AI Superpowers Stream with Igor Pogany

And we are live. Hopefully we'll give it a few minutes for everybody to catch up and the streams to go all over the web. Quick disclosure, I hear myself twice. But it's not going to ruin our nice, lovely stream because I invited a good friend of mine, Igor, today from the AI Advantage, and I wanted him to talk about his lovely channel and about all the amazing things we can do with AI. So allow me to introduce you to Igor.

Hey, what's up? Hello everyone. Good to be here. This should be fun.

Yeah, it's gonna be amazing. Amazing. So Igor and I, we met at the GTC and it was an amazing time. So I know Igor, but you guys don't know him. So Igor, please introduce yourself.

Yeah. Hello everybody. So you know I'm a fellow content creator. I suppose that's how we met. But for everybody out there, I spent pretty much every day, every hour of my life looking at brand new AI releases, making sense of them, researching them, testing them, and presenting back to my audience on YouTube what we find along the way. It's been a real journey across the last two years. But it's interesting. I don't want to go too deep on how it, how it all started and everything, but it was really interesting because like I was one of the first people on YouTube to upload around this topic out of genuine interest. I had five other YouTube channels before this and on like December 1st, I think my first upload on this channel was like December 4, 2022, now 2022 and just uploaded ChatGPT videos because I thought it was fascinating. And then all of a sudden like January 23rd and February 23rd came around and it just exploded. Exploded. People caught on to the. Caught up to the fact that hey, this is actually a game changer. And yeah, here we are. Made it my full time thing. I built a team of, what is it, 10 people now and we're just trucking along, researching AI, teaching everything that relates to it. But it's very different, it's a very different deal than what you do on this channel, Maria. Because we go after the non technical audience, we kind of, we assume that people don't know how to code, whereas with you it's sort of the opposite. So it's interesting. We are sort of in a similar ish niche, but not really at all. It's definitely interesting to talk to you about these topics because you have a different perspective and I can't wait to chat about whatever you bring up today or whatever comes up in chat.

Yeah, absolutely. So here we Usually make AI, we don't use it, but today we have an exception.

So you have a channel for two years and you have multiple million viewers type of videos and basically way more subscribers than me. So congratulations.

I mean, not that that matters, but I don't think that's even true.

It just indicates that you're really, really researching your field and people are enjoying your work. That's why you're here. And if you guys have any questions in the chat, please leave them with a few stars in front of. Let's use three stars so I can see the questions and I can detect your messages between one another there, too. So, Igor, I wanted to ask you something. You actually answered somewhat okay. You said it's your genuine interest, but I want to see if there was any other component that made you start your AI journey. So you had channels in different topics and then what makes you. What made you think that, you know, the next one should be AI?

Yeah, bet. So, like, I think, you know, to really answer that, like, I, you know, I don't want to do the typical long answer thing where it's like, yo, on a rainy day. Back in 1998, I was. I was sitting in Austria with my brand new phone and. And like, But. But it's true. Like. Like, even as a little kid, when I got my first phone, my first intuitive reaction was like, opening it. Because back then you had to, like, open the. No, that was even before those phones. I had like this, like, black and white. I had the indestructible Nokia. Remember that one with the 3310 or something? And then, like, usually the kids would go in and be like, oh, my God, snake, right? And they would play. I was the kid that went into the settings and had to open every single sub menu, change the language, see what changed. Later on when cameras came around, I got obsessed with like, ooh, trying the different megapixel counts and how is the quality and what are the settings. I spent more time in the settings than in the games or in the contacts or in anything. I just like tinkering with things. I like figuring things out. I had this, like, my mom got me this robot building magazine when I was nine, I think, where every week you get. It was really cool, actually. Every week you get like a new component and you got to put it together. You learned the basics of putting robotics together, and the magazine explained the steps. And all of a sudden you had this robot who could follow a black line on the floor, and you had a little remote control and it had a speaker and you could put presets in it. So that's just me. I always love this stuff. So, you know, as I went through life and as I kind of, you know, entered adulthood and all of that, whatever I did, this part always stayed constant. And whenever I found a new tool, I just wanted to tinker with it naturally. I wanted to see what it can do, I wanted to break it, I wanted to push it to its limits. And then it turns out that it's quite useful to do that and then share some of the findings with others, because then you save other people the time of doing it, and I just do it naturally anyway. So, yeah, that's really the answer. It's that. That's the core kind of driver.

So you're a curious person from birth. And I'm just pulling out my headphone. I'm not going to be able to hear you while I speak, because you know, why.

So, folks, super happy you're here. Thank you for joining us. We have lots of comments here. We. We have people from Ethiopia, we have folks from Brazil. We have. Do we have our regulars here? Let's see. Badger. No, Badger is not here, but we have Nick, who is a member, and he thinks he's seen you before.

Doesn't he review all the new AI that's coming out and if it's useful?

Hopefully, yeah, that's the whole tagline. I was looking for a way to differentiate myself from the avalanche of AI channels that has showed up over the past year here. And, like, the thing that I found is just the essence of what I just kind of said. Like, I. I like using this stuff, even though it might be useless in some cases. Right. So, like, I built this whole show that I named AI news that you can use. It sort of has a good flow to it, by the way. We came up with that on the. On the trip around gtc. We went to another conference afterwards. And like, a friend kind of, I was describing to him what I do on the YouTube. I was like, oh, so it's like AI news that you can use. I'm like, that's the line. We're gonna use that. But, yeah, that's what we do. We kind of like, filter out the drama and all that and just focus on the tools that are available today that you can actually put to work. And, yeah, it's fun. So I hope it's useful. That's my hope.

Yeah, it sounds like it. If folks from the channel know you, then, yeah. And Badger is indeed here. So Igor the reason for which we have gathered here today is what is the coolest and the most exciting thing you've ever done with AI?

Okay, let me think that. That's a big question, so I know.

