Large Language Models Bootcamp Information Session

Key Takeaways

I think we are good to go and uh I hope all of you can see my screen I'm going to go ahead and get started okay welcome everyone this is our uh session we do the session every two weeks uh just to give everyone an overview of what our large language models offering um offerings are um so I will go ahead um and get started my name is Raja akbal I'm this uh Chief data scientist and one of the instructors uh at data science Dojo um and I'm going to basically walk all of you through uh our training program the curriculum and uh uh and toward the end we'll be answering any questions that you may have um so we have been around for a long time uh for one of the oldest and uh highest rated rated boot camps in industry and our our approach has been actually um distinct and very different um I come from Academia myself uh got my PhD long time ago uh when machine learning was not mainstream worked at industry and uh the approach that we t uh took um or we take in our trainings is uh we take a more practitioner approach right so we are not here to learn purely textbook techniques we are here to actually learn um techniques in the context of business right so and you know think of it as you know yes you not need to know science part of it but you also need to know the engineering side of it so that's the approach that we take um um pretty much you know any any country uh someone from every country on the planet has attended our trainings more than 3,000 companies globally uh you know we sort of lost track of how many people 11,000 plus graduates uh and and so on so let's Jump Right In right so how how did this boot camp actually start so the boot camp actually started when we last year um around this time actually last year early last year um uh Chad GPT had come out and everyone was uh you know all the Enterprises they were rushing to implement um large language model applications uh we started helping companies uh the the Consulting arm or the services arm of data science Dojo started helping companies Building Solutions and for the most part they were rag Solutions retrievable augmented Generations Solutions and then we realized that while many of the use cases are quite easy to understand quite um approachable um you know just summarize this small piece of text for me or or you know generate a quiz out of this course material and so on right so those cases are easy but when you actually decide to deploy something in an Enterprise it is incredibly hard companies they um there are challenges that may not be obvious when you are using chat GPT or B or Bank chat for that matter um and I will at a very high level summarize right so you know there is cost associated with that like J GPT may open or B or Bing chat they might make you think that hey this is all free but when you ENT adopt a solution like this for Enterprise it is not all free so you know there is cost associated with it how many tokens and they Char you per thousand tokens and all of that um uh the context window which model you can use and which model cannot uh you cannot there are regulatory challenges right so you are you are dealing with uh you work in an in some industry that is regulated Healthcare Finance uh legal um and for that matter any industry right so uh you know you have your pii data you have uh your intellectual property that uh that is at risk um how do you handle that uh inference latencies your application requires answer uh or a response for from your model in uh very quickly fairly short short amount of time how do you do that or uh or maybe for that matter having that wisdom when to use a when do you need to build a low latency application versus a high latency application data governance is a big deal uh whether to use open- Source model or not um you ask the same question again it's a it's a it's a non-deterministic responses so your response can responses can vary how do you evaluate your new responses correct compared to the previous response uh how do you stop hallucinations how do you update the source of knowledge uh how do you evaluate and uh you know your prompts are brittle promp uh prompts changes uh uh the prompts change and the response changes drastically even a small you know uh a change a single word for that matter a single you know punctuation it changes and then the response changes drastically how do you handle these situations so in in real world in Enterprise there are some serious challenges in adoption of large language models and then while we had started implementing all of these Cutting Edge Technologies we realized well others need to learn what we have learned uh in while implementing these systems and that's how this boot camp actually uh started we have three different types of programs um we have some self-based courses that are out there free uh feel free to take take those courses our YouTube channel uh more than 100,000 subscribers go check it out I mean we have a lot of webinars that happen besides this webinar um every other webinar is actually uh free learning we uh you know we are we have built a a big community on LinkedIn close to 3,000 subscribers now uh and you know we keep posting content there then there is live instructor Le courses short duration courses the next one that is coming up is next week which is large language models short duration five six hours course where we will be actually giving be calling it