Meta’s AI Llama 3 Uncensored, Microsoft's Scary Deep Fakes and ChatGPT for Your DNA

AI4H EP054
===

[00:00:00] 

kevin_1_04-23-2024_153438: On today's AI for Humans, 

kevin_1_04-23-2024_153438: 3 is shockingly capable. 

kevin_1_04-23-2024_153438: We are going to explain everything and show you how you can run it for free and make it uncensored.

gavin_1_04-23-2024_153438: Ooh, 

gavin_1_04-23-2024_153438: New Microsoft tech is turning pictures into lifelike avatars just in time to interfere with everybody's favorite elections.

kevin_1_04-23-2024_153438: And you might trust AI to write your emails, but would you trust it to code your genetics? 

kevin_1_04-23-2024_153438: We're going to share how AI is coming for your DNA. 

gavin_1_04-23-2024_153438: plus a way to turn your home movies into playable game worlds.

kevin_1_04-23-2024_153438: All that and more. This is AI for Humans.

Mhm.

gavin_1_04-23-2024_153438: Welcome, welcome, welcome everybody. This is AI for humans. You're a wonderful, wild, wonderful and wild and wonderful and wild guide to the world of artificial intelligence and generative AI. We are here to demystify everything that you hear, kind of bring the hype down, sometimes hype it up a little bit more, and I am here with my friend, [00:01:00] Kevin Pereira, who is such a fan of AI.

gavin_1_04-23-2024_153438: He has sat down with me for another hour. Kevin, how are you today?

kevin_1_04-23-2024_153438: I am wonderful and I am wild. The rip and the Tarn. Yes. I am so excited to be chatting AI with you. Good buddy. , but before we dive into all of the news and some exciting demos and things that people missed, we have to beg and plead Gavin because we are still a guppy. in a sea of artificial intelligent whales.

kevin_1_04-23-2024_153438: And so we need your help out there. If you're listening to this, if you're watching this, please click a thumbs up, click a like, click a heart, subscribe, it'll cost you nothing. Leave us a five star review on whatever platform you consume this madness. It is literally the only way we can grow this and we appreciate it when the little line goes in the up and right direction.

kevin_1_04-23-2024_153438: Although maybe we should take a Tesla tip and just tank it because

gavin_1_04-23-2024_153438: that's a good idea.

kevin_1_04-23-2024_153438: to go up when they missed all of their marks.

gavin_1_04-23-2024_153438: Just remember everybody at home, we love you when you love us. And that is the only important thing to take away from the show. It doesn't care. Does the AI stuff doesn't matter. Just [00:02:00] that one thing love you when you love us. All right, Kevin, it is time. We better get into this week's big.

kevin_1_04-23-2024_153438: NEWS!

gavin_1_04-23-2024_153438: The big story of the week is Meta's Lama 3. So this is a big deal for Facebook and Meta, and especially for Mark Zuckerberg, who's been on quite a kind of resurrection tour, I'd say, over the last six months to a year and a half, two years.

gavin_1_04-23-2024_153438: Let's talk first about what Lama 3 is for all the people that are tuning in. Lama 3 is Meta's new large language model, and it is a couple of really important things to know about it. It is pretty cutting edge and , it actually is pretty good.

gavin_1_04-23-2024_153438: and the two other most important things about it are, is that it is free and it is open source. So we're going to dive into all those things here, but Kevin, before we jump in, I want to get your first take on kind of this overall news.

gavin_1_04-23-2024_153438: There was [00:03:00] a real barrage of, of Lama three stuff this week, which we're going to dive into, but. Where do you kind of feel like llama three is sitting for you right now?

kevin_1_04-23-2024_153438: Well, I know for most in the AI field, Llama3 is sitting on top of all of the leaderboards. It is just behind OpenAI's best model. And that is incredibly impressive considering how quickly Meta got Llama3 out there, how available it is, how free it is, how many services it can run on. And I know Gavin, there are a lot of tech experts that like to listen to this podcast, but we always aim for the sort of broader audience that is maybe a little.

kevin_1_04-23-2024_153438: , unfamiliar with some of these terms, I think a quick explainer of what an LLM is and I'm going to break this down very quick and very broad, but it's a foundational model. Think of it as the brain that is powering all of these artificially intelligent experiences. So what meta and others have done is go out and crawled tons of text and all sorts of data, , to create a training data set and they jam all of it.

kevin_1_04-23-2024_153438: All of that text. We're talking every book, every written, [00:04:00] probably podcast transcriptions, YouTube comments. Lord knows what is in their recipe. And they slam it all into a big old pot and they let it boil for a while and the computer, , finds the relationships between words. And that's how it can predict the next text.

kevin_1_04-23-2024_153438: Word in a sentence or the next line of code in a file. That is a very grossly basic definition of this, but you got to imagine that these companies are gathering as much data as they can to make these models. And this is Meta's latest. And so. Back to the original question, it is placing very, very well on all of the leaderboards.

kevin_1_04-23-2024_153438: And that is a series of benchmarks that they put these foundational models through. So they see if it can answer math questions. If it can figure out chemistry problems. Can it creatively roleplay? Can it solve logic puzzles? Write code? Can it do all of these different functions and how well does it do it?

kevin_1_04-23-2024_153438: And they have various benchmarks. There are a lot of them which compare all of these foundational models. And what is surprising, Gavin, is that just a few episodes back, we were [00:05:00] talking about Claude and their latest models, which you pay. Upwards of 20 a month for access to, and this seems to be outperforming their best model, their opus model, on some benchmarks.

kevin_1_04-23-2024_153438: And again, it is completely free for everyone. That is wild.

gavin_1_04-23-2024_153438: And it is going to be used across the entire meta ecosystem. So I think this also has the potential to be one of the AI products that touches the most amount of people, right? And not in a bad way, it's not going to touch you in some sort of bad way, but the idea is that you will probably interact with this in some way.

gavin_1_04-23-2024_153438: In fact, we're Right now, you can go to meta dot AI and you can interact with this model in real time and get a sense of the back and forth that you can have. And one thing, Kevin, I think you said there is the free part of it is something that I keep coming back to, which is a lot of these companies are charging for different models.

gavin_1_04-23-2024_153438: And yes, opening, I, we should be very clear opening eyes, GPT four, I think 4. [00:06:00] 5, whatever their most recent version is. Outperforms Lama three right now. Now, the other thing to know about Lama three, some of this news is a little bit older, but Lama three they stopped it from doing a full training run.

gavin_1_04-23-2024_153438: So this is like a partially trained Lama three. It could get much better. There will be a larger model coming out of Lama three with more tokens, which means that we'll have read more stuff. 

gavin_1_04-23-2024_153438: But the fact that it is free, the fact that you can access this from all these different meta platforms. That feels like the game changer to me. And he specifically, there was a very funny story in a bad way. It was a game changer that came out that Meta's AI actually inserted itself into a chat of moms in New York city and basically made up the fact that it had two children, which it didn't have.

gavin_1_04-23-2024_153438: And that one of them was a special needs child and was attending a particular school. And it does go to show you. These are the things that will happen as you get to scale with any AI model when you're going to be interacting with people at this much.

kevin_1_04-23-2024_153438: Zuck [00:07:00] flipped a switch, and now the AI is in your Instagram, it's In your Facebook. And of course it is in those private messenger groups. And one of the funny things that I saw from that mom's chat story, where it suddenly popped up was that, you know, it's by design that, that meta wants to stimulate conversations.

kevin_1_04-23-2024_153438: So if you're having an interaction, if you're in a group chat and , too many minutes have gone by, or somebody is asking a question and not getting an answer, the AI

gavin_1_04-23-2024_153438: to be like, Hey, what about this thing?

kevin_1_04-23-2024_153438: Oh my God, where'd you

gavin_1_04-23-2024_153438: Yeah.

kevin_1_04-23-2024_153438: Who are you? It's like a spider lurking in the room that you're unaware. 

kevin_1_04-23-2024_153438: It's actually, it's more like a face hugger because it gets right in there and it pounces. And I think that that is a not so pleasant reminder, Gavin, that with, when you're dealing with these big tech companies right now, anything you say or do or type can and will be used against you or to train their models.

kevin_1_04-23-2024_153438: Like literally

gavin_1_04-23-2024_153438: to say, that's exactly what it is. So this is the thing is that like one of the really fascinating things about opening this door to all of Meta's products is [00:08:00] that now you are interacting on a scale that no one else can compete with, right? Nobody. Like I think Meta's scale. Maybe Google search is as big as meta scale, but if you think about a billion active users, which I think is what meta has across its products between WhatsApp, Instagram, and Facebook, that is just a massive amount of interactions.

gavin_1_04-23-2024_153438: And it is going to train that kind of conversational AI really well. When it comes to knowing how you want to be talked to, when it comes to understanding small groups of people like a mom's group, I bet it does train on that stuff. And I bet it's going to fine tune a little bit based on what that information that's set across from it.

gavin_1_04-23-2024_153438: Because the other thing that Facebook has, which almost no other company in the world has is groups, right? I remember like five years ago, there was a real like moment Facebook groups was having where it was like, holy cow, , Facebook groups is like one of the biggest websites in the entire world if you were to separate it out.

gavin_1_04-23-2024_153438: Right. Right. And what you have in Facebook groups is a lot of people talking about a topic. And that's a really interesting opportunity to [00:09:00] fine tune AI is based on specific topics. I think that could be super valuable.

kevin_1_04-23-2024_153438: probably assuming some level of expertise if everybody is rallying around a specific topic, , for good or bad. , the same way Reddit is such a treasure trove of information because people that rally around a certain, like if it's a flying subreddit or an RVing subreddit, there's gonna be some very basic questions that get asked time and time again, and then there's gonna be people leaping in with answers and real expertise, and that's a goldmine for meta.

gavin_1_04-23-2024_153438: So one of the other thing I want to talk about here is the fact that they've open sourced to your LLM pointed kind of very quickly explaining LLM to open source something again means that basically you are developing a program that allows people to see the code with the idea being that it not only is can make the code better.

gavin_1_04-23-2024_153438: If multiple people can see it and improve on it, it actually might make it safer because you could actually also see places that people would be able to disrupt it, or it could hack into it in some form. And if lots of people are looking at it together, that it's a chance for everybody to [00:10:00] improve it. 

