How to Learn AI with a Free Bootcamp!

You're going to see this everywhere AI PC , right , and you know you can have like a decent AI PC that's just on your network , that's not necessarily used as like a window station , but more just , people use it as inference , but you could work it into your business right , Like just not even online , but just in your office , because you can be making prompt

documents and leveraging other things to improve just your , your workflow . You don't have to go use Microsoft Copilot or Google Workspaces , Gemini , and you're going to get a lot more flexibility and control , and that's something that I think that I'd like people to get out of it .

As always , with me is my co-host , Chris Miles , at BGP Main on Twitter . We have a very special guest with us tonight who is my cat ?

So I guess , andrew short introduction in case people missed your last episode and then let's roll right into it , man .

I don't know , but I keep opening the mail and there is like wads of cash and I don't ask any questions . But , uh , you know , it's just , it's just how it works . But , um , I recently just bought a , a church , an old church , and I'm turning into a technology center . Um , and if anybody has postcards , I'd love to receive some .

Um , I have an address somewhere . I was supposed to have like a title card so I could show it up here , but uh , yeah , somewhere on my twitter we'll get in the show notes .

I was like well , well , I'm going to , I'm going to send it somewhere as a postcard . And she was like no , no , you're not .

But yeah , I just , you know , I don't know why I'm doing it , but I'm doing it .

Is it just because , like hey , gen AI is a thing now , or did you actually have like a kind of a plan that you put together for this ?

They wanted to build a project , they wanted to know what , like what to do , and I would give them ideas , but they they couldn't get started and I didn't realize there was such a gap there , and so what they really needed was somebody to take them through to a degree and then give them flexibility to to discover on their own what they could do .

But the reason I'm doing the Gen AI Bootcamp is totally for selfish reasons , totally different , which is just like I've been taking Japanese lessons and I keep building out little Gen AI learning assistants and I want to just squeeze more time in there and I'm like , how can I do that on company time ?

And so if I do a bootcamp and Gen AI is hot , hot , hot , then I can go ahead and do that on company time . And so if I do a boot camp and it's still and like jenny , is it's hot , hot , hot that I can go , I can go ahead and do that .

But you know what language learning and uh works really well because , like natural language processing and lp uh I gotta use some initialisms here to sound super fancy but it like it's just a powerhouse for lms , ims . I don't know if it's novel , but it's a good use case and I think that with everybody really interested .

I'm not sure if people are interested , but they feel obligated to make sure they don't miss out on this and learning about Gen AI and their tool belt , because we don't know if it's going to take people's jobs or not . It's not , but people are worried about it .

Uh , and and and the whole boot camp thing is is , um , and we've said this before right , like the best way to learn something obviously is to do something , and you really like the use case thing is so , so important , right , like just applying what you're learning and doing something with it , it's probably like the best way to learn something .

So language learning is definitely definitely the way . I think you know we're both learning , we're both learning Japanese . So I'm , yeah , we're , we're , you're building some really cool apps with it and I'm asking chat GP dumb questions .

I'm sure there's some people that self-study , but like having a , a real human that knows the subject matter and can guide you through it , is like super useful . But you know , it just alleviates a lot of the the busy work . It's like imagine having our teacher every day , you know , having to do drills with me .

That's not , that's not a good use of their time , right . So right , you know , I think . I think it kind of meets in the middle . It's not perfect but it's definitely working for my needs anyway .

So my question for you is how did you go into preparing you know yourself for this Gen AI boot camp and how did the content kind of come out of that ?

So the experts that are in academia and know how to build machine learning models and stuff like that , to them they're just like yeah , this is a thing and it's very novel .

They know how to apply it , but a lot of people in the Gen AI space don't have I shouldn't say they all , but a lot of them don't have that AI ML foundational background Like the capital models .

Yeah , they're more like practitioners of like oh , I have a model , I play with it and for them it's very hard for them to find a fit because they might not understand how it fits into the landscape , right . Find a fit because they might not understand , like it , how it fits into the landscape , right .

Or they're utilizing gen ai when it's very expensive and ineffective at specific ml tasks where traditional machine learning is more effective , or , uh , you know , building a simple neural net or using something like xg boost , um . So I think there's that kind of muddying going on . Today . I think it was um eight of us andying going on .

Today I think it was AWS and I use all the clouds . I just happen to be wearing this because it's warm , but today Matt Wood , who was the VP of AI over at AWS , has departed . He's been there for 15 years , which is really interesting . So in the internals we were like , oh , does that mean that AWS is deprecating AI ?