So should I answer that for myself or like, for something that you might be interested in?

No, for you, for.

And to be fair, folks, Igor did not prepare for this stream. We actually, we talked about it about a month ago, but yours truly here took too much time to book it. And then a few days ago, I'm like, igor, my streamyard subscription is about to run out, so would you like to join me on a stream in a few days? And he said, yes. So.

That'S good. But I think both of us are pretty good at kind of shooting from the hip and we have this very fluid communication style where we can really easily adjust. And I think I have a good answer. So it's a hard one because there's a lot of like, cool things. But like, personally. Personally, the thing that blew my mind the most or. No, the thing that has been a net positive for my life more than anything else I've tried. It's just using ChatGPT as kind of a therapist counselor. I really. And I know it's. I know there's a whole discussion around that, but. But I just feel like it's a great supplement to everything else we already have in life. Even if you have a loving family and supportive friends and maybe a partner to go through to talk through things, I think there's still value in having this completely neutral opinion that, hey, if you. If you want practical advice, you can tell it that and it gives it to you. It's sort of. I don't know, it just. It just works for me. There's no judgment. There's no. It's just. It just helps. You can kind of just brain dump if you have a yet. Yeah, it's true. But you can kind of just brain dump everything that is on your mind. And if you have a lot of things going on, it doesn't have to be just personal. Like, I have, you know, I have 10 people to overlook and I have personal relationships, I have family relationships, I do different activities, I plan my travels and like, spontaneous. I don't know, life gets complicated sometimes. And sometimes it just really helps me to dump it all out with the Voice feature and ChatGPT and then I have a few prompts that I like to run on top of it. But there's nothing really. Nothing really set in stone. I suppose people will be curious. So Maybe one thing that I do is I just dump into it and then I ask, what are some actions that I could take to remedy some of these worries? And another one is, what are some blind spots in my thinking that I'm missing? I found that to be really powerful because then it can sort of like, poke holes and take different perspectives. A lot of people do this thing where they set up, like, a Persona with it and then they talk to the Persona. But that's a different way to get to these blind spots. Right. But if you just generally ask, like, what are some blind spots in my thinking? And you give it your thoughts, then, yeah, I legitimately use that regularly, but there's not much to teach there because you just hit the mic, you just brain dump into it, and then you ask the questions that are on your mind. And yeah, if you're aware of your thoughts, you can transfer that into the phone. And that has been the biggest net positive for my life.

Awesome. Awesome. I usually use tarot cards for that.

Question, and then I pull them out.

Look, this is the interesting thing. I've been talking about this particular use case to many people, actually. And generally speaking, like, this is. No, I haven't conducted a study or anything, but this is just my personal experience. This seems to be more. Way more useful for guys than it is for. For girls or, like, women, like, straight up. Because, like, maybe you.

I believe you.

Yeah. Yeah. I don't know. Something about guys venting to AI is just like a match made in heaven. I don't know.

Yeah. If we have a problem, like, we just call one of our girlfriends and it's like, you can't believe what he did to me. This is terrible. So, yeah, there you go.

There you go. I don't really have a girlfriend that I can cry to, and I think a lot of. A lot of men will relate. So, you know, guys. He listens.

I guess for guys, it's embarrassing to call their friend and tell, hey, I had a tough day. Just man up, dude.

It doesn't. Yeah, it doesn't. It doesn't work that way with guys. You know, go out and you, like, make a backhanded comment or something and you're like, oh, she's a bitch. Anyway, like. But, like, we have our own ways of handling it. But this is. It's just really useful, especially with giving practical. So that's like a net positive, but it's not, you know, it's not like this fancy, sparkly, amazing thing that you might have been hoping for.

But, hey, there's a lot of people with mental struggles nowadays, especially in the past few years. And if it's something that can help you without being embarrassed, without going to like a psychiatrist and paying a lot of money, that's probably a cool thing to do. Right? We couldn't do it earlier before we had these like fancy large language models capabilities.

So I agree, it's cool. But let me, let me be clear. I just wanna, I just really wanna make this point even. Like I, you know, I'm a big fan of, of personally I'm just super interested in psychology and psychotherapy. Just, just like personally. But, but I, you know, I go to therapy since a while and I, I have friends and family and all that. I just, I'm just saying it's a great supplement to it too. Just breaks this, this like neutral opinion to it. So, so I think it can be both. Yeah, it can, it can be a replacement for people who might, who don't have the means to engage in therapy, for example. But also as a supplement. It's great.

Awesome. Cool. So we have a few questions from the audience. Let's start with the first one without my earphone. So what were some of the biggest challenges you faced while learning to leverage AI and how did those experiences shape the AI advantage? Oh, wow, that's a nice question.

I think I have an answer. That's a great question actually. Really good job there. I think the biggest challenge is translating your life experience and how that forms your communication and teaching that to others. Because at the end of the day, a lot of the usage of these tools comes down to communicating with the damn thing. Right? I don't know. In your case, if you're prompting something like Sonnet 3.5, that is really good. In the code generation, you're going to need a little bit of context and experience on how to talk to a fellow coder. You're going to need some technical knowledge to do that properly and that you can't really teach. Right. For example, I know a lot of people just hop into ChatGPT and they're like, oh my God. Social media marketing solved. This is amazing. It's going to write my TikTok scripts. I'm going to go viral. You know, I never wanted to do this, but with the help of AI, I can do it. And then they go ahead and do it and they produce the trashiest content in the history of humanity. And then they're like, oh, AI is terrible. No, bro, your social media marketing skills are, I'm sorry to break it to you, but like you need to use it elegantly, you need to use it, you need to have a base understanding of what you're trying to do.

