Building AI LLM Apps with LangChain (and more?) - LIVE STREAM

foreign what is happening guys how are y'all doing today good to be back on the live stream I know it's been a little bit of time but there has been so much happening in the world of AI and machine learning literally been drowning and stuff in terms of keeping up and and building stuff but I figured why not just uh do a live stream with you guys because I love you guys any we can go and check out some of this technology uh together there's been things like baby AGI gpt4 um what else has there been Auto GPT we've been seeing stuff coming out there was like a new model released by meta yesterday kind of insane that that it feels like there's a new model literally every couple of days there's some big new announcement and it's it's absolutely insane so um hopefully we can go mess around I'm gonna try to get back into the live stream process every couple of days ideally I want to do it every day but that that's a big ass given um I'm literally going to be flying out to Southeast Asia next week and then again the week after and it's just a it'll be crazy but we got a bunch of people here so let me say hello how you doing Saddam how you doing Abby and nodo how you doing Antonius I'm very good hope you're doing well yourself didn't realize it's 11 A.M in India but that kind of makes sense it's 3 28 p.m here in Sydney at the moment and I got like a little bit of free time so I figured why not let's get this stream running srida how you doing morning in India that is absolutely amazing Kevin what is happening dude glad to have you here and work day so definitely gonna miss the live oh no shattered that's right I'm gonna keep doing I'm gonna get back into the live streaming a lot more so um and this will go up straight away afterwards as per usual so don't stress if you don't get to catch the entire live stream it's going to be up straight away afterwards I normally just bang it right up so everyone can go and check it out how you doing karisha how you doing Shields 128 here in Trinidad oh that's awesome what's happening 128 you need to be getting some good rest to be able to go and do some machine learning stuff 728 in Germany coffee ready let's go how you doing Conrad how you do how you doing YX Nagi 7am in North Africa yeah where else are you guys from I don't how many people are on the stream now oh damn 36 okay we uh didn't realize how many people tuned into these but what is happening guys loved my party look all right it's a backstory so yeah I'll show you the photo of the screen in a sec but like that story behind that that um the Pokemon outfit if you haven't gone and seen it go and check it out in the community post so there's this forum that our marketing team decided that they were going to run so specifically a data Ai and machine learning Forum purely um around like IBM Tech and so the whole there's a whole bunch of stuff around it but one of the biggest things and why I was I was so excited was it was the first ever event that I was emceeing so I was like running the entire keynote slash opening and they're like Nick how do we spice this up and so I'm like well yeah let's get an animated suit and somehow this devolved it like started off as like a data and code suit and then it went to like a Pokemon suit and I had the mullet rolling around as you know uh and yeah so that that's that's what we ended up with funnily enough it was like so tight I could barely move so I went to squat and like it started tearing so we didn't end up going with it but um that's the backstory behind uh the Pokemon suit um hopefully you liked my party look might have to bring it out we'll do some doctor disrespect Stars I just got to get like a bulletproof vest and then we can run with that um 229 in Japan how you doing 11 A.M in India hi Nick what's your opinion on x dot AI we should go check that out because like I haven't gone and followed up with it like there's been so many things coming out so far so like my goal in these sessions from now on is ideally to show you what I'm learning as as and when I'm learning it because I've literally got a list it must be like 60 items long that I want to get up to speed on like I want to learn a new programming language I want to pick up rust I've been watching primes live streams and I absolutely adore them so I want to go and pick up a new programming language and for us looks like a pretty interesting one um you left 30 but like what time in Chicago 6 30 a.m in England awesome your neighbor country Jakarta Indonesia I'm actually coming to Indonesia soon coming to Jakarta soon at 12 30 p.m in Vietnam I'm going to be coming into Vietnam soon as well that's awesome awesome to have you guys here alrighty cool well I reckon uh let's get into some stuff and let's take a look at what we're going to take a look at today so I'm literally gonna you can follow along with all the stuff that that we're getting up to and um ideally it helps he helps keep you up to speed with some of the the machine learning stuff that's coming out thick and fast because I literally cannot keep up with that without doing a little bit of coding and stuff every single day um can I share the roadmap for becoming a data scientist it would be a great help you know what let's do that first let's go and take a look at what I think people should be learning to become a data scientist Alejandro 10 30 p.m in San Jose California awesome what is happening all right let's jump to the other screen transition now alrighty cool so I'm gonna do a little bit of stuff on Excalibur draw so we can start mapping some stuff out now keep in mind that being a data scientist there is like so so much and there's so many different types of types of data scientists right like there's people that focus on specific niches it's like a spider web on my camera it's people that focus on specific niches specific Industries so like if you just have a good basis in data analysis and some of the stuff that I'm going to show you you will effectively be in a really good spot so if we take a look at like I'm gonna rather than show you a road map I'm just gonna paint a picture of what I think a data scientist should have the data scientists and if I zoom in on that a little bit so what do we think so typically if we focus on the foundations and then we'll do a little bit of python type stuff so in terms of the mathematical foundations you need linear algebra you need uh stats statistics and probability and what was the other thing that you need calculus as well it's good to have an understanding of these three so like whilst you're probably not going to use them all day in day out it's it's super useful to at least understand how they all work and where they all fit in like the cool thing about calculus now even though like it's a fundamental underpinning framework when it comes to building machine learning algorithms building deep learning algorithms a large majority of it is automated for you inside of Pi torch and inside of uh tensorflow Auto grad helps you do that as well Jax is another Library that's helping you do that um if that's something that you'd like to see let me know in the comments maybe we can dive in um so inside a linear algebra ideally you want to understand what a scalar is what a vector is and what a matrix is you want to understand different transformations scaling you also want to know matrix multiplication and then there's like a bunch of stuff that that is useful you probably won't see it as often as you think but we've got Matrix decompositions or effectively Vector decomposition so typically you're going to see eigen D comp and this is effectively like the equivalent of like factorizing a matrix or a vector so that's the way the best way that I can explain it so we're extracting out similar components across a specific uh plane or a specific line within a a dimension all right so eigen decomposition you should also take a look at SVD so those those couple of things so these are foundations right so these are not things that that will be I I it's probably not going to be the most practical thing but understanding these is going to give you a foundation for what comes next um so stats and probes then you're probably going to want to go and take a look uh I'm just thinking of my memory part so understand uh different types of data so binary categorical you might also want to take a look at continuous data or numerical and that's going to break out you might also have a discrete that's continuous you also want to take a look at sampling super important hypothesis testing you want to understand the difference between parametric and non-parametric data and parametric and non-parametric I know I'm going through this fast but we can we can always come back um so what do we have so binary sampling hypothesis testing you also want to understand that the idea of a random variable and different distributions So Random variables and distributions so if we sort of stem out of that so we'll know that a discrete variable will take its value from a a specific distribution and we can effectively extract a probability prob Mass function and if we have a continuous variable we're going to get a probability density function out of that probability density function one of like my favorite blogs to go and read for a lot of this