Now , all there is is Gen AI , which I don't think that's the case . I think it's just coincidental . But Gen AI is definitely , again , it's just a subset of very specific machine learning models , right , or like put under that umbrella . But let's talk about how I prepped for getting up to speed .

So I did already have some machine learning skills behind me Before I even had this company . I had a couple of startups and I was doing machine learning pipelines .

I didn't realize I was really doing machine learning it was more like classical machine learning and things like that but I already had some working skills at building more efficient machine learning pipelines on AWS and other tools like that . So I already had a bit of that foundation

. But for this it just felt like Web3 , but there's like all these fricking new terms and you dig , dig , dig and you find out that the term was just like overblown as the description of what it was right .

Like a marketing term or something I don't know , but it's like obscurifying the utility of what it was right , like a marketing term or something I don't know . It was like obscurifying the utility of it . Like RAG is Retrieval , augmented Generation why do we have to call it that ? All it is is a no , I know what it is .

It's just like a way of going and retrieving data and bringing it back and putting it into the prompt before the , the , the agent or LM response applies , and so at anything that it goes to , if it goes to the internet , if it goes to a database or whatever , that's all it is Right .

But like , when you start reading about it , people don't describe it very clearly . And then it's like it seems like it has to be with a vector store , and then it's like it seems like it has to be with a vector store .

So the vector store part of it is is literally just like a very fancy mathematical matching algorithm , right , like the whole vector store thing , where you take documents , plug them into vectors , turn them into math and then , when you build your prompt , you turn the prompt into a vector and then you compare them and try to find the closest match for the docs

you have , I guess , right ?

But the part where it gets kind of more complicated is that how do you know like these two words are related , or these four words are related , and that's where embeddings come into play , because the embedding is the algorithm , the way to describe that relationship . And so when you store them in a vector store , what's the relationship ?

Is it because the length of the word ? Is it because they look similar , like similar letters and spelling ? Is it because they're in the same topic ? And then , on top of that , an embedding could be optimizing on multiple things , so it could be like a combination of those things and you have a bunch of numbers .

So that's my , my simple explanation , that I understand it , and somebody that knows vector storage would be like I don't , like you're . You're missing out all this information , right ?

Well said because people don't realize there's like lots of different kind of embedding models out there too that'll based on sentences , based on topic , based on there's there's uh , video ones and image ones and stuff , and it's all chunking that same .

There's multiple modalities , so , right , you have vision , you have audio , you have text , you have molecular . You know , there's other other ones there that I like .

I'm sure there's other , like video , which is technically just images , moving images , and then when you look at those other ones besides LLMs , those are just like models that already existed before Gen AI , right , like we already had vision models .

Is that the marketing ? Like the , the , the , the fact that you can use common language to engage with all this data ? Like before it was just machine learning , it was just like pattern recognition and everything .

What would it be 10 days from now , based on the data that you have , and it'll spit out a number . Or it's like tell me , you know , based on this , like classification , what's like ? If I give you this word of an animal , can you classify it and tell me , is it a , is it a mammal , is it a reptile ? And so that's the classical stuff .

Now the interesting thing is that llms can do these simpler types of classical machine learning tasks . Um , but it's very expensive compared to like . Even if it's not that expensive , it's just like it's pennies . It's dirt cheap to learn classical machine learning and utilize it for those things .

But with llms you don't have to understand how a machine like a machine learning pipeline works . You don't have to know how to build an ML model or work with XGBoost , and so it makes these other simpler ones more accessible at a higher cost .

But once you understand the costs and stuff , you learn a little bit more and you can utilize those other simpler models , which is what people need to do when they're learning Gen AIs . To go into those simpler models .

I don't know how much you're willing to share about the structure and whatnot at this point , but I'd love , I'd love to understand kind of what your , what your thinking is or planning is about , like who's coming into it , who's the ideal student and like where you want them to leave . Maybe they just start there . We can work our way through that .

So I've built out a Gen AI roadmap , and it uses what I call a maturity model .

The idea is that you talk about where somebody is in the maturity of the knowledge that they have in Gen AI , and so the first place that people usually end up are AI-powered assistants , gotgpt , anthropic Cloud , lama , mitstroll , you know , et cetera , et cetera , and so just in that little that area , there's a lot to do .

