The AI Sandbox - OpenAI o1, NotebookLM, Testing Veo, Autonomous Browsers and more!

We're going live. We're going live. Welcome to the AI Sandbox. We're going live. Welcome to the AI Sandbox. We're going live. Welcome to the AI Sandbox. We'Re going live. Welcome to the AI sandbox. We're going live. Hosted by Matt Wal, Sat by Mad Wolf. What's up, everybody? Yeah, so I was playing around with a new intro that was actually done with Suno. I saw in the comments people were guessing Suno or Udo, but yeah, that was Suno. And the videos were made with Veo Vo. So, yeah, just having some fun trying to make the live streams a little bit better with each one doing stuff like that. If. If you joined me on the last live stream, then you probably already know how this one's gonna go. I'm kind of all over the place. I like to let the comments take me where the comments want to take me. So as you ask questions and talk about things you want to know more about, we just sort of steer the stream wherever you guys want to take it. And that seemed to work well last time and was a lot of fun. So we'll be doing that again. Hey, from Germany. We got a couple people from Germany. Nice. So we're going to go ahead and get started. I've got Dylan hanging out in the chat. You'll see him as dw. He's the. The main moderator on here, and he'll be kind of keeping an eye on any of your questions and feeding them to me as the show goes on. Last time I planned on going for two hours, ended up going for three hours. We'll see how this one goes. There's a little bit less news to talk about this week because it's, you know, getting into the week of Christmas. So I figured this one will be less about news and a little bit more about playing around with different tools and, you know, seeing where things go. So cool. We're going to go ahead and get started. Dylan, AKA dw, there, he's. He's hanging out in Slack with me, so he'll feed me any of the questions that pop up that I should be answering. But I want to kind of kick it off by talking a little bit about OpenAI's O3, because I put out a news video on Friday, and the news video that I put out on Friday was always recorded on Thursdays. I send it to my editor on Thursday night. He edited it while I'm asleep. And then when I wake up Friday morning, I have an edited video that I put on YouTube with all the news. And as we know, oh3 came out on Friday, so that didn't actually make it into the video. But obviously the biggest news in the AI world over the last several days has been open AI's O3. I've been hearing people claim that this is AGI. People have talked about what a monumental leap this is. So I'm actually interested in hearing from the comments what other people think about O3. Do you think that OpenAI finally hit AGI and they're just not releasing it to us yet? I have my own opinions. I don't actually think this is quite AGI yet, but I do think it was a pretty big leap. So I'm going to go ahead and, and share my screen here. I've got a bunch of sort of O3 related stuff up on the screen, but again, I do want to hear what you guys think. What you guys think about it. Cody Creed. Was that Cody Creed? Yeah. Cody Creed was already asking about OH3. So that's, that's where we're going to start. And then we're going to get into sort of playing around with some tools. For the most part, I don't come into these shows with a huge agenda. I don't plan on coming into these live streams with a huge agenda. I. I plan on coming in with like one or two things I want to talk about. And then after I've talked about those things, we'll go wherever you guys want to take the show. So that's, that's kind of the goal here. But yeah, let's start with O3. So they announced this on, on Friday here and some of the benchmarks are really, really interesting. We can see that everything is this huge leap in the benchmarks. So competition math, it was getting a 96.7 where, oh one was getting at 83.3. PhD level science, we got a leap from 78 to 87.7. The research math jumped to 25.2 over 2.0, which is insane. They managed to. What was this test called? Was this the ARC one? I believe this is the ARC test where they give it like these three squares and then the output they have to put this blue square in and then they give it another input and they want it to see if it figured out the pattern from 1 to 2. And I guess this one did a really good job in this benchmark as well. Seeming to figure it out. Now. One of the things that I found interesting here, and it's actually not in this screenshot, but it's In a different screenshot that I found and I just realized I'm blocking the screenshot. But there's this screenshot right here where it shows the ARC AGI semi private evaluation here. And we can see that IT scored a 75.7 under the compute per task. So I need to show a different screenshot of this one because this one doesn't really do it justice. There's got to be a better screenshot somewhere. But there's one that actually shows like the price per task of these things. And the price per task is like way up there. Let me see if I can find it again. Anyway, let me see. Maybe there's a better screenshot within their, in their stream here. Basically they're not showing it in this screenshot but on the bottom here, the compute per task there's another version of this chart that actually has the cost to run it. And to get this 75.7% on the O3 low series it cost about $20 per task. But this 87.5%, this high, high compute 03 series here, this one actually costs something like 5 or $6,000 per task. So if this is AGI and we did hit AGI. Well at the moment AGI is only going to be able to be used by the extremely wealthy because it's like 5, $6,000 per task to run it at this level which I found pretty interesting. We have never tested a system that so something to keep in mind. But I do again want to hear your thoughts on this. When it comes to large language models, I always really struggle at benchmarking and doing like testing, comparing the different LLMs against each other. So would love to hear your guys thoughts. Yeah. So Justin says the chart I saw said 1000. If you look at this chart right here, I need to find the better chart. Let me see if I can find it. Let's see what's the best way to do this.03 AGI benchmark. Maybe I can find it like in an image search here. There's got to be the one that shows the pricing along the bottom somewhere. It's a, it's a. Basically the point is it's a logarithmic scale. So like this was like this one right here was $10, this was a hundred dollars, this is a thousand dollars. So if you were to sort of chart this, the price goes up dramatically. It's not like a straight line pricing. So if this was. If this right here was 10 and this right here was a hundred and this right here was a thousand this right here would be like 10,000. And we can see that this 87.5% is actually closer to 10,000 than it is 1,000. So I don't know if OpenAI, like, buried the chart because I can't seem to find it anymore, but there was a chart that showed the actual pricing along the bottom. Oh, here we go. Here's one. Let's see. So you can see here, at this price point, it's $1. Here, it's 10. Here, it's 100. Here, it's a thousand. So it's a logarithmic scale Here, it's not a straight line scale. So if there was another line right over here to the right, this would be a 10,000. Right. Based on these jumps, this would be a 10,000 over here. And so this 88% is actually closer to the 10,000 mark than it is the 1,000 mark. Let me see. I'm checking out some of your comments here. Boris says, how do we know Matt is not AI? I mean, I can promise you I'm not AI. I've met a lot of people in person. There's probably people on the stream who have actually hung out with me in person. That would probably attest to the fact that I'm not AI. Let's see. Harvey. I'll be at ces. I will be at ces. So I'll be floating around at ces. Would be cool to meet up with some people. Yeah. Alan says there is also a problem. There's no clear definition of what AGI is. That's been another issue with AGI, is everybody's kind of got a different definition of it. Let's see. Doing one task could power my home for over two years. Yeah, that's right. It's. It's insanely when it comes to the energy use. It's really, really crazy. But one thing I did want to show. So we've got the ARC AGI semi private here. Let's see. I've got this HRC paper pulled up here because I was curious, like, what does a human score? Like, if a human was to take this test, what would a human score? Along this chart here. And according to this research paper here, we estimate the average human performance lies between 73.3% and 77.2%, with a reported empirical average of 76.2% on the training set. So an average human average is about 76.2% on this benchmark here. So we can see that this O3 low model, the one that costs. I don't know, what would this be? Like, maybe $30 or so per task? This one actually performs about as well as the average human. This one performs above as well as the average human. So if AGI, if the definition of AGI is that it's got to perform above the level of the average human, then, you know, maybe we've hit AGI. In my mind, AGI is more like. And this, like, everybody has their own definition, but my kind of definition of it is when it performs at a level, you know, greater than all humans, right? So, like, no matter what sort of niche task you give it, no matter where, you know, where you ask your questions, it's going to perform better than the smartest human in that field. So if you were to ask it a question on biology, it should outscore the smartest biologist, right? To me, that's like AGI, that's when I feel like we've hit AGI is no matter what term you give it, what question you give it, what sort of niche you ask the question in, once you've. Once. It can answer any question in any field better or equal to the best human in that field, that's when I feel we've hit AGI. I don't know if that's making sense, but that's. That's kind of my thinking on it there. So I don't necessarily think we've hit AGI. I also think the cost of where this is right now is really high. But one of the things that I wanted to test on this stream and I. And I'd love it if you guys gave me some questions to ask, is I've been playing with the new Notebook LM features. I don't know if you've logged into Notebook LM anytime recently here, but they've redesigned it, they've overhauled it, so it's a different look than even when I was playing with this last Monday. And I've got the interactive mode now, which means I can interrupt the podcast and ask it questions. So what I did was I uploaded the Transcript from the O3 video, the early access safety testing text that they put up on their website, the deliberate alignment text. And then I uploaded these two PDFs, the ARC PDF here, and then the Frontier Math, which is one of the other benchmarks that it really kind of crushed it on. I put all of those into a notebook here in NotebookLM and had it generate a podcast so that we can essentially ask it questions about OpenAI's O3. Somebody said it sounds like what I'm describing is super AGI or maybe asi, which is artificial superintelligence. I wouldn't see it as that. So when. So in my mind, and obviously definitions keep changing in my mind, ASI is like, all right, now it's able to do all of these tasks smarter than the smart, better than the smartest human. But it's also creating new. It's creating new coding languages, it's creating new thoughts that haven't been thought of by a human. It's sort of gone beyond a level that humans are capable of, not sort of matching a level where the smartest human is capable of. But again, everybody's gonna have a different definition. That's kind of where I've been looking at. Yeah, ASI has consciousness. I don't know if I would say ASI has consciousness, but I would say ASI is to a point where it is so smart that humans no longer even understand what it's like talking about anymore because it's so far advanced beyond human understanding that it's doing things that humans can no longer understand. That in my mind is where we is where ASI is. Let's see. Dylan sending me some other questions here, but I'm trying to keep it on topic a little bit better than last week. Last week I was literally jumping from topic to topic to topic like every two minutes. So I'm trying to stay on topic and then I'm going to jump to other topics here. So sticking on the O3 theme here for a second, I do want to play around with this notebook lm. I haven't used this interactive mode a whole lot, but I do want to play with this on stream. So if you have any questions you want to Basically ask about OpenAI's O3 or O3 mini, let me know because I want to ask it to this notebook LM here and see what it says about it, If it'll play for me eventually. A lot of times if I'm streaming stuff slower.

Hey everyone, welcome back. Ready for another deep dive? Today we're tackling something that's both fascinating and a little bit mind boggling. AI and what happens when you push it to its absolute limits?

Yeah, it's like, oh, hey there, what's up?

Do you think that OpenAI's O3 is AGI?

That's a great question and what a lot of people are asking.

It really gets to the heart of what we mean by AGI or artificial general intelligence.

You know, the idea of a machine that can basically do anything a human can do.

It's not just about being good at one task, like playing chess or solving.

Math problems, but also understanding things in the real world and being able to learn new skills.

Right.

It's that adaptability and ability to reason broadly that's the key to AGI.

So is 03 there yet? It's tough to say definitively.

It's definitely a huge leap forward. We saw it crush benchmarks that were previously unbeaten.

Yeah. Like that ARC benchmark we were about to talk about that.

It achieves scores that are comparable to humans, even exceeding them in some cases.

And the math and code challenges, too. It's showing incredible progress in reasoning.

What's fascinating here, it's not just following rules, it's understanding them and applying them. Opex. Yeah.

What is the ARC benchmark? Why is that in? Why is that an important benchmark to pay attention to?

That's a fantastic question. And it's right on point with where we were headed.

Yeah, we were just about to dive into it, so it was perfect timing. You chimed in.

So the ARC benchmark, it stands for Abstraction and Reasoning Corpus.

Think of it as a collection of visual puzzles.

But these aren't just simple picture puzzles, like the kind you might see in a magazine.

These puzzles are all about grids of different shapes and colors.

And the goal is to figure out the underlying pattern or rule that governs how those shapes change.

Like cracking a code, but a visual one.

It's designed to really test an AI's ability to reason abstractly.

That's why it's in. It's not just about memorizing patterns, but truly understanding them.

The why it's important comes. Oh, go for it.

Okay, I get it. So let me ask you this, then. When it comes to O3, a lot of people are asking, what is the actual use case of O3? I kind of understand the coding part of it. If it gets really good at coding, it can basically build software for us. But outside of the coding use case, what is O3 really good for?

Okay, that's a really insightful question, and it's something that we should definitely dig into.

It's true that the coding abilities of models like O3 are pretty obvious. It can write and execute code, which is hugely useful.

But you're right, there's a lot more to it than just writing software. It's not just a coding machine.

We were about to talk about how these models do on absumec reasoning tests like the ARC benchmark and how that relates to general intelligence.

We'll also get to the math Stuff which shows some of its broader problem solving abilities, but also some limitations.

And this idea of deliberative alignment where the model learns to reason about safety, that's another important use case.

So, yeah, we'll show you that these models have a bunch of potential uses, not just writing code.

It's about how the reasoning abilities can be applied to different fields and how it impacts safety, too.

Okay, so with that in mind, let's.

Get back into it.

A lot of benchmarks focus on tasks where AI just needs to follow rules or memorize lots of data.

Oh, yeah.

Huh. I have another question somebody asked why it can't solve a simple puzzle like a wordle yet.

That's a really interesting question. And it highlights a key difference between different kinds of tasks.

Yeah, we've been talking about these incredibly complex benchmarks, but something like wordle might seem easy by comparison.

But that's the thing. Easy for a human doesn't always mean easy for an AI. They are good at very different things.

It's true. These language models, they're really good at finding patterns in large amounts of text or at visual tasks like we've been discussing.

But something like wordle, it requires a different type of reasoning. It's about trying a word and using the feedback to narrow down the possibilities.

And it's not just about what words are possible, but what's probable. Right. You have to think strategically about what to try next.

Exactly. It's an iterative. Oh, yeah.

Huh. So would you say that an AGI would be able to solve a wordle, so therefore we know this isn't AGI?

That's a really smart question, and it gets to the heart of how we define AGI.

Actually, it's a good point. If a model can't handle something like wordle, does that mean it's not truly general intelligence?

I think what we need to remember is that he isn't just about solving any specific problem.

It's about being adaptable, learning new things, and reasoning in a way that's similar to how humans approach things.

So even if a model struggles with wordle, it can still be incredibly intelligent in other ways.

Like we were talking about about before with the ARC benchmark that really tests an AI's ability to learn new things on the fly.