llms for everyone and what we will explain is U um no matter what your background you don't have to be in technology you can be a doctor a lawyer an accountant if you're interested in how your industry or how how llms work for the most importantly you know how are they different from other things that have happened in the past um you uh we talk we'll talk about it and how all of these Technologies they are disrupting um your industry or your work and and so on so that's a very bner level course that uh you know people who are in technology they can attend uh to get a high level idea if you so it's a it's a good mix between the technical and uh the general overview so check it out uh it is there and then we have another uh uh full day course coming up on L chain but right now I will focus mainly on the large language models boot camp and we recently added this remote option to it the entire boot camp this was a long standing ask since we started uh last year uh it happen it so happens that we are only boot camp in Industry uh at the moment I'm you know the only boot camp in the world uh at the moment so the demand is there but the demand uh in this case many people they wanted uh to ask or they were interested in finding out um if they there is an online option so for the next boot camp that is going to happen uh you have an option you can attend that in person but if you can for any reason if you cannot travel to Seattle to attend the boot camp there is a uh there's a remote option you know will uh we have multiple cameras in the room and all the interactions multiple mics I mean the all so it's almost like you're remote but still in class right so you don't have to travel if you do not if you cannot right so so it's basically that uh that uh mixed uh a hybrid mode it is a 40-hour immersive immersive program I will show you what the curriculum looks like shortly um uh so the boot camp is going to cover as I said it's a comprehensive curriculum uh pretty much uh a lot of lot of uh uh things that you would otherwise need um a lot of things uh um that so U let me actually walk walk through the curriculum in a moment but all the uh everything that you would need to build uh a large language model application in foreign Enterprise we actually cover that in our in our boot camp um and then it's uh think of it as 50% Hands-On 50% Theory so we will cover a topic go to the Hands-On exercise then come back you know discuss the theory then Hands-On exercise and so on and on the last day of the project there is going to be a uh there's going to be a rather last day of the boot camp um there is a Hands-On project that we will be working on uh these are our boot camp Partners the list is growing uh you know we have the leading names in Industry they are our partners they are actually helping us uh helping us in different areas these are some leading names in industry uh in the Gen space when you attend the boot camp uh we give you credit for all the software so the boot camp actually includes the cost of uh licenses and subscription so uh you can actually have a browser enabled laptop uh we have set up our sandboxes as we call them uh you know uh online coding environment uh any virtual machines any uh any any compute clusters that you need for any of the you know any gpus we actually provide you that because we know that figuring out the logistics during the boot camp it it is hard so uh so you'll get all of that uh and then as on a as a side note your breakfast lunch you know um everything is actually taken care of when you are on site in person um you know we we know how to take care of people um so and then we also have guest talks from industry right so it is very important to for all of us to understand that uh you know technology can only take you so far and understanding and appreciation of business side of it how do you actually uh how do you uh build a solution that actually has a business impact so for to that end we have some uh some of the uh industry speakers who are actually uh who are there to give talks uh uh some are uh fairly technical some talks are I would say less technical um and more from business side so it's a good mix of uh it's it's a good mix of uh you know uh technical talks and Technical learning plus a business side of it um I think that is there let me actually keep going here um so I will talk let me let me go over the curriculum here um this is a reference architecture uh that I will ask you to take a look at and then we'll talk about uh it uh like component by component so if you look at this at the core of a large language model at the core of the large language model application you have your your uh well the language model right so you know just a this is need to zoom in okay zoom out maybe a little bit I hope this is visible to all of you so at the core of it you have n llm think about llama 2 llama 3 you have um gp4 gbt 3.5 you have uh you know uh Falcon you have uh anthropic cloud is there so many of the some are closed Source some are open source um so at the core of it your llms are there and then on top of that you need a vector database and I will uh you know talk about it uh you know why would you need a vector database you you have you know you are um you're creating embeddings or semantic representation of your documents uh we'll talk about embeddings we'll talk about you know how do you store those embeddings in a vector database then um different kind of tools