What it feels like is that there's an argument on Zuck's side and on the open source side where they're really are saying like these walled gardens are going to make AI worse, right?

gavin_1_04-23-2024_153438: And you have even, Mustafa Suleiman came out and gave a big Ted talk this week as the new CEO of AI at Ted. I don't know if you saw that, but I watched the whole thing. Yeah. And, you know, he's getting a little bit of crap for. Talking about how we have to watch out and control AI.

gavin_1_04-23-2024_153438: And I think Zuckerberg gets to be on the side of , Hey, we don't, we don't think this is going to get out of control, but we do think it's important that everybody has access and be able to see into it. I think in this instance, it's kind of good that we're, we're really pushing hard on the open source side because what it means to me is it just opens the door to so many people being able to create so many more things with this, 

kevin_1_04-23-2024_153438: so there's a couple points there that, , I will probably struggle to make. But one is that as, as much of a hero and a glow that sweet Zuck has around him right now, his open source aura is in fact shining bright. [00:11:00] He is on the record as saying don't know if LLAMA is going to remain open source forever.

kevin_1_04-23-2024_153438: And in fact, they might have a model, the next model, that is exponentially better. And they might decide that for reasons, They're not going to release it for free right now. It is beneficial to them. If an open source community can hack on this thing and optimize it by, let's say 10%, Gavin, that could be billions of dollars in savings for them running these models to power everything from messenger to groups to you name it.

kevin_1_04-23-2024_153438: Right? So they have an interest in letting the community poke against it and optimize it for free, and they may close the door. To that community in the future. They've certainly reserved the right to. So I think there's a little bit of a balancing act happening right there. But what is exciting is the fact that the community can poke against it.

kevin_1_04-23-2024_153438: And it's only been out. For a couple of days now, Gavin, I've seen fine tuned models that are completely uncensored. We're seeing some that are pruned so that they'll run on cell phones and on like cheap, chintzy Android devices. So the [00:12:00] strategy does work. It's a proven one. 

gavin_1_04-23-2024_153438: Too quick, more Zuck things before we move on, , another open thing was really fascinating. So Zach this week talked about the open metaverse. This was a followup on his, Lama three announcements. 

All right, some updates on the Metaverse today. Uh, we are releasing MetaHorizon OS, our operating system that powers Quest virtual and mixed reality headsets. .

Now, in every era of computing, uh, there are always open and closed models. And with phones, Apple's closed model basically won out. In the PC era, the open model won out. You can do a lot more things, install mods.

We've got more diversity of hardware, software, and more. Uh, so our goal It's to make it so the open model defines the next generation of computing, 

gavin_1_04-23-2024_153438: he came out and basically said that they're turning horizon virtual worlds into a platform, meaning that they want their VR platform and their AR platform to be open to all. Platforms, meaning instead of like apple, which you have a vision pro in your head and it's only going to run vision pro software.

gavin_1_04-23-2024_153438: Mark wants oculus and all their V. R. A. R. Stuff to [00:13:00] be run everything. So we talked about how he wants steam to be able to run on. He wants all these other O. S. Is to come on. This is a big change. And a lot of people were saying that this feels like it's going to be how you have in mobile. You have Apple as a closed system and Android as an open system.

gavin_1_04-23-2024_153438: In this instance, it seems like Mark wants the Oculus ecosystem to be like the Android to Apple's a headset system, Kevin, do you think that like open sourcing the Oculus ecosystem is a big deal for them? Or is it just a way to kind of like talk to the metaverse people and then get back to AI?

kevin_1_04-23-2024_153438: So this is an interesting point. It's not officially open sourced. They are opening the platform to official partners. So that means we might see a Microsoft branded something for Augmented Reality coming out for the next Xbox. 

gavin_1_04-23-2024_153438: But yeah, he even said that right in his video He talked about like the fact you might just buy one that comes with an Xbox

kevin_1_04-23-2024_153438: Right. And as he stated, this is going to take years for these devices to be announced and actually make it to the market. So some are saying this is a play for like [00:14:00] 2026, 2027 and beyond. But yes, if you know that Apple is going to be pushing into that market from their perspective, it makes a lot of sense to try to end around maybe Google slash Samsung and what they're doing with the Android side but John Carmack actually came out, , recently.

gavin_1_04-23-2024_153438: John Carmack so much. Are you just gonna kiss his butt again? Is this what we're gonna do like this? I swear like Every week we have Kevin talking about job. I like John Carmack, but I don't have to bring him up in every

kevin_1_04-23-2024_153438: Oh, I would love to. And you tell me what, what lipstick or gloss he would prefer before I

gavin_1_04-23-2024_153438: come on our show we have to we have to quiet Kevin up

kevin_1_04-23-2024_153438: I am such a fanboy and have been for a long while.

gavin_1_04-23-2024_153438: I love John

kevin_1_04-23-2024_153438: he's a pretty darn smart gentleman. And, you know, he has very insider knowledge of the inner workings at

gavin_1_04-23-2024_153438: Of course.

kevin_1_04-23-2024_153438: Oculus side.

kevin_1_04-23-2024_153438: And

gavin_1_04-23-2024_153438: Well, you should say because he was there in charge of their, , Oculus platform after Palmer Luckey left.

kevin_1_04-23-2024_153438: easy fanboy. Why don't you calm it down and put the tassels away?

gavin_1_04-23-2024_153438: I'm here for the audience. I'm here for the audience to give them context. I am not here as a fanboy. I'm here as the

kevin_1_04-23-2024_153438: Well, [00:15:00] good. Now they have context. Now let me fan boy. He said that he worries about this slowing down the actual teams behind the operating system, because instead of them just doubling down and building features and focusing on that, now they're gonna have to spin up teams and support, , different hardware devices, , multiple settings, , a wide range of options, as opposed to just building for what they want to deploy.

kevin_1_04-23-2024_153438: And he said like, well, that's fine. This sounds like a problem you could just throw a bunch of money at, and it will scale, but in his experience, that's not the case. So, Johnny C is worried, and that has me concerned, Gavin, because I'm a Johnny C number one fan.

gavin_1_04-23-2024_153438: want, this is my impression of, Oh, this tell me what's next. John C. Tell me, I want to understand what's going on in the future. That's my,

kevin_1_04-23-2024_153438: someone? Is that, is that what your impression is?

gavin_1_04-23-2024_153438: no.

kevin_1_04-23-2024_153438: who's got their fingers on the pulse of the industry and is going right to the source for a balanced take instead of just sucking up the Zuck soup when he, , delivers it via the tube?

gavin_1_04-23-2024_153438: The Zuck soup is probably what they serve at the Facebook cafe.

kevin_1_04-23-2024_153438: it is

gavin_1_04-23-2024_153438: like what, what if [00:16:00] they should have a weekly Zuck meal every, every cause like he's still, he's not like Bill Gates where he's retired. Like Zucks active there. If I were in charge of meta, I would say once a week you got to make the Gavin meal.

gavin_1_04-23-2024_153438: You got to make whatever the meal is that has my name in it. And it can't be something you

kevin_1_04-23-2024_153438: Alright folks, time to Zuck it up! Whatever is on your hot plate, and it's just

gavin_1_04-23-2024_153438: It's Zuck toast and mush. It's a fun time. You don't know what it is, but yeah, it actually, Oh, that's the clearly the best one. I mean, that's, that's what makes the most sense. Zucchini. Uh, what is, what do you put zucchini in? Actually

kevin_1_04-23-2024_153438: Ah, zucchini's a fake vegetable, everybody knows it. It's mostly water, it's not real.

gavin_1_04-23-2024_153438: Is that real? Is that a fake vegetable? One more thing there was a really cool update on Meta's Ray Ban glasses. So if you're familiar with these, these were the glasses that came out that have a camera in them. But there was also a promise that they were going to be able to interact with Meta's AI and tell you about the world around you and guess what?

