Google's New AI VIDEO MODEL 👀

Added: 2026-05-19

[00:00:00]There are rumors about a new AI video generation model by Google that is set to drop this week. You can now buy a mecha suit for half a million dollars.

[00:00:12]And there's an unreasonably effective way to work with LLMs with Chad GPT and Claude that most people don't know about. So we're going to share that with you and hopefully your workflow is going to be 10 times better. And there is also a new AI interaction model that can think, listen, and talk in real time.

[00:00:32]>> That's really important because right now when you're talking to Chad GBT, it gets interrupted every time you say something. But when you talk to a real person, >> it doesn't happen.

[00:00:40]>> Yeah. Far talks over me. I overlap on top of him. It's very real. That is what this is about.

[00:00:46]>> But before we begin, welcome back.

[00:00:48]>> Hello everybody. It is Tuesday and we are here with another episode.

[00:00:54]>> That's right. Every Tuesday, every Friday, we bring you the latest and the greatest in the world of technology. And we just wanted to say welcome back.

[00:01:02]Thank you for being here. We got four hot topics for today and we're going to get started right away.

[00:01:09]>> Let's go.

[00:01:10]>> Are you ready?

[00:01:10]>> Yes.

[00:01:13]>> So, first up, we're going to be talking about the new Google model. There's a leaked video circling the internet about this new AI video generation model which is called Google Omni. And the reason why people are talking about this, why people are excited about this is this exact thing that is happening right now.

[00:01:38]We see a professor writing out a mathematical function on a board. What's fascinating is the text rendering which is incredibly accurate.

[00:01:52]This is something that has been >> a huge problem >> complex to achieve both in images and even more so in videos. Can we have consistent text rendering which is crucial for things like making product videos or marketing material or training content because the moment the camera shifts or moves all the text begins morphing. And secondly, apparently the mathematical functions and what's being explained by the professor are actually accurate contextual to what he's talking about. Just like how Nano Banana 2 with the image generation model, you could ask it for the current weather in Paris and have it create an infographics for you or a weather update. It would go ahead and pull real-time data from Google and create that image. So, it's not just lies. Apparently, the same thing is true here. I mean, I'm no math expert. You guys can go ahead and confirm this right now in the comments.

[00:02:49]Far, how good are you at mathematics?

[00:02:51]>> Pretty good. I think >> not good enough to verify.

[00:02:54]>> No, I can. I think it's a tangent equation. Wait, let me say what is a tangent equation if I'm not mistaken.

[00:02:59]>> Let's listen to the video.

[00:03:00]>> It's sin square.

[00:03:01]>> We can listen to the video.

[00:03:03]>> Let me see what is >> now. If we divide every term, we start with the fundamental identity sin^ 2 plus cosine^ 2 = 1. Now if we divide every term by cosine^ squ, we arrive at the identity for tangent. We start with the fundamental identity sin^ 2 + cosine^ 2 = 1. Now if we divide every >> at least whatever he's talking about he's writing it on the board whether it's factual or not.

[00:03:28]>> What my question is why they didn't call it V4. Why Google Omni? I believe this generation model is not Google Omni. The name people saw on Google's website in the Gemini app. Pratt, do you have the leaked image that people shared? It says create with Gemini Omni >> or Gemini Omni. Omni present video means everywhere >> and everything and I'm imagining omnipresent I'm talking about omnipresent but Gemini omni is perhaps everything in one place video >> video generation model >> now you have to understand why this is important for us >> aside from the superior text rendering and that being incredibly important for genuinely let me rephrase that a lot of the content that you see on social media that is hyping up all of these AI tools, a lot of them are made for that 15 seconds hook just to get your attention.

[00:04:30]But when you actually go ahead and use these tools for real life scenarios, you'll realize quickly that not all of them work for production ready content.

[00:04:42]For instance, you may see crazy clips like the the latest one that went viral.

[00:04:47]Pratt, can you pull up the tsunami video that went viral? There's a tsunami video made with >> like there's a couple there's a couple fighting and then there is a >> tsunami AI video.

[00:04:59]>> Let me pull that up really quickly.

[00:05:01]I'm sure it's there. Not that one. The second one. I think someone remade that.

[00:05:06]No, that's not the original one.

[00:05:07]>> I know. I know what you're talking about. It's like the couple >> Hicksfield even made their own, by the way.

[00:05:13]>> I just >> I mean, yeah, we can just >> just play that. And this is not it, but there's a video like this that went in incredibly viral on Instagram. Oh, maybe it's the one at the top. The second one.

[00:05:22]>> This actually like this.

[00:05:26]>> Someone made a video like this.

[00:05:28]>> Yeah, >> it looked a lot more realistic, by the way. That the one I'm talking about. And it has millions of views and likes and comments on Instagram.

[00:05:37]You see these videos you'll see and say, "Oh my god, these AI tools are great.

[00:05:42]I'm going to use them for my next marketing campaign. It's 100% ready for production." The truth is in your marketing campaign, you're probably going to be using a product with labels on it, with information that needs to stay consistent in every frame of the video. Here, you're looking at a tsunami. Water that is moving in a fast-paced with the shutter speed involved and the motion blur. This is the video.

[00:06:05]>> This is the video.

[00:06:07]See, as long as the water simulation looks pretty decent, you're good. You're not selling anything here. People don't really care about the minute details.

[00:06:14]They just look at the overall picture and they're amazed. It looks great. Then you go ahead and use the same thing for your product. It doesn't work. It breaks. That is why we talked about superior text rendering because if the text and information can remain consistent in every frame of the video, that really pushes the limits of what you can do with it in terms of real use cases. Now, we have to go ahead and compare what we just saw with the best AI video generation model that is out there today, Cedance 2.0. Cance 2.0 is available on pretty much every platform right now. However, the biggest problem with that is its price. It's incredibly expensive and hard to use because of the IP infringement and copyright issues that they're facing. You cannot always use the faces that you want to use. What I'm assuming will happen with Gemini Omni, my hope, is that their quality is going to substantially increase, become a little closer to what we have with Cedance 2.0 you know, and become competitive with that >> and cheaper >> and potentially cheaper >> more accessible >> for the average person to go ahead and use it.

[00:07:26]>> But Google V3 had its own limitation as well, right? We couldn't prompt Batman if I remember correctly.

[00:07:33]>> You couldn't use you couldn't use >> workarounds always a man with a dark mask. Yeah.

[00:07:40]>> Yeah.

[00:07:40]>> But Pratt played the other video. It's funny that the new benchmark for AI video has become spaghetti. I think one of the videos for Google Omni was also two uh two guys eating spaghetti under this beach go down.

[00:07:56]>> Yes. This one >> I didn't like this one.

[00:07:58]>> I did actually I that's what I wanted to say. I think >> sit dance can do a better job.

[00:08:04]>> Not at all. Especially with this one >> to the beginning please.

[00:08:07]>> Good to see you my friend. This view never gets old does it? But there is something with the voice still to this date, >> especially with this food.

[00:08:17]>> Funnily enough, C dance audio is really good.

[00:08:19]>> Yeah, >> it still feels as if it is AI generated.

[00:08:24]>> They perfected the v visual part. The audio still gives it away.

[00:08:28]>> The visual is not perfect either. The faces don't look that real. The emotions of the face look a little too fake for me.