like large language models caching why do you need a caching how do you set up a cache uh then um Frameworks orchestration Frameworks uh like llama index or Lang chain uh you know you would need them then there is a concept of guardrails right so you need guardrails to make sure that your uh the model prompts are compliant with your requirements and then the model responses your application responses they are also compliant with your responses then you are going to be uh you know there's this deployment part of it you you have to deploy your app in some some sort of you know on U some kind of host um we actually cover over this entire ecosystem and I'm going to actually step by step I'm going to show you so let me go to our learning platform and uh start here so if you look at this um we start our uh we start our um uh start our discussion with attention mechanism and Transformer architecture uh and actually step back and we start our discussion with uh with the high level overview of uh you know what is going on in Industry we start with uh um you know very simple idea right so how are llms different from well what was happening in the past right so you know what makes them um uh what makes them uh demonstrate sematic understanding then uh how do you store so we talk about the very high level I talk about the idea of embeddings how do you store those embeddings why do you need a vector database uh what is rag a very simple naive rag example what is retrial augmented generation we talk about that we talk about guard rails we talk about risks we talk about prompt engineering we talk about how an llm works and uh all of that right so this is the first two hours of kicking off the boot camp we set the breath of topics so by the end of this uh the first session or first two sessions you uh you understand what the bigger picture is and then we go and start drilling deeper into each of the topics then we talk about you know text representation how do you represent text as vectors and then we talk about embeddings and you know uh we talk about uh you know uh semantic embeddings we talk about the encoder decoder architecture and we talk about the Transformers and attention mechanism and once we are done with that you know we will have for every topic that you see for instance right so after attention mechanism we going to Vector databases so we will cover Vector databases in depth and after every topic whether it is embeddings whether it is you know attension mechanism whether it is uh Vector databases we have these exercises uh built into the platform if you notice this right so right now I'm going to click on this and when I click on this this exercise actually opens right here we provide you all of these uh so uh vb8 is our partner on the vector database side so you can see that we will give you access to a a server and a vector database and then you can see that uh you know everything is right here uh in your uh within your learning platform uh and by now you have you know what a vector database is we have covered all the theory we take about two to three hours in about 2ish hours on discussing the theory of vector databases and once you done with that then you actually go how do you create a collection how do you import data and how do you do a vector search and all of that once you're done with this well how do you perform a hybrid search how do you uh do Vector compression how do you multi- tendency in vv8 right so if you look at this uh what we have done here is we have lowered the barriers so we are we have completely lowered the barriers uh as a matter of fact I mean many people have attended this training who have minimal uh coding or minimal python coding ability uh and that's okay right so if you're even if you're a product manager I would strongly recommend actually uh we are looking for a product manager for a product that we have built I mean and then finding a product manager uh for a product like this it is one of the hardest things actually because uh these products are quite nuanced and if you are a if you are a tech already a technical product manager you need to have these kind of details because no longer is it just about managing the uh customer Custer and product timelines uh you actually now dealing with a very different Beast where you you have to understand as a technical product manager because you're taking making these tradeoffs around uh you're making these trade-offs around uh a lot of product decisions and a course like this can actually help you uh even though it's a fairly technical course but it is easy enough that you simply you know you run run run you run the Jupiter notebook and uh and that's it right so um you run this Jupiter notebook you don't have to make any changes to the code so you understand conceptually how things are done and if you are a Savvy uh developer you can still actually go and make changes and uh you know get uh what you wanted to get out of it so it's it's a it's a good balance you know if you're less technical that's okay if you're deeply technical that's okay too and we have options for everyone so um what do you do after that you are going we are going to talk about you know Advanced drag so now that you understand uh you know how what is Vector database so what is what is retrial augmented generation and then we talk about this in this session a very detailed understanding of uh you know uh what can what can uh a complex rag system look