stuff is actually machine learning Mastery it's like it's so detailed in in terms of what it goes through but it is also hyper practical so understanding the stuff is super useful right so all of this types of linear addressed out some probabilities to calculus you want to understand like fundamental um you typically want to learn like derivatives are all about derivative calculus right and you want to understand basic derivatives you also want to learn the product rule uh the quotient rule and the chain rule what else do we need to know from calculus I think that's pretty good for now there is more stuff when you get into like matrices and and dealing with linear algebra but if you understand the fundamentals around calculus and how we use the gradients to and specifically the gradient descent algorithm to find a global Minima to be able to minimize our loss that's going to give you a good idea all right so those are fundamentals now let's in terms of getting into practicals right so you ideally Want To Learn Python then in terms of Frameworks or that my top Frameworks scikit learn you also want to know ideally if you want to get into deep learning you're probably going to want pi torch or tensorflow and then what else so the other ones that I like are stats models so this is super good if you want to do um like ordinary least squares regression and uh like logistic regression and if you've ever done any any coding in uh the cool thing about stats models is it gives you like a similar style interface so you can literally go um OLS past three x variable pass through your y variable and then you literally go and run dot fit and then dot summary and it will literally output the uh like an almost like R like summary when it comes to actually building your regression models um so that's that's pretty much it right so like that this is the fundamental math around it then you ideally want to go and get Python and get yourself familiar with some of these Frameworks and then I personally think it's really important to have a good grounding in storytelling and presentation because you want to be able to communicate your ideas so a big part of what I do is like using these fundamental concepts but explaining it in a way that anyone can understand which is why this stuff is so so important okay so I know we we digress but hopefully that was a little bit useful so it shows you the um what I think is probably a good grounding in terms of getting up to speed with data science status becoming a data scientist looking at Deep learning stuff any questions around that is there a big difference in the roadmap for a machine learning engineer it's like very very similar right um so a data scientist I feel is a lot of the time is going oh my aircon just turned off uh let me jump back over to the other camera so I think a lot of the time it's going to be very similar machine learning engine is going to be very much concerned with building the algorithm so like let's say you've got a deep learning framework it's going to be around building the algorithm structuring this getting it ready for deployment and and making sure that it's going to work when you get it out there inside of applications but the lines are very blurred when it comes to becoming a machine learning engineer versus becoming a data scientist like let's actually go and take a look at some um some job listings so we can actually go on ahead and see because what you'll see is that it is very very similar I think machine learning engineer might have a little bit more of a deployment Focus but it's still going to be really close um that that's in response to your question Jordan let's actually go and take a look because I always talk about it but I probably haven't shown it so we want to take a look there that's storytelling and presentation we are going to get into Lang chain I promise um so if we actually go into like LinkedIn right and let's type in machine learning engineer let me zoom in on that so you can see it right machine learning engineer what are we going to get out of this let's check out all the job results all right so senior machine learning engineer at this company whatever that is all right so what are what are they asking for in terms of what you'll do leave the design development integration and support sorry key key point right design development integration integration typically means deployment and getting this out there so you can build it into other applications and support a machine learning based sensor processing capabilities this typically means you're probably going to be dealing with like iot based sensors to build machine learning models typically around fault analysis screw effectively screw-up analysis um develop software so again a little bit more of a software engineering Focus but again whenever we're talking about building full-blown software this this can blur the lines between machine learning engineer and software engineer as well um automation so this is again ml Ops getting stuff repeatedly out there and into production uh manage model selection that still could be done by a data scientist training data source and integration well data sourcing can also be a data Engineering Process right now necessarily machine learning Engineering Process integration integration could be um well that could be like ml Ops but again that I typically say integration falling into the realm of machine learning engineering but keep in mind you've got like wizard machine learning Engineers that cover all the uh wizard data scientists that will do all of this machine learning Engineers that will do all of this including the Eda including um doing the analysis what I think people typically associate as the difference between a machine learning engineer and a data scientist is a machine learning engineer typically sits more so on the I.T side of things whereas a data scientist is going to be working a little bit more closely with line of business users they're going to be talking to line of business and finding out about stuff that's happening things that may be influencing the data that's that's typically what I find um required qualifications you probably need every single qualification Under the Sun um interesting that they're asking for C plus plus go python or rust like interesting I mean I think everyone's probably going to have different requirements regardless of where you go my headphone have a loop in that can anyone code if you've got like a loop in your mic or like headphone line it drives me insane um all right cool so like you can sort of see like if we go to Deloitte they've got a job job ad see this is uh this is machine is this machine learning engineer no this is senior consultant um see I mean you still get the Blurred boundaries right lead data scientists slash lead machine learning engineer lead machine learning engineer let's just take a look at this one what's this asking for um excellent programming fundamentals yeah that's fine you probably get the same again so take a look so this is asking for Azure Cloud native tooling so it's asking for Dev and ml Ops again that's deployment integration type stuff again Docker Docker compose Docker swarm kubernetes you probably wouldn't be expecting a data scientist to have a full-blown full stack understanding of those particular type of Technologies although again when whenever I sort of preach this stuff the the stronger your skill sets in all of this the more appealing you're going to be to a potential employer so super important to know okay cool that's enough I think on taking a look at the foundation or the the roadmap but I think a lot of people ask for that so instead throw it in there okay um all right let's take a look [Music] had both mics on my bad um what we're going to do now is have a look at Lang train so I've been playing around with it I've been playing around with Chad GPT as well gpt4 got an interesting video coming up soon um but for now let's go and take a look at Lang chain we'll go and do read through some documentation yeah I had two mics on my bad I had the one on this camera and then the one on here is it is it gone it should be gone this is concerning Gone Gone yes thumbs up thumbs up in the chat if it's gone you can tell I'm a beginner stream okay I hope it's gone anyway it's not God involved this is going to be interesting in the replay 145 in the Philippines I'm in the Philippines next week I'm so pumped I've never been there before Yoni can we someday soon have a class on GPT for all yeah maybe we'll do that tomorrow TF or open AI what do I like and why two very very different things so tensorflow is a deep learning framework open AI is like um open AI is obviously a company that builds software and specifically machine learning algorithms but that they are two very very different things so tensorflow is going to be a deep learning framework that allows you to build models and algorithms from scratch very similar to Pi torch open AI is pre-built stuff so you're typically not building stuff up from scratch inside of open AI that being said a lot of developers are picking up on it now because if we actually take a look at I just saw someone asking about the forza machine learning project I feel like I'm taking on new projects and still picking those up but um I feel like I love to developers uh