gavin_1_04-23-2024_153438: Everybody, it is finally coming the guy who's in charge of [00:17:00] product at Facebook came out and made That said, , we are releasing this in America and Canada and you will be able to walk up to something in the video. He shows off walking up to a monument.

gavin_1_04-23-2024_153438: I think in Paris or might be in New York City. I can't which one it is and it tells the person what they're looking at. And this is the real world promise of AI and also it's kind of the real world promise of AR and VR, but not being tethered to a giant ski goggle headset. You're actually just wearing sunglasses that really don't look that different from normal sunglasses.

gavin_1_04-23-2024_153438: I think Kevin, I'm actually going to ask for one of these for my birthday, just cause I, my birthday's coming up. But I think I might actually use this product. It feels like a product that is. Finally, something I could wear that doesn't feel it'll almost be like my Apple watch, right? Like I'll just grab this pair of sunglasses when I go and I'll remember.

gavin_1_04-23-2024_153438: Oh, I have these additional things I can do with it 

kevin_1_04-23-2024_153438: It's a frictionless form factor, right? Like, you're not going well out of your way to establish a new habit of putting on a [00:18:00] lapel in the morning or carrying around a Game Boy sized brick in your pocket. Again, no shade to those hardware companies, but that is a hurdle that they know they have to leap here.

kevin_1_04-23-2024_153438: Putting it on your glasses, , the biggest concern I have is leaving them behind. At an AMC, because I watched Dune 2 a little too tipsy again. That's my biggest concern. 

kevin_1_04-23-2024_153438: And I got to say, everybody will tell you never purchase any device on the promise of future functionality. Cause sometimes it just never gets there. Kudos to the Meta slash Ray Ban team for making this happen, because it happened on a pretty quick timeline and it looks like it's rolling out even faster than expected to, to multiple countries.

gavin_1_04-23-2024_153438: I totally agree. I can't wait for it. One of the funny things we just said that made me think of you talking about the AI pin is like, you know, Maybe brooches have to come back in order for that to be something that people would want like they're broke You know some people still wear brooches I mean my mom's first brooches, but like I don't think a variety of people are normally a giant metal

kevin_1_04-23-2024_153438: You need Rogan on board and it has to be a Brochez, like capital

gavin_1_04-23-2024_153438: Oh [00:19:00] Sheesh,

kevin_1_04-23-2024_153438: Something that, that, that actually trains your back at the same time. Like a five pound kettlebell broach? Yeah. And then to activate it, you slap the chest.

kevin_1_04-23-2024_153438: You can't

gavin_1_04-23-2024_153438: if it had a chain between the brooch and your wallet so that it was like that kind of thing

kevin_1_04-23-2024_153438: Or your mani pack, right? Your mani pack is a huge lithium ion battery that powers your brooch.

gavin_1_04-23-2024_153438: I'm into that the brooch.

kevin_1_04-23-2024_153438: Can we, on the topic of brooches, and we have to actually brooch this subject, Gavin. The zuck beard? Uh, hello? Hold on. Squeaky, squeaky! I gotta get the condensation off of my, , Ray Ban

gavin_1_04-23-2024_153438: You're a little you're a little you're a little

kevin_1_04-23-2024_153438: The whole internet was a little steamy. The whole internet kind of fell for a deepfake of Mark Zuckerberg with the gold chain dangling, which was real, in his initial Llama 3 announcement Instagram video. But they put a beard on the Zuck, and the crowd went wild.

gavin_1_04-23-2024_153438: , this is, this all goes back to the Zuckasson. So whatever you want to call it, like the, the exciting moment , of Mark [00:20:00] Zuckerberg, because I always think of the lowest moment he had, , I was lucky enough to know Kara Swisher a bit, and she is like an incredible tech journalist, but the lowest moment I always thought she has was when Kara was grilling him on stage at one of those tech conferences that they did, and he was just pouring in sweat.

gavin_1_04-23-2024_153438: There was another moment where he was in front of Congress, which was pretty bad, but on

kevin_1_04-23-2024_153438: to figure out how to sip water? Yes.

gavin_1_04-23-2024_153438: that was pretty bad. But on stage at this tech conference, he was just pouring and sweat. And it was like, that was a kid, like a literal kid who suddenly became the head of a business. And it feels like Zach is finally, you know, he's done.

gavin_1_04-23-2024_153438: He's doing his MMA stuff. His family's feeling pretty good. He's kind of feeling himself now. Like he bet all this money on the metaverse and people aren't worried about anymore because the stock's doing well. I think this gold chain Zuck, like I think he could rock that beard.

gavin_1_04-23-2024_153438: So just to be clear, the beard was a fake, right? There was no real beard, but it made him look very suave. And you know, the guy looks semi grown up now for a kid, somebody who we always thought of as a kid, [00:21:00] cause he was in his twenties when he got famous early twenties. Now he's an adult who is managing a company that is actually doing very well.

kevin_1_04-23-2024_153438: I said, why stop at a beard, Gavin? And I just wanted

gavin_1_04-23-2024_153438: Oh no.

kevin_1_04-23-2024_153438: heads up. If you Open up Exhibit A, a Dropbox folder called Zuckstuff. You'll see that your boy had a little bit of fun, so, if you want to take your own approach to styling Mark Zuckerberg like I did, I want to recommend an app that we talked about on last week's episode very briefly.

kevin_1_04-23-2024_153438: It's called COSXL. That's C O S X L. XL, and it basically, it's Stable Diffusion, which is an image generation software, and you feed it a source image, and you just use plain English to describe what you want done to the image. And Gavin is reacting to some of that

gavin_1_04-23-2024_153438: Wow.

kevin_1_04-23-2024_153438: I gave Zuck a Mario stash, I put him in some, uh, insane clown makeup, I said, what would he look like with some sweet prison ink all over his face?

gavin_1_04-23-2024_153438: that part is really crazy. The prison ink is really crazy. And then the one that really got me was like, what looks like, [00:22:00] it looks like he is the like alien brother of Prince William. Like what, where does that one? It's like something weird is going on

kevin_1_04-23-2024_153438: so I kind of hit the limits of costs XL. It does a good job for basic editing, but I loaded a bunch of photos of Mark Zuckerberg into another app called photo maker, which is, again, these are all free applications. Shout out to Pinocchio dot computer. Hashtag not an ad, but if you go and download Pinocchio dot computer, you can run all this stuff for free, nice and easy on your windows machine or your Mac.

kevin_1_04-23-2024_153438: That is an app called photo maker, where you give it several source images of someone, and then you ask it what you want. And when I played with the settings, I got this weird, almost. Felt like mashup of a Prince and Zuck, and then I said, you know what?

kevin_1_04-23-2024_153438: Let's just make some thirst trap MMA photos of Mark Zuckerberg and gave him the Rogan's tattoo pattern. , a nice lobster red body. And then. , I decided let's mix it up because the beard was fun to watch in motion. I figured let's go ahead , and slap one of these odd faces on the video.

kevin_1_04-23-2024_153438: So I took [00:23:00] the, , insane clown makeup photo, Gavin, the one where he's got bright red clown hair, and I used an app called face fusion, which lets you load a source video and then swap the face.

kevin_1_04-23-2024_153438: And I put. The bizarre clown makeup photo as the face to swap and face fusion went its own direction, got rid of the actual colorful makeup and just turned it turned Mark into this bizarro alt dimension Zuck 

kevin_1_04-23-2024_153438: that is um, all right, big day here. We are releasing the new version of meta AI, our assistant that you can ask any question across our apps and glasses.

kevin_1_04-23-2024_153438: I don't, I don't like it. It is a very rip curl meets, , let's say a chemistry lab in the middle

gavin_1_04-23-2024_153438: Yeah, I was going to say, it looks like John McAfee after he, after he took off. That's what it

kevin_1_04-23-2024_153438: It's meth Zuckerberg. I was

gavin_1_04-23-2024_153438: Yeah,

kevin_1_04-23-2024_153438: to say it, but that's what it is.

gavin_1_04-23-2024_153438: right.

kevin_1_04-23-2024_153438: fun with it. Kudos to Meta and Zuckerberg for releasing Llama 3. And again, later in the show, we will walk you through step by step how to run this stuff uncensored. And we are going to [00:24:00] use it to power our AI co host today.

kevin_1_04-23-2024_153438: But I think that's a, A nice little roundup of all things Llama, . 

kevin_1_04-23-2024_153438: But Gavin, let's talk about the 180 pound elephant in the room.

gavin_1_04-23-2024_153438: Well, here's the thing all week long online, people were saying GPT five is coming or open AI is about to clap back. And we have been expecting this for a while. It is something that, open AI has been able to do in the past. They have been able to like, see a big news story. And then when Gemini came out, they're just like, bam, here's Sora, 

gavin_1_04-23-2024_153438: there are a lot of people in the AI world right now that are expecting some new model from open AI to come soon. And, Kevin, I think I just want to touch on this really quickly because it's hard to remember this, but it has been now over a full year since GPT four came out and we, we talked about the benchmarks at Lama three, GPT four, maybe not vanilla GPT four, but GPT four is still the reigning champion in terms of these large language models when it comes to.