like uh we have uh we have um uh um a module on prompt engineering we have once again the same way we have these Labs uh built-in Labs uh for prompt engineering and let's say if I go let's say if I open up this prompt hacking lab right so the same thing I click on it pops up in my uh screen and then we go through this uh all of these labs and we leave you with a lot of work that you can do on your own uh we spent um around three to four hours on fine-tuning uh so far we are doing uh um llama 3 came out recently but uh you know we need to update the curriculum but uh llama 2 a 7 billion 4bit quantized model we do a prompt uh um fine-tuning of a Lama 27b 4bit quantized model uh we give you the GPU Cloud which we start with actually on the theory side we start with fine-tuning we start with uh um transfer learning fine tuning and you know setting the context we talk about quantization what is quantization what is low rank adaptation explain what Q quantization Lura and Q Laura what they to do and then we go and give you a data set give you a GPU Cloud um you know again uh V actually um bear the cost of that GPU Cloud we give you that GPU Cloud give you the uh the boiler plate code you go and run this you modify your Lama 2 model and compare it with an un fine-tuned model uh and then compare the two um we have a session on uh conversational AI um and we this is a great session because now industry is moving towards more domain specific models so this is a session on domain specific models we spend almost an entire day about sixish hours on Lang chain Lang chain is this orchestration framework you know uh we go into uh every possible detail that is out there on Lang chain if you look at this uh we talk about language models prompt templates you know uh chat models output Parcels we talk about all of these we talk about retrieval we talk about chains uh let's say if I go inside chains here you can see we talk about simple sequential chain summarization chains Etc when we talk about memory uh different kind of memory you know fixed length buffers uh you know or some kind of uh you know more Advanced memory uh approaches uh we also talk about multi-agents we also talk about you know this agentic behavior of Lang chain uh and then we have labs around that uh all of that happens in class you can see um we and of course I mean you can see that this is a lot of content Lang chain itself can be you know a few day worth of learning but what we do is we give you we set the foundation we give you an idea of what Lang chain is giving you enough in each of the categories and then uh and then leave you with these Labs so you can be practicing on your own uh uh then uh we are uh you know we are now incorporating some of the newer uh features like uh lsmith not lsmith actually Lang graph it's something that is uh that is something exciting basically think about this some agents interacting uh together it's almost like instead of a sing single AI agent you multiple AI agents that are interacting with each other um what else is there uh let me see did I show you that's so the same drill even Lang chain you click on it the lab pops up we give you the API Keys we give you our own openi subscription and you use it and you can see that we go through it we uh basically you will by the end of uh this session you will actually actually understand how applications are built and uh the plumbing is done then when we go here what else is remaining we are going to observability and guard rails right so we also talk about monitoring your model uh once you deploy it you have to keep monitoring it uh how do you set up guard rails we talk about that um we also talk about productionizing uh how do you deploy the model or deploy an app uh end to end uh we also have a session on AI cover governance uh data Andi governance uh because uh for those of you who may have interacted with some llm applications right so once it is um you whether it is rack pipeline or you're building a fine tune model um the data that goes in into building these models you have to really really make sure that uh the data is uh um you have to really make sure that the data is uh it does not have any personally identifiable information it does not have any intellectual property and and uh uh how do you actually make sure that your your solution it is uh it is compliant with your company policies or perhaps regulatory policies within your industry um on the last day of the boot camp we have uh we have a um a project we give you once again some resources some tool some some code uh your own VMS and then everyone Builds an application uh end to endend um everyone starts and of course we give you boiler plate code otherwise it wouldn't be possible uh within a day but we give you uh pre configured pre set up uh repositories and you take those repositories and you deploy them and you leave with your own working application deployed in streamlet Cloud you know you have your own application URL then we give you exercises hey how do you incorporate this chain how do you add this guard rail um how do you uh install buffer memory how do you um can you can you add a Wikipedia agent in in addition to a ducto agent so those kind of things that give you an idea um uh they they give you an idea what does an application look like and once again I do not want to intimidate people uh because people tend to think that you know hey I don't know any coding we have had Fe a lot of success