work with me: music & chat

ស្រាន់ានិត្ស្ម្ន្លាន សេរាន់ម។ okay so if you're joining this live stream now or you're just watching me uh just to give you context of what i'm doing i decided to stream what i'm doing at work well i'm just working but so i decided you know let me stream what i'm actually doing um and just chill at least to hang around and chat with all of you if you if all of you are online right now somewhere just wanted to test out all this streaming stuff and see what that would look like so currently at work well at home working and happy to chat anything about ai or just any topics really but what i'm doing right now is just I'm doing what I usually do at work, which I can tell you a little bit about what I'm doing at the moment. And then just being able to chat with you all while I'm listening to music as well. Also trying to work and get some work done while you're just, I guess, with me listening, then that's even better. So let's get the work done. So what I am doing right now is I am like the code on my screen right now and I'm playing with like different configuration let me know if it looks really weird on like TikTok or any of the other vertical platforms because I'm experimenting with like streaming it vertically and I'm also streaming on Twitch, YouTube, Instagram and TikTok right now and I'm just trying to experiment how it looks like on all the platforms I know for like YouTube and Twitch it should look good because you should see the full screen but for TikTok and Twitch it should be the vertical looking thing but what I'm doing here is there's a blog that not a blog actually what I'm gonna be doing is I'm writing a blog but what I'm doing right now is I'm going through this code base which you can see on my screen and what I'm doing is it's a code that I wrote a while back for the ai agents and if you're not familiar with ai agents this is also like another good opportunity for me to maybe introduce that to you have context so ai agents kind of allow you to automate a bunch of stuff so and also let me know if the volume maybe is overshadowing what i'm saying as well i'm experimenting with like making sure that you can listen to music but also not overpower my voice so what i'm doing with this code it's three agents and let me just go here to show you what it looks like so it looks like so yeah i have three agents and there's a infrastructure mapper there's a security analyst and then there's a report writer so essentially what they do is if you've used aws and you have like resources on aws for example that you usually would use for i don't know whatever anything that you're deploying to aws in this case so if you have an app you have to deploy to aws you have to spin up some resources on aws it could also be gcp but in this case it's just really aws so when you enable those i guess functionalities usually sometimes if you forget to close them out they'll of course eat up your credit but bottom line is this agent allows you to be able to just kind of explore all your services that you have spun up on AWS and basically just tells you you know which ones are not secure what you need to secure it gives you like a security report essentially so you have a infrastructure mapper that will map all your infrastructure that you've spun up in audio interview us to make sure that you know it looks good and then you have the tasks which are like the tasks to be assigned to those agents to do and those are ai agents which is basically agents so it's just like a person you are the job you have a job description and what's expected out of you so that's what i'm assigning here to the agents and then i'm giving it tools which here is the aws tool which is a function really that essentially gives access to the agent to log into my aws Get all my resources that I'm using on AWS and then have another agent that actually goes through all my infrastructure and looks at where potentially vulnerabilities are. Please, I can't speak today. but it maps up and shows vulnerabilities that could be out there and tries to kind of give me suggestions on how to fix those so that's the goal of um of this so pretty easy agents but what i'm doing is i already ran this it looks pretty good and i think i have an example here report so here's a report and let's actually see it on a markdown section so because he generated like an executive report for me and this executive report basically just looked like tells me exactly what's wrong with my with my account in a sense so it's like hey this account has some issues and i practically to make this work i went ahead and added some just things that i know that this is wrong it should flag this just to test to make sure it did flag so did identify some things that are wrong and they gave me things that i should fix so security components risk description it gives me the risk uh for example this one is saying uh unintended public access for s3 buckets and impact also likelihood high priority critical and it gives me also the steps to fix it as well and which it goes to another agent kind of goes online looks up documentation from aws that is recommended to fix that this could include things like blogs on how to fix some of these vulnerabilities so that's the goal of this but what i did i we wrote this code it worked pretty well so the idea was to partner with of course obviously somebody some company which it's obvious at this point uh and we're gonna write a joint blog which is what i'm working on right now is writing a blog explaining how this code works um so that's what i'm doing and that's what ii'm gonna be working on on the back in the background

and in a second here i'm just gonna let

the music play and i'm just gonna be quiet and work until somebody say something in the chat then i can jump on and say hello or something but um yeah feel free to chill with me listen to music while i work um if you have something to if you want to say hi please feel free to say hi happy to chat cool let's see so that's all for now so i'm gonna turn off my my mic and i'm gonna turn up the music