If you don't have that, we're not at this point, that will come in the future at this agentic point where you just kind of hit a button and it does its best and it does it well for you. No, we're at this point where it's very much an assistant. And if your social media marketing skills are not up to par, not even the best assistant in the world is gonna remedy that. Right? So that was the biggest challenge to me because you can teach how to structure prompts, you can teach how to break it down into building blocks, how to add multi shot prompts for it to follow a template, you can teach certain techniques and differences between a reasoning approach and a normal prompting approach and a goal based approach that works better for these genetic frameworks. You can teach all that, but you cannot teach life experience. You cannot teach how to be a. You cannot teach the basics of development. You kind of have to pick that up on your own. You cannot teach the basics of social media marketing. You kind of have to pick that up on your own. So that has been a real struggle because I'm trying to translate some of these like semi technical skills to people who have no experience with them. But often there's also this real world experience components to it which you, which you cannot teach. So that has been a massive challenge. And what I found and how I solved that eventually is I just clearly communicate that you have to see it as an assistant and like pick the things that you're already good at. And there's this, actually this is one framework that recently came up in the community which I think is super helpful.

It's sort of think of prompting as if you hired a very skilled freelancer from a different part of the world and all you could do is send them one email. One email, the entire assignment, everything you wanted to do, wanted them to do, you had one email. No back and forth, no corrections, no revisions, nothing like that. You have one email, they're really good. But you have one email and you need to communicate everything. If you manage to do that and if you manage to set up your prompts in that way, you're going to be successful. Because let's be real, like Maria, if you hired a freelancer, how would that email look? It would probably not be a one liner with like, hey, make me awesome web app, good luck. Right?

It would probably, yes, my emails are very Long.

Oh yeah, there you go. So like. Yeah, and that works because it gives the context they need. Maybe ideally you would in the end include like, hey, here is free other websites that are built just like the ones I described. I want you to kind of copy the layout of these. And hey, what about like you're going to write the copy? Sure. But like here's, you know, three more examples of copy that I really like. Right. If you do that, the freelancer is going to do a better job. But for you to write that email, you need a little bit of like life experience and transferring that is not so intuitive. So AI is kind of an amplifier and if there's nothing to amplify, it's hard to teach.

Yep, agreed. Awesome. Thank you so much. We have another question with three more stars. Where can I find, where can I find out more detailed information about costume GPTs? They stopped working as they should since the beginning of Sum24. The topics are specific prompting and second is game design.

Interesting. Okay, so let's tackle. There's a few things in there. So first of all, they stopped beginning since summer 24. I think you're referring to the fact, to the point in time where they switched it from GPT4 to 4.0 in the background for all GPTs, which like slightly degraded them arguably. Obviously 4.0 is like, you know, supposed to be fully multimodal one day, but.

It'S a more efficient model that is cheaper for them to run and, and arguably a bit worse than GPT4, but only in specific use. You have to get very specific to really nitpick that. So my question would be like, why is it necessarily worse for you? Because from what I've seen for myself and for the entire team and all the various GPTs we have, I always like to include them in the workflow if they, even if it's not necessary, just you know, to get practice in and to get everyone on the team kind of up to speed. So like we use them a lot and we still use them like nothing about our usage has changed from before and after. You know, arguably like maybe a single digit percentage degradation in quality maybe in certain use cases. Okay. But I think they're still, they're still as good as they were on release, which is like not great, but useful. I think there's so much like on the GPT product, the ceiling, the potential that that product has and that kind of, that interface is I think immense. I think this is the agentic future. This is going to be the feature that will turn into this agentic idea. Eventually I think these GPTs will morph into just, you know, tell me what you want to get done and the GPT will get it done for you. And now we're in this very primitive base version of that. You can set up instructions, you can give it a knowledge base that, you know, doesn't really fully process. I would be careful with that. And then you can have some API calls that you activate from the GPT. Now, to answer your question here, that was a bit of a long preface there. But to answer your question, I think they're useful, but I think they're only useful if you custom built them for specific use cases. I don't think the idea of the GPT store really worked. Maybe there's like a handful of exceptions. I don't want to bash any GPTs out there. We have a few of our own out there too, and they're great. But like, I think generally the most value within GPTs is if you custom build them for one specific process. You solve one problem that you run into regularly and you set it up that way. You set up the prompt exactly the way you want it. You give it templates in the knowledge base and you give it actions to interface with your workflow. I'll give you a simple example. So like regularly we look at repurposing content and just creating. We create like internal summaries also. Those are two different GPTs. So for example, we take like the transcript from a YouTube video and then we throw it into OpenAI. Now we have automatic pipelines for this too, where you don't even need a GPT and it just happens. But with the GPT it's still really useful because everybody on the team can do it, right? You don't need to go into bake or N8N or whatever and tinker in there. So let me keep it simple with that workflow, we use it all the time. And what you want to do is you. The actions are not that big of a game changer as many people make them out to be. I would say it's really nice for it to interface. For example, with our notion, we organize everything in notion databases. So we set up the GPT that does the summaries in a way where you can just add the transcript and then it presents you with a menu of keyboard shortcuts. And on the keyboard shortcuts I bound different prompts that we like to use within that workflow. So then the user just goes in, drags and drops the text file file which we also automatically generate. But he just drags and drops the transcript into the GPT and then he gets a menu of like four prompts that you could run on top of it, all within our custom crafted, templated and everything. Then you hit one of those prompts and if you like that result, we have another keyboard shortcut. Usually it's five. That sends it over, activates action and sends it over to our notion. So all of a sudden this workflow that usually would involve like six steps of copy pasting and moving things around, you can just open up the GPT, put the transcript in there, hit a button, hit another button, and it saves the result to our notion database. And I think things like that, you know, it makes you a little more efficient every single time you do. But this is a thing we do almost daily and it's great to have a GPT for that. So those would be some tips and tricks if you want to learn more about it. If you check out our community, the whole thing is like a AI learning community. And one of the courses in there is building GPTs. I think it's like 8.5 hours. Everything I talked about and more is in there, but I think they're useful. They just have to be purpose built. I think if they're general purpose, it's kind of besides the point. And yeah, I think the summer changes didn't make that big of a difference.