Okay, I'm going to go ahead and cut that off there because, yeah, those. The voices are not the best. Right. It is kind of boring. I see a lot of comments going, yeah, this is like the most boring couple in a podcast ever. That's true. I also saw people ask, Is O3 out yet? No, it's not out, but O3 was demoed on the 12th day of OpenAI's announcements. That was the final big announcement they dropped, was that O3 is, you know, working and it's passing all these benchmarks and we should have access to the mini version early next year and then the full version a little bit after that. But obviously they need to get these costs way down. You know, there's a chance that OpenAI is going to drop a thousand, $2,000 a month plan just to use something like O3. I've heard rumors about that. Whether or not that's actually going to happen, I do not know, but have to see how, how that plays out, because the cost right now needs to come down by quite a lot before it becomes beneficial to most people. Somebody's asking, how will we rate benchmarks for ASI when ASI will be smarter than the smartest humans? Yeah, I think that is a very, very good question, one that I don't actually know the answer to. I mean, honestly, I don't even know how, you know, they're creating benchmarks for AGI right now because it seems like every time they create a new benchmark, there's tools or new large language models that manage to, you know, really blow those benchmarks away and then they go back to the drawing board and create new benchmarks. So when ASI comes, it's, you know, it's a whole new mess. I have no idea. Cody said, Can O3 create recursive code where it improves itself? I do not know yet. You know, that's something that, well, we'll have to find out once we get our hands on it. All right, so let me see, is there any last things I want to talk about with O3? So I guess overall, my general consensus with O3 is a lot of the people that are out there going, oh, we finally hit AGI, those are just like, you know, hype people. Like those are hype tweets or hype YouTube videos to obviously get the click. I don't necessarily think we've hit AGI yet. I also read something recently that some of the benchmarks that they, that they passed with flying colors, they actually trained some of the benchmark questions and data into the model so that it could pass those benchmarks really, really well. And had they not specifically trained on those benchmarks, then they may not have actually passed some of those tests as well as they did. So there's also that, like, how do we keep these things from actually understanding what the benchmarks are training on that specific set of benchmarks and then doing really well because they were sort of trained on that set of benchmarks. Which is something that I did hear O3 did on one of the benchmarks. I don't remember which one. I think it might have been the ARC AGI one. Yeah. Justin said ASI will write its own benchmarks, ace them and then say see, I told you. Yeah, probably, probably. Let's see here. So Jason way here says O3 is very performant. More importantly, progress from 01 to 03 was only three months, which shows how fast progress will be in the new paradigm of reinforcement learning on chain of thought, scale inference, compute way faster than pre training paradigm of new models every one to two years. I mean when you think about that fact, that is pretty mind blowing that they went from 01 which people were blown away by, to O3 which is blowing these benchmarks out of the water in just three months. Like that timeline. If that timeline continues, that is just absolutely mind blowing. John Hallman says when Sam and us researchers say AGI is coming, we aren't doing it to sell you kool aid, a $2,000 subscription or trick you to invest in our next round. It's actually coming. And then let's see. Noam Brown says today I'm excited to share with all of you the fruit of our efforts to create AI models capable of truly general reasoning. OpenAI's new Zero1 model series. Oh, this is from September. This is when they announced OH1. So they were showing. That was only back in September. Here we are in December and we already have 03. And then Noam says this is the result I'm most excited about. Even if LLMs are dumb in some ways saturating evals like Epoch AI's Research Frontier Math would suggest AI is surpassing top human intelligence in certain domains. When that happens, we may see a broad acceleration in scientific research. So the previous state of the art model was getting a 2.0. The new model is a 25.2. Trying to remember what this light blue versus dark blue was? I don't know if somebody knows. You can mention it in the comments. There is a difference between the dark blue and the light blue. Maybe the dark blue was the mini model and this is the full 03. I'm not 100% sure. I don't. I don't remember exactly. Yeah, so Ms. Reclusivity says it was the ARC AGI test. They trained on the training set, not the test set. Okay. Yeah, I do know that they mentioned that they specifically tailored it to sort of be good at that test. All right, so plasma pug production says light blue is a high in inference time. Dark blue, I guess, is the lower inference time. Which I guess, would that mean that the light blue is the. This, the low tuned model and the dark blue is the high tuned model? I'm guessing that's what that means. Clay says dark blue is one shot. Light blue is going at it like a mofo over and over again. Okay, that also makes sense too. So this, the dark blue was when it answered the questions, basically the first time they were asked. And then the light blue is when it was basically re asked over and over and over again until it got it right. I suppose. Somebody's asking about Huckleberry Hound, and I'm not actually sure what that reference is. I don't know. Let's see. Huckle. Huckleberry Hound. I have no idea what that's referring to. Yeah, you're gonna need to give me more context on what happened with Huckleberry Hound. I. I don't. I don't know what that means. AI, Huckleberry Hound voice generator, Huckleberry Hound. AI, which is super rare. Which means it's like a crypto. NFT thing or something. Yeah, I. I don't know. I know Huckleberry Hound is a cartoon. I don't understand the reference for AI. Like, what's. What you're talking about there. Ignacio says, what do you think of the new Nvidia supercomputer that fits in the palm of your hand? I mean, I think it's pretty cool. It's not really designed to like stick inside of a, like a desktop computer, like. Like mine. Let's see here. I know I've got the. I know I saved the news article here somewhere. Yeah. So this thing here, the Jetson Orin Nano, this is actually designed to be put inside of robots. It's not like a. A supercomputer that you necessarily like plug into the PCI slot on your PC and run run AI on. It's more designed to be installed into various robots and used to train the robots with AI it can run large language models and things like that. But, you know, the cost, the fact that it's 250 bucks to basically get like a supercomputer that can power robots. I mean, that the cost is pretty mind blowing how quickly the cost of being able to do this is coming down. Let's see, another thing that I did want to do on this Stream was I wanted to play around with vo, veo, vio, vo, I think. I don't know if it's VO or vayo. I keep saying veo, but I hear everybody else call it vo. So it's probably vo because I have a tendency to pronounce things wrong and then I just stick with my improper pronunciation and then people just have to deal with it. Let's see, what am I missing here? All right, I need to, I need to just stop paying as close attention to the comments and trust that Dylan is going to send me over the comments I need, I need to respond to. All right, let's go ahead and move on from O3. We can maybe circle back around to it later in the stream if people want to talk about it some more. You know, I guess just sort of giving my final wrap up thoughts on O3. The main model is going to be super expensive to run, so they got to figure out how to get those costs down. I don't necessarily believe that it's AGI yet, like a lot of the headlines claim that it is, but I do think it is very, very impressive that it's blowing away benchmarks that other tools have not even gotten close to what they're. What this one is getting. Also, you know, it's another open AI thing where maybe it's like Sora, where they show it off, get everybody excited about it and we don't hear about it again for nine months. I don't know. Like that's what happened with Sora. And then when they finally released Sora, it was like a stripped down turbo version that wasn't nearly as good. You know, they could do the same thing with this O3. I'm hoping, I'm hoping they figure out how to get the costs way down and then give, you know, maybe pro users the ones on the $200 a month plan access to it to mess around with it. But knowing how open AI was basically for most of 2024, they would show something off and then wait what felt like forever before finally giving us access. Which I understand why they do it, right? They have to do all their red teaming and make sure that all the safety stuff is in place. And I get all that. I just want to play with the toys. I just want, I want it, I want to play with it. That's. That's what I want. I just, I want it in my hands and I want to use it and I want to see what it can do. Ignacio is asking how would they get the cost down Well, I mean, there's several ways they can get the cost down, but I think, you know, just looking at, we're looking at with like Nvidia and how those supercomputers, the price keeps coming down and down and down and down. I think, you know, the, the, the various GPUs and, and the various computers that they're running these training models on, those costs just keep getting lower and lower and lower year after year. So I think the, the cost to actually run these things is just going to, to get cheaper and cheaper as the, you know, as the actual hardware gets. All right, let's jump over to Veo real quick. I'm going to close out of some of this stuff. I'll leave my news page open just in case there's something else I want to look at. Let's see, labs, Google video effects. So let me know if there is anything that you want me to test inside of Veo and I will start tossing some prompts in here. Let's see. Can I get it to work? Did they boot me? No, I signed in with the wrong account. One sec here, one sec. Making sure I sign in with the proper account. But go ahead and toss some prompts in for veo. It does take a good like four minutes to actually generate with this VO tool. So what I like to do is open a few tabs. Let's see. Duplicate, duplicate. It lets me generate up to four at a time. If I go five, it'll, it'll stop me. But I can throw some prompts in here and we can see what it does. A couple notes. It will not do names of people. So if I try to put like Will Smith eating spaghetti, it'll say, I can't do that. It will avoid. It is a little overly sensitive. I was trying to do an egg falling on the floor and cracking, and it wouldn't let me use the word cracking for some reason. I put an egg fell on the floor and cracks. It wouldn't let me use the word cracks. So sometimes it is a little oversensitive. But, you know, if you want to give me some prompts, I'll test them and see. See what we get out of them. I know more and more people are getting access to Veo, but if you don't have access and you want to see in real time what it generates, we can do that. Now let's see. Mysterious drones flying over New Jersey. Funny thing is, I've actually prompted that one before, but it made like UFOs. All right, let's do Mysterious drones flying over New Jersey. Let's see. A man dressed as a duck arrives to deliver 10 pizzas only to trip on the way to the door, spilling pizza everywhere. That's a lot of things in one prompt. Let's see if it can do it. It's only what, eight seconds? So it's got to squeeze all of that into eight seconds. Let's see. Will Imagen3 have a subscription service like Leonardo AI? I have no idea. But right now you can use Imogen. I know it's probably pronounced. Imagine I still call it Imogen. I have a hard time changing words once I've set them for a while, but you can use it for free. I don't know what Google's long term plans are as far as pricing on that. Any news on when VO might be publicly available? That's from Lo Fi Chillin Beats, from what I understand, within the next couple months. They did not tell me a timeline, they just said it'll be available publicly within the next couple months. Right now the trusted testers have access to it to see what it can and can't do and give some feedback. Yeah, Emre says it's not just about hardware. The software is progressing too. That's. That's right. I did mention there's many ways to bring the costs down and hardware is I think one of the more obvious ones. But yeah, the, the software is getting better too. Let's see, what do we got? Dylan? An extreme close up shot focuses on the face of a female dj, her voluminous black curly hair framing her features as she becomes completely absorbed in the music. Sure, let's give it a go. Let's see, our first video is 59%. Our second video is 51. And then our third we've got 3%. All right, the realistic cinematic video of Meteor Drop in C. Alright, I'll just copy and paste that in. And I can only run four at a time, so I'll take more prompts once we get this first one generated here. But I need to keep talking. I don't like dead air. So you know, keep, keep putting stuff in chat that I can respond to. I'm seeing a lot of prompts. I will, I will throw in more prompts into here once some of these are done. You can see we, we've got them sort of all over. 83, 75, 25 and 15%. All right, so mysterious drones flying over New Jersey is almost done here. I'm going to actually hide myself for while we watch these. So I'm not blocking the screen. And I'll go full screen with each one. So here's the first one. I mean, you got a drone. Whether that's New Jersey or not, hard to say. But that's definitely like a close up on a drone. Here's our second one that looks like a DJI drone just kind of hovering over a field. Again, no, nothing makes it obviously New Jersey, but it's a drone. Here's another one. Does that look like the New Jersey skyline? I don't know. I don't even know what the New Jersey skyline looks like. But there's a drone flying through a city there. It doesn't look too bad. And then why does that look like a tank? That's like a close up of a drone in a city just sitting on the ground, not even flying. So those are the four that I got out of. Mysterious drones flying over New Jersey. Let me type another prompt in here. A sleek futuristic cityscape at dusk. Illuminated by holographic billboards and glowing neon lights. The streets are bustling with humanoid robots and humans wearing advanced exosuits. Alright, let's. Let's try that. Did I not copy it? Copy. Start over. That's what I need to do. And then paste. And there we go. So that's one starting. We've almost got our man dressed as a duck here. All right, let's look at our. For whatever reason, it only generated three. But here's our man dressed as a duck and pizza just kind of falls. It doesn't seem like he actually tripped and dropped the pizza. It was more of like an explosion of pizza falling. And then let's see, what do we got here? This is just a duck. Not really a man dressed as a duck. And then he kind of turns into pizza. I don't know, how would you describe that? It's a duck walking and then pizza sort of trips. And then he sort of merges with the pizza a little bit. And then here's another one. This one looks like a man dressed as a duck holding pizza. So we're on to something here. Man dressed as a duck holding pizzas. Does he actually trip? There you go. There's our man dressed as a duck holding pizzas and tripping. It got the count wrong. There's only five pizzas there. And I think we asked for 10, but that's like the closest one right there. He's got five pizzas and then he slips or something. Doesn't look like he slips. More like he just falls on the ground and drops all the pizzas. All right, let's throw something else in there. Majestic eagle soaring over snow capped mountain range at sunrise with soft golden lighting reflecting off icy peaks. All right, We've almost got our extreme close up shot. The face of a female dj. Her voluminous black curly hair framing her features as she becomes completely absorbed in the music. All right, let's see what we got. So there's somebody bobbing their head to some music, but that's really all we get out of that. I guess we asked for an extreme close up. Sh. And she does have voluminous black curly hair. So I mean, I think that follows the prompt pretty well. Here's another one. It's an extreme close up. She's got curly voluminous hair and she's listening to music with lights sort of flashing in front of her. So that one seems to follow it as well. Here's another one. I mean, they all seem to pretty much follow the prompt. Another close up. Curly black hair bobbing her head to the music. And then finally, what do we got? The realistic cinematic video of a meteor drop in the sea. These look cool. This one looks like a meteor flying over the sea. Oh, wait, wait, there it goes. Oh, that looks sick. That's actually a pretty good one. All right, what else? Here's another meteor entering the sea. Okay, that looks pretty good too. Here's another one. These are all pretty decent. I'm impressed. I'm impressed. I'm easily impressed though. And then here's one coming in from a different angle. And then the splash happens before the meter even never hits. So that one's a little wonky. But the first three were all pretty dang good. Honestly. Alright, a chocolate cake literally begging to be eaten. Literally double literally. Okay, let's see. Where can I plug that one? So these two are still working. So I'll start over on this one, plug it in here, and then let's start over on this one. And we'll do a lone explorer scaling the side of a massive waterfall, water crashing around them with sunlight filtering through the mist. All right, so I'll go ahead and put these in there. Alright, so now we've got a sleek futuristic cityscape at dusk, illuminated by holographic billboards and glowing neon lights. The streets are bustling with humanoid robots and humans wearing advanced exosuits. Here's generation number one. We've got humanoid robots. I do see like exosuits on humans. I mean, I think it. Got it. Let's see, we got the futuristic cityscape Dusk. Hard to say. Illuminated by holographic billboards. We definitely have holographic billboards everywhere. Streets bustling with humanoid robots and humans wearing advanced exosuits. I think it got it. Let's see. Here's another one. I see a lot of