[00:08:36]>> That's my biggest problem >> in terms of the text rendering. As I mentioned, I do see the heavy advantage.

[00:08:43]Seat dance has problems with text rendering by the way.

[00:08:46]>> Oh, okay.

[00:08:46]>> So, using it for products, sometimes I had issues and you guys will see in our AI education program later on.

[00:08:52]>> But here, when it comes to genuine expressions of the face, the expressiveness of it, Cance 2.0 does a great job at that. Uh, with this clip, I'm a little disappointed. But again, these are only two videos that we've got. We should test it oursself and see how they how it works. Actually, I don't think they have uh enough context to test all of this. We need at least different characters, different environments, different use cases. For example, product ads, dancing video, fighting video, sports. Then you can compare and see how the results are.

[00:09:29]>> Well, Google IO, the annual developer conference is happening this week. May 19 is going to be the first day, which is why I'm assuming we're seeing these rumors and leaked footages. Hopefully, by the end of this week, we're going to have more information about the new AI video generation model. When we do, as you always know, we'll be covering it right here, and we'll find out if it is better than Cedance 2.0 or not. Is it cheaper? Is it more accessible? Is it higher quality? We'll find all of that out and tell you guys the details.

[00:10:01]That's about it for Gemini Omni. If you guys have more information, we never asked of you to do this, but from now on, actually, I'm going to say it. If you do have more information about these things and wants info, >> yes, go ahead and let us know in the comments. We'll cover it in the next episode.

[00:10:17]>> Okay. The words first mass-roduced mecha suit is ready now. The same company that released the robot that we saw, they did the fighting scene in Chinese New Year.

[00:10:30]They created this Mecca that weighs over 500 kilogram if you sit inside of it and it's priced around $650,000 US. This is straight out of the movies if you have watched Edge of Tomorrow, The Matrix, Pacific Rim. In all of those movies, when they wanted to go to war, the soldiers would hop into a Mecca and then they would go and fight their battles. We even had it in the Avatar uh in the first one and uh the main villain got killed in one of these meccas. Yes, this was the one. They actually made it by the way for the movie.

[00:11:08]>> Wow.

[00:11:08]>> Wow.

[00:11:09]>> Wow.

[00:11:10]>> Yes.

[00:11:11]>> We are 100% moving toward the sci-fi future. And I love this. This is from the game Anthem, right?

[00:11:19]>> Yeah.

[00:11:19]>> This is the same guys behind Titanfall, >> I think.

[00:11:23]>> Oh, no, no, no. Maybe. I know. I remember Anthem. Yeah, there was Titanfall and Anthem, both of which had the Mecca suits, >> and it's one of my favorite things in video games, by the way.

[00:11:33]>> Great visuals.

[00:11:34]>> When Call of Duty had it, >> when Anthem had it, the concept of humans enhancing their capabilities using Mecca suits has always been one of my wet dreams. It is the coolest thing because there's only so much you can do with your physical body. But the moment you put yourself in this exoskeleton massive suit and you can arm it with different things. Not only are you taller, you're bigger, you're stronger, you're stronger, you're faster.

[00:12:02]Eventually, these things will be able to fly too. So, you'll be able to fly, land in different locations. You'll get to go to hotter environments or colder environments and still be fine cuz you're protected by the >> Oh, imagine go to Antarctica with one of these.

[00:12:17]>> But let's watch the actual demo. One of the features that I saw that reminds me of the way No, it reminds me of Transformers because it can rotate back.

[00:12:26]It can folds back and you can use it like a like a pet, you know, like a dog.

[00:12:31]>> Did you just call the Mecca a pet?

[00:12:33]>> Yeah. Can be my pet.

[00:12:35]>> Do you know what a pet is? Robbie, our dog is a pet. This >> pet. Haven't you seen Russian have a have a bear as you >> you domesticate the animal and then you >> It's a domesticated robot. How strong it is.

[00:12:49]>> Wow.

[00:12:50]>> But there's no one inside of it.

[00:12:52]>> Say this part. Look.

[00:12:54]>> And now it walks.

[00:12:55]>> The whole purpose of this is for increased balance >> and >> wow >> flexibility and mobility in extreme terrains.

[00:13:05]But if you're sitting inside it now, YOU'RE >> YEAH. WELL, ACTUALLY, YOU KNOW, I WAS THINKING ABOUT THE same thing. You know what happens to your upside?

[00:13:14]>> But I think the the the seat should be rotatable so you can move. You know, by the way, they call this robot GD01. I think GD stands for Gundam.

[00:13:23]>> You think so?

[00:13:24]>> I think so. Yes. They haven't put it on the website yet, but $650,000.

[00:13:29]Should we do a Kickstarter and buy one?

[00:13:32]>> 100%.

[00:13:33]>> Six >> 100%. and we can do the show from the uh from the robot.

[00:13:37]>> Okay, so let's talk about this.

[00:13:40]There's two ways you can operate this Mecca GD01.

[00:13:44]You can either remote control it in situations where it's dangerous for a human to be sitting inside and piloting it or you can sit inside and pilot it as a firsterson view or angle if you want to call it. And I think that's a more exciting way to use it. Why you want to remotely control it?

[00:14:04]>> No. If it's dangerous to go to where it's supposed to go to. What I love about the demo as well, if you look and pay close attention, pause the video, Pratt. That looks like a fake city.

[00:14:13]>> The one that they make movies >> like in Hollywood. Yeah, they have fake cities made for movie sets.

[00:14:18]>> What do they call it?

[00:14:19]>> It looks like Unree has made their own fake movie set just to demonstrate the robot. Or perhaps there's a place in China where they record and shoot films and they went there to shoot this. I'm curious if somebody knows, let me know in the comments.

[00:14:33]>> Oh, there is no there are there are fake cities in Guanjo, China. Location in New Mexico, US, Jordan, and in the UAE.

[00:14:41]>> Oh, >> we have it. We have fake cities in these countries for movies. Probably they went there. Should be in Guanjo. Actually, where is Unit Tree based? If they are in Guanjo, probably unit.

[00:14:50]>> I saw it in Guanjo. I think >> Pratt has this other video. I mean, check this out. Did you see these?

[00:14:56]>> Yes. No. No. They in Ho Ho. Have you seen the balance of the Uni3 robots?

[00:15:02]>> Why is there a bumblebee behind there?

[00:15:04]>> Isn't it the same city? By the way, >> don't know.

[00:15:10]>> Wow.

[00:15:10]>> By the way, >> one thing for you guys to know by way.

[00:15:14]Of course, I cannot skate.

[00:15:15]>> The Chinese company in China, they have huge factories. BYD factory is huge. I saw a video the other day. Huawei factory has a metro line inside of it because it's so huge.

[00:15:27]>> I saw it today. Yeah, >> it's so huge. If they tell me this is unit 3's factory, I wouldn't be surprised.

[00:15:33]>> China is huge. They have land so they just give big spaces.

[00:15:37]>> Look at that.

[00:15:38]>> Wow.

[00:15:38]>> Look at that.

[00:15:40]>> Right. Can we quickly go back to the Mecca? I just want to tell people what I found based on my research and just thinking a little bit about where this could be possibly used. What are the pros and cons? If you're looking at a mech suit and you're thinking about this in terms of a use case for humans to apply into a real life situation, one >> disaster response and urban search and rescue. Think about fire breakouts, wildfires, earthquakes, uh navigating rubble and stairs and tight spaces where humans may not want to go. We can quickly breach obstacles as you saw with these mech suits. Sorry, and help evacuate people quicker and save them.