Awesome.

Okay, cool.

So these type of stuff I don't know much about. If you guys are noticing. I'm sitting here like a deer in a headlight, so. But it's great. That's why I invited Igor.

No, but that's amazing. Hey, open up VS code and you'll lose me very quickly, you know.

Yeah, well, I don't use VS code.

Oh yeah, there you go. See, that's the first like. Yeah, I forgot. That's the first tell of HobbyCoder.

It just shows you that I don't like Microsoft.

Okay, what do you use? What do you use right now?

Whatever text editor that is open source and free and you know, doesn't require a lot of setup, it doesn't have a lot of buttons and is not controlled by Bill Gates.

Fair enough. Yeah, yeah, I actually agree with that ethos. Let's move on.

Yeah, awesome. Well, we have a bunch of other questions. One of them is about chain of reason. Now, I must warn you, Igor, some of the questions you will see here in the comments are by very, very smart people who ask sometimes questions that I know nothing about. Okay, so I'm just pulling them out from the hat and hopefully everything is clear. So this is.

This is great, Pedro. Thank you so much for asking. I think he's referring to a paper that was published in early 2023, I believe, which basically discovered that if you. If you go into ChatGPT and tell it the simple phrase of things step by step, a lot of problems, especially in the STEM fields, became way, way more manageable. And the percentage chance of the LLM actually figuring out the right answer actually like multiplied when it came to basic calculations, for example. So if you just tell GPT4, oh, hey, you know, two plus two times four, it will mess it up. A lot of times if you tell it's two plus two times four things step by step, all of a sudden you'll get like the correct answer almost every single time. It's like, what is this black magic? The black magic is basically that if it goes ahead and has this intermittent steps before it gives you the answer, it has a chance to kind of mull over its thoughts and, you know, to lean back in its share and to think about what it tells you before it tells you that it just doesn't predict token by token. And on this whole idea, they actually. One of the recent releases, 01, and namely they released two models, OpenAI 01 Preview and 01 Mini. These models are built upon this thought and they built the whole model around this chain of thought idea. Now they're actually really good, but it's very early for this line of thinking. And to answer Pedro's question, I think it's an amazing approach. And I actually just uploaded a video yesterday on the channel which is one hour long, which is very long for my usual uploads, which talks only about this brand new model and how to like practically use it. Spoiler. My favorite use case is rewriting text because it really considers the full. It considers the full.

The full, like the full idea of the text you give it and also surrounding ideas while rewriting it. Rather than just going token by token and replacing the words with others or maybe replacing the sentence, it just things holistically. So I think it's super powerful. I think it's super early in these models that are specialized towards that. But I think they're definitely a big part of this future where AI will just perform actions on its own because it's probably a good idea to let it think for like 20 seconds before it actually does something or presses a button. I like what you did. Good.

Green screen set up quietly putting some yeah, some of the things you're talking about in the background. I shared a link as well.

Yeah, thank you. So if you care to learn more about that. Yeah, essentially OpenAI kind of took that idea, built a brand new series of models around it and we're just like in the early days of this and it should be very interesting because as I also talk about in the video, it's just, you just get different. Like the examples are just higher quality and they're not like 10 times as good as. But they're literally the difference between a person kind of just.

Speaking their mind without thinking before it and somebody actually contemplating for a few seconds and then answering a question. It's not going to be ten times better, but it's going to be better consistently and it already is with these models. So yeah, I would recommend it. And also they're really good for code editing and code review, code generation. I think Sonnet 3.5, the consensus is still that that's just. Yeah, the best. But yeah, I think it's super powerful. It's the future.

Sort of perfect. Perfect. The present and the future and all of them all at once.

Yeah. You know what else it's good at, Maria? It's, it's really good at like telling stories. You want to tell a story? Do you have a paid chat GPT account?

No, but I'm using the. I have a bunch of videos upcoming with the Quinn model if you're.

Ah, yeah, yeah, the, the, the Chinese Express, I think. Yeah, yeah.

It's like free and it's available on.

Yeah, it works.

So you're talking in the background. Hold on, Igor. Sorry. As usually.

I turn on your video by accident and I don't understand. I don't hear you talk, but I hear you in the background. Holy smokes.

Yeah, no worries.

Technical issues stream.

Yeah, sorry guys.

Oh, we have a. Hi from Germany. Can I say hi?

I do. Yeah. Let me, let me pull it out. Okay.

So you're in Portugal right now, right?

Yeah, yeah, yeah.

Pretty close. There you go.

Yeah, pretty close.

Hi from Germany.

Yeah.

Yeah. I grew up in Austria, so like, you know, Germany's. Hi from Sweden. What's up, Matthias?

Nice. Awesome.

So we have a follow up from Pedro. I'm referring to something like Prof. Of synapse.

It's funny that you pulled this up back to back because actually matthias here is Mr. Professor Synapse in our community and you pull them up back to back.

Awesome. So talk to one another in the comments.

Yeah, Matthias, you can go. Go ahead and go Ahead, Matthias runs a lot of the events in there and like he's like, oh no.

No.

Yeah.

I don't know. Like we talked about in the community also with Matthias, we talked about this and we're like professor sign up sort of got replaced by 01.

I. I know what you mean. There's like a difference. But basically, I don't know. I think it's a good tool. I think it's a very beginner friendly tool, like Professor Synapse, if you set it up. Maria, by the way, I have to fill you in here. Professor Synapse is like.

Let'S say a prompt that you can give ChatGPT where you change its behavior so that rather than just answering it will spawn an expert that will ask like some extra questions. And it will, it basically goes into this process of like, okay, let's think about this a little bit before we give you an answer. And it helps people do that really easily. It's really intuitive.

