Learn Cybersecurity with Generative AI

Key Takeaways

what's up YouTube I'm Brandon Caroll and I'm joined today by Mike Chambers how you doing Mike I'm good thank you thanks for having me on your show Brendon thanks for being here now Mike you are a AIML specialist I know you're a developer Advocate but yep always seems that way and your picture's on the screen yes that's me yes that's Mike so uh today in our video we're going to be talking about generative Ai and using it as a tool for cyber security learning because I've been reading a lot about this lately and people are using it because I think of the interactivity and so being the specialist you are and large language models and how they can be used and where they can benefit us I thought it would be interesting to talk to you about it to see you know what your thoughts are on using it as a tool and kind of how it all works sure so let's just start with that what are your thoughts on using generative Ai and large language models as a learning tool yeah absolutely look I think they could be very powerful to be used as a learning tool to be able to especially with chatbots right you can go to chatbots and you can say just help me understand some Concepts um but I think at the same time as you're using them is just as long as you understand what they are so they're um they're large language models they have a reasonable understanding of language um but where is it getting its data from so as long as you sort of like have all of these pieces aligned in your mind then yeah I think they can be a pretty power tool yeah okay so let's just start from the beginning on this and sort of lay it out there because I I've used chat GPT to ask questions and get answers back and I think it's pretty interesting but I know that I'm interacting with something else that's a chatbot on the front end what do we when we're when we're using generative AI as a learning tool what am I interacting with on the back end yeah absolutely so so well what you're not interacting with is a search engine which is showing you results from the internet necessarily it might look a lot like that but that's not really what's happening so um large language models which are the generative AI language models which power things like chat GPT and other chat Bots they have been trained on masses amounts of text massive amounts of language so initially they start out and they really can't do anything at all then you start to push data in you start to push language in so all of the pages of Wikipedia everything that we can scrape from the internet LLY is what we'll find is put into these models and they use that to start to learn about language and how language is constructed and then we keep training them and we keep going and they get bigger and they start to understand some ideas about the world like the way things work and then we keep adding information and we keep adding more stuff in they get even bigger and then they start to exhibit some of the properties that we can see which are the amaz amazing emergent properties of being able to properly answer the questions that that you have so in a nutshell um it's taken all of this training data it's codified it codified it it's got the statistical relationships between words and it can do a reasonably good job at approximating that it knows what it's talking about okay okay so that sounds pretty cool and it sounds like it's it knows a lot or it can figure out a lot I guess is what I heard um so what I did hear also from that is that you feed data into it so like for example if somebody wrote a blog post about a Cyber attack and how to mitigate against it today um that wouldn't necessarily be in there until it's been given that data is that right that that's absolutely right yeah so training large language models training generative AI models is a huge undertaking so um large amounts of computes long periods of time and actually it's quite costly to do so these Foundation models are created um and they are trained at a point in time and it's whatever data was chosen by the developers and obviously they couldn't choose future data because that's not a thing that we've figured out how to do yet so they take all of that data that they've got at that point in time that they want to use and they train the model on that so yeah it's not going to know about something typically which happened yesterday unless we start to augment it and build it into a different kind of application gotcha okay so it sounds like if we were to want to learn something that was related to like a certification and the certification has been around for a little while odds are that the model will probably have that data in there and I'm getting some pretty good information back from it that's quite possible yes I mean we need to also check and make sure don't necessarily completely rely on the information coming from it but yes if it's uh if yeah like you say certification lots of people write lots of things about certification so likely it is likely that it knows quite a lot about that okay and if I am using it to learn I think that cyber secur is a really really important area of ab it and and and working in Tech and so obviously the information that I'm getting back I want it to be the most accurate that it it can be is that ever an issue when we're working with these models sure so um again so it's understanding how what the model's doing so it's got all of that understanding of language now when you go and ask it a question what you're doing is you're prompting the model so you're constructing a prompt um and you're sending that in all that models actually doing and I I actually find this fascinating and and super super exciting all it's actually doing is predicting what the next word might be and then it will predict what the next word might be and then on and on and on it goes so if you're used to like using some chat Bots for example then you see that the generation comes up word at a time that's not like a special effect that it's doing to make it feel like it's typing it's actually figuring out word after word after word and what what the uh what the output might be so it's really uh performing a mathematical operation as to statistical operation to think what likely is the next word that I need to show and so in some circumstances it might just show you something that feels reasonable rather than is actually correct people call these hallucinations it's a little bit unfair really it's just an accuracy problem back in the old days of machine learning as far back as like last year we'd be training like a image classification model and if it said this is a dog when actually it's a cat we'd say that's a accuracy problem and that's exactly what we're talking about here as well so again it's a machine learning model it's not your friend necessarily it could be um so just be aware that it's using mathematics to try and figure out what to say next okay so I guess moral of story is could be used as a as a tool to maybe augment your learning to to add to your learning um you're still going to want to find avenues that are going to get you the latest and greatest when it comes to cyber security threats and and things like that but um there's a lot more that you can do with this than just ask it questions and get answers and learn from it absolutely and there are definitely ways that you can augment it build it into larger applications and then start to get some authorative truth from it as well so there are things you can do okay well hey this has been great talking about using generative AI for for learning about cyber security but before we go Mike where can our audience go for for more information to learn more about generative AI absolutely well um on the screen here we've got the AWS community space so community. AWS and if you go for/ generative aai you get to this page and yes that's a picture of me um but we have lots of content up here about generative AI including the course that we put together about large language models awesome thanks again for being here and giving us your insights we're going to actually talk in another video about some of the things that you can when it comes to cyber security with generative Ai and maybe what some of the possibilities are so tune in for that video if you've liked this video make sure you like And subscribe to our channel so that you can uh see more videos just like this one and Mike thanks for being here with me today really appreciate your Insight on this thank you very much thanks