OpenAI's 12 Days of AI Surpirse, Breakthrough AI World Models & More AI News
52:16 Share

Hey, merry shipmets, everybody. OpenAI is gifting us twelve straight days of product releases and updates.

That's right. Exam open promise, twelve live streams. We also make a sort of video generation, the new reasoning model and so much .

more of stuff. My little stocking santa women because i've been extra note.

That's not what we're doing, Kevin. That's not what we're doing here.

You you're right. What I meant to say gabbin was the holodeck is here.

Yes, the google announced gene to and world labs came out of still, both of which are really interesting, new ways to imagine prompt to world simulation.

We've got all that, plus new video, audio and image generation tools that you can .

use right now. IT is a pack week, and IT is only getting crazier here, packed like my native little sam. No, is not that. IT is A I for humans, everybody. Yes, everybody IT is here twelve days of chemist opening eye has gone out and said that over the next twelve starting today, meaning that we don't know what the thing is yet but starting today um they are going to release twelve different live streams in each day are promising a brand new thing sammon himself tweet that starting tomorrow attending epa ifc we're doing twelve days of open a eye each weekday we will have a live stream with a launch .

our demos and big ones and socking suppers on the biggest update Gavin, you're bearing the lead according to some ChatGPT users, they saw code that replaces the voice mode button with .

a snowflake I O. We have .

the most incredible .

new thing. IT is prompted snowflake button and you are going to love and don't what ellan is gona .

have nothing to say about sam altman and implementing a snowflake.

But well, my gotta do think so. Let's dive in to some of the things we expect that we're going to see. First of all, sora.

So sora has been in the news quite a bit. We didn't cover IT last week, but there was a big kind of dust up around. A bunch of artists gave access to sora for the first time through their account and IT wasn't exactly approved.

But then, Kevin, the other big thing that kind of sitting out there is o one, the full version, which is the reasoning model. Up until now, we've only been using a one preview and it's very good. But oh, one, that the four a monty, let's call IT, is actually quite more robust. What else do you think we might be seeing out of this?

No clue what is actually going to be revealed beyond sora and probably the full of one model. We know that open the eye is working on their own browser that probably is far away, but would be interesting to see they've supposedly been testing a model that is ad supported to make full fledge open eye free for everyone. Now sam botta has said he hates advertisement and doesn't want that. But then again, trillion dollars to train some of these models. Lets get some .

calls of money. Maybe need the money.

IT might not be the sexist of updates, not to objectify the twelve days of shipments, but I want Price cuts. I want the cost of real time voice agents, which they've been pushing and iterating, I want those costs to go down because as we will talk about a little bit later in the show, it's clear that sort of voice, an agent's behavior, are the two next frontiers that we're going to seen in twenty and twenty five.

It's just really expensive to build with real time voice, which is something i've been trying to do. I want to see less guard rAiling of those voices gave, and I want to get more expressiveness. No, I do. I want.

we don't say good luck. I don't know, want to allow more openness to the voice is to see that's .

that's a good point. I would like to see that though to me that's the tiny thing. IT doesn't require a massive mega tn announcement just then listening IT up a little bit so the voices can be more playful and more performative. I think that would open up more exciting use cases. I don't know if you have access to a yet, Gavin, on the mac application, the ChatGPT APP, what you can give IT control over other applications on your device.

You I think I might, but I haven't actually played altha at I do.

They give IT and they take IT away. Oddly enough, sometimes I have access to IT and IT. Sometimes that goes away. It's just not there. I would fully expect we're going to see that rolling out for the entire user base, and i'd love to see further advancement in there.

In sofa right now, I have a little icon that I can click on my ChatGPT APP, and I can tell that, Gavin, go run cursor, go take control over this terminal window, take a look at this text file and read IT like you can give IT specific access to certain apps, but you cannot let IT control a web browser. IT refuses to recognize zan APP, and even still IT is reading the apps that you give IT access to but not interacting with them or interfacing with them. I would love to see a step towards that.

So I have two big predictions here. I think one thing we're going to see is um a formal AI. The announcement that comes out of opening eyes, there has been some rumors around this into your point, like really controlling computers, which they know kind of anthropic drops with computer mode, like I think they're going to see something that's more robust there.

I also think this is more of a hy post kind of announcement, but I would bet that they're going to demo s some sort of next in frontier model and IT is not going to be available. But I I think that that's going to be in these twelve days. I would be kind of shocked if they didn't use twelve live streams and have one of them show off what the next big thing is.

I mean, there's a big news here that ChatGPT just hit three hundred million active weekly users, and that's up a hundred million from just a couple months ago. So like ChatGPT is crushing. And I think this is opening ice saying we are ready to dominate this world.

We are ready to kind of move forward to this thing and they can have to put up or shut up pretty soon, right? Because there's been a lot of people saying that like D P T four is great, but we haven seen a lot of updates. Ince, like you know march, April this year.

In fact, one thing that still music ever, which we probably will get through thing is GPT four rows image capabilities. We didn't think about like you know a new daily, but there's going to be some new image capability, too. So I would suspect a new image model, not just sora, but some maybe sora has an image model attached to IT, which we've for room is up.