But yeah, it's just a prompt that pushes it in this direction.

And.

Yeah, and I think that's that I just, in my usage, I don't know, I just switch between sonnet and 4o and 1 mini O1 minis. Actually I prefer it to 01 preview. Yeah but I mean those are specifics.

So just out of curiosity, this is my question, it's nobody else's question. How many hours a day do you spend communicating with AI?

Ooh, actually not that many. Not that like how many, how many hours a day do I spend with AI is like probably all of them. I'm pretty much like, no lifing this whole thing which is, you know, it is what it is. I don't know, I'm 30, I can still do this.

But yeah, I don't know. By the way, it's something also that I'm kind of trying to change because it like became my whole identity to do nothing but work over the past years. But like I'm working on it, I'm working on it. You know, I signed up for a stretching class last week. It's okay, I have, I have a second hobby now too. But no, like for real, how much do I talk to it like every day?

I mean it's a bit unfair because like we test a lot of these new tools, right? So like for example, today we sat in a room for three hours and we just looked at all the new releases and we can kind of like going in and testing these things. So if that counts then more, I guess then I would say maybe Maybe an hour a day or something. If I, If I were to disconnect myself from all the advantage activities and I just, I don't know, maybe were a content creator in a AI unrelated niche, then I would do. Would probably be using it around 10, 10 minutes a day on average. 10, 15 if I have a project, you know, it's just open for like a few hours and it just assists me, but then it ramps up. So maybe then the average goes up. But yeah, most days I find myself opening a tab of Claude or ChatGPT. Yeah, definitely.

That's awesome. Yeah, well, I'm inspired because most of the day I'm not working. I'm trying to. It's very. It's very, very hard.

Especially after you're just finishing studies and stuff. Right?

I finished already. I don't have an excuse anymore.

It's gone. Oh, no.

Exactly. I'm just waiting for my results.

But yeah, it was an important question. I was wondering, because it takes so much time to learn everything and be able to articulate it, that's arguably harder than learning it. How to explain it in a way that people like myself can understand. So.

That is true. That is true. But then I think you're quite good with communication, so, like, you shouldn't. I think you're a natural.

I would say, okay, if anything goes wrong, I start laughing and I blame my hair. I'm blonde, you know, what can I do?

I'm not sure that helps with debugging and stuff like that.

It's a great excuse. You'd be surprised.

Yeah, I can't do that.

Awesome. No, you can't. No, try.

Maybe if I go blonde.

Even if you go blonde, it's not an excuse applicable for guys. As far as I'm concerned, when I'm driving and I'm doing something stupid on the road, nobody will curse me. You know, they look at me and see, oh, blonde girl. They'll continue driving. You know what, what can she do? But if my spouse does the same, it's not gonna be the same.

Yeah, that's. That's a live. See, AI AI doesn't judge, guys. That's why you can talk to your AIs.

It's trustworthy only. Only busy drivers. Okay, one more question from a channel member, Nick. I have a paid account, I'm assuming ChatGPT. What are your thoughts on point one? It's so slow. Not sure that I notice a big difference in quality, though I agree in.

95% of the cases it's gonna be the same. But 20 times slower. So you want to avoid those cases. Now, what are the good 5%? As I said, the whole video that Maria pulled up and played in the background.

Accidentally. Yeah, that whole video goes exactly into what are the different cases. Let me highlight two of you that might be useful. Although I don't have any further context on your life here, Nick. But one of them is really rewriting. It's just so good at considering the entire text while before it gives you a result. Whereas GPT4.0 does consider it. Right. It's in the context window. It knows about everything that you gave it and the entire conversation, but it just doesn't.

It's a bit. It's a bit careless when it starts writing and it just goes token by token. Whereas O1, especially 01 mini for rewriting is really good if you just go ahead and you want to improve a piece of. Like, if you. I know many people use AI for this. They brought something and they just want to make it a little better. Right. If you just go ahead and prompt 4O with make this better and you prompt the same thing inside of O1 mini, make this better, and then you compare those Results, I believe 7, maybe 8 out of 10 times, you will find that O1 mini did a better job of that. So that's one case where it's really good rewriting. Another one is translation. And that's because, Maria, you will notice, like, translating things from foreign languages can be quite hard sometimes. Especially something very. It's really because it's not just the words. There's so much. There's so much emotion, there's so much background, there's so much context. There's so much culture within the language. Yeah, yeah, it's. You just can't pull up a dictionary. And O1 is really good at this because it's like, okay, here's the phrase, here's what it would translate to verbatim. But now, you know, let's look at what this phrase really means and how it would be used within a text. It looks at all that and then all of that extra thought it infuses the translation with. And all of a sudden you get something like, I translated. I tested it on a lot of these very unique German sayings that can't really be translated, or it's very hard to translate them. And if you throw them into Google Translator, you get something, it makes sense, but you would never use it. It's sort of nonsensical. If you use 01 Preview for that, you will find like, I had Multiple wow moments where I was like, hold up. That's a. It's a slightly different phrase. But this. This transfers the emotion. It transfers the thought behind it. Whereas for all would just take the words and translate, you know, and just replace them. So there you go. Those. There's two use cases if you want more. In that video I spent an hour kind of blabbering about you can use it for. But yeah, I agree, it's way slower. And in 95% of the cases you really don't need it. And then you just have a slower model that you're paying for that is equally as good or bad, which is like, what am I doing?

I think you brought up a problem that is not just in the realm of AI when it comes to language and translation. It is more.

In a linguistic field in the natural language processing category. So when you have sentences. I'll give you a nice example. You can say in English, they are hunting dogs. So is somebody hunting dogs? Or are these dogs belong to a breed of hunting dogs? Right. It's very hard to decide which word is a noun, which of them is a verb and all these things. Unless you have the.

Linguistic understanding and the context. Right.

What is it? Or you have the thesaurus in a separate tab open. What is that? That's one of the weird words I never wrap my head around.