picking up open a because it's really easy to use like um like if you take a look at baby AGI I saw I saw I'm just a kid I didn't know that no that's cool man it's good this is why we're here to answer these questions um it's it's really about like building stuff and open AI is really about building stuff on top of their Frameworks right so using their embeddings using their llm models like if you want to do image generation using Dali 2 so on and so forth ml Trader I'm going to give you a sneak peek as to what I'm doing with ML Trader because there's stuff happening right now it's so sick um but yeah that that's those are sort of the differences so I'd recommend learning a deep learning framework but also be aware that apis but they're a little bit easier to pick up um open Ai and all that stuff how cool does this look this is just my water bottle but anyway uh okay let's go and do some stuff I don't know very much okay it's going to be back sorry I'm sorry all right what are we doing so how you doing Kevin long time no see okay so what are we gonna do let's um we've hijacked my uh my what's called my whiteboard so to do a game plan so what I want to learn today is I wanted to so where's my to-do list we needed to go and check in on the YouTube video that I'm currently making so YouTube you'll get a sneak peek hopefully you guys will like the behind the scenes we also got to take a look at some of the stuff that we need to build for the course and we also got to do some study and so inside of study we are going to be doing some what did I say well we're gonna do uh Lane chain here and I've also been doing um Andrei karpathy's uh llm like build a GPT from scratch video like I've been going through it like I got home and literally powered out two hours of coding last night because it's a little bit stressed so that helped me I code when I'm stressed I don't know why that helps reduce my stress uh but that that's what I did yesterday and I want to take a look at that as well what are we doing so Lang train we're also going to take a look at GPT from scratch from Andre and the rej of course uh YouTube all right let me show you what's happening with the YouTube video that I'm building right now so if I go into Quant connect so I'm building a well I use chat GPT to build a trading bot so I go and I deployed it yesterday it's only paper trading at the moment but I know everyone's been asking about ml Trader I want to get back to it as well I think it's running here so geeky fluorescent pink keeper models so this is running I think I've placed a bunch of Trades while I was sleeping yesterday I'm gonna have to come back because it's getting super hot in here so I'll be right back promise foreign I'm back it's literally 29 degrees at my desk right now my fan is saying it's 29 I am burning up um what was I going to say so we'll come back to the chat yeah this is actually running live so it's it's it's built a trading bot if we go into live it's uh probably not doing very well at the moment but I'm gonna I'm making it be like a huge yeah I need to get some elevator music um still setting up the live stream so sometimes this opens sometimes it doesn't live view come on buddy here we go no it's not gonna do it maybe huh it's a bit of a pain anyway this is running live I'm not going to show you the performance of it but it was effectively a live trading algorithm that was built purely based on gpt4 um which I'm not gonna spoil the surprise you're gonna have to go and watch a video eventually once I finally put it all together and work it out but it's running live and it was actually doing trading I've had to do really really small amount to to be able to get this up and running it's it's kind of crazy what's possible there like I did some minor tweaks because I wanted it but um you'll effectively be able to see what had this performing a little bit later reaches 45 degrees in summer in India I know but this is like literally at my desk it's 29 degrees like sitting right here like you can see my face is sweating I'm burning up um but yeah this is what's coming soon so that's gonna be the YouTube video so we've got a like I've started planning for it actually guys rate my thumbnail let me know what you think of this I think it's in here trading Bot video what do you reckon about that hey as if that's not awesome sweet right uh anyway okay cool so that's us taking a look so uh I think we're fine there that that's all happening we're happy with this we're gonna come back to that all right course we'll talk about that later all right link chain let's go and take a look at Lane chain could have sworn that I started coding some of this already the city YouTube dear where are we on link chain think I had it in my llms photo it's a CD 15. just activate that a limbs back with Slash Scripts backward slash activate go to double check make sure that I don't have my API key Expo so we're gonna have to do that in a sec Jupiter lab YouTube is telling me now it's a good time to place ads here but I'm not going to and I'm probably should do something around that um I know I don't have link chain here okay all right I guess we're gonna have to play with Lane chain from the get-go so let's oh no I've got link chain it's there let me just double check I do have an API key in there so let's just uh let's just not expose that API key dot Pi I'm gonna I'll bring it back in a sec API key I'm just setting this up so it's not exposed and I will set this to from API key import API cool all right that's not exposed okay let's cut let's do this so uh we can close that close that close that sorry name chain let's actually go and take a look at some link chain doco Lang Shane it's so cool how it works because like I I went into it and I talk about it in a sec but like just the way that this works and like how it's using Vector databases to pass context I didn't even realize you could do that um when I went into Auto GPT and baby AGI like I'll show you where that actually comes in um but basically all right let's let's have a read here nope I'm zooming in on that foreign chain is a framework for developing applications powered by language models everyone loves the language model hands up if you like a language model we believe that the most powerful and different differentiated applications will not only call out to language models via an API but will also be data aware connect a language model to other sources of data so so important right because if you don't have connections to other data it's not going to be able to give you appropriate context very very important be agentic is that a real word agentic apparently so comparative more that behaves like an agent okay don't need to learn something every day well that's why we're here to learn allow a language model to interact with its environment super important because otherwise kind of pointless right the Lang chain framework is designed with the above principles in mind this is a python specific portion of the documentation for a purely conceptual guide to Lang chain C here for Java what is what's here yeah that doesn't give us more there are two main value props the lane chain framework provides components users but alright cool interesting so we got the idea so they're basically gives you the ability to have data associated with your llm and be agentic I don't get the agentic bit but we will dig into that all right so what are people doing with it on Twitter um that's how I get a good idea as to so if I type in Lang chain what's happening here so Lang chain learn about length chain one plug creating my first AI agent so many ads what are people saving to notion reads annual reports and that answers queries via the tools super interesting how did this guy do it all right so he annual report is forwarding queries to a lang annual report tool so he's got an annual report tool and that's going to a lang chain also he's saying I don't have he doesn't have access to GPT for I do we should go play around with that as well in a sec um okay so like the the the association to or what's this saying so baby Ajo by default baby AGI just executes things with an Alm response by changing the execution chain to be an agent with tools yeah this is the big thing right like you need to be able to go and interact with other stuff so do we need to build the integration to other stuff so agents what is it tools toolkit python guide yeah all right so I actually did this I I'll show you what I played around when I when I tried it first I actually used it to go and do exploratory data analysis very meta right so I used Lang chain to go and say uh what are the columns in my data frame which is kind of interesting because if you wanted to do like basically build a virtual data scientist you kind of could with Lang chain you could go and pass it a CSV and go hey do stuff with it and it will just do it but like I still think that these um like they're still more needed right like jira is a really good one because if you need to raise a juror ticket but like exactly as that guy said so maybe this is a project that we need to do after I finish the trading video but like imagine you could just go hey tell me whether to buy or sell um a specific Company's stock like tell me if I should buy or sell Tesla right now tell me if I should um rebalance my stock portfolio tell