So maybe that will come. And you know, everything is just to come of transition to our next stuff. I bet that they're going .

to show off more sore as world simulator stuff is. You can see the the arms and hands and legs kicking against IT with all the deep mind staffers, yes, and they got caught in a chi mnh.

That's right. We have ve also to remember the sam and opening. I just recruit three new members of deep mines. So google continues to lose burst out of its A I org.

And in fact they they recruit the personnel in charge of multi model training for deep mine, which is a big deal, because multi I model training is this whole idea of, like how we see the world, or how the A I. The world. So to that point, I think when they talk about sora, one of these things is gonna about word simulations. And so in that instance, we had two really other giant stories that drop this week in the idea of A I as world's, but maybe care for our listeners and viewers. Maybe we wants to talk a little bit about what A I as world's m even means, like you you want to give of the til D, R.

And what that ideas yeah, right now, if you imagine a call of duty or fortnight or row blocks or any of these big, massive games, they are powered by traditional game engines. These are pieces of software that take three d objects. They slap a texture on them, so, uh, cube can look like a create, if you will.

Uh, they power the sound. They allow you to interact with that using your controller or whatever the input method is. And so these are complex pieces of software that render these interactive worlds that you can journey through gross distillation. But here we are.

Well, instead of a traditional engine putting all that geometry and handling the lighting and the user input controls, what if you used A, I to do that? What would that look like? What would that feel like? What new capabilities would that unlock? And while we are very much at the pung stages of that generation of technology, these A I powered world engines are already producing some really incredible imagery and what looks like actual interactive behavior based off inputs from the user. Now, Gavin n you let up when I said at the pung phase, why are people hlubis ating mass effect worlds and halo and these massive desert? Where are we just .

polluted ating pung? Well, oh, that's a good question. We could just have if we made pung giant hot dogs going back in the the three space that seems like A D you make that.

Can somebody make that? So I think you're right. What I was smiling about the wrong thing is that I think we might actually be past the pung stage with what's just come out. So there are a couple big updates that just came out. Jenny two is the update to genie one.

If you're a long time, listen, you remember us talking about gene one, google deep minds, a video game simulator or created tool, back in the day they announced genius one because I was not released, but was a sight scrolling two d video game generation. You could put in a prom to say, like, you know, plumber move through super and IT would come up with a kind of a version of the plumber through looking through a super. What genie two takes that two d environment and now has made three d prompt to video games.

Again, this is not out. This is a blog post. And google deep mind is very good at shipping blog post.

So we know that they are out there as a lot of I companies are. But Kevin, you look at this post is pretty crazy. Like you can see right away that like basically there's A U I. They figured out like you know, you can walk through as a robot through a 3d environment or you can walk through much of things。 One thing that really shock me was the one it's like a power boat, like there's a power boat on the water and it's it's getting the physics .

of the water kind of right me. That's .

pretty crazy.

A lot of these demos thus far have been most like a map painting that you can kind of whittle around and they go, wow, look at this. What we're seeing here is coherence as the camera moves about. So it's rendering the world, but not forgetting the world that it's supposed to be in.

And the interactivity with objects within that world. So believe that are not a character walking up to what looks like a door in any other engine might just be they're walking up to a painting of a door. It's like the old lady tunes title painted on the canyon, right?

It's not a real thing. Well, here, the road runner can run through the tunnel. The character is interacted with.

They move with an animation cycle. There seems to be physics on IT. They're casting shadows. They walk up to the door. The door opens, the character goes inside. I still say like punish phase because we don't know if IT completely falls apart after that, but we are very quickly going from proof of concept. Could you even do this to now were seeing interactivity, physics, entire world's halcon ated from a single steel image?

That's right. And on top of that, there's another big announcement. A company that came out of stealth this week called world labs and doctor fa leaves behind this, who you may or mean out.

I've heard SHE is like an AI pioneer. SHE worked on some of the early image net generations of things like that. Her company basically is doing this exact same thing.

But in their mind, what they're doing is that you can take a single image and then walk around within that image. And more than that, you can add physics to that image. You can change the lighting of that image in that world.

Now, if this one you can demo slightly. If you go to their website, which will link a to our show notes, you can actually walk through these worlds. They can stop you a little ways into IT. You can't like go fully because this is a very expensive thing to render. So if you're going to continually render IT, it's pretty hard to do.

But they have a lot of videos showing people doing that and kept the thing that this one blew me away was like when you watch in IT, you can do this in the demo, you can change the lighting of the scenario, and the lighting actually follows along with a spotlight. And one of the videos s they throw a bunch of of basketball, and you see the physics of the basketball move. This is like at the very beginning stages, that that holistic idea, right? Like the idea that you could step into an environment, and it's not program to do this.

But IT understands, because he knows what the world is supposed to behave like that as you move forward. This changes, this changes, this changes. And and that feels fundamentally massive when you think about generated video or generating worlds at large.

And not just video games, I think that's an important thing. Body, this is like a bigger deal. Just video games. This is like how A S will probably in going forward and that's like the secret sauce to all these companies making these they're looking for more training data.