Just with prompt engineering , right , you could do a lot and never ever learn how to programmatically work with LLMs , and so that's like level 100 . Even my mom would want to use that , right , and so I almost feel like I could do six weeks just on that

. Then there is the next step . There's managed services . You know how to program , but you don't know how to work with , let's say , hugging Face or Python or PyTorch . If you get conflicts , you just give up , right , you don't want to download a model . You just want to be able to work with models .

So one step there , and so that would be something like Amazon , bedrock , gemini or working with Cohere's API . So I feel like that's a level in its own and that could be over six .

And then there's the more top level , which is like next level , which is like okay , I'm comfortable with code , right , I can work with Python , I can work with Hugging Face , I can download a model , I have CPUs and GPUs in my house that I want to utilize and I want to have more flexibility in my models .

I don't just want to be using these stock built ones and I'm worried about security and I'm worried about these other things that come with . Like if you use a proprietary service . Like if I use ChatGPT and I get , I'm used to it in my workflow and I forget how to program priority service .

If I use ChatGPT and I'm used to it in my workflow and I forget how to program , now I'm stuck If they jack up the prices . I'd rather have something I have more flexibility with . So that's another level . And then there's a level beyond that , which is like I got to deploy this for enterprise .

I got to know how do I right-size my workload and what would it look like at scale ? Do I need AI inference for that , like an AI accelerator , and what's the concurrency and what's the deployment model look like , and et cetera , et cetera . And what's my technical path from startup to full size ?

And so it'd be nice to teach all those four levels and you could do that over six weeks , but the thing is is that it'll be challenging , because I also just want to build apps and have this little ecosystem , so I'm going to have to figure out a way to do it , and it's probably gonna be messy , but you know , if it's all there , then I think that's that's

what matters , and we have a lot of people on on the same thing . It's going to make it like the . The road will be very well traveled for everyone else to follow .

Oh , I'm curious to get your input on what you think will actually be the outcome of this from a from a business perspective . Like , how are people actually going to use these LLMs ? Like , do you feel like there will be a lot of businesses just using the off the shelf ones , or do you think they'll actually put in the effort to build their own ?

And is there a specific or a specific reason why you think they would do one or the other ?

You're going to see this everywhere aipc , right and uh , you know , you can have like a decent aipc that's just on your network , um , that's not necessarily used as a like a window station , but more just , people uh , use it as inference , um , but you could work it into your business right , like just not even online , but just in your office , because you could

be making prompt documents , uh , and leveraging other things to improve just your workflow . You don't have to go use Microsoft Copilot or Google Workspaces , gemini , and you're going to get a lot more flexibility and control , and that's something that I think that I'd like people to get out of it . So that's probably one of the bigger things .

It's just not feeling that you have to use the cloud for it even though I'm the cloud person , that there's other options there , and I think edge is going to be a big deal . So I just want people to get more comfortable with the edge .

I don't know why .

Even that part of it can be difficult just to do technically , and also , of course , making sure that you have , you know , a graphics card or something that's able to do it as part of the boot camp .

I mean , is that something that you're going to want to try to solve , you know , for the students , or is that something that is that is bridged too far , that it's just not gonna be part of it ?

But I just don't have the hardware Right . And then the other one is like I had the hardware , but it's not the best Like it's like you know , it can do it , but maybe just on CPUs .

And then there's the person like I have my gaming computer or I was trying to do mine like crypto mining , but now I guess I can utilize this for work for for Gen AI , we can do that . So we kind of like those four levels and you know I want to try to satisfy all four as best I can .

But that's going to be a hard thing because it just going to be things that people can't participate in because there's that hard restriction . But you know , but at least we get the exposure to it . But yeah , I mean , it's just , it is what it is Right Like I can only do my best to try to support all levels .

But that's another challenge , like I need to make badges .

And so when we did the AWS Cloud Project Bootcamp , the Terraform one was just a single badge because it was smaller in scope , but the AWS Cloud Project Bootcamp had four badge levels and the idea was that the farther and harder you pushed and the effort you showed , the more likely you could get that Red Squad badge which was subject to me to issue .

The other ones are extremely well-defined , but the other ones were the last one . There was like push , push , push , push . And so here I have to kind of figure out how the grading would work , the badge would work , when you know , not everyone can work across all those four things .