me if I should like is now a good time to watch And subscribe to Nick's YouTube channel yes uh but like that's the type of thing that that I think is is super interesting right so let's actually go and take a look at this so to install I had to run a bunch of stuff so exclamation mark pip install Lang chain open AI chroma DB and tick token so if we go and run that uh not jira tickets you can see it actually has been triggered all right while that's installing because it looks like it's going to take a while let's have a chat no Echo so like there's a button here I don't know if literally like if I don't hit that button then that means that this this microphone stays on which gives the echo so thankfully I just went in here that this is still installing you can see it right there there alrighty cool let's check the chat so what's happening here James Briggs yeah loves Nick's videos I had to jump in thanks for jumping um to learn mldl from scratch is this much maths that I've discussed adequate I think it's a good start like I'm still learning every single day so like if there's anything that you should take away from this like it's going to be a journey like I can show you my um so like I probably told you guys before that like I learned using memory paths to like memorize stuff but like you can literally see like this is all the math like it just keeps going like and I'm continuously adding to this list of stuff that that I want to be able to learn like I was memorizing the formula for like a gaussian distribution just so I know it um because like when you take a look at things at maximum like maximum likelihood estimation it's going to be critical to be able to understand how that framework fits in so it's a really good start it's probably going to get you like 80 of the way there there's always going to be more stuff but if you get that 80 so much easier to keep going forward how you doing David Bell what's happening your Nick live coding and working is such a blessing it's just showing people that the real works more faults and errors and even people of specific knowledge IQ are still human yeah like like I look up to so many absolutely amazing Engineers but like I feel like sometimes it's a bit stressful when you see these people just like build from scratch and build like super fast without errors and you're like what how on Earth like it it it's good to see what that level of performance and eliteness is because it gives you something to Aspire to but I think it's always useful to show your own process as well which is why like if I screw up I show you like because I don't want you to get this warped notion that like the developers out there are all perfect they all get it right completely every single time that they know how to build everything absolutely perfectly like I'll memorize Frameworks and like build mental associations to stuff so that I can remember it and I don't have to come back to it but every single time I have to go and build a data loader in tensorflow I'm like oh how do I go and process strings like I know it's it's something that I always forget how to do like but there's always going to be stuff that that you need to go and refine and that that's just what development is like you not everyone's perfect and and take your time it's a learning process you're going to enjoy it I think you use track GPT for Generations yeah I've been playing around yeah you can use chat jpt for Generation but check jpt is also uh tuned on instruction-based data sets that means that we can pass it in set of instructions and it will give us a set of responses what country do I live in I'm based in Australia baby AGI versus Auto GPT I've only played around what have I played around with what's this one I think this is baby AGI what is it I mean it doesn't really say it just says create Panda's data frame agent we'll go and take a look at what's being used behind the scenes but I think it is um hey when will you be coding something live maybe related to I think that's towards James Sprigg uh definitely a robot advisor would be fire yeah we're gonna be awesome we need more likes on this stream only 16. yeah I mean if you want to drop me a like feel free um what else am I ever going to finish the forza live stream I want to get back to it I just I'd like I need to get some time so um but yeah I really should get back to that uh hey Nick what's up after a long time you're on YouTube live I know it's been taking a while but um I'm back not bad no audio again getting used to it the audio should be back on let's start again so we are importing OS so import OS the next thing that we're doing is we are then bringing our API keys from API key import API key and then what we've got is we've got our we're setting our API key am I still muted it shouldn't be muted I know there's a bit of delay between this back yeah cool awesome all right so then to set our API key so we can run OS dot Environ and then go and set our open API key I think my sweat is clogging the mic I think my sweat is clogging my mind at the moment but uh we're gonna power through okay so then what we can do is this is where the agent spits comes in so from langchain.agents import create Panda's data frame agent so this is presumably using this pandas data frame agent so you're able to go and do exploratory data analysis using our pandas data frame agent I don't know it does does what is this used behind the scenes is it using its own this agent calls the python agent under the hood which executes the so this isn't specifically using a um like a like a what's called like a baby AGI or like if I go what's the python agent here LM agent yeah so this must be using its own thing is it up prompts inside of it I don't know it doesn't say we can dig into that a little bit later all right so basically the cool thing about this is that we can use it to do exploratory data analysis with the pandas data frame agent without having to go and do it ourselves so we're importing our own llm so from langchain.lm's import open AI from langchain.agents import create pandas underscore data frame underscore agent and then we're importing pandas to actually do our Panda's data analysis now I think I got rid of my um data set that was in here so let's go grab one so let's just go jump onto kaggle and grab a data set let me sign in [Music] I like spaceship Titanic so let's grab that one and let's grab the data let's grab train so this just a data set it's like the um the original Titanic data set but it's spaceship data set all right so we've got our train data set started dying on my water there um so I've downloaded it so it's inside of here so we just want to grab that and go into my D drive YouTube and then llms and I'm just going to drop it there I'll share all this code afterwards so you guys can pick this up I mean it's not much code that I wrote but I'll give it to you anyway um so you can see that I've got my trained I can't that's not zooming in but I've got that train file there so now if we go and take a look inside of our folder you can see that we've got train.csv so that means we can go and import it up now let me show you what this would look like without having um without having Lane chain right so let's say for example I wanted to go and calculate some correlation on a bunch of data sets Okay first up let's just add in a couple of cells here minimize this let's call it Lane chain link chain all right let's Lane Chain's gonna pick up there all right so let's paint a story so what would this look like without an llm so we'd actually first up reading our data set and I have an imported Panda so let's run that now audible okay cool make sure you shout at me when my mic is muted because otherwise I'm not gonna know well I'll probably end up coding for like an hour and we'll have no audio all right cool so that first up that's our data frame now loader right so I can go and do exploratory data analysis on it so you can see I've got my big data frame I can type in df.head so let's say for example I wanted to get the columns that I've got inside of this specific data frame I could type in df.columbs that will give me my set of columns right now watch what this would look like if we were to do it with Lang chain so first up what we would do is create an agent so we'd specify a new variable called agent and we'd set that equal to create underscore pandas underscore data frame underscore agent we then go and pass through an open AI instance and we'd set our temperature to zero I've got to dig into like I remember messing around with temperature a while ago but I really got to dig into that a little bit more if anyone has a good good definition let me know in the chat and then what we're doing is we're actually passing through that data frame there and we're spreading verbose equal to true so if I go and run this now and then I can then go and run a bunch of effectively natural language based commands against that data frame so do you start to see how we're blending data with our llm so rather than just using a raw um rather than just writing the code ourselves it's effectively going to do it for us and it's going to structure the thoughts and actions so let me buy this so let's actually go and say um what columns are available in the data frame right so you can see it's going to kick off a chain and keep