humanoid robots and it's definitely like Dusk. I'm not seeing the humans with exosuits, but other than that it's pretty solid. Here's another one, very similar. We've got the neon signs everywhere, we've got humans. That one's not wearing an exosuit, but we've got humanoids and maybe some humans with exosuits. And then here's our last one. This one's sort of panning over the city and can't really make out what's going on with the people. Like I can't tell which ones are humanoid robots and which ones are people in exosuits. But we do have our futuristic city. Not too bad, right? It came out pretty good. Shine if I thank you so much for the 20 bucks. I really, really appreciate that they say love these lives. Matt, thanks for everything you do. Really appreciate it. Thanks for tuning into the lives. I have a lot of fun doing them and it still blows my mind that people actually enjoy watching them. That's how I feel about my YouTube channel in general. Like I just get on record myself having fun with AI tools and people keep watching them and I'm like, but why? What's so special about these? I'm just kind of having fun on camera. All right, what do we got here? A majestic eagle soaring over a snow capped mountain range at sunrise with soft golden lighting reflecting off the icy peaks. All right, here's number one. Look at that. You got the snow, the sun reflecting off the icy peaks, the eagles, the mountains. I mean nothing to complain about with this video here. Did we get it all? Majestic eagle, snow capped mountain sunrise, golden lighting reflecting off the icy peaks. I mean, got it all. Here's another one, a little more like animated cartoony looking here I'm gonna hide myself again. But I mean, yeah, that one looks very like CGI to me. Here's another one. Pretty good. We got the eagle. The eagle just sort of changed directions on us like mid video. Like it's head change positions so it's facing that way and then boom, all of a sudden now it's facing us. So that's a little wonky. But it's got the snow capped mountains, sunset coming over the mountains. It's got a lot of the Elements. But the eagle is definitely wonky. How the head just sort of changes position. Here's another one. A little bit of funkiness in the head again, but pretty good. I definitely think the first one did it best. So if we check out this one here, this is, this is probably the best one. And start it from the beginning. Eagle flying over snow capped mountains. Sunset with the sun coming up over the mountains like. And the eagle doesn't randomly change direction on us either. Pretty good. All right, let me plug in another prompt over here so I can have some going. I've gotten, I'm getting a lot of prompts here. Mischievous cat here, I'll just copy and paste them. Oops. I'll just copy and paste them in and then we'll read them as they're ready. That way I'm not sitting here just reading every single one as it comes across. So create that one. Let's get rid of our eagle. Let's do this one. And then our chocolate cake literally begging to be eaten. Literally. All right, let's see how literal it got. So we got a chocolate cake. Not seeing it literally begging to be eaten, but it's a very, very close up of, I guess a chocolate cake. Here's our second one. Just a close up of a chocolate cake. Not literally begging to be eaten, but definitely a delicious looking cake. Once again, we only got three out of four for some reason. I don't know why it does that sometimes. Here's our next one. Cuts into the cake and then look, it looks pretty moist. I mean that looks like a delicious cake, honestly. But none of them actually literally begging to be eaten. Alright, let me plug another prompt in here. Now we've got a lone explorer scaling the side of a massive waterfall. Water crashing around them with sunlight filtering through the mist. This looks like Lara Croft from, from this angle here. So she's scaling a waterfall and I mean it looks pretty good. Let's see, what was the prompt? Lone explorer scaling the site of a waterfall. Water crashing around them, sunlight filtering through the mist. Yeah, I think it hit it all. Alright, here's the second one for that prompt. So we've got the waterfalls we can see. I guess somebody, something. What the heck is going on? I don't know what's going on with that one. That one definitely just turned out weird. All right, here's our third one. We've got somebody scaling a waterfall. It's a pretty close up, so it's kind of hard to make out that it's a waterfall. But we got a Close up of somebody climbing a waterfall there. I don't know what's going on with this, like, thumbnail here. Let's go ahead and take a peek. So there's our waterfall. Not seeing anybody scaling it. It's just more of a waterfall. But again, that's what's really cool about Veo is that it gives you four every time. And usually one of them is pretty good. Like this first one up here pretty much nailed it, right? Like this one right here. Like, can't complain about that. It pretty much got everything in the prompt in it. All right, I'm gonna throw another prompt in here and let it get going. I'm gonna stop reading them as they come across. All right, so now I'm waiting for more prompts to finish. All right, mischievous cat wearing a tiny jetpack and goggles accidentally activates a time machine, causing chaos as it zips through different eras. This should be fun. All right, so there's the cat zipping through different eras with some funky goggles on. Okay, that was interesting. It sort of changed to like a rocket ship at the end. That's pretty hilarious. All right, mischievous cat wearing tiny jetpack. Okay, can't really make out the jetpack. You got the time machine zipping through different eras, though. That seemed pretty good. All right, here's another one super close up. Hard to make out the jetpack, but you see it sort of zipping through different eras and then it randomly changes to like the weird tunnel or something. Here's one that looks more like a Pixar style as it zips through. I don't know, it doesn't really look like the scenery is changing that much on this one. And then finally we got this one here. It's almost a little too. Oh, there you go. You got some, like, Roman soldiers, a dinosaur cat. Oh, this is like my favorite video I've ever seen. All right, so the cat randomly turns into wearing a dinosaur head, and then there's a dinosaur leaping, and then there was like a gladiator at the end. Like, this one just went off the rails super quick. Oh, that's amazing. That is so amazing. Okay, all right. That's an awesome one. All right, let me. Let me go ahead and toss another prompt in here and we'll move on. And I think I'm kind of running out of prompts to test. And then we can sort of move on to a different topic after this. All right, so a first person view of a man driving a car. Realistic car inside driving in New York City. Many cars and people on the streets, sunny weather, Pixar styled. All right, let's go ahead and take a look at our first one here. So definitely a first person view. It doesn't really look like a Pixar style to me, but we definitely have a first person view and it looks like a more like old fashioned car. So driving. So we got the first person view. Realistic car inside. Driving in New York City. Yeah, maybe I can see that many cars and people on the streets. So it kind of got everything but the Pixar styling. All right, here's another one. This one looks like the Pixar styling, but it's not the first person view. It's actually looking into the car. I can buy that. This is New York though. I mean it looks close enough. Missed the first person view on that one. Here's another one. First person view. We got that. We could see the inside of the car. People and cars walking a city that looks similar to New York. I mean that one did a pretty good job. Debatable on whether this looks like Pixar or not, but pretty decent. This one definitely looks like a Pixar. Right? That looks like a Pixar. Back of somebody's head. First person car, other cars on the streets, New York, I think. Yeah, I think it kind of nailed it. These first two did pretty good. The first one missed the Pixar. The second, the fourth one, I don't know, it looks Pixar y. And then what do we got here? A boxing match of two humans on the surface of Mars with the sun above. All right, I do know when you put two people in they tend to get wonky. So let's see what happens. Here's our first one. Actually not too bad, honestly. Better than I thought it was gonna be, if I'm being honest. Little, a little grainy, a little noisy. I don't know what's with all this extra noise in here in the middle, but not too bad. Here's our second one. They look like they're made out of plastic or something, but not bad. These are actually better than I thought they would be with two people in them. These people are just more staring at each other. I don't know why. It did a man and a woman. Every single one it did a man and a woman, which is interesting. This one looks very much like a older game, cinematic. So that's what you get with a boxing match of two humans on the surface of Mars with the sun above. I mean it followed the prompt on all of them. Alright, I'll go ahead and close out of this one here, a mechanical earthworm made out of shiny metal is digging into the ground in a cinematic and dramatic ground level shot. All right, so for whatever reason, it didn't generate two out of the four, but let's see what we got out of the two. So there's our metallic sandworm. Looks very funky. Sort of splits into two at certain parts. Like right there, it kind of got a little funky, but mechanical. Right. So mechanical earthworm made out of shiny metal. We got that digging into the ground. A ground level shot. So it got it all. It's just the earthworm kind of looks a little funky when it almost merges into two. That one looks pretty good, though. I mean, that's solid. That looks like it kind of follows the prompt perfectly. Mechanical earthworm made out of shiny metal. Let me stop this one. Mechanical earthworm made out of shiny metal digging into the ground. It's not really digging. It's more just like moving through the ground, but it. It seems to work well. All right, what else do we got here? A first person view. Oh, we already did this one. All right, so I've already done all of the prompts that Dylan has fed me here. Here's the last one we haven't seen yet. A 360 shot of a person dodging bullets wearing a black oversized jacket and dark sunglasses, Matrix style. All right, let's see if it actually does the 300. Matthew Stevik, thank you so much for the. The $20. I really, really appreciate it says. Just saying thanks. Thank you for being on. I really, really appreciate everybody who likes to hang out on these streams. All right, so that's the first one. Didn't really do the 360. Here's another one. We got the bullets flying, but we're not really seeing the 360 camera. I don't know if it understood the 360 camera thing. See, here's another one. Let's take a peek. This one, the camera is sort of rotating around him a little bit. This looks like it could be from, like, a Matrix video game or something. It's got the little, like, health bar kind of thing up in the top, which makes me think it's modeled from, like, a video game footage or something. Is that Sarah? No, that's not supposed to be Sarah. I saw the glasses and thought it was supposed to be Sarah Connor. Hey, we got a 360 camera, but I don't know what this lady's doing running backwards in the middle of a dystopian sort of rundown city or something. So it didn't get the 360 very well. Person dodging bullets. Three out of the four got the bullets. Black oversized jacket and dark sunglasses. I think all of them got that. So just kind of really messed up the 360 shot, really. All right, I'm going to do one more prompt here. Oh, no. Two more prompts here. Okay, we got two more. And then, Dylan, we could probably move on past this and maybe start talking about something else. And then maybe later in the stream, if we want to, we can circle back around and do more VO videos. But I'm going to go ahead and plug in these last couple prompts that Dylan's feeding me here. There's one of them, and then here's another one. Oops, wrong button. Create video. That's what we want. So we'll let those go ahead and generate and then I will do a quick peek at the comments here. All right, let me ask real quick. In the comments, do you want me to keep going with more VO prompts, or should we go test some other tools? I'm happy to kind of go either direction if everybody wants to keep on going deeper and testing Just a ton of VO prompts on this stream. Like, I'm good to go with that, but we can also dive into other topics. There's other tools that we can try out. But again, like I mentioned at the very beginning of this stream, I kind of start with a rough idea of some of the things I want to talk about. But then after I get through the rough ideas, I want to just go where the comments take me. So if there's a specific task you want to understand how to do, if you want me to, like, do a quick tutorial on how something works, if there's a tool you want me to check out, some news you want me to comment on or read about, I'm willing to go wherever you guys want to take it. But I'm also willing to just keep on playing with more video prompts as well. Cool. So I'm getting the go ahead and keep moving on. Yeah, VEO is free right now. It's just I know a lot of people don't have access to it, so I figured I would demo it and use prompts so that anybody who doesn't have access can see what some of their ideas look like. The DO browser. Somebody pointed me to the DO browser earlier today. Was that. That might have even been you, Alison. I'm not sure who pointed that out. Let me go over to Twitter here. Somebody linked me up to the DO browser earlier. Yeah, it was you, Allison. Here's that DO Razer agent mentioned on the live stream last week. See? Okay, let me see. Are you done yet? No, I'm still waiting for these two to finish. Did I see the acquisition for from Perplexity? I think they're moving into the second Brain thing I did. Let me see, Let me pull up the news real quick. No, that's not it. Yeah, they acquired Carbon. So Perplexity acquires Carbon to connect AI search to your work files. I mean, I think that'll be pretty cool. Basically, it looks to me like the same thing that Anthropic recently did where they did their. What do they call it, the MCP model Context protocol. It seems like the same concept as what Anthropic did here. Enables developers to build secure two way connections between their data sources and AI powered tools. So what I'm gathering out of this Perplexity Carbon deal is that Perplexity acquired Carbon so they can build in this same sort of model context protocol thing and connect their, their language model into your own data sources. So let's say for instance, you have a company with a large database of information about your customers inside of it. You can connect Perplexity or, you know, Anthropic CLAUDE to that large database and then when you ask questions, it will actually use what's in that database for context to respond to your questions. That's what this MCP is from Anthropic. And that is what it seems like Carbon does as well that Perplexity just acquired. So it sounds like we're going to get kind of that same functionality inside of Perplexity soon. Let's see, what else? What else? Sachin Sakin. I'm not sure how to pronounce the name. You have the most useful AO channel on YouTube. Thank you, I appreciate it. Chimera. Thank you for the $10 Australian. I really like it or I really appreciate it. Thank you. They say try this. An over the shoulder conversation between two people. Blurred background night, artificial light coming from two directions. All right, I'll go ahead and plop that one in as like another one to test before we finally move on. Let me go ahead and duplicate this tab here, toss this prompt in and let's see what we got out of some of these others. All right, so a highly realistic IMAX style video capturing the catastrophic moment of a massive asteroid striking a modern city. The scene begins with a serene urban skyline showcasing tall skyscrapers. All right, so here's our first one. We see the asteroid moving in slow motion. Unfortunately, the city was already destroyed before the asteroid hit it. Here's our second one. All right, we got a modern city. Doesn't look destroyed yet. The asteroid is going horizontally for some reason, and then it explodes in what looks like an atomic bomb. So definitely struggling on this. Here's our third one. This looks like a futuristic city that's already been destroyed once. And then the asteroid comes in and nukes the city. So definitely struggled with this one a bit. Once again, we have a city that already looks like it's fallen over and then an asteroid. Actually, the asteroid dropping on that one looked pretty decent. Unfortunately, the city was already destroyed when it hit. So that's. That's that prompt there. And then We've got a GoPro mounted on a dog's back shows the canine running through a field leading a group of soldiers. The view is low to the ground and moves quickly, capturing the dog's focus determination. Alright, so this one doesn't look like the GoPro's on the dog, but we can see a GoPro strapped to the dog. I mean, it got everything pretty decent other than the fact that it's the dog's point of view. Let's see. Here's another one. This could be from another dog, right? Like there's a. I don't know if the dog would pan sideways with this dog, but you can see a GoPro strapped to the back. Low angle with soldiers. So it made a video of a dog wearing a GoPro, but it's not actually showing us the footage from the GoPro. Here's another low angle. This looks like it could be a dog wandering next to this dog, maybe. I mean, that looks pretty decent. That kind of looks like what you may be going for. And then finally, I can see the GoPro in the shot of this one, but it's almost like there's two GoPros. So there's a GoPro on the back and you can see the GoPro on the front of the dog from the GoPro that's on the back. Other than that though, I mean, it actually did pretty well. And then I've got this one here still going. Oops, I accidentally left. Try this in the prompt. But we'll see an over the shoulder conversation between two people. Blurred background. Night artificial light coming from two directions. Moody feel. Film grain shot on a Sigma FP12 bit 4K. All right, let's see what happens. So we've got the blurry background and two People talking. Is that the same person or is that a different person? I guess that could be the person that's on the right here. And it just kind of cuts to the shot of them over the shoulder. Conversation between two people. I mean, it all looks good. Hard to tell