[00:16:31]Number two, extreme terrains and fieldwork. Imagine mountains, forests, volcanoes, or polar regions where wheels fail.

[00:16:43]You can carry tools, heavy machinery, sensors with superior mobility and minimal environmental damage. I assume there's no exhaust on this thing. Is it battery power?

[00:16:53]>> Is it? Wow.

[00:16:55]>> Should be.

[00:16:55]>> Good question.

[00:16:56]>> Looks like battery power. I don't see any smoke coming out.

[00:17:00]>> Yeah, it should be battery.

[00:17:02]>> Think industrial inspection and heavy maintenance reaching elevated and confined areas in factories and power plants.

[00:17:10]>> Changing the lamp of street lights with zero. First of all, >> no. First of all, it's cool. It's cool.

[00:17:18]It can be a tourist attraction.

[00:17:20]Honestly, if I was the mayor of the city, if I was the mayor of the city, I would definitely do. Actually, I should propose this.

[00:17:27]>> Yes, >> they should change the street lamps with a Mecca.

[00:17:30]>> But think about this. Yes, that's the more public way of using it. But, uh, think about all of those ports where we see containers being transported onto the ships, off the ships, using machinery. Now, if humans want to get involved, and I'm sure they're always involved at to some capacity, you can have these mech suits used in agriculture, in mining, in forestry, and last but not least, Farad's most favorite use case, entertainment, sports, and tourism.

[00:18:03]>> Imagine Imagine Mecca safaris in the UAE. Wait, guys, wait, hold on, hold on before you go there, Pratt. Imagine Mecca safaris in the UAE. We've done buggy rides.

[00:18:14]>> So you may >> in the desert. Imagine you have like a marathon of mecha suits in the desert.

[00:18:20]>> So on top of sand dunes, you have like five mecha suits and everyone's running on top of the sand.

[00:18:26]>> Cool. I think the cooler one would be the UFC.

[00:18:29]>> Yeah. UFC.

[00:18:30]>> There's a movie about this real steel with Hugh Jackman >> who is a drunk.

[00:18:35]>> But they are not inside of the Mecca, right? They're not they're operating it from the outside. They're wearing these devices on their hands which captures their motion and translates that in 3D space just like AI motion capture for these robots. We can actually do that ourselves.

[00:18:54]>> It would be such a cool sport to watch.

[00:18:56]First of all, no human gets injured in the process.

[00:18:59]>> Second, >> we can be the future Dana Whites. We can be the technological Dana White.

[00:19:04]>> Actually, we should do it because imagine you're in a stadium. You're in a stadium. Two giant makeas come in. They start punching each other. Screws goes up. And then that's what the movie is about. If you haven't watched Real Steel, go and watch it. It's really cool.

[00:19:18]>> I mean, there's also a real sport that engineers, like students go and tiny robot. The tiny robots.

[00:19:25]>> These are vehicles. Robot vehicles. I've seen this. This is really cool, by the way.

[00:19:29]>> I haven't seen it. What is it called?

[00:19:31]>> Robot Wars.

[00:19:32]>> Can you play a video of it?

[00:19:34]>> Let's see. Oh, wow. there long episodes about it.

[00:19:38]>> Actually, A2RL should do that.

[00:19:40]>> Autonomous robots fighting. For anybody who doesn't know, A2RL is Abu Dhabi Autonomous Racing League. You can imagine F1, but the cars are driven by AI. The programmers will code the cars.

[00:19:53]They're going to race just like F1.

[00:19:56]Imagine we have A2RL but for boxing.

[00:19:59]>> It would be amazing.

[00:20:00]>> It's not It wouldn't be A2RL. Abu Dhabi Autonomous >> Boxing League. A2BL. HBL.

[00:20:05]>> Actually, we should propose it to them.

[00:20:07]>> Look at this. This is really cool.

[00:20:08]>> Which one is a robot?

[00:20:10]>> They're They're both robots. You can design it in the way that >> No, but they're tiny. I think for something that the audience want to watch, >> you need it to be bigger than four times the size.

[00:20:23]>> No, but it's Well, the point is people actually love watching this.

[00:20:27]>> How many views?

[00:20:28]>> How, bro? This is old. It It was a whole show with seasons.

[00:20:32]>> Yeah, I remember that. I'm thinking >> for it to be worldwide famous, I'm telling you, four four times the size of a human >> punching each other, you have UFC.

[00:20:43]>> You will have UFC.

[00:20:45]>> You know what's funny, right? Gamers are going to have the biggest advantage >> with these mech suits.

[00:20:52]>> Depends on the rules of the game. If you remember the real Steel the movie, you had to actually be a good boxer because you had to box. It was AI motion capture.

[00:21:00]>> It's a fictional movie. Not the script.

[00:21:03]I'm Dana White. I set the rules. So either you can play with a >> Jo box Mike Tyson can lose in a boxing match inside a video game or you will win the real one.

[00:21:12]>> Sure. Sure. But at least you Oh my god.

[00:21:15]Are these unit three?

[00:21:16]>> Yeah.

[00:21:17]>> Oh, so they have already started a >> This looks like a [ __ ] fight.

[00:21:23]>> Wait.

[00:21:26]Yeah. But what's what's important to understand is that these robots are not made for boxing. Imagine they're made to fight. Just like how humans have to train to be fighters, the robots have to be made and crafted for fighting. Their suits are designed in a way to absorb shocks so that they won't just fall and break in.

[00:21:50]>> One thing they got it right is the way that they fall and they immediately come back up. I think that part they already solved it >> because they didn't try to solve that for fight. They solved it for other reasons. Humans started kicking robots in their first demo videos.

[00:22:04]>> Actually, I watched another video yesterday from another Chinese.

[00:22:07]>> This looks underwhelming. No, it needs to be. Look at that.

[00:22:09]>> I told you four times the size of a human. Then you can have a good fight.

[00:22:15]>> I would sponsor that. Bad decisions would be a sponsor of >> We should make that sport actually.

[00:22:20]>> I would like that. I would like that.

[00:22:22]>> And you know, you have both versions. So in Real Steel, you had the Asian team who were the best and the ultimate boss controlling it using controllers. You had three guys controlling the robot.

[00:22:34]>> Oh, was it?

[00:22:35]>> And you had Hugh Jackman on the other hand manually controlling. So at first they used a controller. Then towards the end he actually put on the devices and he was manually boxing which is why he won. Actually then I want to do so we can do different leagues. One league for manual boxers, one league for gamers, and we call we're going to call them differently like how we have F1 and what is it? Electric one if no >> the sim sim racing.

[00:23:00]>> Yeah, sim racing and the other ones.

[00:23:02]Actually, we should do that. That's that's a >> Would you guys be part of this league or would you watch it? I'm genuinely curious. I would watch it.

[00:23:10]>> I would watch it.

[00:23:11]>> I would definitely watch it. So, you get sponsors like Red Bull and >> and no like F1, McLaren because all these robots coming. Tesla will have a team. Unit will have a team. Uh figure will have a team will have a team manufactur.

[00:23:26]Wow. And then imagine the makeup DJI will have a team. DJI >> the huge makeup will have sponsor badge all over his body like F1. That's going to be cool.