I don't even know what thesaurus is.

You don't know the thesaurus? That's like the. That's like the dictionary of synonyms and antonyms.

I'm an immigrant.

That's one more reason to know the thesaurus and chat. Do you guys know the thesaurus? Come on. This is. You guys. I bet a lot of people here know the thesis. Please, can you. Can you. Can somebody back me up on this? It's a book or electronical resource that lists words in groups or synonyms and related concepts.

We are waiting for confirmation in the comments.

Please. Just like crickets.

Thesaurus. It sounds like a kind of dinosaur. Igor. I. I don't know.

Would you be surprised that the logo of the thesaurus actually is a dinosaur?

No.

Awesome.

Yes. Nick is just like thesaurus.

Yeah.

Did I pronounce it correctly? I haven't said that in a while. I'm starting to enjoy that word a little too much. Maybe we should move on.

I wouldn't even know.

But yeah, you're right. Yeah. Language is difficult and we take it for granted just just because we like speak it on everyday basis. It's just like the. Like the electricity grid. We don't think about the fact that this damn thing, like, keeps, you know, society alive, but it does, like, turn that thing off for like two days. And good luck.

Especially in Canada in minus 40.

Yeah, you'd be running so quick. Yeah.

So cool.

Keep that on screen, by the way, please. I like having, you know, mixing thesaurus in my corner here. It's good.

Perfect.

Perfect.

Nick, you're going to stay here. Okay, Now I have a question. I'm sorry for all these questions, Igor. It's not even a discussion. You barely have time to drink.

It's like an interview.

Okay, so my question is based on something we talked about about a month ago and you told me that you're looking for easy ways of web scraping. Now my question is, what exactly do you want to web scrape? What kind of data are you looking to get? Is it videos? Is it text? Is it a bunch of lyrics? Is it encyclopedia pages? I don't know, what exactly is it that you're looking for?

So my initial answer, which is not going to be helpful at all, I'll say it anyway, though, is all of that because, like, you know, often, like, it really depends on like, what you're prompting, but context is king. And you just. If you have an automatic way, like the idea behind me asking you was if you have an automatic way to scrape a page and can then automatically pull all of that context in and give it to LLM along with a prompt, then you can just get superior results because then it has more info to, you know, formulate it to answer around. Now, concretely, I've been thinking about this. Like, one thing that I'll be super interesting interested in doing is, is scraping like a little bit of online information around a specific person. It could be as simple as scraping their LinkedIn or something like that. Because we've been in the community, we've been discussing this. We're always like, kind of like hunting and looking for new ways to use this stuff, right? And one of the things that caught my personal interest is kind of a meeting preparation workflow or GPT or automation or whatever, where, you know, you're meeting somebody, maybe you know, you're familiar with meeting with sponsors or like other YouTube creators that you might know nothing about or whatever life throws at you. All of a sudden you have this meeting in your calendar and you know nothing about the person. Wouldn't it be great to kind of just go ahead and like, not just Google their name, but have this sort of tool where you can put in the name and it goes Ahead and, you know, maybe uses the SERP API to look for Google and Maybe looks at LinkedIn, pulls it all together. And then maybe let ChatGPT, you know, do a little meeting preparation for you where it pulls in the notes from the meeting and it pulls in all this info from the Net and then you can, yeah, you can kind of like get a little bit of context on the person you're talking to.

I thought that would be super useful. That's. That's the thing that we just discussed, like today. That's why I'm bringing it up.

But it's an amazing spying tool.

I guess. Fair enough. Yeah.

Like, I can call it stalker simplified. It's gonna be incredible.

Teach me how to. How to scrape, you know, all the government agency databases. How to scrape that?

Actually, it's funny you mention it. About a half a year ago, I learned in school about a. How do you call it? Oh, it's an SQL injection. It's a hacking technique that gets you to break into a data set, like a database online.

It only involves a very simple command that you type in the URL. For some reason, yours truly here decided to test it on a official Canadian website. I think it was Weather Canada. And I thought, there's no way it's going to work. It worked.

It worked. And I've seen the data set, so all you need to do is a bit of hacking. Okay. After the stream, I'll tell you what to do.

Maria, you're going black hat real quick here. I was just looking to automatically look at people's LinkedIn profiles. Jesus.

Hey, you talked about government sites, but.

That is, like, really interesting. Like, I've never. Yeah, and then here's the thing. If you combine that with, like the AI brain, you could be like, oh, look at all the weather. And now I'm not sure what you could do with the weather actually. But. But you see what I'm saying, it's like there's real power in, like, taking some data like that and then combining it with the right type of prompt. And that's why I was asking about scraping, because I just, you know, I might have, like, some knowledge and expertise on this side, but like, on the. How to pull the data inside, I just have none.

Yeah. Right now I'm thinking about a nice scraping tutorial, and this one is using a tool that makes everything unlimited. It can scrape sites that cannot be reached otherwise. Very nice company I worked with earlier. They're called Bright Data. Actually was in their head office. I met their CEO. It was really, really nice. So we're going to have a nice call up with them sometimes in the future. And I told them, yeah, I'll reply to you with a nice use case, but let me talk to Igor first. Like, I want to see what is it that people who use generative AI actually want to scrape before I continue with this tutorial.