Funny, you know, that is specifically something that the gene two paper checks as well and their tweet thread about IT says, like we finally unlocked training data for these new models and so that yeah clearly that's going to go there. I liked that you mention that not for video games specifically, because some of my favorite world labs, examples that saw were full motion video of humans walking, and then they replaced the world behind them.

It's like an impact vfx swap. If you're listening to the audio version of this podcast, do yourself a favor, go into the notes or check IT out on youtube, because you should see these videos. They are really, really impressive. So this is sort of real time world simulator. And I know where in the middle of shipment s now maybe a sara came out this morning from underneath us, but tencent has been shipping and now there is an open source video model that I think some of the Cherry picked examples are just as impressive as anything that open eye has shown.

Your tiktok are going to good work, everybody. Your tiktok are being use very ten, said the company that owns take talk at a bunch of other the startups has released a brand new open source. That's the biggest thing. They open source is day. The video model IT is pretty good.

I feel like it's maybe lagging a little bit, but the fact that it's open sources incredible and you can go try right now there is a company called file no no add here, but that I was able to actually do this on myself. I I made generation, and I actually each other. And you can see, Kevin, what came out of IT was definitely not coordinate y or by ferry.

But IT was a pretty good look at like two random guys. High five then hugged each other. IT did cost me fifty cents for that one generation and did take, uh, twenty minutes to get IT.

So like we are a ways away this, if you in order, you can run this locally, but you really have to have A B C step. Like no one's gonna be running this on a forty ninety. But the fact that this is open source and that is being released into the world is a really interesting sign for we go for video generation.

Again though, this is a chinese company and we've talked about this on the shower bunch. I'm going to talk about clean later on, which I joined, and got to play with the emotion brush. Chinese video companies are going to be more advanced probably until we is america, either say, okay, train on everything, or maybe a couple of generations down the road, because these chinese companies have no trouble training on any data that they have. In fact, I joked about tiktok, but it's very likely that many, many mini tiktok, if not all tiktok, are part of this training data that I would imagine.

oh yeah, I think the only solution is to put a terror on video generations again. And imagine if that fifty cent generation actually costs .

you fifty dollars. Yeah, now that would be a big deal, would probably make amErica lot of money in that.

I digress. Just a joke, not politics. I swear, so many are saying that with these tools advancing as quickly as they are, hollywood is going to be completely out of business, and there's going to be a brazilian different creators out there. There is going to be so much noise, this is going to be so disruptive. This is going na be way too much content but our sweet heart, dear friend of the podcast, zac neither, came out and said, I don't want you know that he's a huge fan, huge fan.

Wow, that's great. I'm so glad maybe you get a coat from him at some point you think you can get on the .

volunteers can get like a quote to put on our presley, get quoted now through a in the same way that when you know, we all cause like every single person has a pretty good camera on their phone, pretty good camera on their phone. Every single person a pretty good movie camera yeah like yes a legate movie camera um and yet you we don't have you know millions of awesome movies just being uploaded out of people's pockets.

I I I feel like the AI represents a another another tool that will help us make movie sauce on my hope. So I mean, that's the point that I think we've had and and many others have. But it's great to see professional directors, capital artists, reaching the same conclusion, which is like having a professional grade.

Cameron, your pockets certainly LED to everybody sharing their story for Better or worse, all across social media. But I didn't fundamentally change triple a blockbuster movies. IT made IT cheaper for some to do IT. You still have to be highly motivated and real IT as the tool that IT is to make the piece of art, to make the thing.

Yes, are you ready for a counter attack here? Kevin have been contacted by skynet and ready for a counter attack here. Here's the counter attack of my mind, not an attack.

But like, I think you're absolutely right. And zx, absolutely right. Like they're warn a million, you know new triple a blockbuster movies that came out with the phone camera.

But guess what did come out? tiktok? Guess what did come out? Real people making content that was specifically like camera related content because he didn't make a lot of sense for a Normal person to go out there and shoot like a two hour movie with a camera.

I do think these I tools will make a new sort of generation of people making stuff. Right now. People think of IT is a ice lap, and that's really happening in a big way like this. When I say I slap, I mean the idea that like there were being flooded with A I content, whether it's A I influencers on tiktok or it's like some of the very funny, weird A I generations we see all over the place yeah I think .

that almost two different things we know the floor is going to raise, and individual creators with very little experience are going to be able to make a certain level of quality of something. So IT democractic zs creation, something you and I have discussed numerous times. But does that make actual movie creators out of everybody that starts to create content? And I think there's that concern of, well, i'm a professional now.

Someone can do what I do and it's like I I don't know that I buy that either. I think this goes to sort of sex argument. Is that just because you have the yes.

Kevin, counter point two is going to be coming up later in the show, Kevin, because in a ic which you did there, we have a ten minute batman movie that was clearly lifted. A lot of I P, I would say is a really interesting point to that. To counter point to this thing.

i'd look we'll table the conversation then that's totally fine. I'll just say that a team of capable creatives today more often than not make something Better than a single person who is experimenting with a tool. I don't think that's gonna change.

As the tools get Better, the ability for one person to make something amazing does rise. But the ability for a group of people or a group of professionals to make something way Better, the bar is just gonna keep raising along with the floor. I imagine, I guess, is my point that makes sense.