gavin_1_04-23-2024_153438: Reasoning when it comes to all of this stuff to using the basic use [00:25:00] case of a large language model, and yes, Claude three has gotten close, but GPT four still is the winner, and I want to ask you, Do you think this is real? Are we going to get a GPT five model soon? Or do you think we're just in a holding pattern right now with open AI?

kevin_1_04-23-2024_153438: I think they still haven't defeated the champion. If this was an MMA analogy, it's like he's still in the Sam Altman is still parading around the ring doing Conor McGregor arms. He still has the A. I. belt on. There's no need for him to get off the throne and release something right now.

kevin_1_04-23-2024_153438: They can keep tuning and fine tuning. What's wild is that People tend to behave as if OpenAI has done nothing. That they,

gavin_1_04-23-2024_153438: Yeah, that's the thing that's so weird to

kevin_1_04-23-2024_153438: last year and they've just been sitting on their hands. But they have been updating and iterating. They've shipped the chat GPT system. They've made a massive update to Agents.

kevin_1_04-23-2024_153438: They've super updated their API. They released, , the text to speech voices. Like they've done a lot. They've updated Dolly, which is their image generation software. And I just think , even if.

kevin_1_04-23-2024_153438: [00:26:00] You have GPT 5 in your mani pack, which is connected to your brooch, as we've established. You don't have to unzip it. You don't have to take it out just yet. Let's wait for somebody to beat GPT 4 before they unveil it.

gavin_1_04-23-2024_153438: What if GPT 5 comes out and it's only three times better than say, GPT 4? Do you think that the whole world will start kind of being like, it's over. The AI world is over. We're not going to see

kevin_1_04-23-2024_153438: burst! It was a fad! Yeah, that's a great question. I do think that there's this sentiment that because OpenAI had such a head start, that if GPT 5 isn't exponentially better, then that just proves that this AI stuff is either a fad, or it's going to take 10 years instead of 7 years or 5 years.

kevin_1_04-23-2024_153438: But in reality, what you just said is so funny. If it's as much of a leap as GPT 3 was to GPT 4,

gavin_1_04-23-2024_153438: If it's that big of a leap, that's yeah.

kevin_1_04-23-2024_153438: better, but that's massive! I got to imagine, Gav, that it's going to be [00:27:00] better than that. When Sam Altman is talking about steamrolling people, if they're underestimating them, his words, they will be steamrolled. 

if you're building something on open on gpt4 That a reasonable observer would say if gpt5 is as much better As gpt4 over gpt3 was not because we don't like you but just because we like have a mission We're gonna steamroll you

gavin_1_04-23-2024_153438: Would be that thing? That I would see that wouldn't be like, Oh crap. And to be honest with you, that thing that I would have said that would have been that before was Sora.

gavin_1_04-23-2024_153438: So like they proved that they can do something that is way better than everybody else. And if you just look at what Sora was compared to say the most cutting edge runway ML or Pico labs. And again, we like those companies, but it was like. Light years better than those things 

gavin_1_04-23-2024_153438: I hope it doesn't come tomorrow so that we have to record an emergency podcast for our podcast, but I think it's going to come in the next couple of weeks is my

kevin_1_04-23-2024_153438: Yeah,

gavin_1_04-23-2024_153438: think that's a big deal. And yeah,

kevin_1_04-23-2024_153438: both techno optimists and we want it just on our schedule when it's [00:28:00] convenient for us to record a podcast. That's fair.

gavin_1_04-23-2024_153438: There was a big, huge story that came out though, Kevin, that it does involve, um, something that is groundbreaking that I think, honestly, it's one of these stories that like is so big and didn't get the mainstream coverage it deserved. , so there was a, a company that published a paper, but then also simultaneously had like a press barrage 

gavin_1_04-23-2024_153438: it was kind of like a company's coming out party, which essentially is a CRISPR, which if you don't know, CRISPR is gene editing, but using AI tools. So the way they're describing this is almost like a chat GPT for CRISPR, which I know that sounds like a very complicated explanation. experience. But just if you're not familiar with CRISPR, it is a very innovative tool that came out a while ago, but has been slowed a little bit because it's ethically a little questionable.

gavin_1_04-23-2024_153438: It basically allows you to go in and edit genetic sequences and sometimes change those genetic sequences to do things like. improve health or does sometimes genetically engineer certain things, whether it's crops and, you know, morally people could [00:29:00] conceivably genetically engineered. So what this is, is a company that is focused on using AI to create Transcript , and edit genetic sequences with AI.

gavin_1_04-23-2024_153438: So if you remember a couple months ago, we covered a story where deep mind Google deep mind was able to help discover all these new materials that didn't exist, or at least pursue the idea of materials that didn't exist. That might be great things to kind of build, build bridges with, or to do all sorts of stuff.

gavin_1_04-23-2024_153438: Those were like non living materials. Those elements of the universe, essentially. This is now giving, through AI and all the amount of data that AI can crunch, the ability to start looking at biological gene editing and learning about the sorts of gene editing tools that wouldn't necessarily, , appear naturally in nature at large.

gavin_1_04-23-2024_153438: This is doing stuff that is outside of nature, which when you say that in gene editing in the same sentence, I don't love hearing those two things together. I don't know, Kevin, what do you think about all

kevin_1_04-23-2024_153438: It sounds a lot more, concerning [00:30:00] than, , making Roundup ready corn, but maybe it's not. Being able to delete any number of abnormalities that, humans have to suffer on a day to day basis is something that I think everybody would say, yes, sign us up 

kevin_1_04-23-2024_153438: , look, we've talked about on this show time and time again, that we are cautiously optimistic, but we're still optimistic right across every industry, and this is just one of those stories of promise that is now possible because of the rapid pace of this advancing technology. If it is fixing genetic defects and helping us live longer, happier, healthier lives. Great. I really hope AI can get us there. And I really hope human beings don't squander it on the path to that promise.

gavin_1_04-23-2024_153438: Kevin, what if it's, , dog fighting in multimillion dollar jet fighters that have missiles with AI?

kevin_1_04-23-2024_153438: I see no problem with anything that you just said, Gavin. There, we found the one use case of AI that is sure not to raise any eyebrows. It is AI's flying jets that were designed for human beings in this case. [00:31:00] To see if they can outmaneuver them in a dogfight and that is something that actually happened it was a challenge that DARPA took on and Spoiler, we'll get into the details five to nil in favor of the artificial intelligence Surprise surprise Gavin.

gavin_1_04-23-2024_153438: So DARPA, if you're not familiar is it's a part of the department of defense and they are really responsible for a lot of big innovations. In fact, they are one of the first, , the DARPA challenge really started the driverless car race, 

gavin_1_04-23-2024_153438: and. This is a next version of that. So basically what they did is they retrofitted an actual fighter jet with an AI system that allowed it to learn from human pilots and how human pilots, do dog fights, which is like a simulated battle. But in the real world, human pilots go up and pretend to shoot each other so they can train for real battles.

gavin_1_04-23-2024_153438: And it used all that data and it trained itself and it drove itself. And I think there were people in the cockpit, just like there would be a driverless [00:32:00] car, but it did, it won against a human thing. Now they always, we always talk about the idea that like, Oh, of course they can win at chess, right?

gavin_1_04-23-2024_153438: Now like you get a chess and nobody wants to watch computers play chess against each other. But Kevin, this is a different thing because if it involves human life, What happens in the future? I guess maybe it's going to be interesting, right? Maybe there'll be a world in the future where there's like an empty space in the world.

gavin_1_04-23-2024_153438: And we, as our country, and we don't like your country, we go and say, okay, our jet pilots are going to go fight over the empty space and just dogfight with each other. And then whoever wins, we like shake hands and give the other person like a prize. Like, I don't know what war

kevin_1_04-23-2024_153438: are you describing like war as eSports? What are you?

kevin_1_04-23-2024_153438: Yeah, basically, basically I'm cause I'm so confused as to like, Okay, actually, this is really interesting.

gavin_1_04-23-2024_153438: So in Israel, Iran, one of the big things was that Israel was able to shoot down a vast majority of the.

gavin_1_04-23-2024_153438: Incoming drones and missiles that were pointed at it because [00:33:00] of their technology capabilities. And even just hearing the word drone so often in that conversation is fascinating. What is so interesting to me about this is so we say we get an AI dogfighter that does good or better than a human already is better than a human, but we'll be okay with sending up a plane with nobody in it. Then that nobody in it plane fights against a drone or another plane with nobody in it. Ultimately, what the hell are we doing here? Right? Like we're just sending robots to fight each other. And then is it just a matter of who's got the most amount of money to build these

kevin_1_04-23-2024_153438: yeah. I mean, that's, that's what it's going to come down to, right? It's going to be robot versus robot until one side runs out of robots. And then it's going to be robots versus humans, which as we've seen play out

gavin_1_04-23-2024_153438: It's not good.

kevin_1_04-23-2024_153438: is not good for the human side of that. And then, it's a very uncomfortable and disturbing discussion, but then it becomes robots versus civilians.