in mind this is using the open AI API in the background so it is going to be consuming API tokens and take a look so it steps us through this so we've got a thought so I need to find out what columns are available in the data frame so that's our initial objective and then it actually goes and creates a number of actions so first the action input is we go and run df.columns and it actually outputs that out and then in terms of the thought process I now know the columns in the data frame so it gives us our final answer the columns in the data let me zoom out columns in the data frame are passenger ID home planet cryosleep cabin destination my head's covering that age VIP room service food court shopping mall Spa VR deck name and transported pretty cool right so like that effectively allows us to go and leverage temperatures similar to Randomness with temperatures 0.0 interesting Thanks James um so that effectively allows us to go and explore the data frame without actually having to go and do like right out our line so over here I had to go and write df.columns here I just had to say what columns do I have available let's say for example I wanted to go and calculate correlation I could type in df.core assuming if I've got missing values what are we getting could not convert string let's just grab um df.select D types and then include oh actually let's go exclude equals object and then run dot core all right so that gives us our correlation right if we did this and if we went and took a looks at anything which is let's say in the high negatives on the high positives we would assume that is high positive or negative correlation so food court has a point uh so correlation of 0.13 against age dropping more pretty low against Spa 0.12 so a little bit higher VR deck 0.1 so not too bad looks like there's stronger correlation between the food court and the spa so if you spend more at the spa you'd probably spend more at the food court and vice versa likewise with the VR deck um what else so that probably means that if you're spending money at like all these ancillary Services you're probably like splashing cash getting out there foreign James has got an answer for us on temperature the temperatures similar to Randomness with temperature equal to 0.0 is meaning that means that the model will always be the most probable next word increasing temperature means less probable okay so it's effectively how strict we're being so whether or not we're sticking to the prompt or not okay so super interesting typically more temperature means more creative and messy outputs so for code you probably want temperature equal to 0.0 creative writing 1.0 super interesting Thanks James you're my boy all right cool so we've got Lang chain so all right so we just went and calculated correlation right is dying uh so df.select D types allows us to go and do that analysis but we could also go is there any strong correlation between columns using agent dot run and we should effectively get a new agent executor chain which goes and does the same so if I went and ran this let's go and see so the thought or effectively the objective I should look at the correlation between columns this is saying that we need to use python all right take a look super interesting so this is going through the same thought process that we did right so the action input is going to say use df.core but it's getting a response back which is saying could not convert the string to float exactly the same as the response that we got the next thought is I should remove the non-numeric columns exactly what I did right so I went up here and I said I want to select D types it's gone and done a little bit differently but it's gone and said to select it I've said select e-types and exclude anything that's an object down here it's going the opposite it's going I should remove non-numeric so it's actually gone and identified the non-numeric columns to passenger ID home planet destination and name and it's excluding those then running the dot core method on it which should effectively give us the correlation but see here it's run into another error could not convert string to float so it looks like it's missed one of the columns which still had strings so take a look it should have effectively then gone and added that particular column which I believe is cabin to the the chain so you can see it's gone and added there so IBG pt4 by adding by just selecting the appropriate details but no no like I'm kidding this is ridiculously amazing right like so it's able to go go through that thought process to go and generate a response but keep in mind we've said give us columns that had high correlation so there would be one more Associated step to this tell us which of those columns have highest correlation not just return correlation to take a look I know now know the final answer final answer there is weak correlation between some of the columns right such as cryosleep and transported age and transported and VIP and VIP and trans autumn there is a weak correlation between some of the columns such as cryosleep and transported agent transported and VIP and transported right there you go so it's actually going to given us some of the correlation and you can see it's finished the train and given us our response pretty cool right so let's say that you like you wanted to empower citizen data scientists or citizen analysts this is the type of thing that you'd actually be able to go on ahead and do like give me another one like what do you want me to test out so I could go agent dot run um and say uh let's give it a trickier one so build me a machine learning model super meta we're using we're using llms to build our own machine learning models so build me a machine learning model that um predicts Weather or Not of the predicts the column transported my head's covering that transported right so this should effectively go and try to build a machine learning model that predicts whether or not the passenger has been transported or not so be interesting to see how whether or not this works so let's give it a crack God my open AI bill is going to be brutal this month okay so it's creating a train test split it is what's it doing this is going way faster than I can process the colors are not very good on the eyes it's not right look at the core between VIP and transport interesting was it crap VIP and transporter 0.12 is still okay I would say that's weak correlation what did we get VIP and transported is VIP a numeric column hold up DF oh we've still got our executive training happening down here all right so that's finished I said this ad actually hasn't created this hasn't returned could we ask it to save the model the final answer is to create a machine learning model that um but I go and save the can we chain on a separate step to this like if I said what would this be could I just then go and say agent dot run save the model as a pickle file I'm going to come back to your question John the data frame is oh so this just saved the data frame it hasn't actually saved the model interesting okay so so it doesn't maintain context and maintain context between the data frame but save the machine learning model that was trained was trained as a pickle file I don't know will that work no it's not picking it up it's just saving the data frame it is saving it though kind of cool so it's actually executing the code I guess we could do that there's still a little bit of context that needs to happen here so I reckon if I had gone and said save like added it onto your save the model as a equal file and ran that line how would that go look at the core between VMP and transport I'll come back to that hit it save that looks promising modeled up equal so what would model would have been logistic regression so you can see it went and trained a logistic regression model and then down here I went and saved it as model.pickle so if we go into our folder now we've got a model.p equal that's right there pretty sick all right John your question so but it's not right look at the call between VIP and transported so it said what did it actually say there is wheat correlation between VIP and transported transported how is it saying that is it are we missing there is an alignment here it's it's looking a bit Dodge what data type was VIP df.d types VIP was an object yeah so it didn't drop VIP hold up what is VIP is a binary so could we not just go as a type int it must have dropped missing values that's the only way I could have done that um so if we go through here somewhere along the line it would have dropped missing nope core yeah how did I calculate the correlation with that how did it right if I run this am I going to get uh what's called or unless it's just ignoring okay so it's different between mine so I've gone and excluded anything which is currently classes and object this is actually just excluding certain columns that is still factoring in the columns which are objects but could potentially be binary so if we actually go and take a look the correlation between v i between VIP and transported that is weak like interest did a pickup cry asleep and transported prior sleep and transporter it did interesting okay so I did pick that up okay well I mean it clearly works like it was just gone and done it a different way um and I've been a little bit lazy in excluded