And guess what there another barriers raising, Kevin, and that amazon has jumped into the state of the art A I model release race. This is like amazon has come out with a new series of nova models. So nova AI, that's our new um we have games, we have a chat.

B T. We have anthropic clad now we have nova and allama. All these different families keep getting bigger. And Kevin, this is like amazon kind of big jump out into the world of of trying to kind of make their own thing.

I will say there was a great video that came out where new AI chips are called train um the train um chips. And by the way, apple has just said they're gonna using the amazon chips to work on some of their stuff. So that's A A big deal. But I want you to play this this released video and listen to some of the gybed ish that is said .

with in a power and high band with memory are combined in a server with advanced virtualization, enhance security and high performance storage and networking. Now imagine connecting these servers with an ultrafast dedicated chip interconnect and then deploying them in hyper scale clusters with a pet bit scale network fabric.

What you get is a silicon server and data center architecture that delivers state of the art performance for front to your models. Did you imagine a Gavin? Did you imagine hyper scaling had a big fabric that .

I wrapped around myself and I wore like a row to Christmas morning? yes. So again, those videos are not for Normal people. They're not for people like us.

But there is a thing if you release something like making a kind of at this understandable by the world of large things. Amazon coming into this just shows you the actual weight of how big this races. Each of these massive companies need something on amazon to spend a lot of money investing in anthropic. But the fact that they are also releasing their own model, much like microsoft has released their own models, might mean that they want a little bit of like freedom from just the startup s and they want to do the thing .

internally as well yeah and they ve already taken up the micro O B U I P A naming convention because when you say it's amazon noa, you mean there's nova micro, nova light.

nova pro nova is nova light IT?

Yeah not as many as no a zero. You're going to I just .

wait off of three of them. That's my question.

So it's technically four models that came out of this thing. Obviously, the microbe super light, super cheap, super fast, all the way to premiere, which says coming soon. But it's the most capable multimodal model. And IT has complex reason, and they also advertise that is the best for distilling models, Gavin, which is where you take a more capable, powerful model and you poke against IT to just train and distill a particular type of knowledge that you want in your model.

You learn. Watch in the video. You saw Kevin. Kevin just did A A hand motion that you probably used for something else. You're out there watching the video.

Very good.

Do you know what, Kevin, that might be just the hand motion for a come and join our show and and subscribed to our podcast because this is the important time that we have to say you're watching on youtube. Please subscribe. If you subscribe is the only way we grow. If you're this podcast, please share IT with other people or also give us a five star review on whatever podcast platform on we grow because you all listen and sure with other people.

I do want to mention that like we do get more podcast down those each and every week, which mind blowing to me. So thank you. If you're hearing this and you haven't shared yet, know that fellow listeners are, and you need to do your part as well.

You need to get up about how there really out there. IT really does mean the world to us, and IT really does help us. gross. If you can take two seconds to share IT posted to your favorite read IT, send IT to a friend. That's the way we keep this train charging along.

Let's keep this train charger along. We're going to japanese announcements from eleven labs, one of our favorite voice A I companies. They're been a really good job of shipping themselves lately.

The biggest thing that came up just couple of days ago is conversational agents. We've talked a lot about A I agent on the show, and there's a bunch different definitions of them. But in this instance, it's basically allowing you to create a specific voice to interact with certain people in specific ways.

Hi, what would you like to talk about today?

Introducing conversational A I with eleven labs, build, test and deploy, all in one platform for the most, a way to communicate with technology.

Hey, can I check the status of my order? Yes, of course, let me check for you now. Your order is due to arrive around two P.

M. Obviously, the business applications of this are massive because you could create agent that allows you to answer the phone, or an agent that .

allows you to do sales calls. The biggest thing is that even if you have no idea how to create anything, you can go sign up for eleven labs hashtag, not an ad. You can spend up a conversational agent and with basic english, with basic language, you can give a voice of personality.

You can give me a rule set. And what's really fascinating about this, Gavin, is that it's it's a eleven labs performative, a voice capabilities bolted onto whatever model you want to use. So when you go and you create an agent and you say, here's the system prompt, here's how IT greets the person, here's the voice they should use, there's actually a little drop down for which large language model you want to use. And you can plug in claude gami GPT or a custom large language model. May be you want to go with a no nove draft, nova dark, maybe .

nova .

I P A nova table. yeah. Again, a new amazon model. You can do that. And knowledge bases are really important. So whether you're building a character for a video game or, like you said, a customer support agent, that agent needs to know everything about your business and about how your product works or what your company standards are, your mission statement, you can plug all that in.

And when I say plug in and I don't mean higher, an engineer to hike their glasses up and write the code, you can click and drag A P, D, F. And now this agent will know IT. And that is so powerful, IT works over a telephone if you want to activate that. So people can actually receive and make calls with these agents and then you get full analytics for everything as well.

So IT will automatically analyze your call and you can plug in ah I was going to say KPI, you can plug in, in the metrics that you want IT to analyze the call with and let you know like was this customer happy? Did that we do a conversion? Did the did the quest giver give the key?

Whatever the thing is that you're building, you can build A I agents super easy with them. I wish this was an ad so desperately. Gavin.