ទំទំ ទៅទៅ ទោទោ

ស្រូងស្ក្លើង្នៅលើម្ម។

ទ្ទ្ ទំទំ ទៅទៅ ទោទោ ទៃទៃ ទែទែ ទេទ់ ទះទះ ទ់ទេ ទៀទៀ ទ៊ទ។ ទ៌ទៈ ទ។ ប្រុង្ស្ល្នៅប្ម។ ប្រុង្លាន់ប្ម្ន្ត្ប់ត់នែលាម។ បក្វានបៃតែរាមនៃរាតៃនោះ។ ម្រាន ប៉្សះតោមែនិត។។ ស្ញុំបោក់។ ក្កះបែវាម បេ� ទ្ទ្ ទំទំ ទៅទៅ ទោទោ ទៃទៃ ទែទែ ទេទ់ ទះទះ ទ់ទេ ទៀទៀ ទ។ ទ៊ទ។ ស្រុងស្ក្លើង្នៅលើម្ម។

What? I

just realized some of the chat is not coming through on my end. So I just went to the platform, for example, I went on YouTube. I see Miguel Bello, you asked me a question, what are you working on? Tarell, what's up from Germany?

what

i'm working on yeah i am working on some code i wrote uh actually end of last year um it's supposed to be for a partnership that we're doing with a company here and especially these ai agents that kind of audits like does a security check on your aws infrastructure so if you have some aws stuff that you spun up maybe some services so sometimes maybe you don't configure it correctly and you might actually get some vulnerable i can't say that word vulnerability issues uh maybe i should just find a pseudonym for it but you might have some issues with it but you might not realize that you have some issues with your platform or your configuration or maybe you just want to optimize it make it better so these agents essentially they run a scan of your AWS setup all your settings that you've added on there all your configurations What it does is it goes ahead and generates a final report which I have here as an example. AWS account with some services that we know were wrong to begin with so we expected it to fail and let me just make sure the voice is not kind of like the

music is not overpowering my voice

but yeah so we added some services that we knew were wrong on our AWS account and then we ran these agents to make sure that they identified and they did identify it and we got some like this report right here you see there's like a full report with like issues it found and ways that we could mitigate it On the part of mitigating it, we have an agent that kind of goes online, searches the web to find actual answers to kind of help rectify some of those problems that we might have with our configurations, for example. And that's why it links, you can see here, like it references where you found the information on how to fix that, for example, that RDS instance here. So the issue here was like public accessibility, which increases exposure to unauthorized access. So restrict the network access and enhance encryption and tells us how to do it with like solution there. So initially before the AI agents, probably what you would have had to do here is maybe hire a consultant that would come look through your entire infrastructure and figure out for you, these are the things you're doing wrong, let's fix these are the remedies, but the agents now can just do it pretty much pretty fast in a few minutes with less money in my opinion. um of course there's still a lot of like human in the loop there so human needs to be there to verify and make sure that whatever is there it's actually good um and let's see uh we do have uh just want to make sure that this is good and uh supreme uh you mentioned that are you a full-time content creator oh man i wish uh i have a full-time job actually this is just that my job is more like a content creator at the same time i'm a content creator outside work but my job is devra which is essentially it rotates devrel so if you've heard about devrel or developer advocate that's what i do i've done that several now several years so what it devrel is is uh maybe i should just kind of like spin up something here to explain what it is right so let me see escaladro spin this up because i like explaining with diagrams so why not right let me do that right now um spray on my screen and actually let me share another screen here i'm gonna add another source and i want to capture window capture a browser source let me do that oh that's not the one okay never mind never mind delete that source man you can tell i'm i'm starting to learn how to stream i'm not a content creator like that i usually record youtube videos and just share it um maybe this is best so uh yeah let me just run xcode ស្រូងស្ក្លោ។ That's not the issue. අපි ස්තූතියි សុន្រាន់ស្ស់។ ស្រាប់ប្នៅ។ ម្ម្លាបស។ ក្វុង្តែលាម។ ទឹក្ក់ក។។ It's kind of like chill and get some work done.