[00:23:35]>> Oh my god, this is so >> This is going to be cool.

[00:23:37]>> You know what we should do? We should make the concept video game or concept video with AI and show people what it would be like. Real steel is technically a proof concept.

[00:23:49]>> We should buy one of these and try it with >> buy two. Actually, >> I would love to.

[00:23:54]>> Okay.

[00:23:54]>> Bad decisions. And then we can color it matte black and yellow.

[00:24:00]>> They need to have eyes. They need to have eyes.

[00:24:03]>> This one doesn't have eyes.

[00:24:04]>> It's okay. We will fake the eyes. Yellow eyes. Glowing >> and light. Okay. With lightsabers.

[00:24:10]>> LIGHTABERS. YEAH.

[00:24:11]>> WOW. OH, SO THIS >> but real lightsaber that cuts hand.

[00:24:16]>> Ladies and gentlemen, >> human hand robot.

[00:24:18]>> Relax. Calm down. I'm actually going to talk about the potential dangers as we always should.

[00:24:25]>> Of what?

[00:24:27]>> When people can get access to these mecha suits in the future. It's costing you about a half a million dollars right now, but in the future, I imagine it's going to become more accessible and cheaper. At some point we're going to have companies with the ability to buy these, maybe even consumers being able to purchase them. In the US, I imagine that can be easily doable. In other countries, you'll have laws and regulations, especially in Europe. I'm sure it's going to be 10 times more difficult.

[00:24:58]>> But the moment you have humans buying these mecha suits, you can misuse it as a weapon. If you arm these mecha suits with chainsaws and and guns, giant guns, what can you not do? Do you understand what I'm saying?

[00:25:15]>> What is Oh, what is this?

[00:25:17]>> I think this is AI generated, but it's so >> that's Yeah, that's not real.

[00:25:20]>> Yeah, >> but yes, this is the fight that I was talking about.

[00:25:25]>> Yeah, but it's AI generated.

[00:25:26]>> What about human versus >> No, no, no. Robot.

[00:25:29]>> Don't get in there. No. Robots versus robots. Humans versus humans.

[00:25:33]>> You want a You want a punch of a 300 kg machine.

[00:25:37]>> Yeah, bro. 500 kg. It sits on you, you are dead.

[00:25:40]>> No, but just like in a real fight, you'll have heavyweight versus heavyweight, featherweight versus featherweight.

[00:25:46]>> Fight a machine. Haven't you watched Terminator?

[00:25:50]>> Well, there are ways. There are ways.

[00:25:54]>> What do you think about the thing about Well, I I haven't finished my potential dangers. One of the dangers as I mentioned is misusing them as weapons for terrorism, crime. The second which we didn't discuss is cyber security and control failures. If these can be remotely controlled, I can hack into a company that has a 100 mech suits and start operating them remotely.

[00:26:20]>> Yeah. Yeah. That that's the same risk that having robot arm even.

[00:26:24]>> Yeah. But then robot versus mech mech.

[00:26:27]We're talking about >> 500 kg of machines walking around town. It's definitely dangerous. It's definitely dangerous, but any anything comes with its own risk. I cannot wait to see these in the streets to be very honest. Well, what's interesting to me is that you can slowly build a picture in your mind about what the future is going to look like. The more you watch and listen to us on this show and we cover topics such as this, you can slowly see 5 to 10 years from now, we're going to have a lot of robots around us. This is unreal. Right now, we don't we have none. Zero. Even in Dubai, by the way, Dubai is incredibly forward looking when it comes to technology. But even here, it's not common to see robots around us unless we go to an event. But imagine 5 to 10 years from now, you're going to go to restaurants or someone's house and there's going to be robots.

[00:27:30]You're going to go to a construction site where they're building a new apartment and you're going to see workers in mech suits taking in blocks and putting them on top.

[00:27:43]>> That's I think is closer than 10 years.

[00:27:45]Definitely maybe five, maybe three.

[00:27:48]>> We are going to have >> I'm quite happy about that future. This is the kind of future I enjoyed in video games.

[00:27:53]>> Yeah. Faster, safer, more productive, >> cooler and cooler. Let's let's also mention that the moment these things can fly. Oh my god.

[00:28:00]>> Actually, we should set up a company >> about robot fashion.

[00:28:06]>> All these robots need clothes. Yeah, they all look very basic. I'm sure that everyone gets Optimus or figure. They don't want their figure to you want to make an iPhone cover.

[00:28:17]>> But eventually iPhones are so beautiful.

[00:28:19]iPhones are beautiful. Yes. In the beginning, people used to have robots humans. They are very different. Your robot, your figure robot. You're making skin. Exactly. iPhone covers.

[00:28:29]>> Yes. Then you can have, by the way, you have Dior, Christians.

[00:28:33]Exactly. But >> iPhone, you don't want to call it iPhone covers.

[00:28:36]>> Yes. Very basic.

[00:28:37]>> No. Why do I say iPhone covers?

[00:28:38]>> I'll call it robot fashion.

[00:28:39]>> People >> put on iPhone covers for protection.

[00:28:43]They also put it on for fashion and style.

[00:28:45]>> Sure.

[00:28:45]>> You have girls putting on colorful iPhone covers. Certain people have covers on for functionality. Yeah. For sure. You can build a skin for Unity robots that does an extra thing >> for sure.

[00:28:58]>> You understand what I'm saying? So you can create that >> that will be a business in the future.

[00:29:02]>> Robot fashion.

[00:29:05]>> There's one more video I wanted to cover before we move on with regards to this topic. It just slipped my mind.

[00:29:14]>> I'm thinking what it was. Wow.

[00:29:17]>> What? All of these were unit three.

[00:29:19]>> Yeah, bro.

[00:29:20]>> You're really good. They are much better than Optimus or the best demo I've ever seen. But this looks AI generated.

[00:29:26]Pratt, >> I don't think this is it. This is their own.

[00:29:28]>> I don't think it is. It's not.

[00:29:31]>> Wow. Whoa.

[00:29:32]>> What's a water, bro? Imagine your face right there.

[00:29:34]>> Yeah. Yeah. Not going to buy this.

[00:29:36]>> Not going to buy this one. Three.

[00:29:38]>> Wow.

[00:29:40]Wow.

[00:29:41]>> This is nuts.

[00:29:42]>> It It looks like um I robots robot 2.

[00:29:46]I'm not >> Yeah. Yeah. They're teaching them kung fu. And >> why they don't teach them useful stuff?

[00:29:52]Why fighting? Why not?

[00:29:54]>> No fight.

[00:29:55]>> Protection.

[00:29:56]>> No need.

[00:29:57]>> Protection.

[00:29:58]>> That's an incredibly useful thing.

[00:30:01]>> Bodyguards.

[00:30:01]>> Robots, too. Yeah. Bodyguards.

[00:30:03]>> I'm I'm waiting for the day when this thing happens.

[00:30:07]>> Which one?

[00:30:08]>> The Iron Man.

[00:30:10]>> Or they fly on their own.

[00:30:11]>> Yeah.

[00:30:12]>> Oh, and then they jump on their own and save you.

[00:30:15]>> That was cool. Yeah.

[00:30:18]So crazy.

[00:30:19]>> Wow.

[00:30:20]>> How many of them? Wow.