I give us as money level this point.

I'll take credits. Please tribble your fun credits into my my santa stocking, please sama is gonna ve room. Gavin, i'm all thrown y and excited for this conversation relations and you are ten kinds of distracted is going on at your end.

The other thing that eleven labs did is they dropped a and no book L, M competitors. So you know, we've thought about no, but L, M, that is one of the best in google, has ship to the AI space. You can upload A P, D, F, A thing like that.

And IT have two AI cohoes talking about your thing. So h eleven labs has been a similar thing. And I uploaded last weeks youtube episode from our show and IT says at picking out the voices and it's like rendering in the background. So right now i'm waiting for IT. IT may not come out for a little bit here, but the idea is that, clearly, there is a use case of this idea of A I explaining stuff that you can upload, or at least stealing stuff and voice again, makes IT much more accessible for people.

In fact, so much so that no book I am themselves has been integrated in the spotify rap to this sharkey and I can play a little bit of that with all the other one's rendering because what was interesting to me was that he comes up on a if you look on the, uh, minis, but IT comes up as your wrapped the podcast on spotify. So this is spotify rap, which just dropped this week. I go into this and what it's done is created a five minute podcast of my rap. So here i'll play this and this is coming straight through the spotify APP. All right, music lover is bucket up.

Welcome to your own personal twenty twenty four. Spotify wrapped deep .

dive s visiting from google's no book I M here to, well, unrated this .

amazing year of music you've had thousand eight hundred and minutes and minutes in the .

top seven percent of listeners globally is serious .

dedication. IT goes on.

And you want to get a sense of if you do IT yourself, I would say it's an OK version of a google. Does IT doesn't feel that specific. It's not that far off from just taking basic inputs from your wrap like it's not giving .

me like I the most of Christmas future have just visited you through the magic rectangular portal in your hand and you're already naing IT. You just got two humans sounding things, having the conversation about your specific listening habits, about how many minutes you consume your favorite. You have a gift delivered from the acoustic records clock from the either, and you are like.

and okay, it's fine.

So i'm going to set up for apple music.

You know what they listen. I I appreciate the fact that, first of all, google was able to allow spotify, the users and they were able to, to make that happen. So and IT might be the first kind of big introduction to this product through a lot of people to spotify, spotify A I D J.

We've talked about that on the show too before, and it's interesting, right? Like I kind of talk you through whatever the music is. I thought that was a really cool thing, all right.

So now let's go back. Jn, FM is with hope in Chris. So IT generated a thing with hope. And Chris, this is a version .

of our podcasts a last week. Do you know? I know hope. I know not kid.

And like I, you do so much work on to eleven lads. And yeah, you use hope for so many projects. I already know what hope is a sound like.

so here we go. Okay, i'm going to play you then. This is generated about five minutes in real time .

or of the digital age. Well, not quite. Today, we're discussing how A I agents may actually be our collaborative partners rather than our replacements. That's an interesting perspective. Can you elaborate on .

what exactly AI agents are and why they're generating so much buzz? Sure thing. A I agents are essentially so me is going to skip ahead a little, but you can slide through and i'll skip ahead and just kip to a different part text to image generators that have been making waves. That's part of IT. But IT goes much further.

Companies like runway and stability AI are pushing the boundaries of what's possible. So books, what did you put into the magic machine to get this podcast out? So i'd literally .

put in the link to last weeks episode of A I for humans in the youtube. So, so clearly, what is doing is taking our conversation, not basically talking about who we are, what we are like. The thing about the book is, if you put a nobo this in a note, m IT will often say, like, well, we're talking about last week episode of A I for humans or we're talking about.

This pok say, have for humans were they talked about this, and in this instance, IT just kind of jup right into the stuff that we were talking about. I will also say, hearing that for the first time, and I first other people do this, and there are people who said, this is like the voices are not as expressive, and this feels like there might be a little bit more like plug and play, where as no book I am might have had a little bit more tween to the sauce that went into IT. All right, now we're moving onto a something we don't Normally do in the show, but there is a fair amount of IT happening. What's elon doing, Kevin?

Is that the name of the segment?

Long, long this way after man. What's your down?

What are doing this weekend?

Like I just see you like the neighbor from what like you. What do you like? I go on on over there. We don't do this often because we don't like to and we'll keep IT quick. But elon is suing OpenAI for the fourth time now, I believe a for going for profit once again.

But the other side of the stories that he's claiming some anti competitive, violating federal law nonsense that's going on between OpenAI and microsoft and other competitors and investors, he claims that someone who previously invested in X A, I did not further their investment in X A, I because they were explicitly told not to buy like a microsoft in the room saying, hey, do not ay nice with our competitors. We will see if this panel out but meanwhile, while he is trying to club or OpenAI in the courtrooms over here, he's striking deals with video over there. That's the other hand, if you're just getting the audio version.

So he's also strike a deal with video, Jenny wong, gs company that is continuing to crush the salary picks of the AI digital gold rush he's going to for a billion dollars worth of chips in january, delivered in january. So you an is really going for this in I do think the one important thing to say here about the elon watch that is kind of important even though we don't love to focus on this stuff, is that you on now sits next to the new presidents of the united states at his resort in florida. And I think that is going to be a bigger story in the A I space than we expect.