object that the true false is fine replace with one zero yeah cool all right interesting so that at least gives you a little bit of an idea as a as to what is possible with Lang chain I wonder what else we can do with this like what other agents are available what agents are there what's the difference between an agent and agent executor take agents and tools and use the agent to decide which tools to call and in what order thank you nope let's go back to agents the various tools what are the tools multi oh okay here we go oh wow we've got a ton Wikipedia API let's go check that one out oh sick let's try this so from langtrain.utilities import Wiki API wrapper testing it out let's um let's try building a streamlit app completely unplanned but are we doing for time yeah we're okay uh what are we doing with this let's let's jump back into the chat and then we're gonna we're gonna try building some stuff the coding is great but the interpretation is dodgy fascinating yeah I think I mean this is kind of something that you need to be mindful of like whenever you're using large language models right like if at if you keep it keep in mind that there's a chance that stuff is not always going to be perfectly accurate then it definitely helps you be a lot safer which is a little bit of a pain because it means that you always need to double check and sort of question your results so I think it's good but again you always need to be very mindful of that our alarms are not great at numerical understanding however combining Wolfram has shown some amazing results I reckon let's go check out the Wolfram one let's try building a um a what's called like a stream lit up with that yes please definitely make a stream lit up all right I guess that's what we're doing next I'm really enjoying this live stream uh what else is gpt4 training set also available for download I don't think it is like um I did a bunch of search it's not actually available in that many spaces so uh just be mindful of that uh Lorenzo hey buddy how you doing what else no audio thankfully I solved that one what did you think of that I mean let me know in the chat like I thought it's pretty cool but like obviously I think my example kind of sucks because it's like why would you use an llm to do data science like if I am a data scientist although super amazing if you don't have the data science skill sets because you can definitely go and pick it up but I think using some of the other tool sets which I just realized there are makes a lot more sense that being said I'm gonna give myself a little bit of a plug guys if you do if you're not like a hardcore coder and you do want to get up to speed I don't know if you've checked out but like I've probably dropped it in the chat or in the community post a bunch of times you can actually go or like I've set up a link and a page now which will actually take you and give you a free link to be able to go and do my free python course so if you actually go I should probably have text on the page but if you actually go to go.courses from Nick forward slash I'm going to show it on the screen soon forward slash I think it's free python forward slash python you can actually go and pick up um that free python course which I'm going to add to so like you know how I showed up some of the stuff that I think is important when it comes to doing data science so like some of the more Matthew type things I'm going to be adding some of that to the python course so it won't just be a pure python course it is continuously evolving um to be more and it also helps support the Channel having you guys on that that email list because then I can interact with you a little bit more closely so I'll show that on the screen now and then let's go build a stream letter yeah cartoon Chan what is happening doesn't matter better late than never so uh what do I want to train yeah so if you wanted to go and do the free python course you can go to go.courses from Nick forward slash Python and pick that up um I am doing a lot more like businessy type stuff oh I know for 150k sub pretty sick right like um I thought I think that uh it's all because of you guys like like the reason that we're at 150k is because of you guys so um thank you so much thank thanks for calling that out and again if you want to go and do Learn Python for free by all means go and jump onto that go to courses from nick.com forward slash python plus I'm also making it some T-shirts that I'm going to be doing live giveaways with why is this light turned off there we go cool got the blue light back all right cool um what are we doing we're gonna go do some let's go build a streamlined up and see where this goes okay so what do we need so we need to open up an environment so I'm gonna go into my D drive and we're going to go see the YouTube CD uh what do I need to go into dear also how cool like I personally think that that God of course is from so I'll give you a little bit of a top secret so if you actually go and subscribe to this or if you jump in it'll give you a free course to actually go and do the free python course at courses from nick.com which is something that I spent a ton of time setting up um and I'll actually show you it let me let me jump in so even though it says 267 like someone reached out to me and they're like oh my God Nick it's 267 dollars no this is for free just make sure you go through this link and it'll give you a free call uh code to be able to go and do this like you do not need to pay the 267 for the tech fundamentals course which covers like setting up Jupiter doing python don't pay the 267 please guys this I want to give this away to you guys for free um it covers like crud so like that was the friend like create read update delete conditions and logic I've got a bunch more coming it just doesn't show because I haven't gone and uploaded the videos but let me actually show it to you in the back end I don't know have any of you guys seen this before I probably haven't turned it to you so if I go in and show you the course nope oh these are the other projects in the in the full stack mL of course but to Tech fundamentals let me show you foreign yeah so like that there's more stuff that I got planned so I had working with numpy powering pandas I'm gonna add like a statistic section a math section maybe maybe a little bit of calculus you let me know what you want um but it covers things like you end up building like a z-scoring function handling standard deviate or calculating standard deviation calculating mean from scratch um going through a bunch of Loops so again if you want to check this out completely free just make sure you sign up at this link so you don't pay the 267 please don't do that I've set this up for it to be free for you guys and plus I actually give you a bunch of data science advice in a bunch of emails as well so did I build the courses from Nick website from scratch I thought about it but I didn't so it's actually um a bunch of websites combined together the combining together I did which was a a little bit of a pain to say the least but this is powered by thinkific so no I didn't build that I thought about it but I was like actually started building it then I don't have time to manage it plus eventually I like one day I want to go and build my own like course platform that I can use to share stuff with you guys but that that's further along down the track um okay what are we doing now we're gonna do some uh some streamlit stuff let's go do it all right so we are going to CD into our llms folder let's activate that environment llms scripts someone's calling me let's decline that Dash activate um probably call them back a little bit later and I add PCA MDS and cross validation two to the python course or to what Naruto let me know um all right cool what are we doing here so we want to build a um have in mind blanks we're going to install streamline it so pip install streamlit let that install grab a little more water also what do you guys like do you like long live streams or like short ones because like in the past like I've always tried to keep them like under an hour but I don't know do you like longer ones PTA MDA and cross Valley I'm not sure what the MDS is but um we might be I might actually add the PCO on because that might be good for like uh linear algebra great answer maybe I'll add that I'll I've got to do a little bit of work on it anyway um so llm so we are inside of llm so I'm just going to type in code dot to open this up inside of vs code hopefully it doesn't show my API key nope it's not going to close that close that okay so we are now going into our let's create a new file we're going to call that app.pi and we are going to import streamlit as St and then we can run it by running streamlit run app.pi one hour you like one hour ones under an hour the long live streams under an hour [Music] how you doing Maxine glad you're a new schooler one hour seems fine longest feels like a podcast yeah multi-dimensional scaling maybe yeah I don't mind unless it's useful content are you are you guys finding this useful let me know if uh boronia we can always uh mix it up okay so we've got our streamlined app running but we need our Lang chain notes let's just Chuck that in there I know we might finish this tomorrow but are we doing for time how long we've been going for now