whether or not it looks like it was shot on a Sigma FP 12 bit, 4K. I don't know. I don't know how to determine if it looks like a Sigma fp. Let's see. Here's the second one. We got the over the shoulder. It's at night. Two people having a conversation. We got cars moving in the background. I mean, not bad. Looks pretty good. Follows the prompt. Here's another one. Two people over the shoulder talking. And then it. I like how at the end, it sort of racks focus on that other person. See how the person in the foreground is blurry? And then right before the video ends, it sort of racks focus and they come in focus. That's a pretty cool effect there. I like that. Here's another one. Some bright lights shining into the camera. But we do have over the shoulder of people talking. More bright lights shining into the camera. Seems to have done that prompt pretty well, in my opinion. Cool. All right, so I'm gonna. I'm gonna move on past video. Maybe we'll circle back around later in this video and play with it some more. If. If you guys want to, I'm gonna. I'll go ahead and maybe download this DO browser. See. See how it works? So I'm just downloading it on my other monitor over here real quick, and then we can. We can give it a go. Hopefully it's not a giant scam and I'm getting scammed live. You never know. You never know. All right, I'm installing it. So is the DO browser. Is it a Chrome extension or is it its own browser? I'm a little confused. See? So I just put this little extension on. Here's our DO browser. Give me one sec. I messed up. I have multiple Google accounts, and I every once in a while log in with the wrong one. All right, let me close this and then open it again. All right, so, welcome to do Browser. This is an AI agent that lives in your browser that can perform tasks for you. You can also type DO in the address bar and type your command to open the side panel. You can press Control K on Windows or Command K on Mac. If you press the microphone button, you can control the current tab hands free by speaking. Try these examples. Order a black seep, black Sheep sandwich from Suvla. Buy toilet paper on Amazon. Create a bucket on S3. Summarize the contents of this page. Alright, so if I go do in the address bar. Okay, so it opens up the do browser extension. Let's just try. Buy toilet paper on Amazon. Is something happening? Do. No, that didn't work. Do browser. Buy toilet paper on Amazon. All right, somebody's gonna need to tell me how to make this do browser thing work. Let me try putting it down in this one. Buy toilet paper on Amazon. Okay, now it's actually doing something. It didn't work when I put it into my chat bar, but I'm hands free right now. It's. It just went to Amazon typed toilet paper but then sort of got stuck there. Oh, nope, it's still going. Not touching anything. All right, it found the toilet paper. Hands up. I'm not touching my keyboard or mouse. Okay, over on the right. I'll navigate to Amazon. I'm now on Amazon's homepage. I'll search for toilet paper. The search results are displayed. I found a suitable toilet paper adding to your cart. Now is it actually going to like check out for me? Like, is it going to pay for it? Oh, it's buying multiple rolls. Do I really need 100? Geez. $141 worth of toilet paper. Do I really need that much toilet paper? $168 worth of toilet paper. Alright, so do browser is just going to Amazon and buying all of the toilet paper Now. It's at $197 worth of toilet paper. Is it still looking for more toilet paper? $205 worth of toilet paper. I'm going to basically sell out Amazon with toilet paper. It's literally going through and adding every toilet paper that it can find to my cartoon at $240 worth of toilet paper right now. What the hell? I'm gonna. I want to see how far this is gonna go. I'm sorry, I'm gonna. I'm not gonna actually check out. I will cancel it before it actually checks out. But I want to see how much toilet paper it tries to buy for me. Oh, this is great. 300. We're up to $317 worth of toilet paper. Yeah, Ted says do browsers. Definitely made by the toilet paper company. $390 worth of toilet paper. It's still adding more to the cart. We're still going. We don't have enough toilet paper yet. $424 worth of toilet paper now. Alrighty. $431 worth of toilet paper. Just. Yeah, just sell out Amazon of toilet paper. I want to own the world's supply of toilet paper. Just in case there's another pandemic. Everybody has to come buy it off of me. $543 worth of toilet paper. All right, how do I actually stop this now? We're at almost $600 worth of toilet paper. Let's go ahead and cancel it. Okay, I think it stopped. I think it. No, no, stop. Keeps on reopening and trying to buy more. It won't stop. $608 worth of toilet paper. Stop it. Okay, maybe that's right. Success. 600. I finally got it to stop after adding $686 worth of toilet paper to the cart. Somebody said I was supposed to be more specific, but that was literally one of the examples it told me to test. I was literally testing the example it gave me inside of the original. See right here. Buy toilet paper on Amazon. I was trying the example. No, don't. Oh, my God. It's trying to buy more toilet paper again. All right, aborted request. All right, abort, abort. What do you want me to try? Somebody throw something in to see. To see what? I should have this try. Because apparently trying to buy toilet paper is. Yeah, it tried to buy a little too much of it. Somebody says, yeah, more specific. Tell me what you want me to specifically try. Because this was their example to try. I have 21 items in my Amazon cart. I'm gonna go ahead and clear my cart so I don't accidentally order it all. Here. $686. Delete all items. All right. There. I deleted it all from my shopping cart. Hopefully still says 21 in the cart. Oh, AI, AI. It's going to buy too much toilet paper for people. Try buying condoms. Have it. Download a Top 40 song. Hack the Pentagon. Yeah, See? Find Lana Del Rey concert tickets. Okay, sure. What would you like to do? Find Lana Del Rey concert tickets. There. I'm putting it in this box here. I just moved my head so you can see what I'm doing. All right, so find Lana Del Rey concert tickets. Extension conflict. Conflicting extensions. Okay. Grant permission. Apparently. My password manager is apparently conflicting. Let me try this again. Find Lana Del Rey concert tickets. Alright, so it went to Ticketmaster. I'm on Ticketmaster. And we'll research Lana Del Rey. Retry. Searching. Okay, so it found something there. Found several concert options. Hey, there's actually no, that's not in San Diego. That's something else. She believes cup. Oh, that's soccer. Right. All right, so it found the concert tickets, but it didn't find any near me. Maybe if I did find Lana Del Rey concert tickets in or near San Diego. Actually let me do a different name because I already know this one. Let's do. What's another band I can try? I'll do like Blink 182. Find Blink 182 concert tickets in or near San Diego. Jet Whitaker. Ask it to go to Mid Journey and generate a photo of a dog. Okay, I'll try that next. I mean this is pretty wild that it can go and use your browser for you. It's just also wild that it tries to buy almost $700 worth of toilet paper. All right, so it found Blinkway 82 in Long beach and Vegas, which are both semi close. Long beach is about two hours. Vegas is about five hours drive from me. All right, let me try this other one. Go to Mid Journey and generate a photo of a dog. That's from Jet. He suggested that. Let's try it. Oops, I put a space in Mid Journey. Let's see if it figures it out anyway. Alright, so it logged into midjourney. That's a good site. A photo of a dog. It just typed that for me. I'm hands free right now. Is it actually going to submit it for me? Okay, so it's still executing. We can see over on the right there. Type the prompt a photo of a dog into the input box to generate an image. Submit the prompt to generate the image. So did it submit? Must have. It's creating. Yeah. So there we go. It generates generating a photo of a dog. So it can do that. Nice. So it can go to Mid Journey and generate a photo of a dog for me. So check. Let's see. Add custom search shortcuts like typing yt for YouTube or am for Amazon searches. Is that like a setting in here? Maybe I could just tell it. Oh, it's doing it again. It's creating more photos of a dog without me even asking it to. So does this thing have like an issue where it'll just keep on doing the task you ask it to do over and over and over again? Because I didn't give it the prompt again, but it just went and generated more photos of a dog. Alberto says I think the more specific the more accurate. I mean this do browser is really, really cool, but it's also kind of scary. I'm glad it didn't actually go and finish the checkout for me on Amazon. The image generation photo of a dog is in Progress currently at 35%. I'll notify you once it's ready. Looks like they're ready. And it finally stopped. Let's see. Sofa King Rad. Sofa King Rad. Can it scrape toilet paper prices on Amazon and give it to me in a CSV? Scrape the toilet paper prices on Amazon and give me a CSV file of them all. Let's see. Alright, so went to Amazon. You can see what it's doing on the right, what it's executing right here. So it's searching toilet paper again. I've already got PTSD from the last time I searched toilet paper. Ask it to make paper clips. That's scary. All right, submit search query. I will now extract the toilet paper prices and compile them into a CSV file. So it's apparently trying to do that. Create a YouTube playlist of your videos. Somebody said Silpha King. Lol. It's funny because I just, I read it without thinking about it first, but I get it doesn't seem to be doing the CSV portion. Yeah. So it went to Amazon and searched out toilet paper, but it does not seem to actually create a CSV search for the latest AI news and add it to a Google Doc. All right, sure, let's try that. Let's see if it can interact with a Google Doc pretty well. All right, so now it's Googling for me. So right now it's doing a search for AI news. Yeah. I honestly think most of the stuff that this can do for me it would be a lot faster to just do myself. I noticed. 1, 2, 3. ABC said just do it yourself. Makes too many mistakes. I agree. What is it doing now? It's. Oh, I don't, I don't know what it's doing. Yeah, this makes like a lot of mistakes. Still. It is really cool. I like that we're getting to this like agentic phase where you could just tell it to do stuff and it'll do stuff on your behalf. It doesn't really make sense to do this kind of thing from my browser because I feel like, yeah, I can just do this faster myself. But once I get to a point where like maybe I just use like my Siri or my, you know, Google Assistant on my phone and I'm out and about and I go, oh, you know, order this thing for me from Amazon. I can do it with a voice command real quick. And then behind the scenes it goes and does all these steps for me that I think starts to get more powerful because then you can sort of instruct it to go do things for you when you're out and about and you don't want to forget to do it when you get home, that that'll be cool. But they obviously need to work out a lot of these kinks because I don't want to be like, hey, buy me some toilet paper on Amazon and then come home to see that it bought $700 worth of toilet paper. Ignacio is asking if it's better than Claude computer use. Right now I'd say it's fairly equal. The big difference is that this seems to be only in the browser. So it's only going to take actions inside of Chrome for you, where the computer use can open up other tools. It can open up, you know, Microsoft Paint, it can open up Excel, it can open up some other tools and actually interact with those other tools. I don't know if this is actually doing anything in a Google Doc. It didn't like open a Google Doc for me. So, yeah, I don't think it's going to actually put these in a Google Doc or if it is, I don't know where it's doing it. Let me go ahead and stop this. Let me. One sec. Let me open up my Google Drive folder and see if it actually created anything. Yeah, I'm not seeing anything. Let's go to my recent. No, I didn't. It didn't create anything, so it didn't do anything on the Google Drive side. Cloud computer use takes quite a few steps to get going and I don't have it set up right now. Like I need to run it with within Docker. So maybe if you want me to try that like on next week's stream or something, I can sort of prep it a little better. But I don't have that one prepped and set up to go. Yeah, so this one never actually went anywhere. So that's the do browser. Or do they call it do browser, but it's an extension. So here's some tips. The more specific you are, the better the results will be. If do browser is going off the rails, you can close the side panel or press the stop button to kill it. You can customize the do browser prompt in the settings. Cool. So next week. Yeah, I'll try to set up Claude computer use and we can test that next week and then maybe we'll open up do browser again and compare them side by side in next week's video. All right, so agrochess find the most subscribed YouTube channel based on AI and comment high on their most Recent video. All right, let's try that. Find the most subscribed YouTube channel based on AI and comment high on their most recent video. All right, so it's searching AI inside of YouTube. And it found my friend Julia McCoy's video. I don't know if that's the one it's going to actually click on or not. Let's see. Click on the channel link for Kurzig Sucked in a nutshell. Okay, so it's picking this channel, which is a good channel. I love these videos. Actually, if you're a fan of our work, you probably value rigor and humility. Alright, so it started playing the video, but now it needs to click on the comment section for me and type hi. Nope, it didn't. It browsed back for some reason. Says I attempted to post the comment high on the video. Please confirm if it was successfully posted or a further action is needed. I don't think it posted anything. If you're a fan of. Yeah, I didn't actually post. So it found this video which is actually not about AI. And then before it actually posted hi. It went back to like the dashboard page, so kind of ran into an issue there. Tell it to like and subscribe to Matt Wolf on YouTube. How? I don't know how that would work if I'm already liked and subscribed. Well, I guess I can't subscribe to myself. I can like myself, like and subscribe to Matt Wolf on YouTube. There we go. Let's see what happens. So it's going to find me. There's me doing a live stream right now. And click the subscribe button. I'm already subscribed, so. Well, I don't think I could subscribe to myself, but if I could, I would already be subscribed to myself. And then I will now. Now it keeps on trying. So it's trying to subscribe, but it can't subscribe. You can't subscribe to yourself, so it's not going to work. Have you signed up for O3 testing? Yes, I'm signed up, but haven't heard anything back. Alison says it can tweet for you. Let's try. Write a tweet for me that says we're having fun on a livestream. Come join. Let's see if it'll do that. I mean, quite honestly, it probably would have been faster for me to just go to Twitter and type that and submit it, but. Or about the same. But we'll see. Let's see what is. Are there any other tools you guys Want me to play with? I see people asking if there's any other tools we want to play with. Let's see. I've played with that one already. I've played with that one already. I'm looking at some of my notes I have here for things that we can test out. There's a tool called Backflip AI which does like 3D objects that I thought might be kind of fun to test. Hey, Gen has a new feature that's like Notebook LM where you can create a podcast, but it actually has two people speaking, so it's not just audio. I think that might be fun. Kind of fun to test. So I do have some other tools, but also if you guys have tools, let me know. Here's a tweet you can use. We're having fun on a live stream. Come join. Yeah, I don't think it did anything. Did I abort the request? I must have by accident. Write this X post for me and submit it. Let's see if it'll do that. All right, so it's going to so navigate you to X. We're having fun on the live stream. Come join. Post. Hey, it did it. So apparently if I give it the prompt, write this X post for me and submit it. We're having fun on a live stream. Come join. You can see. There it is. I'm gonna actually put a comment on there so people know what this is. Actually, screw it. I'm just gonna delete this post. But it worked. It definitely worked. So can write tweets for you. It already posted it for me. Yeah, it did. I just deleted it. Do it with a voice prompt. You can use voice prompts. Oh, I'm an idiot when I see this already. Write a tweet for me. Oh, wait. Write a tweet for me. I didn't finish my thought, but it's writing a tweet for me. I will now compose a new tweet for you. And it's thinking doesn't know what to tweet. Let's see. News parody. Creative question for host and viewers. Choice of lack of music on live, good or bad. Asking for a friend last week I said I wish there was some music because I felt like it would, you know, add fill in some of the silence and stuff. But I can be sort of bobbing my head while I'm waiting for stuff. But I got a lot of comments that people said that they preferred to not having the music, so I just put intro music and that's about it. We played with Ignacio says Tripika, we played with Pika quite a bit on last week's live stream. I don't think it's actually going to finish writing a tweet. I think it's stuck trying to write the tweet, but it's not going to actually do anything. It's been taking way too long. Ken said. Can you show us what you use to face swap your thumbnails? I can try. It's a fairly complex process, but I use stable diffusion. So let me see if I can get stable diffusion booted up. I still use the old school automatic 1111. I never really got in on the Comfy UI train. If you don't know what I'm talking about, don't worry about it for super, super nerdy AI image people. But I do use stable diffusion, which is, you know, an open source video generator, image generator. And that's what I use for my face swaps. And a lot of people use ComfyUI to do the local image generation I now use. I still use automatic 1111, which was sort of the OG before. Before comfy UI. Let's see, Tweet and X don't equate. Yeah, that's probably the problem. I'm going to go ahead and close this sidebar. So here's stable diffusion. I don't know how well stable diffusion is going to work while running a live stream because it does use up my GPU while I'm using stable Diffus fusion. So if my computer crashes or something while trying to do