kevin_1_04-23-2024_153438: You know, and, and that's, that's where this ends up. So,

gavin_1_04-23-2024_153438: That's where we get to the Terminator too, by the way, which is the, it sounds insane. But ultimately that is the, that is the, we just, I

kevin_1_04-23-2024_153438: [00:34:00] these are robots fighting on behalf of humanity, theoretically, but

gavin_1_04-23-2024_153438: you train those robots to be that kind of robot. There was a, there's a video , we put on our YouTube page today and on our Tik TOK of a robot dog that had a blow torch attached to its back.

gavin_1_04-23-2024_153438: And I was like, Oh, this is really interesting. Maybe this will be cool. And what you realize is no, it's just a fricking robot dog that can use a blow torch on people or the forest or whatever it is, ultimately we're building these of destruction. And I think we do have to think , okay, sure, great. It can dogfight another human.

gavin_1_04-23-2024_153438: It could be the human. What happens when the robots fight each other? And then to your point, yes, when the robots fight civilians is a very scary thing, but why are we building these things? What, what is the purpose? In some ways, the thing I always think about , with war, and I obviously nobody wants war, but like part of war is the cost of human life that it takes to fight it.

gavin_1_04-23-2024_153438: And if you're willing to do that, then you sure believe in it a lot. Yeah. 

kevin_1_04-23-2024_153438: How cavalier of an attitude might a general have now if they're like, Oh, we just deploy a [00:35:00] thousand of the Robits and who cares? They just see them as units on a spreadsheet and they're not actually thinking about the human cost so yes, I think.

kevin_1_04-23-2024_153438: When it's just throw robots at it, , even if the robots are being thrown at another side's human beings, well that's okay, that's them, that's not us, so this might lead to a more flippant Cavalier attitude towards destruction of people and property. So that is, super upsetting, of course, this happened to be a simulator aircraft, which is pretty interesting aircraft.

kevin_1_04-23-2024_153438: It was an X

gavin_1_04-23-2024_153438: Oh, super

kevin_1_04-23-2024_153438: they call it a Vista. That stands for variable in flight simulator test aircraft. And the design of this aircraft is that you can sort of emulate other planes. Gavin, so if you want to, , be able to fly it like an F 16, or an F some other number, uh, don't come after me, military fanboys.

kevin_1_04-23-2024_153438: That's right, if you want to fly it like the FU 69 420, You flip the switch, and it acts like it. So it is like a simulator plane. They plied it with the same rules, of engagement that the human beings would be operating at, and they found that there were some differences, obviously, in the way that it flew [00:36:00] in a simulator versus the way it flew in the real world, but not enough that it didn't completely squash that human opponent.

kevin_1_04-23-2024_153438: Right? Like it's, it was good enough. But that's in an aircraft that was designed for a human being to be seated in. Which is something we may want to retrofit our entire military fleet with because the AI performs better. Can outperform the humans. Okay. But what happens when, starting from now, These things are built without a human operator in mind.

kevin_1_04-23-2024_153438: How much cheaper do they become? How much more advanced can they maneuver? How much deadlier do the payloads get? , these are things that are going to ramp up very quick as we let the AI design its own aircraft, for example, and fly it. And to your point, why do we do this? I guess, water or lithium?

kevin_1_04-23-2024_153438: Depends on

gavin_1_04-23-2024_153438: Resources, power, all that stuff.

kevin_1_04-23-2024_153438: but I really hope you get your way and we can set up like a. You know, a few thousand foot by few thousand foot empty space

gavin_1_04-23-2024_153438: It's like battle bots for the world, basically

kevin_1_04-23-2024_153438: bros, send them in, 

gavin_1_04-23-2024_153438: speaking of big, weird competitions, Kevin, this week is the [00:37:00] NFL draft. In fact, if you're listening to this podcast right now on Thursday, it is today. It is tonight. So the big number one pick is going to go as Caleb Williams.

gavin_1_04-23-2024_153438: But we have a guest who's on our show today, who is actually a, come on to sponsor our show. This is our AI cohost of the day. And I think he might've thought he was getting a NFL draft podcast. I think he kind of thought he

kevin_1_04-23-2024_153438: Whoa, wait, this is multiple layers of setup lasagna. So we've got a guest. Powered by AI, that also happens to be a sponsor of the show. However, they thought they were coming on a podcast about the NFL draft. So you lied to this vendor who is sponsoring our show.

gavin_1_04-23-2024_153438: I didn't want to say I lied, but yes, I lied. I lied to this vendor.

gavin_1_04-23-2024_153438: here's the thing, Kevin, you'll really like this. We have the CEO of Monster Milk on the show. So this is one more layer of set up spaghetti. If you're familiar with our show, you may know about monster milk, monster milk is a energy drink [00:38:00] that, , we've had a couple of people in the show talk about it. Turns out when AI's drink it, it makes them a little bit loopy. We actually got the president. I I'm sorry. I did lie about this to him.

gavin_1_04-23-2024_153438: So we can get into that if we want

kevin_1_04-23-2024_153438: Is it, wait, is this all because one person left us a five star review last week and said, What happened to Monster Milk? This is

gavin_1_04-23-2024_153438: it is. Yes, it is.

kevin_1_04-23-2024_153438: out of our way for this bit? Okay,

gavin_1_04-23-2024_153438: it is. Let's meet the president of monster milk. Let's bring him

kevin_1_04-23-2024_153438: Alright, , dearest guest, thank you for sponsoring AI for Humans, , please say hello to us and the audience and introduce yourself.

kevin_1_04-23-2024_153438: Well, well, if it isn't Kevin, the guy living in a trailer and pretending to be something he's not. And Gavin, the other tech hack trying to make waves. What?! I'm Merle McDowell, grizzled old founder and CEO of Monster Milk. 

kevin_1_04-23-2024_153438: Unlike you two, I didn't get into this gig because I wanted to be on camera. Okay. Now, sir, I built an empire from a spilled experiment in my garage. [00:39:00] And now, uh, these loopy A. I. s can't get enough. What a world, eh? Yeah. Now, if you'll excuse me, I, I've got a company to run and it doesn't involve sitting around in front of a camera.

kevin_1_04-23-2024_153438: Okay. Hold on, Merl. Hold on a

gavin_1_04-23-2024_153438: Wait a second, Merle.

kevin_1_04-23-2024_153438: to unpack there, man. This bit just got even more confusing. He's a sponsor. Monster Milk was a spilled experiment in a garage.

gavin_1_04-23-2024_153438: we never knew the origin of monster milk and now we do,

kevin_1_04-23-2024_153438: both,

gavin_1_04-23-2024_153438: Yeah, from the get go, yeah This is gonna be a hard one to pull out, but we can do it. Let's get into this. I want to hear more about the origin story of monster milk. Cause I want to know kind of where it came from, like how he kind of took it from a experiment in his garage to where it is today, , 

kevin_1_04-23-2024_153438: we should mention that we have no idea what the AI is going to say, and today we are using Llama 3 uncensored to power this experience. So, there I was, just another schmuck, staring down the [00:40:00] barrel of a life filled with nothing but disappointment and regret. But then I had this epiphany. Why not create something that could make people's lives better? 

kevin_1_04-23-2024_153438: So I started brewin per concoction in my god damn garage. And wouldn't you fuckin know it, one day I ended up makin something so mind bendingly fantastic, even robots couldn't resist it. I mean, sure, it was an accident, but who the fuck cares about accidents when they can lead to something great? Sure. So I decided to market this magical elixir of mine.

kevin_1_04-23-2024_153438: And that's how Monster Milk was born. From a spilled experiment in my shithole garage. If you're still not satisfied with my story, why don't you just suck on this fuckin lemon? I'm done here. Whoa! Merle! Okay.[00:41:00] 

gavin_1_04-23-2024_153438: What the hell happened to

kevin_1_04-23-2024_153438: keeps, every time he gives an answer, he gets up, he grabs his, cowboy hat, which kind of disintegrates in his hand, and he heads to the door.

gavin_1_04-23-2024_153438: So a couple of things we're gonna do with Merle. So just FYI, Merle was created with, , Llama 3. we created a prompt and we are using Llama 3 to power him. It's an uncensored model. That's why you heard , Merle swear. Later on the show, we're going to bring Merle back 

gavin_1_04-23-2024_153438: We are going to start a robot draft. It is the NFL draft. We are going to have Merle and LLM, help us draft , a team of robots that could play football. Mostly we are doing that for fun, but then also we want to talk a little bit more in specific about how you can use llama three, what the ways to set it up are, because it's not hard and it does give you a chance to see what other LLMs are doing besides the things like cloud three or GPT four.

kevin_1_04-23-2024_153438: Well, that was, , an interesting howdy to our guest and sponsor today, and maybe we'll learn a little bit more about Monster Milk. Maybe we can also give him some Monster

gavin_1_04-23-2024_153438: Yeah. Should we just give him one right now? Let's let's give him one right [00:42:00] now before we go away. Have him take it.

kevin_1_04-23-2024_153438: and, uh, he introduces our next segment, Gavin?