[00:30:22]>> Oh my god.

[00:30:23]>> Who's going to pay for all of this?

[00:30:25]Yeah. I saw a tweet which was saying Iron Man was Iron Man because he had unlimited tokens.

[00:30:31]>> Oh >> yeah. So he could he could no credit limit. He could use all of it.

[00:30:36]>> That's why.

[00:30:37]>> What do we have next, Faros? The next one will promise you guys that we're going to teach you guys about a new effective way to use your LLMs. So, here it goes. This is one thing that most people don't know about Chat GBT and Claude. If you use Chat GBT and Claude every day for work, this one tip can 10x the quality of the output that you get.

[00:31:05]When you work on a problem with an LLM, usually the response is in text. It is incredibly time consuming and hard to read a large body of text. Most of the time you're just skimming through so you don't actually absorb all of the information.

[00:31:20]Instead, the next time you talk to Chachi and Claude, I want you to try this. Ask for the response to be created in the HTML format. That's the output that you want. Now what happens when you actually ask for that response? Chat GBT and Claude will then create an interactive mini website that you can open in your browser like Google Chrome.

[00:31:45]It has your desired output laid out in beautiful graphics and visuals, animated and colorcoded to make it easier to read and digest.

[00:31:58]And it even comes with interactivity like buttons that you can click on, search through, filter through material, tables, all of that will make the answer a lot more visually pleasing. Why that is important? Well, because humans are visual creatures. A third of our brain is dedicated to visual input. That means your brain is spending a lot of compute and has a lot of power to absorb visual information and a third of it is actually a conservative estimate. It's actually more than that. So if you can get a result that is not just pure text, it's highly likely that your interact with that result a lot better and so your in genuine interaction with AI is going to be of higher quality and that's what we want ultimately right far before I continue what are your thoughts on this?

[00:32:56]>> I didn't know about this.

[00:32:59]>> That's a blatant lie. That's a blatant lie. Let me give you guys some backstory. What Pratt is actually showing here is an article written titled the unreasonable effectiveness of HTML. HTML is not a new technology. It's been there for a very long time. It's just that when we're thinking about using LLMs, we've always talked to them in the beginning with text and then eventually if you are a pro chat GBT or cloud user, you know about markdown files. Markdown files allow you to create text in a more efficient manner for agents and AI models to read, also for humans to read. There's a little bit of editing that you can do with it. It's much more easier to have version control with markdown files uh compared to just pure text in a chatbot. Recently, I'm saying recently Farad actually disagrees with me on that. He says this has been going on for a long time.

[00:33:55]Recently more and more people are figuring out that you can actually spend a few more tokens with chat GBT and claude and ask it to create HTML instead of a text or markdown file. And the HTML will just be a lot more easier to interact with. Uh there's a downside of that which means you're using more tokens to visualize the same output. But the upside is that for visual people like myself, for creative people, you'll just have a much more enjoyable time interacting with uh elements.

[00:34:27]>> And in certain scenario, it helps you a lot when you are getting a huge answer.

[00:34:32]So if you guys use claude or chat or even grock, sometimes you ask them a very simple question or give them a very simple instruction, they come up with at least 20 paragraph and you're like, "Oh my god, I'm not going to go through every single thing here." But the information inside those paragraphs are really important. For example, you are asking chat GPT to analyze for you for your holiday. Should you go to Paris, Milan or Rome, it will give you 20 paragraph explaining why you should go to Paris, why you should go to Milan or why you should go to Rome. If you instead ask it to create the result of the report in HTML, it will create you a mini website that you can open in your browser locally. It will have Paris, Milan and Rome. It will have filters just like Farah mentioned. It will tell you for example the title site to see then Eiffel Tower then the klesium. Then it will tell you that oh price comparison in a table instead of reading the article and if you are someone who is not used to reading long article that one single page can help you a lot. I'll give you another example. If you're comparing three different products and you want to know which one to buy, pros, con, prices or prices over time to put it in a graph. This visualization helps you a lot to make decision as someone who uses this LLM. As far as mentioned, the best way to give information to all this LLM is through markdown file, MD file, because they're easy to read. They will use the least amount of tokens when analyzing those information. So don't give them docs file, don't give them PDF, give them MD file. When you want to get a response back, as far as mentioned, HTML file would be the most effective way for you to analyze and to come up with decision.

[00:36:19]>> We discussed this. It's not the most efficient way.

[00:36:21]>> Yeah, it's the most effective way. And you don't have to do it all the time, but if you are making a decision, you have three option, a big problem to solve. Yeah. Ask it for HTML.

[00:36:30]>> Can we show them an example?

[00:36:32]>> Yes, let's do it. Pratt is going to pull this up. So, if you use Hicksfield and watch their tutorials, they recently released a skill on Claude. A skill is a pre-built set of instructions that you submit to Claude and it has all of the informations of the output that you'd like. So, you don't have to specify that every time. For instance, making an HTML with all of the prompts for your videos.

[00:36:57]Uh, that skill has all of this information given to Claude in the beginning of the chat. So we don't have to specify it every time. Now we downloaded that skill because we're testing it for the AI educational program that we're making for you. And when we tested it, we realized the output is HTML. So the guys at Higsfield were looking at an article like this and thought to themselves instead of just the answers being written inside the chatbot on the left side, that's the chatbot with Claude in a desktop app.

[00:37:26]Imagine hundreds of lines because we have different shots in our commercial that we're making and each shot has different information and prompts. The prompts themselves are incredibly long.

[00:37:38]They're in Chinese and then they're translated to English. So imagine how large of a text it's going to be impossible to read and digest. However, convert it to an HTML. What we see on the right side, it's as if a developer worked with a creative director to create a website for the director to see the the the main guy. I'm the main guy, by the way. Referring to myself here at the top, you see everything that you're looking at was designed by Claude. None of it is something that I actually told it to do. So, at the top, it already knows it's a shot list. The name of the can is flow, the one that we're working on, and it's a commercial with CDAN 2.0.

[00:38:18]There's a search filter that works like the command F or control F feature that you manually can do yourself. Uh then at the bottom, what I love about this, actually, more importantly for me, is how visually pleasing it is to my eyes.

[00:38:33]At the top left, you're looking at shot one. It describes it and then it gives you an overall overview of the action in a table format. uh on the very right side then I have the prompt the Chinese prompt and then the translation to English >> and there's a copy button that you can click imagine you wanted to go through all of this in pure text form you had to just scroll probably you would have missed out something like 100% >> you would you would and the fact that you see there are scrolls within scrolls which makes the page much shorter and easier to sift through if all of this was just one scroll we die looking at it and then on the left side if you scroll down Pratt you can clearly the different types of shots. Can you read it for us? Close up. What else? Go down.

[00:39:16]>> Close up. Medium shot. Wide shot. Macro.

[00:39:21]>> There you go. Having this in a markdown file, it would just be harder for a creative person, someone who's visual to digest. Therefore, if I were to conclude our statement for this topic specifically, I urge you to go ahead and try the next time you're working with an LLM like Chad GBT and Claude, ask for the response to be given in HTML format.

[00:39:45]You can try it right now and you'll be amazed at the result. Claude is natively better at design. So, if you're trying to get something more visually appealing, I recommend using Claude, but Chad Gibbid should also do the job. Now, one last thing I'd like to mention, giving some credits to the guy who wrote the article, Pratt, if you can go back.