with people who don't know coding because the approach that we taking is that everything is preconfigured everything is set up for those who don't know coding they don't they have nothing to fear because they can actually run run run and you can still still see in the Jupiter notebook right so yeah I'm I'm seeing this I can go down down down down if you look at this up up up so they can still run it and they don't the code is functioning and running at the same time for those those uh who are Advanced um Advanced U uh uh in programming they can actually you know I can save this as I can save this as a different name you know and then once I have saved this as a different name I can modify this I can and uh and it stays in my because everyone gets a dedicated space uh in the in this learning environment you have access to all of this for one year um from the day of the boot camp so if you look at this right so um it's what the way we have set this up is um whether you have minimal coding ability or you have a lot of coding ability it should not matter you know uh the course uh is for anyone who's you who's uh I think it's more your motivation it's more the intent that matters and if you have the intent we have ways to actually teach you um so I think Dan you have a question that is somewhat related to uh what I'm talking about so Dan has a question what are the prerequisits for learning about uh llms any mathematics uh I'm intermediate Python and know some machine learning you're more than qualified to attend Dan uh because we have had people who had very beginner level uh uh beginner level uh understanding of python uh and they were able to um actually attend the boot camp uh and finish the boot camp and learn from it very Su uh successfully um the interesting part about any mathematics right so while mathematics can be helpful uh the fun thing is while mathematics can be helpful the reality of it is that uh uh uh you know unless uh uh is still building llm applications for Enterprise it is still 95 maybe I would say 98% it is still good old software engineering it is you know many we may be tempted to believe that I have to know every single you know uh Matrix operation that is happening in the uh when I'm using Transformers I do not I mean so as long as I am aware of uh you know which library to use and how to use it uh that should be okay but having said that uh we take a middle ground right so some courses in uh large language models they can be deeply deeply uh mathematical some can be this mindless application of libraries what we try to do is we try to find this right balance between uh too much Theory and too much uh you know application so we do talk about uh you know the intuitive aspect of um how uh the math side of it but not to the point uh you know that we try to explain it um in a way that any math that is needed it is anyone who has finished high school they should be able to understand what it is because at the end of day I mean Matrix operations and all of that I mean these are high uh we learn these Concepts at high level and uh somewhere along the way I mean we get sort of disgruntled we think math is difficult but math is not that difficult actually so um you don't have to worry about it right so you don't have have to worry about the math part of it uh in for the most in most cases I'm pretty confident you will be fine okay um let me see what we have and meanwhile please keep these questions coming uh what else do I have here um we talk about the curriculum uh the technology stack so the technology stack that we have is um um we are using uh actually Vector databases I think this slide needs to be updated we brought in vv8 which is one of the leading databases in Industry right so so we use VBA to one of the I mean I was fascinated I mean I'd use Vector databases but when it's when I looked into vv8 I said wow I mean this is this is awesome so uh vv8 is our partner for Vector databases uh we use on the foundation model side we use among others uh you know we use a mix of close source and open source models you can see the different models that we use in our in our during our um uh presentation during our boot camp then there is uh uh in orchestration we have been using Lang chain and now llama index also is coming on board starting next uh next boot camp uh on the on the deployment side uh or you know llm operations or you know monitoring and deployment side we have a few different players by laabs they are more into logging and monitoring and guard rails run pod uh is our GPU resources when we are using any f tuning when we have the GPU workloads uh we use then zml and then now we have Union coming in Union AI they are coming in next time uh deployment as I said we use hugging face and streamlet for deployment uh so you can leave with a working app on your uh that has your own you know you deployed it through your GitHub account uh that's your own working app uh so we have talked about I've given an idea for all of this uh actually let me see if I have missed anything um I think I mentioned on the last day uh we have a project and that we will be doing this is our list of uh instructors uh if you look at this this is uh um um and this is not everyone teaches all the time for uh you know the last boot camp that we did we had uh I was teaching uh LS was there LS teaches Transformers and ion mechanism and you know encoded decod architecture uh offer is from Veta over more