this, then yeah, that's, that's why it's because this does try to use my, my gpu and I believe OBS is also currently using my gpu. So here's how I would make a thumbnail. I'll just walk through the whole process real quick. So I would start in a tool like Leonardo. You can see like here's my library of previous images that I created. And then I would go, let's go ahead and go to image creation. I like to use the flow state now because it just gives me a lot more ideas. And I will disclose I'm a advisor for Leonardo, so I do have a little bit of equity. I need to, by law I need to say that every single time. But it is the tool that I use to create my thumbnails. You just saw the last screen. That's where I created them. So let's just do like a man standing in front of a colorful background holding a microphone. Actually, I'm going to add a white man. Because if I don't do that then it might make it the wrong color and then the face swap will just look weird. So a white man standing in front of a colorful background holding a microphone. Let's see. I'll go ahead and just sign out of this do browser here. Somebody said it's still running, so I'll just make sure it's closed. Oops. I like to do it at 16:9. So let's go ahead and generate it at 16:9 here. Alright, so we've got some images of a dude holding a microphone. Some of them great, some of them not so great. This guy's only got three fingers. The nice thing about flow state is I just scroll and scroll and scroll and eventually I find an image that I'm like, all right, that one's kind of close to what I'm looking for. But I'll do a simple one, this one right here, that's fairly simple to use and work with. There's nothing covering the face. So I'll go ahead and download this image. I've got this image on my computer now. And then if I jump over to stable diffusion, I'll go to image to image here and then I'll do InPaint. So I'll change this over to InPaint here and let me move my camera again. So I've got this as inpaint. And then I'm going to drag this image here into my like inpaint section. And then what I'm gonna do is I've got a preset style here. I'm gonna delete that. But you can see it's got a whole bunch of negative prompts that I like to use on every single prompt here. It's also got some like negative loras here that I've installed that help clean it up a little bit. And then what I'll do is I'll click interrogate clip. Interrogate clip. Because now it's going to look at this image and it's going to try to determine what it thinks this prompt was. And while it's doing that, I'm going to mask out my face. So I'll just go ahead and mask out the face like this to do. All right, faces masked out max, mess out the neck a little bit here like that. And then I've got my prompt up here. Now with a man holding a microphone in front of a colorful background with a rainbow hued background behind him, Multicolored background. I always get rid of stock photography because sometimes that adds like weird watermarks into the image. If I leave that there. And then I trained myself with a tool called Dream Booth. And when I trained myself, somebody said that the way to do it with Dream Booth is to make your like code that you use to generate your own face look like a, or be a celebrity that looks like you. So I used a tool for like, what celebrity do I look like? And it told me I look like Dominic Cooper. So I went, let's see, Dominic Cooper. Apparently the tool that I use thinks that I look like this guy Dominic Cooper here. So I set my like code word as Dominic Cooper when I trained myself into the image. And I have a whole tutorial on YouTube about this. If you want that tutorial on how I trained my face, you should watch that one because that, that actually is a couple hour long process that I'm not going to do on a live stream. So. All right, so I've masked out my face. I've used the original prompt that they gave me here. And then I do change some settings. I go to just resize for inpaint area. I want it to only change the masked area that I put here. Sampling steps, I like to set to about 40. I usually do a batch size of four, but I'm just going to do a batch size of one to keep this sped up here. Denoising streak, I usually bring down to about 7 GFC scale, about 7.5. And then hopefully this doesn't crash my computer when I click generate. But theoretically it should swap out my face. And yeah, it's a little funky, so I'll probably run it again. Usually I have it generate four and one of the four is pretty decent right now it's making me look a little bit special, but that's what I would do. I would generate until I come up with one that I think looks pretty decent. And then there we go. That's perfect. Once I've generated that, then I run it through an upscaler. All right, there we go. That one's good enough. So I'll go ahead and save this image here. And so now I've got an image that I used Leonardo to create. And then I swapped out my face using stable diffusion. And then in my final step, I like to use CREA for the upscale. So I'll take the image I just created here and drop it into crea. And then just click the enhance button. And then it sort of cleans it up a little bit. And then, yeah, that usually takes a minute or so to do Empire Gone. I think you look like a Young Tom Selleck. That's funny. He was kind of more known for his mustache though. I don't know if I've seen Tom Selleck with a full beard other than older Tom Selleck. So that's pretty much the whole process. The only thing that I didn't cover was actually training my face using Dream Booth. And if you go to my channel here, let me go. There's a video that explains how I did that part called inject yourself into the AI and make any image with your face. So this video is where I actually walk through how I trained my face the first time. Joe's asking which AI tools do I have running locally? Mostly just Stable Diffusion, I think I do have Pinocchio installed, but I haven't used it in a while. And Pinocchio is like a browser that has a bunch of like locally installed AI apps in it that you can like one click install. So I've used some of this stuff. Like these are the ones I have installed. Bark Voice, Cloning, Illusion Diffusion, Stable Cascade, a text to speech tool. So some of these, but I don't really use them that often. I mostly just use Stable Diffusion. But this Pinocchio, if you've never tried it, it's really cool. It's got all sorts of like AI tools you can install locally. Like a local version of Comfy ui, a local version of Bolt, a local version of Pyramid Flow, which is an AI video generator. All sorts of cool stuff that you can install. Face Pokemon, Face Fusion. I mean, a lot of these are probably an easier workflow for swapping your face than what I actually use. It's just I've got my system down, so I just do it the same way every single time now. But there are lots of cool locally installed AI tools that you can use here. That's Pinocchio Cray is taking a lot longer than normal. Let's see. So somebody's asking how are people using these types of videos in their business? I really think when it comes to like AI video generation right now, it's still kind of like a novelty. It's still just sort of fun to play with, see what we can get out of it. Use it as like memes on X and Instagram and places like that. I have used it for some B roll, so I have created some like YouTube shorts where I'm talking about a concept and I couldn't find the perfect video for the B roll, so I went and generated a video and use that. So I've used it for a little bit of B roll on some of my videos. So here's what you can see. Kreia did. It actually kind of makes it look less like me, but it's also a little bit more clear and clean. And when you shrink it down to the size of a small thumbnail on YouTube, it doesn't. You can't really tell that it looks less like me. So, like, let's see. Control. So if you make it, like, really small, like what you'd see on YouTube, right. It still pretty much looks like me. It's just when you zoom into, like, full size, you start to notice, like, okay, that doesn't totally look like his facial structure and teeth. It got the bags under my eyes, right? I do, like, permanently have bags under my eyes. So it adds that properly. Mr. Beastified. Yeah, that's kind of what it does, actually. It gives it, like, almost like a Mr. Beast, like, look to it. So that's my process there. Just to recap, I usually start with with Leonardo. The other one that I find myself using quite a bit more and more is Ideogram. Ideogram. I've been using this one a lot more. If you look at my creations, I started messing with this one to make some of my thumbnails as well as, like, the starting image that I swapped my face on. Like I was trying to do a man with his head exploding. So I've played with. So I'm usually switching back and forth between Ideogram and Leonardo quite a bit to get some of my images now. But I would say Leonardo is mostly my go to. And somebody also mentioned Flux and how Flux is really good for creating realistic images. Well, the cool thing about Leonardo is that they actually added Flux as well. So if you want to generate with Flux straight instead of Leonardo, you actually can do that now as well. Ignacio, do you use Grok? What do you think about that? Grok was better when it was using Flux. Definitely. I don't think it's as good as it used to be. So if I go to X and I go to Grok here, it used to use Flux when you did, like, an image generation, and they were really, really good. Now when you go to generate an image that uses Grok's like, own AI image model, which isn't quite as good as what Flux was, it's definitely lost some of the realize the realism. So let's see. Drones flying over New Jersey. Oops. I need to tell it to create an image. Create an image of drones flying over New Jersey. Yeah. One, two, three. ABC says Grok doesn't have guardrails. That is one of the benefits. There are some guardrails. There are definitely some things that won't generate for you, but it, it'll generate like celebrity images and actual people's likenesses and it will generate ip. So if you want to generate Mario or Sonic the Hedgehog or Mickey Mouse or whatever, it will generate actual like copywritten ip. So it's uncensored in that way. But it, it's definitely not uncensored in like the adult way. They're not going to let you co generate like AI generated porn with it or anything. And by the way, if you use X, you gotta follow lists. The for you feed is such a joke. Like I, I create a list called AI is Awesome. And pretty much every time I log in, I either fall, I either use one of these. I've got AI is Awesome, which is my list, which is just people that post about AI. I've got XR is Awesome, which is my list of people who just talk about, you know, AR, VR, XR stuff. And then I've got my YouTube is awesome, which is people who just talk about, you know, growing on YouTube and stuff like that. I've actually made all of these public, so if anybody wants to follow them, you can. But the AI is Awesome is pretty much what I keep as my default when I log into X is just see what all the AI people are talking about. So how do I share the list? Somebody's asking about that. I always have to like find where you share the lists. Profile. No. Lists. Okay, so here's my list. AI is awesome. It's like kind of my main list here. This is the URL for it. Oh, Dylan beat me to it. I put the link in the chat, but Dylan put it in before me. So I've got that list. I've got my YouTube is Awesome list. So if you're trying to grow on YouTube, this one's pretty good for that. Here's my YouTube list. And then yeah, those are, those are pretty much what I follow on X. I try to ignore the home feed because it's all just politics and propaganda. When in trouble, just call dw. Let's see, what's your Leonardo affiliate list? I'm not sure what you're asking there. I'm not actually an affiliate of Leonardo, I'm actually a part owner of Leonardo. I have a very little sliver of equity. Oh, my affiliate link for Leonardo. I don't have an affiliate link. Just go to Leonardo. Go to Leonardo. AI. I Don't actually have an affiliate link that I share. Have you played with Personas on Suno yet? I have a little bit, yeah. I don't know though. I haven't found it super useful because I just kind of make completely new random music every single time. Can we use Flux locally? I'm not sure. I don't believe you can yet. But you might be able to. Let me see something here. Look on GitHub. That's not it. Black Forest Labs Flux. So it looks like the code is available on GitHub under Apache 2.0 license. And this is Black Forest Labs. Let's see. So yeah, you can install it locally. I haven't actually tried this yet, but you can. There's the instructions. I wonder if there's any details here about. Yeah, it doesn't actually say the system requirements, so I'm not quite sure what is needed to run it, but it looks like you can run it locally if you want. I don't know, you might need like a 4090 GPU or something though. I'm not. I'm not 100%. Let's see. So that's Kreia. Kreia has some other new features that I haven't played with yet. I'm actually not a paying member of Korea though. So Kreia has Flux built into it. They have Video generation. I haven't actually played with a lot of this yet though. One thing that's interesting about Kreia is you can actually pick any of the models here. They don't have VEO in it or VEO yet, but you can use hello AI Luma Runway Cling Cling Pro Cling 1.5. So you can actually like pretty easily test a whole bunch of different AI video models all in one place. I mostly just use their enhancer though. Somebody said, was it Sarwar said that you can actually use Flux in Pinocchio. So I haven't tried that yet. Maybe we'll play around with Pinocchio on like a future stream. Because Pinocchio has so many tools in it, right? Like you can open up Pinocchio and there's like 200 different locally installed tools that you can play with. Jim says, do you think agents are overhyped? I mean, I think anything that claims to currently be an agent is overhyped. I don't think there's anything right now that is a great agent. I think the best that we've got that's doing agentic type stuff. I think Perplexity is really good for that. You know the way it sort of does a multi step search for you. And then also Google's deep research is really, really good and it sort of searches hundreds of websites for you to find information for you. Those are the best like agentic type tools that I've seen so far that DO browsers pretty cool. But obviously we ran into a bunch of issues with it. Claude's tool use is very similar to what we just saw with the DO browser. So I don't, I don't know necessarily. I think the concept is properly hyped. I think anything that's currently out there right now is overhyped. All right, what else should I try here? All right, so I have hey Gen and I've got backflip. Let me log into backflip because this is one I've literally never tested. So we'll, we'll try it out in real time here. I don't know if I've got to pay for something or not. Alright, so here's backflip AI. I know last week we were talking a little bit about how to create 3D printable objects using AI. And then later on I came across this backflip. I showed off meshy last week. Then I came across this one which looks almost as good. So if I go start creating, I'll just do a wolf howling at the moon. Let's see if I can get the sort of pose right. So it looks like maybe it starts with an image. Okay, so it created an image. Now how do I create that into a. Let's see, cannot sketch on this tile. I don't know how to turn it into a 3D thing here. Right now it's making me look a little special. But that's what I would do. I would generate until I come up with one that I think looks pretty decent. And then there we go. That's perfect. Once I've generated that, then I run it through an upscaler. All right, there we go. That one's good enough. So I'll go ahead and save this image here. And so now I've got an image that I used Leonardo to create. And then I swapped out my face using stable diffusion. And then in my final step I like to use CREA for the upscale. So I'll take the image I just created here and drop it into CREA. And then just click the enhance button and then it sort of cleans it up a little bit. And then. Yeah, that usually takes a minute or so to do Empire Gone. I think you look like a young Tom Selleck. That's funny. He was kind of more known for his mustache though. I don't know if I've seen Tom Selleck with a full beard other than older Tom Selleck. So that's pretty much the whole process. The only thing that I didn't cover was actually training my face using Dream Booth. And if you go to my channel here, let me go. There's a video that explains how I did that part called inject yourself into the AI and make any image with your face. So this video is where I actually walk through how I trained my face the first time. Joe's asking which AI tools do I have running locally? Mostly just Stable diffusion, I think. I do have Pinocchio installed, but I haven't used it in a while. And Pinocchio is like a browser that has a bunch of like locally installed AI apps in it that you can like one click install. So I've used some of this stuff. Like these are the ones I have installed. Bark Voice, Cloning, Illusion Diffusion, Stable Cascade, a text to speech tool. So some of these, but I don't really use them that often. I mostly just use stable diffusion. But this Pinocchio, if you've never tried it, it's really cool. It's got all sorts of like AI tools you can install locally. Like a local version of Comfyui, a local version of Bolt, a local version of Pyramid Flow, which is an AI video generator. All sorts of cool stuff that you can install. Face Poke, Face Fusion. I mean, a lot of these are probably an easier workflow for swapping your face than what I actually use. It's just I've got my system down, so I just do it the same way every single time now. But there are lots of cool locally installed AI tools that you can use here. That's Pinocchio. Korea is taking a lot longer than normal. Let's see. So somebody's asking how are people using these types of videos in their business? I really think when it comes to like AI video generation right now, it's still kind of like a novelty. It's still just sort of fun to play with, see what we can get out of it. Use it as like memes on X and Instagram and places like that. I have used it for some B roll, so I have created some like YouTube shorts where I'm talking about a concept and I couldn't find the perfect video for the B roll, so I went and generated a video and used that. So I've used it for a little bit of B roll on some of my videos. So here's what you can see. Kreia did. It actually kind of makes it look less like me, but it's also a little bit more clear and