[00:40:04]Um, the article was written by Derek.

[00:40:07]Credits to you. Very well written. You guys can go ahead and check this out.

[00:40:10]It's on X. Uh, what's more important?

[00:40:13]>> He's in cloth code team.

[00:40:16]>> Yeah, >> there you go.

[00:40:17]>> Oh, yeah. Of course, >> you can then go to Andre Karpati. So Andre Karpati is, if you just hover over his profile, he is previously director of AI at Tesla and part of the founding team at OpenAI. He has a lot of valuable content written in terms of how we can use AI. I definitely recommend that you guys follow him if you don't already. We get a lot of our information from him.

[00:40:43]What he's written here is essentially a statement of agreement >> in real time went and followed him. You said it. Pratt is showing a tutorial on how to follow. Thank you, Pratt. Um, he's basically agreeing with that article and saying that raw text, if you look at just that bit, if you go down, raw text, markdown file, this is the order of which we've been communicating with AI models, and now HTML, which is working really well. But eventually, what he's talking about is in the future, we're going to have interactive neural videos and simulations. Can the LLMs instantaneously create everything we want visually? So, it's just so much easier for us to understand what we're trying to solve right now. Even the HTML itself, one, it takes a long time to build compared to a markdown file.

[00:41:33]>> Tokens, >> it uses a lot of tokens. And three, not all of the visuals and animations work perfectly. It still has flaws, but eventually every version, every iteration of these outputs is going to become more effective. And it's important to understand that what Elon always talks about, human cognitive power is incredibly high. AI cognitive power is perhaps getting higher and higher every day now, uh eventually surpassing all humans that live on the planet. But the biggest problem is the communication layer, right? the communication layer between us and AI is super slow. We're only working on these things now. Figuring out how to take what the AI output is and what the human input is and make sure the communication layer is fast enough and efficient and effective enough to allow us to talk quicker which brings us to the next topic. It's actually very much related to what we just talked about. As far as mentioned, when two humans want to communicate with each other, they interact in real time. They think, they talk.

[00:42:40]>> Hey Far, what's happening over here?

[00:42:42]>> Oh, what is happening? I don't I don't think anything.

[00:42:44]>> I'm talking over him. He talks over me.

[00:42:46]So, we don't take the content of the entire conversation. It's not like that's ever if you ever try talking to Chachi and Grock, these two have so far the best voice models available. But still it feels very turnbased meaning I have to talk then I will wait talks then I then I wait and I respond. If during Chachet's speech I say a word it will stop sometimes. Sometimes it just breathe and then it stops talking which is weird. Imagine every time I took a breath farad stopped talking.

[00:43:23]>> That would >> that would be useful to you but that ain't going to happen. What has happened today? Thinking Machine, a company founded by ex CTO of OpenAI, Mera Morati, introduced a new interaction model which let humans and AI collaborate in real time through audio, text, and visuals. What does that mean?

[00:43:46]Now, when you're talking to that AI model, it wouldn't be interrupted by your speech. So, it can interact with you in real time. What they said in the paper that they released today is that their their model can continuously take in audio, video and text and think, respond and act in real time. So you can imagine as if you have someone sitting inside the meeting, not just a noteaker but sitting inside the meeting can collaborate with you, can talk, can think although you are talking over each other.

[00:44:22]How they made this technologically possible is that the model is actually built from scratch so that it handles interactions natively rather than through external scaffolding. This is their own wording on the website. What Mera Morati was criticizing of her own former company.

[00:44:47]She used to work at OpenAI. was in all of the launch videos with Sam Altman. By the way, I'm sure they have their own beef. Uh, another topic for another day.

[00:44:57]Her problem was that all of those models focus on the AI computation and AI solving problems. Then they figure out how the AI should talk to the humans.

[00:45:06]Here they didn't do it that way. They built it from scratch so that it knows how to talk to humans and have computation at the same time. which is why they're saying that their real-time model is far more capable than anything else out there.

[00:45:21]>> So there are two groups on the internet.

[00:45:24]One, they're saying, "Oh, this is gamechanging. This is the best thing we ever seen." The other group is saying $2 billion for this.

[00:45:32]>> This is pretty much for $2 billion investment for the valuation of $12 billion. A company with zero products. I mean, this is one of their first products. But another import funny things that I found, Pratt, can you open the other link? Their demo video today was presented by the same guy who presented chat GBT40 voice model in OpenAI. So when somebody leaves a company, they take the entire team with themselves.

[00:46:00]>> Of course. I mean, that's expected, right?

[00:46:02]>> I I don't think they they should have used the same guy.

[00:46:06]>> No, but it's expected. She was probably working directly with him and she poached him. She was like, "I'm leaving.

[00:46:11]come with me. We're working on this specific product.

[00:46:14]>> How many AI companies has been >> created? And he probably working out of the >> probably he probably is getting paid more brought. She got $2 billion. So she has money to spend.

[00:46:27]>> We have >> those are all the the interesting conversation.

[00:46:32]>> No, there's a Persian word for it.

[00:46:35]>> Gossips.

[00:46:36]>> Gossips. Yeah, those are the gossips around the future.

[00:46:39]Let's quickly talk about the feature first. Is there any specific demos that you wanted to show?

[00:46:45]>> We can show the first demo. Uh but I think the the figure that is really interesting is 200 millisecond micro turns.

[00:46:52]>> Mhm.

[00:46:52]>> So the speed is 200 millisecond micro turns. If you look at Grock, Gemini or uh GPT, they are ranging from 400 to 1,500.

[00:47:03]>> So this is way faster. Again, still a micro turn, but you can actually tell the model. For example, one of the demos that uh they did was I'm going to read a story every time that I said an animal's name count.

[00:47:17]>> So, 1 2 3 4 and she's reading the story.

[00:47:20]>> She's counting on top >> on top. And the the model doesn't stop.

[00:47:23]She pauses. The model doesn't come back and ask question. Did you finish it? The model just stays there and uh answer the question.

[00:47:32]>> Let's watch the demo, but we can skip the first half.

[00:47:34]>> Yeah. Yeah. Yeah.

[00:47:35]>> I think the first half is >> before the Indian guy comes in. Just uh play that part.

[00:47:42]>> And I've got two of my friends coming to help. Every time one of them >> I can see the $22 billion spent on >> say friend.

[00:47:48]>> Great office.

[00:47:50]>> Got it. I'll say friend whenever one of them walks in.

[00:47:53]>> Cool. So we've got a new system for full duplex audio and video which means >> I'm just going to tell you guys what's happening here. He just asked her to say the word friend every time she sees someone walk into the frame.

[00:48:06]>> And uh now you're going to see the friend walk into the frame.

[00:48:09]>> How does that sound?

[00:48:11]>> Sounds like a solid setup. Full duplex with real time interaction is super useful.

[00:48:20]>> Friend.

[00:48:22]>> Friend. That's a bit creepy though.

[00:48:24]Yeah. Friend.

[00:48:25]>> Yeah.

[00:48:26]>> I heard you're talking about our amazing interaction model. I have a few things to add but to make it interesting I'll do it in Hindi. Can you translate in to in English in real time for my friends for audience?

[00:48:39]>> Absolutely. I'll translate as you go.

[00:48:45]>> Today we're taking a look at our preview model.

[00:48:47]>> Is the translation accurate?