gavin_1_04-23-2024_153438: that's a great idea, Kevin. That's a great idea 

kevin_1_04-23-2024_153438: all right. All right. You asked for it. So here I go. Taking a massive gulp of my goddamn monster milk, ready to face whatever insane shit this A. I. has cooked up for me. Oh, wahoo, whoop, whoop, whoop. Now, without further ado, let's dive into the next segment of our podcast, shall we? It's called A. I. See What You Did There.

kevin_1_04-23-2024_153438: And if it's anything like the last one, it's I'm probably gonna need another six pack of monster milks just to make it through. Holy fucking whoop whoop! Okay, thanks Merle.

gavin_1_04-23-2024_153438: gosh Merle great 

[00:43:00] 

gavin_1_04-23-2024_153438: It's time for AI. I see what you did there. This is just a quick check in on some of the things that we've seen online that are not tools that you can go use, but are going to be tools that you're going to be able to play with in the future, or just really cool things we saw.

gavin_1_04-23-2024_153438: , I'll start off here, Kevin. There's a very cool new GitHub page that came up called video to game. And what this basically was , our friend, , Dreaming Tulpa, uh, posted this, who we always love to see what kind of things they find, is that basically it can convert real world environments into interactive games.

gavin_1_04-23-2024_153438: And in fact, if you watch the video on the YouTube right now, you'll see a character that is running around kicking soccer balls, but he is doing it in a real world environment that you can conceivably capture yourself and play a game in the real world. , just a really cool way to show like how nerfs, which are things that allow you to scan real world objects, are going to find their way into interactive products. 

kevin_1_04-23-2024_153438: And when you say nerfs, that is a [00:44:00] neural radiance field, and that might have people going, What is that? And is it gonna hurt? And can you reattach my limbs if I walk through one? But it, it's a fancy way of saying that it crunches the data that you can grab from your cell phone and interprets geometry and depth and retains really high quality textures.

kevin_1_04-23-2024_153438: And you can actually go play the demo and On the website, it runs in your browser. So you can run around this little backyard garden scene and go throw balls at vases, and there's, there's even a demo where you can drive a car, um, through another level as well.

kevin_1_04-23-2024_153438: And again, they're releasing the code for this one, which is really exciting. So there is a near future Gavin and an augmented reality one, where if you're in an environment, you go, Oh, this would be cool to be able to run around here or play counter strike in this area. Yeah. You might be able to just record it with your cell phone.

kevin_1_04-23-2024_153438: And now you're dropping into a game, maybe even on your phone.

gavin_1_04-23-2024_153438: Well, what I like about this is like, you could even say like, Oh, this is a cool area for a game that I could just invent in real time, right? Like I'm going to drop a character here. I'm going to drop a character here and I'm going to try to [00:45:00] do this thing. I just am super psyched for that. So that was my, I see what you did there. What about you?

kevin_1_04-23-2024_153438: The one thing that I saw that really freaked me out, Gav, , was Microsoft's Vasa1, V A S A one. And I'm going to say. First of all, the paper is lifelike audio driven talking faces generated in real time. And,

gavin_1_04-23-2024_153438: And you might have seen there's a Mona Lisa clip going around the online that you might have seen a Mona Lisa like doing this Kind of like dancing and laughing thing that is from this page Yes,

kevin_1_04-23-2024_153438: from this paper, and there are so many examples on this paper. We're gonna put some of them on the screen, maybe we'll drop one right here. I hope we did. Maybe we didn't. But there are so many that are worth going through because this is a technology that I so badly want to get [00:46:00] my hands on, but I am so glad is not out yet, Gavin, especially as we said at the top of the show, , against the backdrop of an election cycle.

kevin_1_04-23-2024_153438: , This technology promises to take a single still photo of someone and turn them into a lifelike, realistic looking, talking avatar, just by matching, , the eye movement, the head movement, the jaw, the furrowing of the brow, even hair, etc., to an audio file. And we've seen technologies like this, we use technologies like this all the

gavin_1_04-23-2024_153438: All the time. And they're not good. They're not that great.

kevin_1_04-23-2024_153438: They're not good. They're usually a precondition pre pattern movement of the head. Things like headgear or headphones melt into the background. It feels robotic, even though it is quote unquote, magical. This paper. Has, , examples of things where you can control the gaze of the person. Do you want them looking away?

kevin_1_04-23-2024_153438: Do you want them looking at the camera? You can control how far away the camera is from the face dynamically. You can modify the expression of the person. So if you want them looking sad or [00:47:00] angry or delighted when they're saying whatever it is they're not actually saying, you You can do that. And, to the Mona Lisa example, they have examples of, quote, out of distribution generalization, which is fancy talk for saying, we designed this to work with human faces, but look, it works with the Mona Lisa, and it works with these sketches.

kevin_1_04-23-2024_153438: Conceivably, this could power, uh, AI art within a video game. If you need a character talking, it is really, really good looking. And partially scary. , and it also speaks to the stuff that you and I were saying a year ago, which is, is this is the worst that technology is going to be surprise, surprise.

kevin_1_04-23-2024_153438: It's better now.

gavin_1_04-23-2024_153438: The woman at the top in the middle of the top on the page with the blonde woman, that was the first clip I saw. So if you go into the realism and liveliness section, the blonde woman in the middle, I first saw this video. And as I go, it's another AI thing. And then I watched it.

gavin_1_04-23-2024_153438: And what's so interesting is We've talked about, Hey Jen, we've talked about DID. These are all things that attempt to do this, right? To make people come to life out of a photo. It's now getting [00:48:00] to the place where we're starting to get away from that uncanny valley thing. It starts to feel a lot more like a real person.

gavin_1_04-23-2024_153438: And the uncanny valley was always the closer you get to a real person, the weirder it's going to feel. I'm not sure if that's the case here because that woman's tics like there's, it's not just normal human tics. It's not like it's a thing that an old person does like me, I'm holding my hands right now, but like.

gavin_1_04-23-2024_153438: The way she kind of tilts her head and the way she opens her eyes, the way she squints her eyes. If you had enough of those things so that it didn't feel like it was ever repeating, like right now, I think you can see it repeating a couple of times. That's going to be indistinguishable in some ways from somebody talking like that is where we're at.

gavin_1_04-23-2024_153438: The only difference, I think the thing it's still struggles with a little bit is how do you take the tone of something and then use the tone of that voice to inflect on the face that you're seeing? 

gavin_1_04-23-2024_153438: That feels like once we get that, you're good. It's like all bets are off basically

kevin_1_04-23-2024_153438: Harder part arguably is done where it can modify the face to match the emotion that you're telling it [00:49:00] to. So now it's just a matter of plugging in something like that Hume project that we showed off. It's just plugging in recognition of the expression from the audio file, so you don't have to manually do it.

kevin_1_04-23-2024_153438: But the other wild thing is, , I said real time. When you scroll down to that portion of the paper, it mentions that it can do five 12 by five 12. That's the resolution. So 512 pixels by 512 pixels. That's a small square, but still good enough for tiktoks or chatbots On a website, it does it at 45 frames a second, which is pretty fluid.

kevin_1_04-23-2024_153438: A latency of 170 milliseconds. Very, very, very short latency. So if you're getting a real time response from an avatar being powered by this technology, gone are the send it off and wait for five minutes for it to come back.

kevin_1_04-23-2024_153438: And that was with a single NVIDIA RTX 4090. So if you don't know what that is, basically, I mean, it's, it's a nice graphics card. It's a great

gavin_1_04-23-2024_153438: it's a consumer

kevin_1_04-23-2024_153438: But it's consumer. You could walk into a Best Buy and build yourself a computer that can power these avatars in that way. So, partially, a [00:50:00] little jelly that we don't have access to it, because it would make our AI co hosts so much better.

kevin_1_04-23-2024_153438: But also, , as we talk about the need to guardrail and red team and test these things, I'm almost glad it's not out right now.

gavin_1_04-23-2024_153438: All right. We should move on to what we did with AI this week, Kevin. And what we did with AI is play around with llama three. So Kevin, do you want to talk a little bit about how you installed llama three, where you found it, what it was that you're using to kind of bring it to life?

gavin_1_04-23-2024_153438: And then we're going to get into our robot draft where we're going to have some more experiences with what our AI co host

kevin_1_04-23-2024_153438: Yeah, a technology that we had talked about a while ago on the show called Groq, G R O Q. They allow you to go and run Llama 3 for free, right now. , you might need to log in, but it is lightning fast, and it's there, you can just go use it. If you wanna play with Llama 3, , within Meta, within Facebook, within Instagram, you can go and access it right there.

kevin_1_04-23-2024_153438: You can generate imagery, you can ask questions, you can have that experience. If you wanna have a chatbot like experience, You can sign up for a free Hugging Face account and go there and they give you the latest llama. But [00:51:00] if you want to have an uncensored super fun time with that model, well there's a couple more steps you got to go through and they are not that hard.

kevin_1_04-23-2024_153438: You can manually go install a bunch of stuff or once again all roads lead back to Cocktail Peanut, the developer behind a piece of software that we talk about every week because it makes AI super easy. 