Yeah, there you go. So it's like personal context is huge, right? Like if you. I don't know if you've seen like the Apple intelligence stuff, right? Like the way they're integrating AI, Basically, it's very simple. They just built like. And this is me kind of extrapolating from it, from the way we do things, but basically they just build a kind of a profile on you and some of your preferences, your writing style, your name, your age, gender, all those things. It builds a little profile on you. And then in their presentations they always refer to it as your personal. I think they use the wording personal. Did they use personal context? I'm not sure, but they always refer to it like, oh, and then it uses your personal context to like write the message for you. And basically you can do that like manually. This is like one of the things that I talk about a lot on the channel. You can set ChatGPT up with your own custom instructions. That's what we call it, that's what it's called within the interface. So if you have a little profile, which is like, hey, I'm Maria, I do this, this is my goal, this is the style I write in, this is the type of work I do. These are the projects I'm interested in. This is my communication style, this is my educational background. And a few more points. If you flesh those out, you can let AI do that for you with a simple generator that I can show you. But if you have that, you can easily copy paste that with any other prompt you send, or you just set it up inside of your custom Instructions and then ChatGPT knows all the details about you without you having to tell it every time. And that can be really useful because it just saves time whenever you're using it, it just has more context on you. And the idea basically behind the whole LinkedIn meeting preparation thing is building a profile like that on somebody else based on what they shared across the Internet. So, yeah, I think that'd be a good one to. Good idea to pursue.

Yeah. So LinkedIn will be relatively easy. It's not going to be very complicated because this is not a protected platform. They didn't block it from bots you don't need to SQL injection tool for that.

And so LinkedIn is relatively easy. Easy, but doing a full research on that person, on their name, like Googling it and getting all their social media accounts. I would say that the biggest problem is how do you know that you're not pulling information from a different person who has the same name? That's the biggest challenge. Because I guess it's. Well, I guess you can have a component that looks in photos and compares them, but in cases where photos do not exist.

Yeah.

So let's say somebody has a LinkedIn account and he's also a researcher and he published a paper on a different, you know, platform. How do you know that it's the same person?

God damn it, you're right. We need a unique identifying number. That is our name. What's going on? Bring on the future. No.

A barcode.

Yeah, Maria. Hashtag 735892. Just like the Discord IDs. Yeah.

The more we talk about all the.

Sci fi novels come alive.

Right? Orwellian reality.

Yeah. For real. Like. Like, how do we track the people if multiple people have the same name? We're gonna give them numbers and then we can scrape them.

Yikes. Yikes. But let's be positive.

Oh, yeah, of course.

There's good use cases too.

Awesome. There was another question I think I missed. Okay. It's from Nick, too. Nick, I want something that will scrape all my stock option chains. I am so sick of my API going down.

I don't know. That's a question for you, Maria. You're the scraping expert here.

No.

I will never know how to do something like that.

What are stock option chains? I know what stock options are, but what are chains?

I guess it's basically automating when you're leaving a certain stock. Like if you're automating, if the stock reaches this and this number, then exit. I'm assuming. But that's just my guess.

Yeah, I think you're right. Yeah.

But yeah, I would. I don't know. I don't use stock exchanges. I don't trust them. We were in Robin Hood for quite some time. We had some. I think our Doge was there.

And then one day we wanted to pull out and we couldn't. They blocked us from pulling out and we really wanted to leave, but they say, no, you cannot leave right now. You'll have to wait. But I want to leave now. Okay, so we got.

Did you buy gme?

Sorry?

Did you buy gme?

No, we only had Doge okay, yes. The cryptocurrency that does not require an exchange.

The good old doge.

Yeah, but now it's on ledger. I'm waiting for Musk to do something about it, because one day he will, I have a feeling. And then I'll be rich. But until then, I have a nice little device where I physically store it.

Yeah, that's the way.

But yeah, there you go. So, I'm sorry, Nick, but.

Yeah, option.

Chains, if you know the details of the option.

Yeah, I just looked it up. Yeah, it's. It's. Yeah, it's all of this. It's like the strike price, etc. Right, okay. Yeah, not exactly sure. Scraping, yeah, that's.

That's not a financial advice.

Not a financial advice.

And don't scrape your. The same exchange where you keep your money. What if they label you as a bot? You know, how are you gonna access your money?

There's a few good comments in here. Maybe there's a good alternative here, Maria. Like, how about Daksh here is asking, like, he wants to scrape Tinder and Bumble profile for hot babes. We do that.

You know, you can, you can pull out these comments, right?

I can too.

Can you? I think so. Try it now.

Oh, there you go. Oh, yeah, yeah, there's comments. I didn't. Okay, nice.

All right.

Nick, just use Selenium or Puppeteer.

Or. Please don't, please. If they detect you as a bot, your money will be locked. I wouldn't do it. Check the terms of service of your site. If it says no web scraping will delete your account or whatever. It's not worth it.

Yeah.

So, okay, how do I highlight? I can't, Dax. But, like, just to return back to the scraping Tinder and Bumble profile for hot babes, like, where do you find Maria? That's a. That's a. That's a viral video idea if I've ever seen one. Like, I scraped 10,000 Tinder profiles.

Like, and went on to get a hundred dates. That title right there. If you, you know, if you make that happen, actually.

Yeah, that will be very useful. Yeah, if you have a Tinder.

Yeah, I mean, like myself, gotta do it. Gotta do it for the project.

Where do you find this comment? I can't even see it.

It doesn't matter. It was a bit further up.

Yeah, tough.

Yeah. But that's just bringing the quality here.

As you notice, I don't film viral videos. I don't know what's wrong with me. I never do. I feel what's useful and nobody watches it.

No. Okay, first of all, that's not true at all.

Fraction.

That's not true at all. You're doing great. And secondly, sometimes there's just these ideas that are like, you know, that are worth pursuing. 100 days fake accounts, for sure.

Yeah.

Like this is a cool one. You know, like legitimately, if you could hook up a phone and just set up a thing that just like swipes. Auto swipes for you and then a little LLM that like rizzes them up in the. In the DMs or something like how the kids would say it these days. Then you know, that's a viral video. Like I think that's going to get a.

It's a Mr. Beast grade.

Yeah, I guess.

Awesome. Yeah. So we are almost out of time, so if you guys have some more questions for Igor, please write them in the comments. And in the meanwhile, Igor, please let us know. How can people find you? Where can they find you? And where do you want them to find you?

