyze your psychological profile and Hyper personalized messages that go directly to you all of a sudden if you're not expecting it you can get fooled right because it's somebody directly targeting your interests now that's the views part of it but you should be on the other side of it you should be using this for yourself to your advantage right and you can do that by customizing these little custom instructions so we won't be going into like every single one of these and you can actually be quite loose you don't need Advanced Training to actually get this right just go ahead and look at that and ask yourself hey this does this apply to me please cross check any proposed strategies against best practices in Talent acquisition this one's pretty un right if you work in HR you're probably going to want this and that's how these have been crafted so there's more there I'm not going to go into all that some people like tell me hey ego are from marketing standpoints it's not so smart to give the people so much just give them one thing I'm like nope you're going to get all the assistants for each one of these assistants you're going to get a prompt generator so for example for the recruitment specialist and this is what you can do when you have your profile okay for the recruitment specialist I'm going to turn this off I'm going to go into a new chat and whenever you ask yourself like hm what can a recruitment specialist do with chat GPT What could they do right do a lot you can start googling you can look on Twitter or you just paste this prompt run it and it's going to give you 30 use cases that you could run today guess what these are personalized on this little profile at the bottom okay so you have these 24 fields that we just talked about 12 and based on these this whole like prompt that I wrote like it's a lot we're not going to get into it but basically it generates prompt templates for you so as you can see enhance interview techniques optimize Talent sourcing Elevate employment branding and 30 more this works in all language models this is a simple thing that I just gave you for free you can go ahead and use this all it takes is customizing this and it's going to generate use cases for you okay so rather than teaching you how to prompt I'm just telling you like read through this you know make yourself a nice tea put your feet up and like read through this and make sure this fits your needs and then you have the prompt generator that's going to find you all the AI use cases that you could be using today okay so you don't just have a personalized profile you also have 30 use cases and I could rerun this it's going to give you 30 different ones pretty powerful stuff and this works across everything okay so we talked about you know the history of the product now where is this leading prompts talk results into existence right but the results I mean it's not omnip potent right it can't do everything so the next evolution of this if I said that gpts are kind of a they're kind of a worker employee then the next evolution of this is this employee getting more capable GPT 5 with the new Gemini model whatever is going to come out it's they're only going to become better from here on outright one two you're going to have multiple of them work together you're not just going to have a web designer and a data analyst you're also going to add a project manager and maybe six more team members and the project manager is going to oversee all those and all of a sudden you're going to have six of these working together solving a more complex task so that's the direction this is going in now how can you use that to your advantage well you need to first of all figure out what is your task how to describe that in words have a little profile like this figure out the use cases for yourself and then when it gets to that stage it's going to be simple it's like oh great I always wanted to work together with a web designer a copywriter and a growth hacker nice let me take the presets let me set up free gpts I have mine over here as we just talked about because you customize it for yourself to kind of have a way of explaining what you do what your preferences are and then going to let them work together and all Su you're going to be working with six seven AI assistants and a project manager in a version of this product that might come out this year next year we don't know so that is where it's said where it's heading I hope this was useful I want to end on one note and that is um we were talking about chat GPT here these are pretty universals um okay so you can have these custom instructions prompt Factory there's a bunch of like prompts and use cases down here that you can explore Styles and down here I have an organizer there's a little video that explains it all if you want um I think it can be found here how to use but basically um I want to leave you with this so two pieces of news I opened with this and I'm going to close with this so the first piece of news is that you know Sam Alman the founder of openi has been exploring deals first of all in chips you know that ridiculous 7 trillion race thing that has been going around sure but more interestingly that they have been looking at building so-called what people called large action models okay so this is all about doing things with language taking language input and also it has you know a p python sandbox and it can certainly do calculations and stuff but what we see with these gpts if you look at them and if you build one of these which literally takes minutes if you have the plus plan you're going to see that one of the fields in there is actions it's very clunky create new you need a specific open ey schema they have a own GPT that does that then you need to host a API endpoint it's not great what this should be it should be a login with Google log in with apple thing and like yeah give it access to my calendar yep creating events give it access to my Gmail I want this little worker to manage all my email drafts like why is this not generating I get I don't know 60 emails a day why don't I have 60 drafts sitting there every single day right and why is that not drafted by the com combination of a customer support agent and maybe a teacher and an instructor and a salesperson depending on what the email needs right why because the product is not there yet but it will be soon and I told you how to prepare for that okay so this is going to evolve in into you know in my opinion into like drop downs and then you're going to let have multiple workers work together and the second piece of news that I wanted to cover um so this was the first one like that open ey is rumored to build like these action models that are going to be based all around the actions not just about the instructions and the second piece of news is actually Gemini coming out now it lacks certain features in here but in my research over the last few days I actually found one use case where it's better and we just talked about this in the beginning it's probably one of the bigger ones okay and I'm creating a bigger YouTube video on it and everything but the one thing that it's really better at is analyzing your own style and then writing in it that's one of the big ones right if you're doing emails if you're doing anything it's really good at kind of like looking at the emails you wrote before extrapolating that style and then reapplying it to new text and it doesn't have that chat GPT stink as I would say you know there's always this way of talking this like these over ad a lot of adjectives as you said then it also always sets up like a question and then answers it like enthusiastically there's just these patterns that like people pick up on by now um so that's the one thing I would recommend in Gemini and everything that I showed you with custom instructions it's Universal if you're using llama if you're using Gemini if you're using chat GPT 3. 54 it doesn't matter you need to find a way to describe what you do to a language model that's going to make your life easier today and in the future that's it thank