clean. And when you shrink it down to the size of a small thumbnail on YouTube, it doesn't. You can't really tell that it looks less like me. So, like, let's see. Control. So if you make it, like, really small, like what you'd see on YouTube, right, it still pretty much looks like me. It's just when you zoom into, like, full size, you start to notice, like, okay, that doesn't totally look like his facial structure and teeth. It got the bags under my eyes, right? I do, like, permanently have bags under my eyes. So it adds that properly. Mr. Beastified. Yeah, that's kind of what it does, actually. It gives it, like, almost like a Mr. Beast, like, look to it. So that's my process there. Just to recap, I usually start with with Leonardo. The other one that I find myself using quite a bit more and more is Ideogram. Ideogram. I've been using this one a lot more. If you look at my creations, I started messing with this one to make some of my thumbnails as well as, like, the starting image that I swapped my face on. Like I was trying to do a man with his head exploding. So I've played with. So I'm usually switching back and forth between Ideogram and Leonardo quite a bit to get some of my images now. But I would say Leonardo is mostly my go to. And somebody also mentioned Flux and how Flux is really good for creating realistic images. Well, the cool thing about Leonardo is that they actually added Flux as well. So if you want to generate with Flux straight instead of Leonardo, you actually can do that now as well. Ignacio, do you use Grok? What do you think about that? Grok was better when it was using Flux, Definitely. I. I don't think it's as good as it used to be. So if I go to X and I go to Grok here, it used to use Flux when you did, like, an image generation, and they were really, really good. Now when you go to generate an image that uses Grox, like own AI image model, which isn't quite as good as what Flux was, it's definitely lost some of the. Realize the realism. So let's see. Drones flying over New Jersey. Oops. I need to tell it to create an image. Create an image of drones flying over New Jersey. Yeah. 1, 2, 3. ABC says Grok doesn't have guardrails. That is one of the benefits. There are some guardrails. There are definitely some things that won't generate for you, but it. It'll generate like celebrity images and actual people's likenesses and it will generate ip. So if you want to generate Mario or Sonic the Hedgehog or Mickey Mouse or whatever, it will generate actual, like copywritten ip. So it's uncensored in that way. But it's definitely not uncensored in like the adult way. They're not going to let you go generate like AI generated porn with it or anything. And by the way, if you use X, you got to follow lists. The for you feed is such a joke. Like I. I create a list called AI is Awesome. And pretty much every time I log in, I either fall, I either use one of these. I've got AI is Awesome, which is my list, which is just people that post about AI. I've got XR is Awesome, which is my list of people who just talk about, you know, AR, VR, XR stuff. And then I've got my YouTube is awesome, which is people who just talk about, you know, growing on YouTube and stuff like that. I've actually made all of these public, so if anybody wants to follow them, you can. But the AI is Awesome is pretty much what I keep as my default when I log into X is just see what all the AI people are talking about. So how do I share the list? Somebody's asking about that. I always have to like, find where you share the lists. Profile. No. Lists. Okay, so here's my list. AI is awesome. It's like kind of my main list here. This is the URL for it. Oh, Dylan beat me to it. I put the link in the chat, but Dylan put it in before me. So I've got that list. I've got my YouTube is Awesome list. So if you're trying to grow on YouTube, this one's pretty good for that. Here's my YouTube list. And then, yeah, those are. Those are pretty much what I follow on X. I try to ignore the home feed because it's all just politics and propaganda. When in trouble, just call dw. Let's see, what's your Leonardo affiliate list? I'm not sure what you're asking there. I'm not actually an affiliate of Leonardo. I'm actually a part owner of Leonardo. I have a very little sliver of equity. Oh, my affiliate link for Leonardo. I don't have an affiliate link. Just go to Leonardo. Go to Leonardo. AI. I don't actually have an affiliate link that I share. Have you played with Personas on Suno yet? I have a little bit, yeah. I don't know though. I haven't found it super useful because I just kind of make completely new random music every single time. Can we use Flux locally? I'm not sure. I don't believe you can yet. But you might be able to. Let me see something here. Look on GitHub. That's not it. Black Forest Labs Flux. So it looks like the code is available on GitHub under Apache 2.0 license. And this is Black Forest Labs. Let's see. So yeah, you can install it locally. I haven't actually tried this yet, but you can. There's the instructions. I wonder if there's any details here about. Yeah, it doesn't actually say the system requirements, so I'm not quite sure what is needed to run it, but it looks like you can run it locally if you want. I don't know, you might need like a 4090 GPU or something though. I'm not. I'm not 100%. Let's see. So that's Kreia. Kreia has some other new features that I haven't played with yet. I'm actually not a paying member of Korea though. So Kreia has Flux built into it. They have Video Generation. I haven't actually played with a lot of this yet though. One thing that's interesting about Kreia is you can actually pick any of the models here. They don't have VEO in it or VEO yet, but you can use Haluo AI Luma Runway Cling Cling Pro Cling 1.5. So you can actually like pretty easily test a whole bunch of different AI video models all in one place. I mostly just use their enhancer though. Somebody said, was it. Sarwar said that you can actually use Flux in Pinocchio. So I haven't tried. Somebody said, was it Sarwar said that yet? Maybe we'll play around with Pinocchio on like a future stream. Because Pinocchio has so many tools in it, right? Like you can open up Pinocchio and there's like 200 different locally installed tools that you can play with. Jim says, do you think agents are overhyped? I mean, I think anything that claims to currently be an agent is overhyped. I don't think there's anything right now that is a great agent. I think the best that we've got that's doing agentic type stuff. I think Perplexity is Really good for that. You know the way it sort of does a multi step search for you. And then also Google's deep research is really, really good and it sort of searches hundreds of websites for you to find information for you. Those are the best like agentic type tools that I've seen so far. That DO browser is pretty cool, but obviously we ran into a bunch of issues with it. Claude's tool use is very similar to what we just saw with the DO browser. So I don't, I don't know necessarily. I think the concept is properly hyped. I think anything that's currently out there right now is overhyped. All right, what else should I try here? All right, so I have hey Gen and I've got backflip. Let me log into backflip because this is one I've literally never tested. So we'll, we'll try it out in real time here. I don't know if I've got to pay for something or not. All right, so here's backflip AI. I know last week we were talking a little bit about how to create 3D printable objects using AI. And then later on I came across this backflip. I showed off Meshi last week. Then I came across this one which looks almost as good. So if I go start creating, I'll just do a wolf howling at the moon. Let's see if I can get the sort of pose right. So it looks like maybe it starts with an image. Okay, so it created an image. Now how do I create that into a. Let's see. Cannot sketch on this tile. I don't know how to turn it into a 3D thing here. Interesting. So I don't actually know how to use this backflip. I just kind of came across it and thought I would try it out so I can get it to draw a wolf howling at the moon. I don't know how to turn it into a 3D object though. Let me try that again. I probably missed something. Probably have to like select one. So this one right here looks cool. Make 3D. Okay. Yes, I did skip a step there. Money. 1, 2, 3. Please give me access to V2. 2. I don't actually have the ability to give anybody access, unfortunately. All right, that's not great. These wolves look like they've got broken necks. Yeah, I still think Meshi did it better. Does say it's refining details, so maybe they'll get better. No, no they won't. They are refined versions of the weirdly contorted Wolfs. But this is free too. I'm not paying anything for this one. Let me see. Let me try something else. A orc standing in a T pose. Oops. Not T post. T pose. All right, well, got an orc. Didn't get the T pose part. That might be okay though. Let's. Let's do this one just because it's looking straight on. Let's see. Eric says, do you know why Eden Necks are closing shop in January 2025? I actually don't know. What is Eden. X? Eden X Shutdown announcement we can't thank you enough for being part of the next shutdown announcement. We can't thank you enough for being part of the ENEX community. Create a passion. I don't. I didn't. I didn't know much about this one. Not even quite sure what this was. So was it an AI movie generator? Yeah, this is one that I've never actually used or played with before. So it could be passion. I don't. I didn't. I didn't know much about this one. Not even quite sure what this was. Didn't tell you why it's shutting down. Here's my orc, which isn't actually that bad. Let's see. I mean, it's not amazing, but it's not bad either. What options do we have for downloading it? So we do have an STL, which means I could 3D print it. I've got an OBJ, GLB and ply. Let me download the OBJ here. So was it an AI movie generator? I want to see what happens if I try to toss it into. What's it called? Adobe. What's the Adobe like pre made animations. Why can't I think of the name of this Adobe Mixamo. There we go. I wonder what happens if I try to toss it into mixamo here. Yeah, this is one that I've never actually used or played with before. So it could. This is a free Adobe tool that lets you like animate character. All right, so here's mixamo. Let's go ahead and upload what we just created. I'll pull this in here. We should have like our little orc dude. And tell you why it's shutting down. All right, so here's our orc. Let's go ahead and rig him up his chin. Here's my orc, which isn't actually that bad. Let's see. N is right there. Wrists right there. Elbows, guess right there. Knees right there. Groin, I guess right there. Let's see if we can get this thing to animate for us. People like to correct me on how I say things and it just makes me want to say them. I mean, it's not amazing. Wrong more often. Honestly, I say a lot of words wrong and I just continue even once people try to correct me and then I just like, lead it and say I'm wrong even more often. Because you know why? If I say words wrong, it creates engagement in the comments where everybody wants to point out that I'm saying stuff wrong. Like, I know inference is the correct pronunciation, but I still say inference. I'm pretty sure it's pronounced Gaussian. Splat. I still say Gaian, but it's not bad either. What options do we have for downloading it? Splat. So, yeah, I just say what I want and this is in my mind. Mixamo. All right, so we do have an STL, which means I could 3D print it. I've got an OBJ. All right, what happened? Unknown error while generating motion. Maybe it's because the wrists overlap a little too much. Oh, I don't know. Let's try it one more time. Wolf. E. You know what's funny about my last name? My dad was talking to his. His aunt and his aunt actually was going glb and ply. Let me download the obj here. Going through her parents like old belongings. So my dad's grandfather, she was going through my dad's grandfather's block. I want to see what happens if I try to toss it into. And she found his birth certificate. And on the birth certificate, there was no e at the end of wolf. So my whole life has been a lie. It's all been a lie. My great. It's either my great grandfather or my great great grandfather actually didn't have an e at the end of his name. And somewhere down the road, either my grandfather or his dad changed it and added the E. What's it called? Adobe. What's the adobe like? Pre made animations. Yeah. So mixamo isn't working with this generation for some reason, which is a bummer, probably because it's not in a T pose and the arms sort of overlap with the legs a little bit. If I had to guess, Why can't I think of the name of this? Thank you, Danny. She said, we still love you no matter what your name is. I'm not going to go and change it and remove the E. I just found that, like, really fascinating that it's only like two or three generations ago that that e was added to the end of My name Adobe mixamo. There we go. Wonder what happens if I try to toss it into Americans. Say herbs instead of herbs. I don't. I don't get the difference. What's the difference? Herbs instead of herbs. I think. I think in the uk, they actually pronounce the H, don't they? We say herbs. They say herbs. Let's see. Aggro chess. You can select a 3D preset before generating inside of backflip. Mixamo here, Part creator. Okay, so there's presets here. All right, so let's do a character preset. Let's generate it again. Maybe that'll use, like. Maybe more of, like a Control Net thing and control the pose a little bit. A little bit. As an American, this is a free Adobe tool that lets you, like, animate characters. Say herbs. Yeah. All right, let's try this one. Oops. I guess I have to convert it to 3D first. I say herbs for herbs. I still. I. All right, so here's mixamo. Let's go ahead and upload what we just created. Say herbs, too. I also say niche, and a lot of people go, no, it's niche. I'm like, yeah, but I say niche. I also say gif. And a lot of people try to tell me that it's gif. I'll pull this in here. We should have, like, our little orc dude, but the G stands for graphics, which is a G. So therefore I call it gif. And in the US we say aluminum, but I still like to say aluminium because I just think it sounds funner. I think aluminium is more fun to say than aluminum. All right, let's try to get another orc here. See if I can get this one to actually dance for me. I need a dancing orc before this stream is over. Somebody's arguing giraffe. Is it G? But. And what is. What does GIF stand for? Let's see. Gif. All right, so here's our orc. Go ahead and rig them up. What does GIF stand for? Graphics. G. G. G Graphics. Right. If it was a giraffe interchange format, then I would call it a gif. And it's right there. Wrists right there, elbows. But it's a graphics interchange format, therefore I call it a gif. Let's see. I lost access to chat. Let's see. Guess right there. Need. I just got a weird message in obs that my chat stopped working, but it must be back now. Okay, I think we are good. Don't know what that was all about. These right There. Groin, I guess, right there. The creator of the gif file format said it's G if not j, if it definitely could be pronounced either way. I mean, if somebody says gif, we know what they're talking about. Or gif. But I just feel like if the first word is graphic and graphic starts with a g, then gif should start with a G. Let's see if we can get this thing to animate for us. Have you seen the new video to audio generator? I've seen a few of those. If you shout out the name of what it's called, we can take a look. All right, let me get into mixamo again here. I want to get this guy to dance for me. All right, so there's the chin, the wrists, the elbows, ish. The knees. And then the groin is right about there. Aggro chess. I understand how backflip works now. It works based on what it sees from the front of the 2D image and tries to make it 3D. It struggles to make Darth Vader's cape, and it couldn't create it all. People like to correct me on how I say things, and it just makes me want to say, All right, wrong more often. Honestly. I said we need our dancing work. Let's see. Any AI separator, separate audio from a podcast recorded on one track. One person on one track and the other on another. You know, I've actually been looking for something similar myself, and I haven't found something that does that yet. I will be on the lookout, though, because it is something I've wanted myself is to be able to split up the tracks. All right, we got our orc here, and it's all rigged up a lot of words wrong. And I just continue even once people try to correct me, and then I just, like, lead it and say I'm wrong even more often, because you know why? If I say words up. So now let's make this guy dance. Let's get him out of the floor first. All right, let's do. What should we have him do? Do the twist. There we go. Now we have our orc doing the twist. This is really fun, by the way, to use as, like, the start of a video to video. So you can like, actually screen capture, like, an orc dancing here in Mixamo or mixamo, and then take that video and then throw it into a tool like runways video to video and actually, like, you know, colorize it and make it look cool. That's like, a really fun workflow. Wrong. It creates engagement in the Comments where everybody wants to point out that I'm saying stuff wrong. Let's see. Like, I know inference is the correct pronunciation, but I still say I can do my follow camera. It would be nice if I don't believe there's a way to, like, download a video of it dancing, which would be really cool. Yeah, it doesn't have a way to do that. I wish Mixamo would do that. Like, give me, like a MP4 file or something with this dance on a loop. Or if anybody knows a good way to inference. I'm pretty sure it's pronounced Gaussian splat. I still say gaian to do that without just, like, screen recording my entire screen to get the animation splat. So, yeah, What else? All right, I'm just looking through your comments. I'll probably start wrapping up here. But before I do, I'm going to take a super quick break about one minute. And while I'm on this break, I'm going to show you a video that Dylan made. And, you know, he wanted to wish you guys a merry Christmas. So this was the video that he made. And I'm gonna play it, and I'll be back in a minute. I just say what I want, and this is in my mind. Mixamo. Hello? Who's there? You scared me, little one. What are you doing up? See Santa? Of course I'm Santa. Now get to bed. All right, what happened? Unknown error while generating motion, maybe. No, no, you're not Santa. Listen, little boy, I'm Santa, okay? I am. I am him. You're a bad man. No, don't you. Don't you point that at me. I'm Santa Claus. It's because the wrists overlap a little too much. Oh, I don't know. Let's try one. Leave the bag, Santa. Leave the bag. Watch, watch. You want to shoot? Oh, do it. You're not going to shoot me. So there you go. That was. That was made by Dylan. I think he actually has a tweet about how he made it. Let me see. I gotta go back through all of his messages to find. Find the tweet where he sent that over. So here you go. If you're curious how he made it, he recorded the voices himself and edited in da Vinci. But he used Luma labs one more time. W Runway halo AI cling. And then made the music with Suno. You found this on the dark web, right? I don't know if you consider Dylan's hard drive the dark web. You know what's funny about my last name? My dad so yeah, I don't, I don't know exactly what the prompt was. Dylan, he may be able to jump in and respond to that. You guys don't realize it, but that was my bathroom break. I managed to, to run to the bathroom and run back. Oh, in the one minute that that video was playing. All right. Was talking to his, his aunt and his aunt actually was going through her parents like old belongings. But I'm totally down to test more tools, talk about some more news. I know since fr, like last week was a crazy AI news week, the week before it was a crazy AI news week. This week is the week of Christmas. Next week is the week of New Year's. So I have a feeling we're going to have a bit of a slowdown. Less to talk about. So I, I still want to do a live stream like this next week and just kind of play with the tools since there won't be a lot of news to talk about. But if there's many, if there's any other tools you want to play with between, you know, now and I don't know, I'm willing to hang out for an extra. Let's see, 11 to 12. 