Yeah, I think there's really two great sources, especially when it comes to staying up to date with all of this stuff and keeping an eye on what's new. That's what most people want to do that we do on one, our YouTube channel. Two, on a weekly newsletter that is also completely free. So on the YouTube channel, we basically do this new show every Friday called News yous Can Use. You can just tune in and me and the team go ahead and test everything and then I present it to you, show you all the results and all my opinions about the new stuff. Again, it's no drama. It's just, you know, like if Sam Altman gets removed as CEO, you won't even hear about it in our show because, like, we just talk about this new stuff that you can actually use him. Getting fired is not something you can click. Right. So that is that. And then in the newsletter, we send out one email a week that gives you one free app that does something exceptional that wasn't possible last week. One news story that actually matters, one prompt that we found really unique and surprised us, and one piece of content that we want to highlight. It's a very simple newsletter format, but it's sort of my idea behind. Hey, if you want to tune out for the whole week and just check out one thing, then you can get the newsletter. By the way, there's a great freebie on signup, so you can, you know, you can get that. But basically YouTube and that. And then if you want more, we have a shop and we have the paid membership where we teach these skills, but no need to get into that if you, if you, you know, you first want to kind of get a feeling for the space, what's new, check out the YouTube and then you can go from there. But yeah, I hope this was helpful and I'm yeah, happy to take any other questions here, if there's any.

So do you have like a YouTube membership program? Basically we.

I built it outside because kind of my vision of it went way beyond what's possible on YouTube. We run like on average like Matthias here that's in chat. He's our events manager and we run 15 to 20 events every single month. So they're like either group consulting sessions or lectures. We have like three districts in the membership which is like one for LLM, one for creative and one for no code tools. And each district runs their own foundations and innovations, events and consulting sessions. And it's just, yeah, just built this whole thing which I'm really proud of and that is quite good at getting people not just up to speed, but like at teaching these skills.

Awesome, man. I'm proud of you too. This is really cool. When you meet somebody at a conference, you don't know what kind of channel and what kind of community and what kind of business they run. You just sit together and you talk about silly things and then you go back home and you check, dude, he's doing amazing. It's incredible.

This girl be SQL injecting Canada. The heck.

No, no, I did not do it. It was an accident.

Yes.

But yeah, I'll see you later.

But I agree this was super fun. Happy to, you know, do it anytime again and then maybe you can come on over sometimes into community or channel. I think that would be a lot of fun. And I see a final question here from Sari.

Yes, you do. I kind of pulled it earlier.

Have you used any local LLMs beside Cloud and GPT? Which one do you consider the best and by what percentage do you think it falls short compared to cloud and GPT?

So I would say if you're a consumer, if you're a knowledge worker, if you're a developer, if you're a creative, if you're a marketer, if you're entrepreneur, basically anybody just using these tools to get like to boost their productivity, AKA you're not trying to build a startup based on these models. Right. That's a whole different conversation. If you're that, then I think all of your needs get taken care of by ChatGPT and Sonnet 3.5. Those two really cover it. Now is there others that are, you know, at the same level? Sure, but they don't have the level of tooling of ChatGPT usually. Right. So if I, if I made. We actually put this ranking together every single month and it's. We also make this, as you might notice, we make a lot of stuff like very freely available, but basically this ranking ranks all the different LLMs. And let me spoil it to you, rank one and two are ChatGPT and Claude, because either the models are best or the tooling around is the best. You know, so something like the large, like Mistral, Large 2 or you know, Llama for Llama 3.1, 400B, they're really good and they, they hold up in terms of, in terms of the answer quality. But you're giving up all the tooling around that. You're giving up custom instructions, voice input, advanced voice output, you're giving up the code interpreter, image upload, image generation, you're giving up the mobile app that works super well. You're giving up like, I'm sure there's more. You're giving up all of that.

While the model is the same. So the model is not the only thing to worry about. It's really all these. You're giving up GPTs, right? We have all these GPTs in our workflows. Like if I go to llama, it's equally as good. Then I'm just sacrificing essential tool to my workflow. So I don't want to do that. So for writing and code generation, Sonnet 3.5 for editing and everything else, ChatGPT, that's sort of really the main stack that I recommend. If you care to look at all the details and if you care to learn about more platforms, maybe you care about the uncensored one or maybe you want a search engine alternative to Google. There's some great alternatives. There's. There's Gigabrain that looks at only the Reddit content and looks at all the comment section and sub pages and everything. And you could, it's freely available. There's, there's Perplexity, which is sort of the main Google competitor, I suppose, in the AI niche. And there's also Find P H I N D which is developer focused. It's a developer focused AI search engine. So yeah, if you care to explore all of those, I'll drop you a quick link in chat to our LLM platform rankings and we kind of. I update this once a month and go into detail why certain platforms are better than others, but this is the two minute synopsis of that. So I hope that was helpful and I think we're on time.

Yeah, perfect. We are. I'm not concerned. It's your time. I'm usually streaming for three and a half hours. I'm too Canadian. I feel too bad. I have comments and I have to reply to all of them. And before you notice, the entire day was spent on streaming, so.

But also you enjoy it, I can tell you.

I guess. Yeah, well, time passes fast when you enjoy.

Look at us. It's been an hour.

Absolutely. Well, thank you so much for coming, Igor. It was a pleasure. Thank you for sharing your incredible knowledge about artificial intelligence and gen AI and whatever it is we talked about.

And yeah, as a final reminder. Yeah. Check out Igor's channel. It is the AI advantage on YouTube. And make sure to say hi. Leave him a comment on his recent video saying, maria told me to come here and say hi.

And yeah, well, excellent. Hope you guys got some value out of this. Hope you enjoyed your time here with us. Maria, thank you so much for the invite and yeah, let's do this anytime soon.

Awesome. Bye. Bye, everyone. See ya.

And.