So we have like four different levels of learning , four different levels of maybe you could say deployment or platform . These terms aren't defined right . Like I'm making my Gen AI roadmap and I'm trying to like okay , how do I , because there's no generic making my Gen AI roadmap and I'm trying to like okay , how do I ?

Because there's no , there's no generic cert on Gen AI , there's not . Like AWS , has I made AWS's one , I made the Azure one , I'm doing the Intel , the NVIDIA one . Right now Google doesn't have one for some crazy reason , and so you know they don't define the stuff very well , right ?

And even you think AWS would like leverage what they the terms and all cloud practitioner to bridge it over . They don't , and so you know I think that we really do need like a generic cert and um as a byproduct .

Before I even start this boot camp , I'm going to finish that production and that's the prereq for the course , so that people you know will have a grasp of what they're looking at before we start building the project .

Because , like you can't , you can't learn any of these certs that exist now and it's like it'll you'll be so deficit in so many areas , more so than previous cloud certifications yeah , I actually just started yesterday your um cloud certifications .

You know , if you're very skilled in in the technology realm , it's probably going to take someone to study . You know , if you're very skilled in the technology realm , it's probably going to take you a little bit less time to grasp some of the concepts and things like that .

Do you feel like this is kind of in the same category , like if you're strong with technology , you're probably going to do better in certain capacity , or are you really trying to ? I know you're trying to teach all you know , kind of grade levels kind of thing but do you think that disparity is still going to exist ?

And Like , if you already have DevOps and you've done MLOps , you know you're going to have an easy time over there and you have the cloud foundation is going to help a bit . But really it's just like you have a very specialized version of machine learning and working with data for these particular type of models . So unfortunately , that disparity is there .

And then it's like then you have this hardware layer and so , like some people don't come from a hardware world , especially in cloud , you know , people have never , you know they've , they've never worked with cuda . They've never . You know they don't even know what cpus they're utilizing , unless they've worked for a larger enterprise .

And even then they just know it as as the skew right . they're like oh , the skew has this and you know , and so , um , you know there's going to be , for , depending on where you are , there's going to there's going to be challenges , right so ? Or even your developer , it's like .

It's like even I'm using , uh , like python and pandas and all these tools and they're just in pytorch and they're throwing all these errors constantly . If you don't get the versions exactly right , it's even more more so difficult than than regular program . You can get through it .

It's , I guess , like learning japanese , like three different language , like character systems . You know what I mean . And and the , the verb comes at the end , you know . So it's a good pairing , yeah that's a good point but you can't . You can do it right yep , that's .

. And actually that brings me to what ? Do you think that and this has nothing to really do with the boot camp ? Uh , it , I guess it kind of does , but it doesn't really , because it's not like , yeah , the presence or absence of the boot camp has anything really to do with this .

Do you think that we're gonna see , or maybe we're already seeing , um , like a gold rush for you know , ai certification , ai learning , ai careers , like we saw for cyber security ? I hope so , because I'm making certs .

And you know what , like , I'm one of the few persons that have NVIDIA and Intel practice exams and making courses for those , and we're not seeing them fly off the shelf .

But certifications are really driven by not just the demand , but also how it's packaged and the authority that's pushing it out , and the way people are using AI is not the people like AWS and Azure want you to use it Right . They're like how do I download , how do I all these other things that I'm talking about ?

So , you know , I'm thinking maybe if I pull it off and I make the right kind of content , it could work , you know , and so then maybe I'd see the demand . But at this point , right now , the certs that I've been covering .

I don't know , it might have reached its peak already . It's hard to tell with these things right ? Do you think that Gen AI will , for one , stick around long enough but , for second , just be diverse enough as a tech to invite that same kind of gold rush to like get people involved ? You know running for it as a career ?

Oh , you can make good money being a comp sci person , or not comp sci . Uh , a security , a cyber security person , I gotta go make good money . That no one's saying that by jni they're going oh , I hope I don't lose my job to jni . I better learn and figure out what's going on with this thing , because I have no idea I'm going to lose my job or not .

Right , like , that's , that's how they're thinking and that's like the number one tag that I'm going to have in the gen I boot camp . It's like , uh , you know , I'm going to teach you everything you need to know so you don't lose your job . Like , because that's whatever . That's people's biggest fear . Now .

It's not to say that the , the , the tools are not useful , like they're definitely are . They're great in your tool belt , but the mindset is different . People aren't running into it like Web3 .

Web3 is like oh , I can make a bunch of money because it's all about money , right , but the conversation is different is such a broad topic that can fit like cybersecurity .