of rag in a box type uh topics um then conversational AI it was kic uh was there um then ran actually was he's the CEO of security AI um and um he talked about the uh the data in governance needs uh and then how do you implement that that uh and and you can see right so Zen was covered vv8 you know Vector databases this was actually a great session once again Hamza was around he T deployment uh and so if you look at this why do we have so many instructors um you want people you want to learn from people not that uh so yes I mean some people who have attended conferences and they know things in theory everyone who's teaching what they uh What uh whatever topic they're teaching they have actually done this in real life they live and breathe they they go to bed thinking about those topics and they wake up you know and they're thinking about these things right so um if you look at this uh you know everyone here whatever they're teaching they do that for a living um you know they have a product or you know they are um they're completely absorbed in this space and which makes that learning exper erience very very different because they know all the issues uh they know all the technical issues they know all the business issues and it it makes interaction actually very very valuable um these are our partners I mean so our partners actually they have been enjoying this journey with us um as as I said we are the only boot camp in Industry um and um and and then um we uh even if after I mean maybe some courses they have started popping up but we have these coll collaborations with Industries right so you know one of the leading players in the respective technologies that you help uh that help uh you in building these applications uh we have our relationships that they they provide our worldclass instructors worldclass Engineers um and uh uh and they are actually building products that are mainstream right so you know it's uh it's basically you get to learn from the best and the brightest uh this is our customers right so I mean if this list is growing I don't think we updated our last uh updated this from our last uh boot camp but you can see that we have people traveling internationally to attend the boot camp as far as from Australia Middle East people have attended within us uh people have traveled right so many of the customers are actually returning customers who have attended our data science boot camp so you can see some of the names uh most of the names should be familiar uh but we have uh we have adding logos to uh to this list real people go check it out I mean uh these people are on our website go ask them what do they think about the boot camp right so then then you should be able to actually tell uh now the next boot camp is actually so this is going to be a hybrid boot camp uh so we will have uh you know some attendees in class because we have been getting an overwhelming uh number of requests hey when are you doing this online so what we will do in this case is we will have a uh we have a we have a good classroom multi camera multi mic type setup um where you know you will have um some of your cohort members who will be physically present in Seattle and then you can be at the in the comfort of your home because many people actually are there are international uh they are maybe in different states I mean they they think that it would be cost prohibitive or maybe they have some responsibilities uh that they that can prevent them from traveling uh so uh we will um you know we are this time we're experimenting with this uh this uh hybrid approach where uh you know you can still ask questions live um you can interact with your peers uh in person uh who are in person and um and you know we have ex we we did a limit pilot uh this time but now we are opening it up to uh to uh general public so if you attend for the Seattle uh at the next boot camp it is going to be you know you have to you will will ask you whether you want to are you attending in person go ahead and register if you want to attend online and we'll send you the instructions so but you will have to choose between Seattle and in person um so that is it um I see a few more questions uh there is I think d and you have this question how long did it take you personally uh to what you needed I I I I'm still learning Dan I mean I'm I'm not I mean this is not fake humility I think the space is changing so quickly uh you know I learn every day I mean so it's it's almost like uh you know this if space is evolving so much um I can give you an idea so we started uh this boot camp back in September and uh the boot camp has I think 40% of the content is actually new content it is changing so quickly the advanced drag module it was added just this uh like the boot camp that ended two weeks ago because we thought you know query routing multiple agents you know um um you know agent interaction between agents and all of that things have been changing very very quickly uh so um so I would say that I've been working on this I mean I've been machine learning space for close to you know I don't know I mean half of my life right so I mean good uh good amount of time my PhD work if you count that right so about sixish years then Microsoft then um like about six years Microsoft and now here so I've been in machine learning for quite some time but for this generative AI actually building things I would say about one plus years maybe a little over one and a half