12 to 1. So my dad's grandfather, she was going through my dad's grandfather's belong. Yeah, I'm willing to hang out for another like 30 minutes or so. I think I have a call at 2 o' clock my time. Let me just double check here. Yeah, so I do have a call. I have to jump on at 2:00 clock. So I have to be done within the next 42 minutes. But I'm totally down to. I'm totally down to test some more tools. And she found his birth certificate and on the birth certificate, let's see what else can we. Oh yeah, I was going to play with hey Jen a little bit as well. So hey Jen has this new like my podcast feature where you can upload like PDF files and it will create a podcast like NotebookLM but will actually have people speaking. There was no E at the end of Wolf. So my whole life has been a lie. It's all been a lie. So let me go to my little. One sec here. Let me go to my like lie. My great. It's either my great grandfather research papers tab here and see if there's any interesting research that I can throw the paper in. If you want 3D image resolution. What's this one? So AV link temporarily align diffusion features for cross modal audio. My great great grandfather actually didn't have an E at the end of his name and somewhere down the road, either generation. Let's go ahead and pull this paper in here. View PDF so I'm going to go ahead and just toss this PDF in and see what it does. See what the podcast looks like. My grandfather or his dad changed it and added the E. No. You got a walkie. Wonky Dragon Dropbox. It doesn't want to let me drop it in. I got to do it the old fashioned way. Come on, you can do this. Open, Open. All right. Hey, Jen's acting Wonky. Come on, come on. Let me upload the PDF. It says drag and drop a PDF, but it won't let me drag and drop it because it twitches out. All right, maybe I won't be testing. Yeah, so Mixamo isn't working with this generation for some reason when this. Hey, gen. It doesn't seem to want to work. Let me see if I can just pull in this PDF URL. Will it actually grab it from the URL? Okay, maybe. So it picked one of the. Which is a bummer probably because it's not in a T pose and the arms sort of overlap at ours. Let's have this woman talking to this guy. Make it less than three minutes. We'll make it 16. 9. Why can't I submit it? All right, well, this heygen podcast feature seems to be pretty buggy. Or I'm totally missing something. Up with the legs a little bit, if I had to guess. So I have a URL here. It lets me click to the next step to pick an avatar and a second avatar, but then I can't submit. It won't let me actually, like, load a PDF file in here. Like it claims. It says larger than 20 megabytes. How big is the file? I'm trying. Maybe I'm trying to upload one that's too large. Let me see. It's 20.5 megabytes. All right, let me try this. Frontier math one. How big is this? I think the option. The issue is that it was too large of a file. All right, so this one's only 1.58. This one here. So let's. Doesn't work that way. So let's try it this way. Okay. Okay, we're onto something. It pulled in my PDF this time. All right, so our two characters will do less than three minutes. All right. Wants me to log in? Sure. Thank you, Danny. She said we still love you no matter what your name is. What did I do? I'm not going to go and change it. And remove the E. I just found that, like, really fascinating that it's only like. Damn it. No, that's not what I wanted. Hey, Jen. All right, all right. Let's see. I got to try this one more time. Pull in a PDF. It opens it twice. For some reason, I'll select this person and then this person. Click. Two or three generations ago, that that E was added to the end of my name. Submit. And it wants me to log in again. Okay, log. And then it started over. Okay, this site's kind of a wonky site. Let's try this again. Open. Okay, it opens it again for some reason. This person, this person. Submit, accept all. Okay. Okay. It's generating something. Thank you. Thank you. 5% ready. So make. Maybe it'll be done before this stream is over. Leon says love the live stream. Thank you, Jordan. Americans say herbs instead of herbs. I don't. I don't get the difference. What's the difference? Jordani Matt, are there any tools available that autocut videos? Not like a narrative video, but more like a highlights video? There's a couple. So herbs instead of herbs, I think. I think in the UK they actually pronounce. Let's see. I use a tool called Time Bolt to do most of my cutting, but really what that does is it cuts out all the silence and mistakes. But it's. This is what I use for most of the age, don't they? We say herbs. They say herbs of my editing when I edit videos. You can see the timeline here. All the red is like sort of silence and mistakes. It tries to find those and cuts them out for you. Hey, look, it's even got my name on there. So that's what I'm using. There's also a tool called Gleam, I think so what it is. No, that's not it. Glean. Maybe I can't think of what it was called, but there's another one. Maybe I can find it on. Let's see, Aggro Chess. You can select a three feature tools. Pretty sure it started with a glass and then podcasting. No, there is another tool that does the same epreset before generating inside of backflip kind of thing. And I can't think of the name, but I will find it real quick. Part creator. This is literally why I created feature tools. Because I forget the names of tools and then I need to go look them up again. What? Okay, well, I can't think of what the other one was called and maybe I never actually saved it to future tools. My bad. Okay, so there's presets here. All right, so let's do a character preset. Let's generate it again. The other one is Opus, but this one is more for, like, making vertical videos. So you can toss any video in. In fact, you can see, last week I tossed in the live stream that I did and have it find a bunch of clips from the live stream for me. So that's for taking long forms and helping you find a bunch of short forms out of the long form. That. That's really helpful. But you can also do, like, searches within your video. Like, find where I talk about making plushies. And then it will actually look for a clip and generate a clip of where I'm talking about plushies. Maybe that'll use like. Maybe more of like a control Net thing and control the pose a little bit. A little bit. As an American, I. Gling. Yeah. Okay. So that's what the other one was. I just needed to look at the comments and people would have told me, let me see, do I have gling on future tools? Say herbs. Yeah, I will be shot. Okay, so I do. But I have it under the video editing category and not the podcasting category. So I need to fix that. I need to make sure it shows up when you search for podcasting as well. So that does kind of the same idea as Time Bolt. So this. All right, let's try this one. Oops. One very similar to Time Bolt. It cuts out a lot of the silences and mistakes and ums and things like that. And then Opus Clip takes longer form videos and cuts them down into, like, short clips that you can use. Not the fastest in the world, as you can see. But also on my last livestream, you can see I generated 53 clips from my last livestream. And if you've been watching my YouTube channel at all, I mean, nobody notices this but me, obviously. But if you look at my shorts, pretty much a new short drops every single day now because I'm just pulling them from my live streams. And that's with the help of Opus Clip, helping me find the life. The clips that are worth clipping, They probably call it because I have to convert it to 3D first. Gling because it sounds like cling. I don't think so. I say herbs for herbs. I still I. Because I knew about gling long before I ever heard of cling. I think gling has actually been around longer. Say herbs too. I also say niche. And a lot of people go, no, it's niche. I'm like, yeah, but I say Niche. Okay. My hey gen video is 51%. My ogre still dancing. Good, good. Let me look in my mind, look up some of the cool stuff I've been looking at recently. If you've never followed any of my old videos where I talk about how I, like, organize what I do whenever I come across anything online that I find really interesting, trusting that I'm like, oh, I need to make sure I resurface this later, either for a video or for a live stream or something like that, I always save it with my mind. You can see I've got this little, like, plus button here that says save to my mind. Anything interesting I come across, I press this little button and it saves it in this, like, giant database of, like, interesting stuff I've come across. And then once I finally talk about it on a live stream or in one of my videos or something like that, I actually give it the tag archive. So you can see everything I'm looking at is not archived. If I just go to everything, you can see, like, everything I've ever saved. And this is just like an infinite scroll of tons and tons and tons of things going back a couple years, but go all the way to the bottom. Like, some of the oldest stuff I saved. Let's. I also say gif, and a lot of people try to tell me that it's gif, but the G stands for graphics, which is a G. See, what is this one doesn't want to load. Oh, it's because I wasn't all the way at the bottom yet. So I saved a post from Luma at one point. Why is it not opening? So this was back in, oh, February of 2023. So therefore I call it GIF. I saved, like, what Luma was doing back then, so I can go back through and look at, like, exactly what I was looking at, like, a year ago. And then. If I click on not archived, it just shows the stuff that I haven't talked about yet. So let's see. Like, here's something that I thought looked interesting. This guy made, like, a projection that understands what's going on in the room around you. And then he has this little, like, Santa character that he can control with his phone, but it actually recognizes where, like, the tables and stuff are. And the game is he has to figure out how to get to this gift by jumping around on the objects that it recognizes in his room. And I thought this was really, really cool. And I'm like, I want to figure out how to play with this thing. And then what else Do I got in there? I'm just kind of blabbering one. And in the US we say aluminum, but I still like to say aluminium. Waiting for some of this other stuff to finish generating. What was this one? How to merge photos in Photoshop. Oh, yeah. I save stuff like this, like little Photoshop tricks because I just think it sounds funner. I think aluminium is more fun to say than aluminum. All right, let's try to get another orc. LTX Studio came across this ad for like this pocket thing which looks just like the plod that I already have. It's like a little attachment that goes on your phone that can record your conversations for you here. Let's see if I can get this one. To actually dance for me, I needed pre holiday gift from Pika, which I think is over now. So I'm gonna archive that. Oh, this is pretty cool. This is from, oh, what's his name? Ian and something. But it can actually see my camera and like my hands could like move this stuff around so, like you could see where the two green dots are. That's like where my hands are on the screen. And if I move them, it sort of like lets me interact with this environment. You can see a little computer vision module up in the top left as I like play with this. I thought that dancing orc before the stream is over. That was pretty fun. So I saved that because I want to learn how to do it, But I want to shout out who makes that Ian See? Ar. Dang it. My mind. Is my mind free? It definitely has a free tier. Ian Curtis. There it is. So Ian Curtis made that, like he makes all this cool, like XR stuff, Including that little like hand thing that I was just showing you. So if you're really into like XR augmented reality stuff, See, here's that hand thing that he made. Here's him showing it off. Pretty cool. Free to use. He linked it up. It's just this Hand Lab Netlify app. All right, let me see. I saw some other questions come in. Let me see what else you guys were talking about. Is my mind free? I just answered that I'm not 100% sure I can open it up in a new incognito window and tell you the cost of my mind. I thought it had a free plan, but maybe it doesn't. All right, so their cheapest plan, seven bucks. Oh, no, they have a four dollar a month plan. I actually don't pay for it. I think they actually gave me access back when they first released it. So this is one that they kind of hooked me up with and then I've been stuck using it. What else? No way. This is real. I'm not quite sure what that was in response to. Let's see. You seem jack of all for AI, but actually, what is your expert at? I mean, I kind of try to be a jack of all trades. Really. What I'm an expert at. What I spent the last, like, decade doing was marketing. I studied SEO. I studied like funnel building and copywriting, and I did paid advertising with Facebook and Google and a lot of those platforms. I spent over a decade working in digital marketing, helping companies do their marketing. And I was always like a tools guy. I was always trying to find new tools to help with the marketing, help with the strategy, help with the automations of companies and stuff like that. And so it kind of got to a point where people started coming to me to ask questions about what tool should I use for this, what workflow should I build for this, what automation should I use for this? And I kind of became like the workflow automation tools guy. And then when AI started to bubble up in 2021, a lot of the tools and automations I was talking about were referenced AI. Right. There was a lot of tools starting to use AI back in 2021. So I started talking about more and more AI. And then over time, it just evolved into like, now almost everything I talk about is AI. Let's see. Lio says, what do you think about AI art? Can people take credit from this AI art? I think that is a very nuanced question with nuanced answers. I think depending on the art, we can or cannot give credit to the creator. I would say that, like, if. If somebody used a whole bunch of different AI tools, right? They used midjourney to create one image and they used Leonardo for a different image. And then they pulled it in Photoshop and figured out how to make them, like blend together. Or they, you know, used midjourney and then like, did a whole, like generated an image and then masked it out and then regenerated and then masked out a different area and then regenerated and sort of kept on working at it and dialing it in until it was the exact image that they had in their mind. Cool. I now think that the person who created it is an artist. They put a lot of thought and effort into working it and working it and working it and molding it into the thing they wanted it to be. Right. If somebody goes to midjourney and says, generate an Image of a cat with a really colorful background and they get a really amazing image and then they start sharing it everywhere and go, look, I'm an artist now. I made this cool cat with a colorful background. I can't claim that person is an artist. You just typed a prompt and got lucky with a really cool image. Same with, like, if you prompt over and over again, like you just hit reroll 100 times until you get the image that you wanted. Are you really an artist? Or you just press a button until you got lucky and found an image that looked good. So it's very sort of nuanced in my mind. I think the level of effort put into getting that image, that work of art that you created is an important factor in whether or not I would myself. This is a very subjective thing. I myself would consider it art if I feel like you put a lot of effort into it versus just rerolling a prompt over and over again. Same with like music videos or videos like what we just showed with Dylan and the Santa Claus thing. Like, in my mind, that took a lot of effort. You saw on his tweet that he used six or seven different tools and he generated the music and then he pulled it all together in DaVinci Resolve. That wasn't something where he just entered a prompt and then 30 seconds later he had a minute long Christmas video. Like, there was actual effort put into that. So in my mind, cool. That's. That's art that he had AI assist him on. Had he gone to like Sora and said, make me a one minute video that has these elements in it and Sora just spit it out. The level of artistry in that went down in my mind. Let's see, are any of my other things done here? All right, so we've got a clip about Frontier Math inside of hey Gen. So this is their new podcast feature that like Notebook lm, but it actually generates people on camera talking about it. So have you heard about Frontier Math? It's this new benchmark for AI models. No, I haven't. What's it all about? Well, it's like the Mount Everest of math problems for AI designed to test advanced mathematical reasoning. Mount Everest, huh? So it's really challenging then.