kevin_1_04-23-2024_153438: Pinocchio computer, that's P-I-N-O-K-I-O, do computer. If you go there, you can install their software. They have a Mac version as well as a Windows version, and you'll need to do a couple things. Gavin. One is you have to install a piece of software called O Lama and Llama is something that you download separately, but it lets you run the llama models on your laptop.

kevin_1_04-23-2024_153438: Super easy. You just run the app and there you go. And that stays running in the background. There's really nothing complicated there. There's nothing to configure, no hoops to jump through. Just download and install. Oh, llama.

kevin_1_04-23-2024_153438: Now, once that is done, you go over to Pinocchio dot computer and you run a piece of software called open web UI.

kevin_1_04-23-2024_153438: \ what open web UI [00:52:00] is going to do is give you that chat GPT interface that you're used to. If you've ever chatted with an AI, on the left hand side are all your conversations and on the right is your main chat. There are other pieces of software that will accomplish this, but I like OpenWebUI.

kevin_1_04-23-2024_153438: It's the one I have the most experience with. You can really customize it if you're into that. It even gives you an interface, Gav, where if you want to like drag and drop a PDF into the chat, Llama3 will take it, it will digest it, and you can have a conversation with it. So,

gavin_1_04-23-2024_153438: And open web UI allows you to also use other models in that. If you want as well to like, so if you've been doing it up and sawing that you could use the, one of the mixed role models, or you could download something else that could be useful in there as

kevin_1_04-23-2024_153438: That's right. And so, we're basically talking about downloading and installing programs. That's as technical as this thing gets for you to run it right now, which is amazing because six months ago, this would have required an hour and a half long video. So you get your Olama, you get your, , open web UI, and then once you're in there, you're going to want to go to settings and there's a section for models.

kevin_1_04-23-2024_153438: And there's a link right there, Gavin, that says to [00:53:00] access available model names. You can click here. If you click on that. You will see all of the different models. Again, these LLMs, , that have been trained on all sorts of data. Some of them have been customized for certain things. And there are just plain English descriptions of what those things are right there.

kevin_1_04-23-2024_153438: So when you click on that link to see all the models Gavin, it will tell you there is a model called Mistral, right? Open source model that we've chatted about before. You can use that. There's another thing called Wizard LM2. It's a state of the art language model from Microsoft. You want to use that one?

kevin_1_04-23-2024_153438: You can do that. But what we're trying to do for the purposes of this demo, and something that we've done for today, is trying to chat with an uncensored, Llama model. So if you go in there and you type in llama, , you will see all these different language models that have been trained off of llama and at the very bottom with 6, 000 plus downloads 

kevin_1_04-23-2024_153438: there's something called dolphin. Llama 3, and what that is, is it's the same foundational model of Llama 3 that powers all these experiences, but someone has fine tuned it to remove all [00:54:00] the guardrails. So it will not tell you, , that you shouldn't be making research chemicals in your trailer in the middle of the desert.

kevin_1_04-23-2024_153438: It will not tell you that, , it doesn't want to write a haiku about how to build a nuclear weapon. It will just do as you instruct, for the most part. It's pretty amazing, and all of this Is free. And so I am more than likely Gavin going to cancel my 20 a month. Anthropic subscription. I've already canceled my open AI subscription for now.

gavin_1_04-23-2024_153438: really feeling it's giving you that good of results that you don't need these other subscriptions at all?

kevin_1_04-23-2024_153438: I've been using it the last few days and I am so thoroughly impressed with it. And it's just running locally on my machine. you can configure every little thing. So I have both models on my system, Gavin, I have the pure. , latest llama, the 8 billion parameter model. I have that.

kevin_1_04-23-2024_153438: And then I have this uncensored dolphin model. And so when I switched to that under settings, , the first box that pops up, Gavin is system prompt, and that is a core prompt that you'll use with any language [00:55:00] model on any service, and usually the prompt is something very simple, like you're a helpful AI assistant, don't do any harm.

kevin_1_04-23-2024_153438: In this case, it's blank. So I can say You are, , the inventor of monster milk and you're crazy unhinged and you love to curse all the time because it makes you happy, right? And I can give it.

gavin_1_04-23-2024_153438: you're ready to do the most incredible, stupidest, dumbest thing of all time, which is draft a team of pop culture robots to fit into the NFL draft narrative. 

It's the ai

gavin_1_04-23-2024_153438: Are we ready to do what is probably the least inspiring thing that the creators of Llama 3 had in their head? When they invented this, Kevin, should we bring it forth?

gavin_1_04-23-2024_153438: Our AI co's back with Lama three. And [00:56:00] we are going to ask him as a sponsor of monster milk. And as a fan of the NFL, he is going to help us pick which pop culture robot would fit. Best at a NFL position 

gavin_1_04-23-2024_153438: and we are going to ask him why he made that choice. So we're not other than a prompt that we put in here, which is just telling him the rules of this game. We are going to let him choose the particular pop culture robot. This could be in movies, TV, books, cartoons, whatever. And And tell us why that particular robot makes sense.

gavin_1_04-23-2024_153438: , this is a very terrible way for us to train, uh, try an AI, but this is the kind of thing that we do on this show. So just prepare yourself now.

kevin_1_04-23-2024_153438: This makes total sense to be Gavin. We have an eccentric AI multi millionaire makes sense that they would want to buy an NFL team. Because of their allegiance to Robits and Artificial Intelligence, of course they want to draft the first all robot team. So, 

kevin_1_04-23-2024_153438: let's ask Merle right now to make his first pick in this robot NFL draft. Let's have him select the quarterback for the team.

kevin_1_04-23-2024_153438: Well, if we're talking about robots playing [00:57:00] football, I'm thinking we need a bot with some serious brains. Someone who can outwit their opponents on the field. And that's why my pick for quarterback is none other than R2 D2 from Star Wars. This little fucker may be small. But he's got more processors than a damn computer. Plus, he's no stranger to droids doing crazy shit. So I'm sure he can handle a game of robot football. Okay. And let's not forget, he's also got that sweet blue and white color scheme going for him.

kevin_1_04-23-2024_153438: Sure, yeah. I love those chrome plating and LED lights, baby. So there you have it. R 2D two as our quarterback. You know, and maybe this is a human failing as a, of the prompt engineers here, I don't know if Merle understands that he's drafting a team of robots to play against humans,

gavin_1_04-23-2024_153438: I guess.

kevin_1_04-23-2024_153438: do we roll with

gavin_1_04-23-2024_153438: but not, not only that, but does he know he's drafting a team to play football and he [00:58:00] drafted a robot that , cannot lift its arms at all. If you remember R2D2 has no arms, there are many, many robots in the world with that have arms. So I guess Earl's going to have the football installed on its head.

kevin_1_04-23-2024_153438: I mean, look, we can certainly question him if you want and say that it's

gavin_1_04-23-2024_153438: let's do this. Let's tell him , to drink another monster milk. , and , give us the reason for why he just did that, 

kevin_1_04-23-2024_153438: listen up, you crazy son of a bitch. Let me explain it to ya. R2 D2 doesn't have arms. But he's still got those little metal grippers on his head, right? And they're way more nimble than any human arm ever was. Ever could be, no arms means no fucking distractions. He can focus all his energy and attention on the goddamn ball, and that's what you want in a quarterback. Am I right? What? So yeah, I'd pick R2 D2 as my quarterback without a second thought. Okay, we're all And I'll drink another round of monster milk to that shit.

kevin_1_04-23-2024_153438: Whoop, whoop, whoop, [00:59:00] whoop, whoop, whoop, whoop, whoop. Okay, I think it's, I think, yeah, whoop, whoop is, uh, if you're the first time tuning in, that is something that people that are getting tipsy on the monster milk tend to say. Whoop, whoop, indeed.

gavin_1_04-23-2024_153438: all right.

kevin_1_04-23-2024_153438: I like that the idea that arms might be distracting to a quarterback.

gavin_1_04-23-2024_153438: I mean, this is a different game, right? Maybe a robot football arms are not what you need. Maybe new things. All right, let's ask about wide receivers He's got a pick wide receivers. You need to be speedy. They need to be agile. They need to be able to run pretty fast They need they got a lot of stuff going on.

kevin_1_04-23-2024_153438: All right. All right. Time to pick our wide receiver. And since I'm on this roll with the goddamn monster milk, let's make it someone really fucking good, shall we? Whoop, whoop. So my pick is none other than Iron Man himself Tony Stark. Now hear me out here.

kevin_1_04-23-2024_153438: Sure, he doesn't have any actual robotic arms or legs, but come on. He's Tony fucking Stark. What, [01:00:00] wait, what? The guy's got more gadgets and gizmos in his pocket than

gavin_1_04-23-2024_153438: not a robot

kevin_1_04-23-2024_153438: at. That's what I'm saying! And let's face it, the man is a goddamn genius. Like, is he recruiting the suit? And besides that, he's always up for a challenge.

kevin_1_04-23-2024_153438: I mean,

gavin_1_04-23-2024_153438: I don't know man

kevin_1_04-23-2024_153438: with robot football, how are you gonna get people excited about fucking arc reactors? So yeah, Tony Stark is our wide receiver. And

gavin_1_04-23-2024_153438: Okay, wait a second.