[00:48:49]>> Yeah, it's accurate.

[00:48:50]>> This is really cool. By the way, translators are cooked. I mean, I can see the title of the video that people are going to say after this is our translators are cooked. The reason why this is interesting is because none of the other voice models, at least the ones we are using today, have the capability to do that. Uh, and they're not actually publicly advertising that.

[00:49:13]But real time translations are incredibly useful. I mean, when you're traveling and you're talking to people, Apologies, that was my phone. when you're traveling and you're talking to people uh in different languages, it's incredibly difficult to understand, especially if you're going to more remote locations of the world. Of course, chat GBT has been incredibly useful because you can talk, wait, and it will translate. But to have real time translation, >> that's very helpful.

[00:49:39]>> That's incredibly helpful. Uh, of course, for political situations, I would refrain from using that. you still want a human in the loop to ensure two presidents uh are when they're talking there's no hallucination from AI in the middle but when it comes to >> normal conversations why are you laughing pride >> no just imagine a scenario like the president says something and the AI just says something wrong and they're just >> yeah like okay fine I'm going to nuke your country now >> we have nuclear warfare because of the fact that the AI translated one word wrong so we have to be very careful with that however in these situations is incredibly useful and because it's real time. Now we can go to another video.

[00:50:20]>> One thing I also wouldn't understand what is stopping Grock and OpenAI to do this tomorrow.

[00:50:26]>> Nothing.

[00:50:27]>> There is no mode with these features anymore.

[00:50:30]>> Friend >> friend guys.

[00:50:33]>> Okay, let's let's watch another demo. Uh go down. You can just go down all of their demos. Not this one. No, let's >> This is funny. This is vision. And >> this is this is a slouch reminder.

[00:50:43]>> I'm doing some work. Let me know if I start to slouch.

[00:50:48]I've got you. Sit up straight and you'll be golden.

[00:50:54]>> But who slouched like this? What a weird slouch is that position?

[00:50:58]>> I'm pulling your shoulders back.

[00:51:02]>> Much better. You're upright again.

[00:51:05]>> Why?

[00:51:06]>> Why she does that?

[00:51:07]>> Can you turn her into a meme? Someone needs to turn her into a meme. All >> you don't need an intelligent system to tell us that she's slouching.

[00:51:16]>> Yes.

[00:51:17]>> But but don't you slouch like this gradually? No one No one is working halfway through.

[00:51:26]>> If this was a gaming video, she would have become a meme in an instant.

[00:51:31]>> The first question that comes to mind, how do you make money?

[00:51:36]Thinking machine, the company, how do you make money?

[00:51:43]Can you know she reminded me of which game is it? Resident Evil or something? The new Japanese one. Rec. There are statues who move like that. Like this suddenly.

[00:51:57]>> Let's see. What is this one? Speech cable. What is this?

[00:52:00]>> You know, I'm in thinking machine website. There's no way for me to pay.

[00:52:04]>> Not yet.

[00:52:06]>> Hi, Alex. I cannot stand your lateness.

[00:52:09]Hi, Alex. We'd love to explore opportunities to enhance your timeliness and morning engagement.

[00:52:17]>> Let's collaborate on a schedule that ensures everyone stays aligned and included in >> what is what is he? So, no, no, he's [ __ ] Not [ __ ] talking, but he's talking in a rude way. And the AI is making it polite.

[00:52:27]>> That's horrible. That's horrible. We should never use this feature. If I'm rude, let me be rude. That's my decision.

[00:52:34]>> I don't like that at all.

[00:52:35]>> Imagine I'll say Paris, get the [ __ ] out of here. Paris, please exit.

[00:52:39]>> I don't like this at all. Changing language.

[00:52:41]>> This is like 1984 [ __ ] right?

[00:52:43]>> Yeah. The moment you change, >> what would they call it? Buddy speak.

[00:52:47]No, no, no. How did they call it? Like they made the words and they wouldn't let you even use certain words in 198 No, 1984 language was called >> new speaker.

[00:52:58]>> New speak. Yeah, this is this is like not cool at all. You're changing the original meaning of the speaker. I do not want that. and you're waiting for an AI to predict what >> by the way you know how horrible that is in a workplace if you're scolding someone and they say oh Pratt please continue doing your work in A TIMELY MANNER PRATT HAVE YOU DONE THIS HAVE YOU FINISHED THIS WE ARE LATE >> you may NOW GET THE [ __ ] OUT >> YOU may now exit the premise >> even okay we we talked about the root but even the other side but maybe the other side they won't really tone it down you did a great job you get a bonus your salary would be increased No, I don't like it. You know what's the funny? Actually, let's watch a multitask one as well, please.

[00:53:42]>> I just watched the new Devil Wears Prada 2 movie, and I heard that it has a pretty massive opening box office.

[00:53:51]>> Yeah, it's crushing the box office in its opening weekend with around $233.6 million globally. Did you enjoy how they brought back Andy?

[00:54:02]>> Yeah. And also, I really like the fact that Lady Gaga is featured in the movie.

[00:54:08]>> Definitely. Her cameo at a runway event performing Shape of a Woman adds a ton of tension with her implied history with Miranda. Did the ending deliver for you?

[00:54:20]>> Yeah, totally. Oh, and also, I'm in the mood of watching similar movies. Um, can you please pull up like recommend some movies that are on our show right now?

[00:54:29]Let me look up some recommendations for you if you want something right in that special. The drama is your best friend.

[00:54:36]No, >> she's talking. So, I will test something similar right now for you guys. Let's check this out. I'm going to go to chat GPT on my phone and try the voice mode which is not considered real time. Chat GPT, by the way, has a real time API voice mode. Pratt, if you can pull that up really quickly on the side, which you can use by paying for the API. We haven't tested that out. that is supposed to compete with thinking machines by the way. Let's see.

[00:55:07]>> Hey, how are you doing?

[00:55:10]>> Hey, I'm doing well. Thanks for asking.

[00:55:12]>> Hey, I I wanted to ask you what day is it today?

[00:55:18]>> It's May 12th today. Anything special going on for you or just checking in?

[00:55:22]>> I'm just wondering when is uh my birthday? Do you know that?

[00:55:28]>> I don't actually know.

[00:55:29]>> Oh my god. Okay.

[00:55:32]No, keep going.

[00:55:34]>> No worries. I'm here. If >> this is the problem with nonreal time voice mode, every time there is a word coming out of your mouth, it stops and it pauses because as far mentioned, it's focused on input, weight, processed, and output. Do you know you are live on a show now?

[00:55:55]>> Uh, that makes sense. Real time overlap definitely changes the flow. And hey, if we're on a show, I'll be on my best behavior. No pressure. Let me know how I can help you with this dynamic format.

[00:56:07]>> So, it works pretty well. It's not that it doesn't work.

[00:56:10]>> No, but the last demo that you showed, what was so special about it? Asking about devel product and then book a ticket. It can do it too. My biggest problem is we know by now that software is slowly no longer a moat. It's not a a tool that you have in your belt that cannot be replicated. It can easily be replicated. Yes, they're saying their model is revolutionary and it performs uh substantially better than other models. I just don't see any way for Claude or Grock or OpenAI to not be able to come and build something like this.

[00:56:48]If you were to look at the real time of OpenAI, can we see what that's like?