It's not just technology specific right , so it's kind of you know . I think it's harder to define . You know what the boundaries are and how malleable this thing needs to be .

I guess I'd be curious from your perspective , like when someone wants to learn it do you think it's going to be best to just like figure out a use case , try to solve a problem and use Gen A how to do it , or do you really want to go back and learn the bits and pieces that go into it beforehand ? What do you think is the better method ?

And I would , I would unfurl it Right and I would go back and make that Gen I roadmap which will turn the Gen I essentials course .

So the point is is that if , if I make the Gen I essentials course and people do that upfront , cause that's what that's kind of more like what people are used to doing , and I'm saying it's the best way to do to how , how people want to learn , but that's not the way that I learned it's it's it's building things and then pulling each part out and then

expanding on them , right ? So you know , I think that that's what's going to happen in terms of boundaries , like now that I'm in it , like when I started I was , I was like we're like where are the boundaries ? I can't feel them Right .

And I think it's because every other day you see something on Twitter like Gen AI look at what it can do now and it's so fuzzy about what it's describing . But now that I'm in it , I can draw the lines of where it starts and where it ends and I just see again , it's the Gen AI stuff that we're seeing online is .

You know , it's that pump up the VC investor money and get everyone excited , and I'm not sure if it's intentionally being obscure , but it's just like you know , people don't know what they don't know , right , and you know , we just keep seeing more of it .

But you know , people feel that it's very like it's over , like it's overreaching , like cloud , right , but I . But now that I mean I just again it just feels like another tool in my tool belt and I think the more we get people through it , that's how they're going to feel about it too .

But then when you start using but you go , oh , here's all the cracks . Do you know what I mean ? Like I was using vercel v0 to it actually worked really well , like to build up the um , that marketing site , for apparently , before I do the boot camp , I'm doing multi-city all-day training and , jenny , I don't know why I'm doing this and anyway .

So I built the marketing website in v0 and I built it in an hour and I was like , oh , this would normally take me , uh , two days to do .

And I was like , I guess I don't need to write HTML CSS by hand because , like , I was using things I don't normally , like I don't like Tailwind , I don't like React , but I know how to use them , but it and I use something new called ShadCN and it worked .

And it was like , oh , I just need to generally know , based on , like , having deep skills being able to work with and having confident that I liked what it was generating . And so for that first one , I was like wow , that was great .

When I started building the second one , I started writing by hand because I was like I'm not doing exactly what I want and and and basically I filled the gaps on on the stuff I didn't like about react like that . I forgot about React and ShadCN and Tailwind CSS .

So you always get that aha moment , but then eventually you're back to tweaking things by hand and then you forget like , oh right , I was using Gen AI and I'm like , why am I not using it ? I'm writing everything behind again .

for bad , I'm trying to get as much as I can right now .

Um , we need to start wrapping up , but I would love for you to first of all , uh , give us a little bit more details about the in-person thing you're doing and then plug any of your AI courses as well , so we can make sure we get them in the show notes and just tell us about it , and then we'll wrap up here .

But you know , I might be going to the states , I might be going all over the place San Francisco , new York , jacksonville , can't say the name of it , but it starts with an R , oh .

I still want to just go and teach and help the community . So you know , I might be coming to a city near you at least it and so I want to get like um , you know , it's like you get like a map and then it's like you got the big bobblehead and they're like going around to the next town .

I want to get that , that , uh , that graphic going there , but right now just a couple cities , and if it does really well , then it's definitely going to expand and that might go on for the next year . So let's see if that roadshow happens . Awesome .

Could have done it on the 10th , but I'm just like I'll give myself an extra week just in case . So january 17th is is the start date for the boot camp , so hopefully everybody is excited , uh , for that awesome , awesome also , apparently I'm doing a boot camp right now . I don't know if anyone knows .

I'm doing like a small , mini , mini camp I didn't know and yeah , for get ops or something . But yeah , very cool , very cool .

Once again , thanks for coming on the show , andrew , it was great to talk to you and , um , if you liked what you heard and or saw this evening , uh , make sure you follow everybody involved on all socials . Make sure you send this guy a postcard several , if you could afford it . And uh , uh , you know , buy our cereal .

Um , do all that good stuff and uh , and we'll see you next time .

You can also visit our website for all of the show notes at CablesToCloudscom . Thanks again for listening and see you next time .