ស្រុងស្មាន់ត្ត់ន្ល្នៅ។ ទ្ទ្ ទំទំ ទៅទៅ ទោទោ ទៃទៃ ទែទែ ទេទ់ ទះទះ ទ់ទេ ទៀទៀ ទ៌ទ។ ទ។ ស្រុង្ស្ម្នៅម។ I think

I know what might have been happening. Let me know if you can hear me now. I think I had two audio sources running let me know if you can hear me I lowered the music a little bit hopefully you can hear me um I don't want to chat if you can't hear me let me know in the chat if you can hear me all right there you are all right cool that's good thank thanks for letting me know man um yeah so what i was saying uh in my little box where you couldn't hear me was um what's your what's your use case what's your use case uh why would you need a gpu i think for me personally i haven't used a gpu locally um because most of my agents around them on my cpu um i don't i haven't seen a need to use gpu on my use cases But if you have one, I mean, I'd be interested to hear what trade-offs you're seeing with using a GPU. I think it also depends on the workload that you have. If the workload that your agents are doing or the tools that your agents are going to be using have a lot of I have a lot of like heavy lifting that needs to be done by the by the agents in terms of like services that it needs to run. So these agents will eventually use tools to run some of the stuff. So if those tools require like very huge compute resources, then I could see how a GPU could be really useful there. I haven't played around with like seeing the trade-off between the two so um really interested in your use case though um have you you know have you have you tried both or are you just kind of uh asking in general all right i do have a few minutes here left before i gotta hop um that's awesome we might be restricted by our client to only scpu oh i see i see yeah i mean that's fine uh i think at the end of the day uh oh you want to use gpus oh that's interesting um we'd love to hear your your viewpoints on that though do you find it i mean you've only used gpu so i guess that's kind of like the the trade-off right um i've only used cpus and it's and it's fine if your clients are requesting cpu then that's fine what i would maybe say is um for long running processes that agents have to run they just have to understand for example like you're doing a bunch of research and generating like a read group report or doing something that really involves a lot of work that could take a while to run i just like just make sure you let them know to be aware that you know it will take a while don't don't just expect it to respond as fast as like a chat gpt or something so that's what i would that's what i would caveat i think andrew ang covered it pretty well when he meant when he was talking about you know if you have a manager and they give you a task don't you know they don't expect like to give you a task and then like 20 minutes later they come to you they're like hey are you done especially if it's a long task right so same way i think we should we should treat agents at least as of now based on the resource and the computer we have even even using something like open ai researcher right the operator that they just released I mean it takes a while they say it could take up to 30 minutes to do a piece of research right so i think it's like patience if you want high quality stuff it's gonna take a minute because there's just a lot of sources to sift through there's a lot of especially with this model's chain of thought has to think through what it's what all the sources are just gathered so the more sources it gathers or the more work it has to do and thinking it has to do then you just gotta give it time because it needs time to think same thing as a human being you know if you're trying to come up with a solution for something It just involves you like thinking about a solution kind of like really from fundamentals think about a solution and then critique that solution that you just thought about and figure out okay this doesn't make sense this makes sense and then iteratively like loop over your solution until you find a solution that you're like okay I think this is robust now let me give this as a final answer so it's still gonna take time regardless of your gpu or cpu in this case because especially if you're using like models that are hosted by somebody else like openai anthropic and them but if you have your own local model then that probably would take even more longer uh time but if you have like your local model and you're using gpus then that could be a little faster you know um if hosted on a gpu but few quick thoughts there but if you have a client of course it's just gonna really depend on their they have what they have what their requirements are security posture what they recommend because at the end of the day it's their product right that you're building for them um but yeah good question i've never done like a comparison side by side with a gpu really uh it'd be really cool to see that with agents ngay ngay cool well it was so nice chatting with you all today i'm just hanging around i'm about to head out and i might stream later or tomorrow for sure tomorrow i'm trying to do this at least once a day uh at least for an hour well today i think i did what how many hours did i just do uh i think that's just about an hour and some change um which is not bad but i think it's yeah i think it's uh just over an hour an hour five minutes about so not bad we'll see you tomorrow um thanks again