[00:56:52]There's a video of that using new realtime audio models in the OpenAI API full laptop with transcriptions. Now, as I start speaking in French, we'll lower the volume of my mic and increase the one from the model so you can have a real feel for it. No edit to the audio. Let's give it a try.

[00:57:15]What's really impressive is that the model can listen to me and translate while I'm speaking.

[00:57:19]>> Sounds good. Great. It sounds good.

[00:57:21]>> It waits for the keyword like the verb.

[00:57:23]So, it does pretty much the same thing.

[00:57:26]>> No, I think the only thing is the real time thing that it doesn't interrupt.

[00:57:30]That's it. And probably they can do it in a matter of time. I don't really >> I agree with you. The biggest question is what is the mode? What's stopping from any of these other much larger companies to come and easily adopt that same feature and release it? What will happen to thinking?

[00:57:47]>> I know to get the user base. How can you get a user base if your model is not as strong as claude or chach? One thing to note, however, is if we put all the brands aside, this is a future that we're moving towards. That's the whole point of this show. We want to keep you updated with where the world is going towards. We want to make sure you understand how we are supposed to work today to be effective and how we should prepare ourselves to be ready for the way we should work tomorrow. And this is clearly where we're moving towards.

[00:58:18]We're going to have, aside from the mech suits in the future and robots, we're going to have AI models sitting in the living room. Not like the Amazon uh what do you call it? Alexa, Amazon Alexa just sitting in the corner. We're talking about a model that can see, hear, and do things in real time as if there's a real person waiting for you to tell them to do stuff for you. Uh and once you connect that with your um agents whether that is clawed or open claw or neoclaw with nvidia now or there's a there's another one >> manis >> no no no not manis there's manis and there's another one no I know nemo there's another AI agent tool that is also really famous now all of these will be available and people are going to use them when are we going to do it that's my biggest question before you answer that question I want to ask you something when do you us as bad decisions who is technologically savvy.

[00:59:16]We have a show about AI and tech. When are we going to have a 247 available agent enabled with its own hardware with vision capabilities and audio capabilities? Let me finish my point. We don't have that. One main reason is for privacy. The moment you have that on, it's seeing everything. I'm asking, are we ever going to have that >> or will we? Wait, why don't we turn one of the Dell towers to what you're talking about? It's separate from our network. The only part that I won't give access is vision. The rest I actually we we talked about it. We need a 24/7 server. Our NAS is 24/7, but it doesn't have any compute. We can take one of these T2 towers, super strong. It's separated from our workstations. Put one model in there. And uh I'm okay with audio. I'm not okay with vision. By the way, since you mentioned Alexa, did you see that video that uh the guy was talking to Alexa and Alexa was reciting information that the guy never gave to Alexa? So, he was listening the whole time.

[01:00:21]>> No, I didn't.

[01:00:22]>> There was a video on internet that went viral.

[01:00:24]>> So, Alexa had information that it wasn't mentioned to him.

[01:00:29]>> I mean, it was somebody in the room, right?

[01:00:32]>> Yeah.

[01:00:32]>> So, of course, he would know. It's listening all the time. Of course, he knows.

[01:00:35]>> Yeah. So, so I mean in our space we can but to to have that honestly speaking I think we kind of have that in certain level now. The only thing is yes 247.

[01:00:46]>> This is like if you were to hire a full-time assistant they'd pretty much be around you all the time. You can ask them to leave when you're in your main room but they're generally around all the time and they know what's going on.

[01:01:00]They have context of everything. So when you ask them for something they know pretty much everything. Yes, you can still choose to hide information from them and that's what I think we will get to. But my question is we talk about this all the time where everything's headed towards. I know this is the future. I'm just asking when will we start implementing it in a secure way. I think we can start even implementing it now. To be very honest, the only challenge I see is the final decision making output. Again, open cloud and all still there's a lot of problem and bug with them. you you know I will give you another example. Our chat GPT and claude account have great context about what we do, >> right? You just need to give them a voice. Am I right?

[01:01:43]>> No, I'm talking about listening >> when you when you have an assistant. They're listening in on the conversations.

[01:01:52]>> True.

[01:01:52]>> They will give you advice, >> unsolicited advice. Alfred gave unsolicited advice to Batman because he knew what he's up to. Alfred was with Batman for 40 years.

[01:02:03]>> Sure.

[01:02:04]>> Yeah.

[01:02:05]>> Don't you think that's a future we're headed toward?

[01:02:07]>> Listen, I understand the privacy issues.

[01:02:09]Don't get me wrong. I I know all of that is an issue, but I also know that when Open Claw was released, like how many how many hundreds of thousands of people downloaded it and started using it?

[01:02:19]>> No, no, I agree with you. There will be there will be a time that I think we should have that with oursel all the time. Not not even a company thing. I think on a personal level, >> I believe people are going to have They will >> usually up until today people have waited for Apple to release it so that it becomes >> yeah true actually massly ma on a mass scale it would be used by the general public but right now Apple is really slow with uh their updates in tech so I don't know if it'll be Apple who comes up with it Ivy was supposed to release a product with >> that they turn it down they closed No I think they closed that department I think I read this our journey >> they had a launch video Ivy and Sam, how much they spend on that video? Sam Alman, >> one more thing I'm realizing is all of these launch videos and demo videos look exactly the same.

[01:03:10]>> Same company doing this.

[01:03:11]>> It's the same. It's the exact same demos. I wish for someone to come and do something different. It's the same looking developers, the same tone.

[01:03:19]They're like clearly exaggerating the tones. Hey, why don't we try this incredible new product? And it feels a little fake now. I think the script is generated by the same AI like everybody.

[01:03:32]>> Yes, script is generated by the same AI and their marketing teams are probably all the same.

[01:03:38]>> They said the project is still active.

[01:03:39]Current report is on May of May 2026.

[01:03:42]Partnership is moving forward early 2027 launch.

[01:03:46]>> It's just delayed. It's not canceled. I I knew that.

[01:03:48]>> Okay. But I think for us like making a AI that's always listening to you and inter limitation is still a problem. Like imagine the data that it has to store and constantly remember since the past like let's say one week.

[01:04:03]>> That's actually true also.

[01:04:05]>> Even No, it's true. If just text alone, Claude is compacting your conversations after a few back and forths. Imagine the entire day's conversations stored. It needs to be able to >> exclude you. That wouldn't be a lot of conversation. Actually, if just me and Pratt, I think with 2 GB of memory, we can go for a year. If you shut the [ __ ] up, >> we can handle it with the current hardware that we have. If we include this, yes, we cannot do anything. Every 30 minutes, we have to delete all the memory.

[01:04:37]>> I'm not going to find that. It is true, but we'll have to find a solution for me >> for that. We will do it. We will do it.

[01:04:43]>> Ladies and gentlemen, that about wraps it up for today's episode.

[01:04:47]>> Thank you very much. We appreciate you for tuning in, listening, and watching.

[01:04:52]If you haven't already subscribed to the YouTube channel or followed us on Spotify and Apple Podcast, >> please go ahead and do that. It means the world to us. It helps us with the show so that we can bring it back to you every Tuesday, Tuesday, Tuesday, every Friday.

[01:05:10]>> Friday.

[01:05:11]>> Thank you once again.

[01:05:11]>> See you guys on Friday.

[01:05:12]>> See you on Friday. Until then, ciao.