years I would say right so that's uh uh but I mean machine learning has has been around um I've been doing machine learning for pretty much as long as I can remember uh so um that's my honest answer I'm still learning uh but uh you know that's that's the case with most of us at the moment I mean the space is evolving a lot but if your question is uh what you needed for llms uh if you are talking about uh you know um how quickly can you start building an application I would say that the by the end of the boot camp if you are already know how to code and you you're already in technology by the end of in the boot camp uh most people they are able to actually build a basic app but uh you know there are scalability issues uh that um we have a product and I deal with those on a daily basis right so hey why is this latency so high um you know have we made sure that this is secure and do we have the right access control and there are so many nuanced issues um um I'm still learning okay uh let me see there are more questions mahantesh with so many so much of changes and element hand out how do you even apply them in real time by the time we decide to use one it would be outdated so mon this is a great question and I think I touched upon this uh I touched upon this a bit we are actually very actively modifying the content we are very actively actually changing the content so uh I think I touched upon this uh that uh very often what happens is that we are in our engineering team meetings we are discussing uh certain issues and um a few weeks later um you know we find out that uh Lang chain or llama index or someone else actually came up with this right so internally I can give you an example right so chunking is a problem right so hey I don't like this approach I wish there was a better way to approach a a chunking of documents and then you find out hey as we expected they came up with you know different kind of chunking approach right so if um is it possible to learn from uh you know whatever the user is uh you know maybe some from from the feedback pipeline uh or uh you know there is this uh uh approach uh uh can we can we use both retrieval augmentation and uh fine-tuning and guess what I mean a few months down the road you find out that there's a paper called raft I mean then someone uh exactly came up with this idea so it's actually um at the same I mean it's it's exciting and also you know I would say humbling at the same time that I mean there how much there is to be learned so uh and we incorporate those CH changes very quickly uh one thing if it is not already obvious is U I'm a practitioner uh as well right so I'm not teaching it based on some curriculum that we uh we found out somewhere we are actually building products with you know billion doll companies relying on our products and we are um you know our thought process has been shaped by you know these uh Relentless uh customer Obsession and this I mean maybe in some cases you know customers who are are not uh for demanding so uh and they keep you honest they actually make you they force you to learn and then when have when we have those learnings uh we share that joy that excitement with our our customers and show them what is a better way of doing this so it's an ongoing Pursuit I 100% agree with you it is changing but if you if you mean by real time within a day of course it is not possible right so we but our curriculum gets updated very frequently and we will leave you in a position that if when things change you are actually you have that mental model and you have that basic Foundation that you should be able to learn um you should be able to learn what you wanted to learn and uh you you are able to identify uh what the new technique is doing the the why part of the technique and you know how can be figured out okay are there any other uh questions anyone okay in that case let me just quickly review if there are any other sounds good so what I will ask you to do is the next course that is coming up please go ahead and take a look uh you know this is our uh this is our large language models BL cam page go check it out okay go check it out uh and then we have these short courses if you like we have this one coming up next week large language models for everyone go check it out this is a very short duration course for really anyone who wants to get started you don't have to have a solid background um then uh the other course that we have is as I said 40-hour boot camp is there that is hybrid now you can attend it online too uh then we have this short duration courses and then we have this uh you know a course coming up on L chain which is sometime next this month actually I'm forgetting the dates um but you get the idea in this CA in this case you basically it's a shorter duration course course in case you cannot you don't have the time to attend the entire uh entire course this is uh you don't have the time to attend the entire boot camp this is on May 14th okay and in case there are no other questions will just give it maybe few more seconds okay uh well uh thank you everyone and I'm looking forward to seeing some of you at the boot camp please feel free to reach out to us through our website and happy to set up a time set up an advisor call and usually if I have my calendar is available I actually try to take the advisor calls myself to make sure that you know we find the right bit so please U feel free to set up it thank you everyone