And the fact the fact that you on is suing open eye is not a great thing from open a eyes perspective because you know, there the person that's going to be leading this country, america, has a habit of like kind of picking favorites. And he's very happy to listen to the people around him, especially those that he dubs like worthy to be hurt from. So this is gonna be a story to keep track of, like, where does the scale get pushed and where does IT not get pushed in the A, I space going forward?

bit? Listen, if you're hearing this in your fill of panic, fear and dread, because you're realizing that a handful of billionaire are going to control our AI sentient destiny. Worry not, Gavin, because an exciting, knew fifteen billion parameter research model was trained by the clubs sorted kind. This is a really interesting company.

This is a little murdy for those people out there who are more kind of ori and not as deep in the A I. space. But this company called news research. I think as you sell the news, I know you research, which has been doing a lot of really interesting research work on their own, and have announced that they were able to train a model in a decentralized fashion, which means that many different s around the world were used to come together and train this model. And its fifteen billion pram ers, which is not no stay of the art like giant model, but still this is a step in the right direction to see a world where open source models could be trained outside of individual some A I safety people might be like, this is frightening as all hell, because the idea that you could train AI models outside of a controlled space really freak them out. Whether at the same point this is, to your point, like um if the government really does start to control and cracked down on this stuff, this does open the door to having IT be a little bit freer in different ways.

I'd love that you wanted to talk about this because this tickles my heart, even though I might make the broader eyes roll. It's called demo or demo d. Coupled momentum optimization allows for a heterogenous mixture of hardware.

meaning all the machines is on social. That sounds a lot like .

the pebble fabric around. That means the hardware environment is mixed, which is important. So you don't have the standardized machine in different rooms in different locations.

I mean, they did use similar GPU in their official training run to keep things consistent. But this is the plot to what terminator up to terminate. five. I know they had three. There wasn't there a force that there?

They arted in the number. So we're probably at this.

at this point, this is the sky at onal. Everybody, because the skies will be nuked. Everybody will be in trouble.

It'll be elon, you punching, I guess, cyborg putin. And there's only like three major people with their AI machines. And that the people, us, Gavin, everybody listening to, watching, we have to ban together. But we don't have super computer cluster.

So we go to our store and sheds, and we go to our direct CT malls, and we blow the dust off of our PS tools and our dream casts, which have just enough juice left together. And we plugged in. And we we get this network of all these machines.

The curig machine is plugged in as well. It's talking to the black tech blender. And all the power combination will train a model that'll be just a little bit smarter and more capable. And that is how we defeat the machines.

gather. wow. I had no idea that my blend tech blender, which is bit in my closet for about .

four years, the ninja air fire, that thing is onna, come in to play in the near future. Actually, I think it's really cool to do decentralized training like this. I think it's massively important. I do think it's a big deal for like an open source flash research community moving forward. So i'm glad we discuss IT.

Alright, everybody, it's time for us to discuss some of the stuff we saw you or other people out there doing with A I this week. Time for a there.

I'm so excited about this. This is one of the coolest things i've seen be done with A I to date. He was done three people over a couple weeks with two hundred dollars and clean a eye credit.

What this is is a ten minute batman movie. And in IT is a essentially a fan fiction, right IT. But IT is ten minutes.

IT is a compelling and incredibly watchable movie. I encourage everybody to go find this. This was upload by a guy and Kevin, the kid.

K, A, V, A, and dash, the dash kid, and he is somebody has been a fair man of work in the at least as upload a lot of stuff to the AI video reit. But this, he said this was done. He was pretty intense.

He said it's been done with sixteen hundred clean credit workflow was pretty much using every available AI tool starting with image and then animated video cloning up breathing talk about three weeks to create with three people. Um so this is an example of what is possible right now in the world of the a video spacing we've talked about prompt to hollywood forever on this show, that the idea at some point build a prompt something and make a movie. Right now this is a also a good example of prompt to hollywood is kind of a myth because these guys did more than just prompted.

They prompted, then they edited than they added voices. And they prompted the voices. They did all sorts of stuff.

But I do encourage everybody to go watch this. IT is, again, ten minutes long. It's like, no, it's not like the best thing you've ever seen in the world. But IT is very compelling and you get a sense of what IT is.

I don't disagree with anything that you just said, but like even us in our position being acutely aware of space a year ago, oh, my working like if you just look back two years ago, impossible. And let's just all be aware that the goal post keep slowly getting nudged as we're looking at different directions.

This five years ago, somebody uploaded to youtube would have been one of the best batman fiction tions ever made.

Assign them in a minute. IT would have been a big deal. Uta have been in a bidding war of the agencies about this ragtag team of three that made insane fan fiction.

And yet here we are in a those lind's, Gavin, we talked about the coco, a commercial which started set the internet a blaze, for Better or worse, a lot of people shading that one. Well, just today, as we're recording this on wednesday, december fourth, a vote phone commercial. They are the client.

The agency was new commercial arts. They released what looks to be a minute long commercial that at a passing gLance, I think if this were on my television, I, O, that's a commercial for vote phone. And it's a spot about moving to the rhythm of our lives. And it's mostly .