uh an hour and six minutes probably wrap it up soon and finish this tomorrow I'm gonna live stream tomorrow as well okay so we're importing streamlit what do we want to do so where was that documentation nope nope nope I wanted to do the Wikipedia one yeah okay so what's it saying what do we need to do um so we need to install Wikipedia so let's just uh get a new thing so we're going to run pip install Wikipedia and then what to from Lang chain for utilities import Wiki PDR API wrapper okay hmm and then let's make this a bit bigger so you guys can see let's toggle word wrap all right so that's our wrapper we can then run Wikipedia equals Wikipedia API wrapper where does the llm bit come into this so this is just returning this notebook goes over how to use the Wikipedia component okay I'm not sure I get what's happening here so this is just returning text Wikipedia dot run enter I don't know let's run uh Lewis Hamilton what's my get my GitHub I haven't uploaded this to GitHub but if you go to github.com forward slash a knick knock knock it'll be there what's happening and then I can go to St dot text let's pass through the text there so would that work uh where's that up rerun okay so this is just returning text how do we make this work with with our large language model so I could let's let's add a text input so input equals St dot text input input type your prompt yeah right okay then if input we would then go and do this stuff we should probably take that out throw it over here and then what we're going to say is if we have an input then we're going to run the Wikipedia agent from Lang chain and then render that back to the screen yes right so we can type our prompt actually I should pass this input back over to here this is completely unplanned I'm literally building this on the Fly Guy so uh bound to be a few errors uh so we can type we should be able to type in a prompt here say um who were the parties in the Battle of uh I was watching 300 I'm spelling that wrong for sure Monopoly how do you let's Google how to spell that Monopoly there we go yeah I knew I spelled that wrong all right so this good run okay so it's returning what are we getting here page summary so the Battle of Thermopylae the engagement occur wow that's uh returning a big block of text okay this is just returning text though so what do we need to do past this is context to Lang chain like this isn't all that magical where what are we doing wrong the tools okay so presumably this needs to get used by the agent okay let's Google this so this is could you combine it with yeah I think that's the whole idea I'm just not very clear how to do it inside of inside of this so would I initialize it inside of so this is getting started there's got to be some better documentation so Lang chain uh Wikipedia agent there's got to be some example and he has to Google stuff hell yeah man that's all I do that's hilarious um all right so this is an example of GPT plus Wolfram Alpha so I like the RPG and a agent Stanford like wait a RPG is in like uh like role-player game agents how you doing now Lewis explains what's happening no I want to see the code behind this I don't wanna this is weird um show me an example example do we have examples okay hold on there's this is just taking us back to the same thing hmm how to combine agents and Vector stores no that's not what I want text loader we're gonna have to dig into this this uh retrieval so wait there's different types of Agents so if we think about this broadly right so there's Bunches of Agents there's tools there's agents there's tool kits so the tools so the toolkits allow you to do processing on these types of data the agents are actually what we're using and the tool sets are basically the stuff that we can integrate into we need to use custom tools what are we building right now we were trying to build a Wikipedia bot tools are depend okay hold on here here the tools that append and then we initialize our agent with tools okay hold up so I think that's it I like that I read it for like 30 seconds and I'm like yeah no that's fine that's pretty much how my development goes like yep that'd be right let's just wing it okay so Wikipedia is our current tool set yes yes I hope that should mean that we're creating tools and then what is tool or class of do we need to import this from yeah okay here we go so um how do we do it with this let me just go back to this initial code go away okay so we're importing the create penders data frame agent great pandas data frame agent so are there other agents that we could just use are they just using a generic agent no we don't need this don't need this let's get rid of the stuff that we don't need so we are using initialize agent to initialize agent does that mean that can take in our tools let's try that let's just do it inside of um Jupiter lab for now let's prototype Jupiter lab you can use this stuff chain with a custom prompt to insert context the stuff chain stuff chain presumably that's under chains okay so what are we doing here prompt template prompt bracing stuff chain Richard foreign stuff trainer am I just like stuff train oh okay hold on hold on stuff document chain do I need an open AI API I think so like to use the llm yeah you do playing train drop trains now surely we can just create initialize agent like just as I saw there llm chain I could I don't know why I got rid of that um let's not open up Outlook we don't need that so we've got a create Panda's data frame agent so if let's just map this out a little bit so plus dot Excel so what we're saying is we have go back to the docker so we have our agents we have chains agents we have chains and we also have so what do we how does this stuff map together chains combine llms and prompts in multi-step workflows right so where is so the llm chain is going to be our open AI bit wait does that mean we've bypassed actually initializing the chain over here because we're just using that specific agent must be because here it's just using initialize agent yeah take a look here we go okay so theoretically I could go and do I don't need to use that I could say load tools Wikipedia so this would be an alias we would Define our llm as such I got rid of llm math is that going to work is this we have no kernel a change kernel let's clear our output because I can't see what I'm doing okay so we want to run that we want to set our API key we want to specify that we want to create our open it let's go Wikipedia so we want to create our llm that's fine load tools is that going to work load tools is not defined where do we import load tools from uh this is not all right here we go load tools is coming from Lang chain agents okay that looks promising guys we might have got this then what we initialized the agent as the agent and that's going to now take in the Wikipedia agent are you guys enjoying me doing this just completely win initialize agent is now not defined because we didn't bring that in Nick you nug let's bring that agent type do we need okay we need to bring that in well you know what let's just bring all this because then we can just bring it into our streamlin app all right so now if we're going to do that okay agent um dot run who won the battle of how do we spell it again Thermopylae paste that in there he's gonna win I just said hold on wait so this is the context how sick is that so it's actually going out to Wikipedia and getting the response the Battle of Thermopylae in 400 BC was fought between all right hold on uh and save that output to a text file right did it save it did not save that's okay you win something you lose something but anyway there you go it's given the response so what so we could theoretically hook this up into our into it is valuable okay cool I'm glad you're enjoying this so we could grab this get rid of that yes yes that is okay so then we would go and set up um so we'd need to go and set our API keys so we'd import OS set our API key so I'd go from API key import API key beautiful that's looking promising so then this is how I a lot of the time when I'm prototyping random stuff what I'll do I'll go and play around with that inside of Jupiter and then I'll go and take it out okay so what have we got done got then what we want to do is we want to load our tools boom we need to Define our llm which is going to be our open AI temp with our temp specified so let's add oh my God I need to add some commentary here what is this so uh let's go and say we are going to import let me make this bigger so you can see it import streamlit for the app then we're going to import all of our lane chains and stuff import Lang chain stuff then we are importing OS import API key and set it to the environment all right let's quickly let's let's backtrack and and take a break as to what we're doing so import streamlit as St so that's going to allow us to have the streamlined app then from Lang chain this is literally what we just learned so from langchain.agents we're bringing in the load tools agent then from link chain we could actually just say bring both of those in a single line and actually we can bring an agent type over here as well so from langchain.agents import load tools initialize agent and agent type and then from langchain.lms import open AI I don't know if I like that load tools is not like a utility an initialize agent