Exactly right. The problems are.

I really don't like the voices. They don't come across natural at all. Are exceptionally tough.

Some can take expert mathematicians hours or even days to solve.

That's intense. So why create such a difficult benchmark? Well, existing benchmarks have become too easy for AI models. I see.

They've kind of outgrown them.

So they needed a new challenge. Precisely. Frontier Math steps in to push the boundaries to see how far AI can.

Really go in mathematics.

But aren't these models already pretty good at math? They are at solving standard problems. But Frontier math includes research level questions across various fields. Research level, that's like grad school material. Exactly. It covers major branches of modern math. All right, I'm gonna stop it there. I can't. I can't watch more of it. I don't really feel like adding the video into the mix really added more just because, like, they're so, like they're very like stationary. They're very like, I don't know, they're set in motion or they're, they're like frozen. But every once in a while you kind of see them move their hands or, you know, go like this or something. But I don't feel like this is any better of an experience than what you'd get out of Notebook lm. Right. Like, I'm fine just listening to this on audio versus watching two fake people say the same stuff. I'd probably rather just use Notebook LM because it's quicker to produce. I also think it's interesting that they add a bunch of like, us in there. Like, they don't have to do that. That was a choice by this model to make them add us and like the sort of filler words that humans would use. They didn't have to do that, but they chose to. And I don't think it's necessary. But that's the new HeyGen video podcast. I wanted to try it here. My opus thing, for whatever reason, failed from my prompt, but this is one that you can see. I did have it cut up tons and tons and tons of shorts from my long form videos, so it's capable of it. I don't know why it failed on my last attempt. All right, let me see. I saw some other comments come here that I wanted to answer and see logical flow. I know you aren't an expert on Quantum computers, but 2025 is likely to be a huge year for it. Do you have interest in building it in more as a topic going forward as you learn more? I do, yeah. I mean, I'm really interested in quantum computing, but it's also one of those areas that I feel like if I tried to talk about it, I would probably come across as dumb because I don't think I would explain it very well. Like, I understand some of it, but it's still very convoluted. To me. Like, I sort of understand the concept that with regular computers you have binary on off one z ones and zeros, but in quantum computing, they can sort of exist anywhere in between as well. So there's not. It's not as binary and it uses qubits instead of regular bits. And like, I sort of kind of get it, but also don't feel like I can talk too intelligently on it yet. But if I dive deeper down the rabbit hole and I have a better understanding, then I will likely talk about it more. But I don't really want to talk about something that I feel still this ignorant. Let's see, Justin. A fi Notebook LM might be the best learning tool available right now, paired with deep research. And you can literally, you can literally learn anything. I totally agree. Deep research is really, really good. And then you could literally copy everything that deep research gives you into NotebookLM. In fact, I would not be surprised if that's an upcoming feature that Google will eventually roll out. You know, generate some deep research inside of NotebookLM, then press a button or generate some deep research inside of Google's, you know, deep research tool, press a button when it's done to send it to NotebookLM and have them make a podcast around it. Like, I bet that's happening. That seems like the next sort of logical evolution of it. What's up, Todd Bergman Rogers. Great to see you as well. See, as I understand there's only a handful of hours that benefit from the speed up just in a 5, quantum computers likely won't be commercial until 2030, especially as Willow won't be commercially available till then. Have I looked at cling 1.6 yet? I've looked at it a little bit. We just recently did a podcast that's not out yet with Tim from Theoretical from Theoretically Media, and we talked about all the various AI tools and he ranked this one very high. I haven't played with it a ton myself. Sure. Do I have like a certain amount of free credits. 326. What should I generate with cling? Text to video. Let me just do a monkey on roller skates. I actually haven't seen cling generate this one yet, so let's see what it looks like. I'll just leave everything default and see what it does. Cling will take hours to make a video. Okay, well, we might not wait for it. It says estimated wait time longer than three hours. Yeah. Okay, so we won't be seeing what this cling output looks like on this live stream, unfortunately. Oh, it looks like I have done a monkey on roller skates. This would have been. What model would have been, though. Is there anywhere where it tells me what model I used? Because this would have been before 1.6. Obviously. I made Trump and elon hugging at one point with 1.5. But I don't. I think it actually. Oh, wait, I think it actually says down in the bottom. Right. But because it's like white on white, you can't read it. So I don't know what model this was made in. And I will know what my monkey on roller skates looks like in three hours. Can I browse away and come back? I'm assuming I can. So now if I go to my AI videos, it's still generating. Okay, so I can close this and come back. Logical flow. Fireship has an okay recent video on it. Yeah, I love Fireship. One of my favorite channels. I pretty much watch every video Fireship puts out. I'm gonna have to wrap up here in just a minute just because I have a call with my accountant in about 13 minutes here. Hours. More like days. Yeah. A 1000 minute wait on hello AI right now. Geez. All right, what do we got here, Dylan? So AG says AI and business marketing, please. It would be refreshing from audio Video Robotics, but AI and business healthcare would be incredible. Yeah, I love to talk about how it works in business and marketing. I mean, my main goal with my channel is to focus on AI that impacts the most amount of people. Right. I don't really want to dig in too deep onto like marginal improvements on large language models or super marginal improvements on AI image or video models. Like, my main goal is to show off. This is the AI that's impacting real people in the real world. Right now. I feel like in the world of AI, there's so much novel AI stuff coming out that it feels just kind of like a novelty. It feels like, all right, that's a really cool tool. I'm impressed by what it does, but I have no idea how I'm going to use that in any real world scenario. And I feel like there's a lot of that happening in AI right now. And so my big goal is to go, all right, here's how you can actually implement AI. That's my goal with the tutorials. That's my goal with the top 10 breakdowns when it comes to the Friday AI news videos that I break down. You know, literally, I have about a hundred news articles every week, and I gotta narrow it down to maybe 20 to share on the video at the most. Right. So I'm trying to figure out what news will the most amount of people actually find interesting without, you know, just showing the two niche stuff that there's only like seven people that are gonna care about that piece of news. So I'm trying to find the stuff that has the most broad appeal, the most impact on the most amount of people. And that's really always my goal with any of the content I'm making is how do I make sure I'm sharing the content that appeals or applies to the most people. And a lot of times AI with relation to productivity falls in that sweet spot. AI in relation to how it can help businesses and you know, workflow automations and things like that. And then also, you know, just for fun. I do think just using these tools for fun is a valid use case. I think going and using video tools just to see what they can generate, or going and using tools like midjourney or AI art tools just to see what you can get out of them and create something that you never thought you would see an image of just because it's fun. I think that's a good use case. And a lot of people like using these tools just for fun as well. So like I want to lean into those areas that apply to the most amount of people. So that's kind of always my goal. My thinking going into every single video that I make and then these live streams, when I do the live streams, they're kind of more all over the place. Like let's get the community all together, let's get a good chat going, let's just see what people want to talk about. Let's try different tools. Let me see if I can help with different, more niche use cases, things like that. So my recorded videos are for more of a broad how do I make this stuff appeal to the most amount of people? My live streams are all right, let's kind of get in the weeds a little bit. That's sort of how I'm seeing the channel at least. AI world is full of coming soon. That is very true. They're a social selling warrior. There are many non tech folk looking to simple, easy to understand applications. That's totally right. That's kind of what I'm doing. I'm trying to figure out how to show people how you can apply this. Like where, where is this actually beneficial to people? Justin says many more use cases for embodiment. Totally agreed. I think we're gonna see the embodied AI. You know, AI basically put into robots and wearables and things like that. I think we're gonna see a lot more of that next year. Hybrid robotics. I have been asking a question. I have not heard it answered. I mean, ask it again. There's a lot of comments in here. So if somebody's asking a question and not getting it answered, ask the question again. Maybe we didn't see it the first time. It's hard to keep up with all the questions. On that note, it seems like the stuff has kind of slowed down. The comment section has slowed down. So I am going to go ahead and wrap this one up. Barry says thank you, Matt and Dylan, for the live stream. Have a wonderful Christmas and New Year. You as well, Barry. Thank you very much. What agent is good at the moment? We kind of already talked about that. I think. You know, the DO browser was interesting, but still has some things to work through. I really like tools like Perplexity, which sort of does like an agentic workflow to like do some research. Google's Deep research tool is really good as well. As far as agents go. Matt LeGrand. You should make a video talking about what channels you like and follow. It's a good idea. I'll probably talk about that maybe on like the next live stream. Break down my favorite YouTube channels. I actually don't watch a lot of AI channels. Most of what I watch are channels around, like, how to make my videos better, how to edit them better. I watch a lot of like, like what's it? Veritasium. And I watch Marquez Brownlee. A lot of. A lot of other tech channels. Justin Ify. You take care as well. Merry Christmas to you. Is there a tool that allows me to make changes to existing image, such as making it look 3D? Yeah. I mean, you can toss it into almost any of the AI tools. You know, you can even. You can do that in Leonardo. You can actually upload a image and tell it to follow that image. I believe Midjourney has something like that as well. So in Leonardo, if I go to image creation under advanced settings, there is. Maybe it's not under advanced settings. Where is it? They have a way to use a driving image. Maybe they moved it. Yeah. So you can upload an image up here. Image guidance. So you can use it as a style reference image to image or content reference or depth to image, pose to image. They've got like a lot of the Control Net stuff that they're building in here. But you could upload a reference image and then, you know, tell it to make that image look 3D and then midjourney has something similar. You can upload a reference image and give it some guidance of how you want it to look as well. Most of the image generators have something similar now. Logical Flow. Thanks for the 10 bucks. I really appreciate it. Merry Christmas and thanks. I also think you're 100% right. AI will develop quantum before quantum develops. AI love your videos. Awesome. I appreciate it. Joka Buju thanks again bud. Merry Christmas. Merry Christmas to you. I miss your ending song. Should I bring back the ending song on my videos? I could always bring that back. I just. I don't know. I stopped using it because it was just like an extra step. But if people miss it, I'll bring it back to my videos. Let's see. Social Selling Warrior Thanks Matt. Love your videos. Watching from Australia. Awesome. Appreciate it. You must have started a bit early today. Justin. Later and Happy New Year. Allison Merry Christmas. Merry Christmas to you. Stephen Porter I love the nerdy geeky side of you. May have a session in the future to script some of these services into meaningful tools. I mean the whole side of me is geeky. I don't think there's any side of me that's not geeky, honestly. Yana 1113 Merry Christmas. Merry Christmas to you. Restful Christmas, Matt from Patrick. Thank you. Awesome. Well, thank you everybody. A lot of Merry Christmases, Happy Holidays, Happy New Year's, that kind of stuff coming in. I'm going to do one more live stream before New Year's. I am going to go live next Monday as well. Don't know if there's going to be anything to talk about. Might be a shorter stream. We'll see. But I am going to go live one more time this year and so hopefully there's some cool stuff to talk about. Hopefully I up level the stream slightly. My goal is each one of these live streams get it a little bit better than the last. Maybe up the tech, maybe up the music and production quality. Just do something to marginally improve these streams each time. So the next one, hopefully I'll have a little bit of improvement. Really at the end of the day it's just sort of like getting in the flow and I want to, I want to keep on sticking to them every Monday. I'm also going to be at CES on a Monday. I'm toying with the idea of actually going live from CES and walking around the show floor with a live stream going. I don't know how that's going to work out, but I might try it while I'm at CES in a couple of weeks and yeah, I'm gonna go ahead and wrap up. Thank you so much again for tuning in. I really, really appreciate you all and hopefully I'll see a bunch of you next Monday. If you have more things you want me to try on the live stream, different tools you want me to test, different news you want to talk about, let me know over on X or in the comments of this video. We'll keep an eye on them and make sure that in the next live stream we go down some of those other rabbit holes. But that's what I got for you today. And thank you so much for tuning in. I do have to jump on this other call in three minutes, so better stop rambling. Really appreciate you guys. Have a Merry Christmas, Happy Holidays, Happy New Year's. If you're not going to be on next week, all that good stuff, I really appreciate you all tuning in and nerding out with me and I'll see you in the next one. Bye. Bye.