A I generator of my life, Kevin, and you should know this at this point, is the rythm of my life in that bad?

I imagine like, uh, like A A toaster pull sitting with hot dogs coming out, like round .

dogs and buns.

But no, listen, if you watch the commercial, is these mini vin yet, which we know AI do very well, A A baby floating in space and then in water being birthed and then the people taking photos of, and it's basically telling the journey, this almost of growing up, and then it's shift to the perspective of the others around IT. But it's these rather beautiful slice of light, yes, that look like they were shot on professional cameras.

It's very well edited. And to what you just said, it's not prompt to commercial, right? If you look at the credits list for this, there is an E, P, and a producer, a creative director, multiple AI artist, F, X, producer, vfx supervisor. The list goes on and on. But if this were done, let's say, three and a half years ago, Gavin, the list of credits would .

be fifty times easily.

easily. And so this is the interesting thing. And I think this is just trying to get to earlier with, like, hey, the floor is gonna ise. The bar is also going to raise as well. So yes, we say IT every week, but I don't mind being a broken record. Get in and start experimenting with these tools now because what you looked at the credit for this commercial, which is beautiful, the spot with the most names on IT happens to be A I artists that didn't exist three years ago. And here we are .

and shut out to a uncanny, a hurry on x to a guy we ve been following for a long time. Speaking of new A I R and individuals, Kevin, there is a really cool tool called a retro diffusion that is now available. It's been available for but a new version.

And what this is is a, is a pluggin that basically allows you to make pixel art with A I and this has always been a tRicky thing for A I. That is cool. Pixel art is like IT always has little kind of funky things with this.

And this guy at real astro polls, who i've gotten know a tiny bit through this chords and stuff like that, has continually worked on this product. And IT is a paid product. If you're looking to make any sort of A I pixel art, I encourage you to go try this and and play auth ID. Looking at like what it's been able to pull off is pretty amazing .

that i've got a mini game for you, you ready to play.

I'm ready to play. What's the mini game?

Listen carefully. Because I don't know this is going to work at all through our sounds of best, but see if you can identify this. 我相信。

Sounds like guy is with a chain he's running after, so a human guy with a chain sauce running after a little rabbit who rushes away really fast.

Pet, in the answer. We look at the judges and they got human with chain. That's actually all we need a Gavin, what about this game?

That sounds like a car driving away really fast.

Oh, I does. And you don't need any rabbits or what anything else that does sound like that. You're two for two. Gavin, what's this?

That's me. Hang over playing this phone for my fourth grade class. All right.

two of three. And bad IT was very close. So what do all these sounds have in common? Gaven, well, they're not real.

This is the future of video to audio generation. Now ten ent, I believe, also does this with their new video model. But this is a dobie new multiple li A I tool. Basically, you feed IT an input video and IT generates the sound effects for you. And now I see that .

that was that sound playing at the end was a woman playing a harb is IT actually picking up the notes that she's playing and that would be .

a crazy IT is relatively thinking back the plus to the fingers, right? So IT is IT is letting the video action motivate the sound that comes out of IT. I don't think it's playing the actual notes on the string, but if I. As that I would like, I would really have disappear. Well, we don't know of the source video if he was even capable, if SHE was using all seven of her A I fingers .

to play and come on, man, you you're just in her playing without even though.

yes, I don't even know if he is real. I don't know if that's A I or actual video. But if you looked down at the other examples given there is clear that they shot a lot of stuff with like a go prose or body camp perspective.

But they have people doing stuff, like washing things and prepping food in the kitchen. They've got borne yard animals. I've got all this weird ootah of IT, but IT is fairly accurately reproducing the sounds from the videos. And that is just the next one lock as we talk about this prompt to hollywood. It's going to be using all these video generators giving IT a thing and having IT automatically generate a soundscape to match your video that is going .

to the world's stimulated thing, right? Because that's the other thing, which is really fascinating image when you're walking through the video game stuff, are talking about the top of the show. One of the things that sound adds so much to the probability of an environment, right? If you can, in real time, start actually flying the world at large, like we're just getting close from closer to being stuck inside our gog's forever. Kevin, and how are we going to stop this?

How are we get stop stuck. I don't want to stop IT, Kevin. I've been outside.

It's overrated. I don't care. There's no reason to go out there.

You know, it's not out there. Gavin. What's the A I for humans newsletter? No, you can only get .

that almost today either.

That IT wasn't. We were a day day.

we were a day late as we with our a room, oh.

because we had to delay IT to pack IT full of neutral licious. So if you have not signed up for the A I for humans newsletter, you are missing out. The line is going in the proper direction there as well.

And we sincerely appreciate everybody sharing that with their friends and family. Go to A I for humans dot show and you can sign up for our newsletter. And that's A I for humans dot show.

Get the newsletter. IT is free. IT is weekly. And what I love is that it's a breezy little read. You can kind of score IT .

in one flick. Also, you can sign up for cabinet to come give you a talk .

at our website if you never want to do talking to to witch.

let's think about what we do with the eyes. This week, I have a couple really fun things. First, I want to just drop something very fast that everybody can go do.