is not a utility it's just direct out outside of that agents um API kid but nonetheless that that's fine well I mean I guess they're defined as as methods because they're not uh camel case but yeah I don't know uh all right and then we're importing our API keys so import OS but that being said the framework is sick and you can always improve that uh so import OS and then we're importing the API kit which I set as a string inside of a separate.pi file then we're setting set the API key beautiful and then we are initializing the LM and the tool set cool and then what do we need to do let's go back to our Jupiter notebook then we're initializing our agent and then we're allowing ourselves to run so then we are initializing our agent which will be here and then we would be so then we would collect the prompt collect the prompt from the user yeah beautiful and then we would effectively be going agent dot run here yes yes I think okay let's give that a crack so now if I go and refresh this good job Nick keep it up thank you I need positivity Aaron 35 minutes damn we're uh take a look who are the parties of the Battle of Thermopylae it's why is it not printing as once as don't make this anyway it's given us the response the battle of the Monopoly and 400 BC 480 BC is for between the aquamanid Persian Empire and the Xerxes one and the alliance of the Greek city-states led by spider under Leonidas one 1991. take a look that's our response how much I think my open AI costs uh to retrieve we'll go and let's we'll go and take a look at my API bill in a sec um but there you go that works I don't think this is right so um how do I normally handle on enter on enter input streamlit text box textbook no text here we go capture enter event uh the value if value this is uh it's it was initial my bad it's initializing because there was initially a prompt here so if I went and refreshed this won't go run all right cool okay we're good um now the next thing that I want to do is let's actually add a title so a little bit clearer what we've got so this is St dot title um Wiki media Lang chain but and we need an emoji as per usual how do we get that bird emoji bird emoji uh it's a little parrot thing carrot emoji so that's our title we can then type in our prompt here and then what else did we want to do uh we just need to work out why I need a multi-line text wrap text uh streamlit text yeah see this is printing over there so let's do it as a text area will that look better let's rerun beautiful Wikipedia Lang chain bot so uh what's another one what do we want me to ask loving the exploration by the way glad you're enjoying it um who developed the first neural network this is going to Wikipedia getting an extract and then actually it should be returning a response the first neural network was developed by Jeffrey Hinton David rummelhart and Ronald Williams in 1986 wait Jeffrey hinton's wrote The Deep learning book yeah he I could have yeah he did what is happening to my text area here yeah we're still getting that weird block just use st.markdown or St dot right instead of St dot text for long text let's use St dot write try that again all right guys what other questions should I ask ask about NBA 2K what specific question give me some examples when was NBA 2K first released and who was the game developer yeah like so like I think we could there you go NBA 2K was first released in 1999 and was developed by visual concepts I think this is pretty sick we've been productive in an hour and a half I think I've achieved my study goal of learning Lang chain I'm pretty I think there's definitely a lot more and I gotta go and revise GPT from scratch but I think that was pretty good who is Nicholas or not let's actually go and ask it I don't know if I'm on Wikipedia who is Famous Last Words who is Nicholas Renault this might just come back with the gpt4 response though so if it does what I have one are we getting anything in the logs back here apparently Nicholas Renard is a Belgian actor No Good Wikipedia search results No Good Wikipedia search result was found for Nicholas or not that's pretty cool uh nice little quick uh build up of a stream lit up there you go so I think that that's we've at least begun exploring there's still like a ton more that I think we're gonna need to go on ahead and do but uh at least we gave it a crack and saw we got to good job glad I didn't quit we don't quit here we keep going well a few times around like let's end this live stream but hopefully you enjoyed it uh all right cool can I combine it with Lang chain agent and then ask questions on the return Wikipedia text yeah that's what we just did yeah I'm jabaloni uh I think I'm just retrieving the text from Wikipedia But there needs to be passed the llm to summarize yeah James I think um now I think it's working so it sounds like we need vanilla lemon then we need an agent so the agent that well we need a tool set we need an llm and then we combine those both to effectively allow it to be an agent um sounds like that's the way that it works again we could dig into it a lot more um RPG agents can't be I think I need to go and check that out I don't know I'd never actually realized that Stanford winner made some RPG agents we're kind of sick also let me know what you thought of the intro on the last video I know it's IBM focused but I wanted to create it for work because it's something super interesting that we're taking a look at at the moment but if you like the latest video that I released just the style I'm going to be doing a little bit more of that stuff so hopefully you enjoy it um what else did I need the open AI API yeah I did uh well we're live past the tools and the llm into the initialize agent object yeah James thanks man I think um that definitely helped like stacking it together seems to get it right seeing the process is valuable awesome glad you enjoyed it good nice do I need a GPU to try Lane chain I don't believe you do because it's actually using the open AI API which means you don't need a GPU but you do need an open AI API key which means it costs money um do I think I'm let's go check my open AI API bill right now um jablani thanks glad you enjoyed it great work worth all this time I should go back and watch from the beginning fine tuning what's my favorite way to host python apps for free it really depends I'm still using uh Heroku at the moment there are some others I answered it inside of the courses from Nick Community page again if if you're just tuning into the live screen you can go to go.courses from Nick forward slash Python and you'll get a free link to be able to go and play or do my python course and that will also give you access to the Tech Community in there so um I check it every single morning uh just to make sure that I'm handling any questions um how to make Nicholas a million for the content thanks so much all right cool let's go and check my open AI API bill I don't imagine it's that much of it you never know all right let's go and check this out I'm glad you enjoyed this guys so if I go uh it's unplatform platform can you imagine it's just like a thousand dollars you're gonna need some help to pay for uh where is it under manage account or billing it should be under billing uh usage actually it's under usage all right so we today's the 19th so I've used 70 cents worth not that bad I'm like I've been playing around with it for a bunch so 70 cents not too bad but again you've got to build apps it really depends like if you're a developer obviously that that's not so bad but good good become more but um yeah hopefully you've enjoyed that again if you do want to go and check out the course you can go to Godot courses from Nick forward slash python that is going to give you free access to the courses from Nick python course which I'm updating every day so if you go to courses from Nick or it's like I can never find where the actual course is but that will give you a free code to be able to go and take the tech fundamentals course so it's on 267 on the website but don't pay that make sure you go through here it'll give you a link that you can do it um let me check because inside of the community I actually had Insider Tech Team somebody else asked this as well uh how do I get into that I think I've got to go into my account foreign community could have sworn somebody asked this as well find render yeah I think render was the one that was recommended anyway it was somewhere in there but um hopefully you've enjoyed this it was a little bit different to what we normally do but uh I'll include all the code for this up on my GitHub so if you go to github.com you should be able to get this along with other stuff um do you want me to share this this mind map thing for the data scientists you let me know but hopefully you enjoyed that anyway hopefully you've enjoyed that live stream guys I will catch you in the next one thanks so much for you guys being absolutely amazing audience and checking this out hopefully you enjoyed checking out a little bit about what's possible with Lane chain um let me know what you want to see tomorrow I'll catch you in the next one a peace foreign

Channel: Nicholas Renotte

Views: 27,195

Keywords: data science, machine learning, deep learning, python, chatgpt

Id: M-2Cj_2fzWI

Length: 97min 30sec (5850 seconds)

Published: Wed Apr 19 2023