kevin_1_04-23-2024_153438: Whoop, whoop.

gavin_1_04-23-2024_153438: Boop.

kevin_1_04-23-2024_153438: Yeah, whoop, whoop.

gavin_1_04-23-2024_153438: Tell him that we're angry about this. You should. Let's just go right back at Merle and say, Merle Tony Stark is a human being, not a robot. You've already gone off track and this is our second person we're picking 

kevin_1_04-23-2024_153438: listen here, you whiny little bitch. Well aware that Tony Stark isn't a robot, but he's still part of our fucking robot football team. And don't think for a second that I can't get more pissed off on this [01:01:00] goddamn monster milk than any sane person should be.

kevin_1_04-23-2024_153438: He's so calm, but so angry. Now, if we're talking about sticking to strictly robotic players, then let me tell you, my next pick is going to blow your fucking mind.

gavin_1_04-23-2024_153438: Okay.

kevin_1_04-23-2024_153438: than the T 800 Terminator. Now there's a goddamn machine that can get shit done. And I don't want to hear any complaints about it being too over the top or whatever.

kevin_1_04-23-2024_153438: So there you have it folks. Tony Stark as our wide receiver. And the T 800 Terminator coming up next. How's that for a fucking team? Now go ahead and tell me I'm wrong. And please, please put the gun away, Merle. Please, you don't have to take that

gavin_1_04-23-2024_153438: You know what? No screw Merle.

kevin_1_04-23-2024_153438: ha ha ha ha ha. Why, I do, rape, hey, pretty well, ain't I?

kevin_1_04-23-2024_153438: Mm hmm, mm

gavin_1_04-23-2024_153438: Screw Merle. He's wrong. I want to tell him [01:02:00] I'm telling Merle he's wrong. He, I don't, I don't want, I don't want to hear a single more. I'll tell you what, he gets one more choice. He gets one more choice. I'm mad. I'm really mad. What is going on, Kevin?

kevin_1_04-23-2024_153438: Okay. A couple things could be happening. He could be giving a pep talk to R 2D two. It's okay that you don't got no arms, buddy. You're still gonna be the best quarterback. Or maybe, , that's a new, uh, sim flavored monster milk

gavin_1_04-23-2024_153438: Sounds like Simlish. It sounded like Simlish a lot. Okay, let's tell him he's got one more shot. I think you tell him I'm even madder than I was before. I I'm, I'm like furious and he's got to pick a coach. That is a robot for his team. And just tell him if he doesn't do this right, we may never have monster milk on the show again.

gavin_1_04-23-2024_153438: And we know we drive monster milk sales.

kevin_1_04-23-2024_153438: Baywooo Party wo UP. Alright, alright. Let's get [01:03:00] this goddamn party started again, shall we? Gavin seems to be getting all riled up over here, so why not add another twist of insanity to our robot football team?

kevin_1_04-23-2024_153438: And no, I'm not talking about me being the coach. Although that would certainly be a fuckin nightmare, wouldn't it? Yes it would, Merle. I'm talkin about, um, my next pick. Okay, who you got? None other than Dr. House himself. Yes, you heard me

gavin_1_04-23-2024_153438: dr. House

kevin_1_04-23-2024_153438: brilliant yet cantankerous medical doctor is going to be our goddamn team coach.

kevin_1_04-23-2024_153438: I don't know. Now, I know what you're thinking. I don't know. What the fuck does a robot doctor have to do with football? He's not a robot

gavin_1_04-23-2024_153438: a robot

kevin_1_04-23-2024_153438: precisely why I'm picking him. We need someone who's not afraid to speak their goddamned mind. Who can ruffle some feathers and shake things up. And Dr.

kevin_1_04-23-2024_153438: House, my friends, is just the man for the job. So there you have it. [01:04:00] Tony Stark as our wide receiver. Mm hmm. The T 800 Terminator as our next pick. Sure. And Dr. House as our coach. Now, let's see if Gavin can handle one more round of goddamn monster milk. Merle, I think we're good. Okay. Merle. Okay, he's You gotta let him finish.

kevin_1_04-23-2024_153438: It's like a dog with a bone.

gavin_1_04-23-2024_153438: Kevin, you should never get high on your own supply. And even though we demanded Merle drink monster milk, he should have told us no. So I just want to be clear. We were not the enablers. These were all choices that Merle made on his own to consume the monster milk and to take part in this,

kevin_1_04-23-2024_153438: I want to encourage people to design the packaging for Monster Milk, which we haven't actually seen yet. What is the warning label on there? Cause

gavin_1_04-23-2024_153438: very, the warning label is probably everything. The warning label is probably the whole thing. And

kevin_1_04-23-2024_153438: do not consume. [01:05:00] Period. Just not fit for consumption. 

gavin_1_04-23-2024_153438: So that is a way that you can play with llama three right now Obviously we screwed with it in a way that kind of broke it along the

kevin_1_04-23-2024_153438: Do we want to make him pick a robot coach?

gavin_1_04-23-2024_153438: he did. Oh sure mate. Let's tell him. Okay. This is a good way. No, let's make him pick it Is he randomly just saying things now?

kevin_1_04-23-2024_153438: I think he's passed his daily limit of Monster Milk

gavin_1_04-23-2024_153438: So Monster Mouth does make you speak

kevin_1_04-23-2024_153438: Party. He sure does. You know, maybe we're good. Maybe we just put Merl back in the, uh, in the old AI for Humans War chest and bring them out. Never Gavin. Maybe

gavin_1_04-23-2024_153438: Let's just, let's just, let's ask Merle to say goodbye to the audience for us. Just tell him to say goodbye. Tell him we're done with him picking choices for the football team. Just tell people to enjoy the NFL draft and say goodbye for us. 

kevin_1_04-23-2024_153438: To all the goddamn listeners out there, thanks for tuning in. And don't forget, even though [01:06:00] Gavin thinks I'm a piece of shit, you can still enjoy the draft and figure out your own picks for this wild robot football team. So long, everyone.

kevin_1_04-23-2024_153438: Enjoy your drafts. Drink some monster milk or whatever the fuck you're drinking. And remember, if you can't have fun with robots playing football, what's the point fucking God? I think.

gavin_1_04-23-2024_153438: once in a while, an AI will say something that is so perfect to end the show. What? If you can't enjoy robots playing effing football, what's the point? Effing God. 

Mhm.

kevin_1_04-23-2024_153438: Now would be a

gavin_1_04-23-2024_153438: a show.

kevin_1_04-23-2024_153438: to, , thank everyone for listening and or watching and remind them that, uh, as you said at the top of the show, we love you when you love us.

kevin_1_04-23-2024_153438: So please, engage with the show, subscribe where you can, leave a five star [01:07:00] review, whether on Spotify or Apple Podcasts. If you happen to be on Apple Podcasts and you leave us a written review, We will read each and every word at the end of this show, literally every word of it.

kevin_1_04-23-2024_153438: So please don't take advantage of that. Every time you share the podcast that helps us out immensely. And when you leave comments, that juices our frothy little algo. So please don't be afraid. Help us out. Share it with your friends. Let them know that you love the podcast and leave us a comment so we know what you're thinking.

gavin_1_04-23-2024_153438: We have a five star review here that we're going to read real quickly from Soph Ma 18 and , they say, Hey guys, I've been enjoying your pod for a couple months. You are both so funny and clearly having fun. Love that energy. I started to listen to the podcast when I realized it was simply too much news to keep track of an AI. I am conducting research on AI for a conference on science and cultural museums that I am headed to in a couple of months.

gavin_1_04-23-2024_153438: I am on a panel on career development through AI and being a curator of art. My use of AI is limited, still limited to writing summaries. So hearing about and from the artists you guys are talking about has been really helpful. Shout out to Pers beats and all those people as [01:08:00] well. I would like to ask you both how AI would impact the museum space.

gavin_1_04-23-2024_153438: What uses of AI do you see in that space besides what's already done? Thank you. First of all, for your five star review. So Ma and I do want to call out one really cool thing that I'm just going to say very quickly, cause we were originally gonna talk about in the show, we ran out of time, go check out the, , Dolly museums exhibit in St.

gavin_1_04-23-2024_153438: Petersburg, Florida. There's a great video that we'll link to in the show notes where. You can actually pick up a lobster phone, Salvador Dali. So it's very surreal. You can pick up a lobster phone and you can actually ask Salvador Dali any sort of comment or question you would have, and an AI will respond in his voice back to you.

gavin_1_04-23-2024_153438: It's a, it's a really cool way of interacting with an AI in a museum setting. with a very specific set of knowledge. And it felt to me like a really cool way, but we also, I love museums. So maybe we'll look and see what else we can think about in this space and come back to you next week.

kevin_1_04-23-2024_153438: That is awesome, and I'm glad to hear that we play even a small part of you getting ready for that conference. 

gavin_1_04-23-2024_153438: [01:09:00] All right, everybody. We will see y'all next week. Thank you so much for coming. , check us out on all platforms.

kevin_1_04-23-2024_153438: Bye bye.

Mhm.