And if you want something to do, impress people like Christmas, this might be a fun thing to do. Google drop the thing called jane chess, which is actually a very cool, dumb thing that you can use to basically create a chess set and play with specific images. You can basically prompt to chess set.

And like, what's cool about this to me is I made one that was like, hot dogs were associated ate the pounds were all little hot dogs, and the queen had, like a hot dog hat. And all this stuff in IT is a very simple, fun way to use A I that kind of make something, but then you can literally go play chess or share with other people. So it's fun to try if you try this all.

I'm trying IT literally right now, because when you go to the lab site, you can choose to make a classic or a creative chess set, and then you can give IT any inspiration. I said, give me a creative chess seat inspired by nantego games. And IT delivered me a fantastic rendition with mario, like a full mario with the m on the cap, wearing his, wearing the crown is the king. Princess peach is the queen.

wow. Really, really amazing that they were the I and .

i'm a shock by that. And i've been banned from the internet.

That was quick.

yeah. Take a look at this. And then I click to generate opponent, which is something that will auto suggest based your first suggestion. And IT says, intendo vers saga games, and share enough. I've got sonic robot nic, and I think that is the character from nights I gotto turn in .

my my google part, maybe didn't get a publicity so that .

they haven't turned off the ipl.

Ly, and I can see the game, and he looks incredible there. yeah. So very cool tool. So that thing I did this week, I saw some really cool videos about cleans motion brush. So clean, if you're not aware.

And the one of these chinese AI video models and the emotion brush was getting a lot of pick up because you can literally take up to five subjects and make them do different things. And IT really does give you a lot of different things to be able to do and action to be able to do when you actually making your videos. Couple things about clean. I got a black friday special, so I I paid for this myself at sixty box for the year.

You get like about it's not a lot of generations for that, but you get like I think about fifteen to twenty and then you can get a free couple, three one's every day that come in IT takes a very long time to render these and i'm not sure if it's just the motion brush, but like each one of these takes eight to ten minutes to get back, which IT makes IT really hard to work with. So first I had this picture I made of of, uh shk holding a little ti sana that I was like here, love to see if I can get the santa. What I want is track to throw the santa a into the air.

So I took shack arms and I made them go up. And on the sand, I may go up as well. And you can kind of mask each section a person and decide what you want to do. And then you can make .

the background stay similar.

So so you can look at the first one, they kind of made the sana like explode upwards and given .

a quarto from total recall the symptoms shirt lives up and there's a full menacing face with teeth on the sale.

right? I never even saw that this. So like there's a second like face underneath.

And then I tried to do IT again, and I didn't really fully get IT. I got Sandy going up, but not to try to show IT up. And I tried that a couple times.

And then one of the other things, as I tried, was, course, did you see these video? The brothers I tried to have there is a picture of a macron, the president france, in trump, where they were kind of like trust, charging his color. I tried to have the two of them hug and then kiss.

And one thing that was interesting, covered about this is in the prompt I put for this one, two men embraced and then kids, and I wouldn't let me do that, but when I put two people embracing because they let me do, you can t right? This is one of the benefits of the ine's models, like do things like this. And again, A I uh artist like the doors brothers use this to make their videos s where they have famous people doing stuff.

And again, this was just me kind of pushing the motion brushes towards each other. So you can see it's like it's a pretty good way of kind of controlling that motion. And then finally, have I took our dominy home last week and I just kind of went crazy with IT.

So our nail was like a robot, terminal robot, holding a turkey you can do in your temple hands. And me just kind of in the background. And as you watch this all kind of a go around, you can kind of see IT figure out three d in its own way.

Is IT struggling with a couple things? Yes, my face definitely changes over the course of IT the like, the kind of graphic lots away. But overall, a really super interesting way to use A I video. I wish that IT was faster, and I think it's it's really powerful. I do think it's more powerful than running ways tool of this same sort and I excited .

them more time with IT. Well, now having spent time with this um even a limited amount of time plus using runway and luma all of these things, sora is in fact delivered during shipment. And I say this knowing full well that I might be out right now as people are listen to this discussion because of the way time works. But if sorry, is just a text to video model. Will that be exciting enough now, Gavin.

I don't know. And my other thing is like how much generations will they give you? How long will that take? And there was a lot of room around A A turbo model that came when that artist thing happened.

And if that's really good and fast, maybe I could be interesting. And I don't know. I mean, it's at least another tool to play with.

They really will also depend on how many generations we yet as they paid ChatGPT user. Like will we be limited to like to? And if that's the case, like IT gets really annoying to try .

that stuff or or will they be announcing a new paid tier? By the way, twenty six gets you basic voice chat and text chat. But if you want advanced image generation and sora, well, that's a GPT plus premium mac and that's .

a fifty dollars a mother or whatever for IT, by the way, which I think if they dropped something like that, probably could get a lot of people to sign up for. IT is not the thing for a company. A lot of money.

agreed. I'm still using twelve different services. Peace mali, if they can give me just good enough across a multi modality. Yeah, you win. You still have my money.

Thank you so much for listening, and we wish you all a happy ship sector. Marry ship mess. Marry ship mess, everyone.