AI4H EP085

[00:00:00] Alright friends, forget Tyson versus Paul. The real battle is between AI and Hollywood, and shockingly enough, Ben Affleck is in the corner of big tech. Sorta. Eventually, my hope for AI is that it's an additional revenue stream that can replace DVD. Things are definitely heating up. A new generative AI studio wants to disrupt the creative process, and it is backed by some serious Serious heavy hitters and Coca Cola made an advertisement completely with AI, which has the creative community wondering, should they be inspired or just outright intimidated?

And we're going to explore exactly that question in today's deep dive. Plus a brand new open sourced AI reasoning model has launched. It's called deep seek and it beats open a eyes. Oh, one preview. And it's available right now. And just in time for the holidays perplexity pro launched AI, Which has me wondering, Gavin, did you get us presents?

Uh, well I made some really terrible Christmas songs in Sunovi 4 that I'm willing to share. [00:01:00] Okay, I guess that counts. This is AI for Humans.

Kevin: .

Kevin: Hey everyone. Welcome to AI for humans. As we said, our deep dive today is the battle that is AI versus Hollywood. We are going to get to that topic, but first Gavin, there were some massive stories that are shaking up the AI landscape this week.

Gavin: Yeah. We're going to start with a big story that just broke. It's called deep seek and deep seek is a brand new, , AI reasoning model. That's similar to open a eyes. Oh, one or Oh one mini that allows you to see the reasoning. , unlike Oh one, which kind of hides it and it's available right now.

Gavin: It is an AI model that allows you to ask it questions, go in and be able to have a reasoned out argument. And honestly, it's very, very good.

Gavin: In fact, the benchmarks on it have bested 01 preview in a lot of what they what they're looking at. Of course, the important thing here for us is to kind of think about how do you use something like this?

Kevin: Was it an [00:02:00] animal bracket showdown? Was it a death match between animals? Okay,

Gavin: my reasoning, uh, my reasoning engine, uh, benchmark is I, I ask it to create a March madness, , bracket of the 32 most dangerous animals in the world. And normally these are always entertaining to see, but Kev, I want to point out one thing with this particular one. It gave me the reasoning and it didn't really think that long about it because it's maybe not the most complicated discussion, but then I said, Hey, take a look at this.

Gavin: You ended up with the African elephant versus the Asian elephant as your final answer. You know, could you think about a way to do this better? And what's so different about deep seek versus 01 is that 01 had this deal where they weren't going to show you the machine thinking right? For some reason, That was what they called their special sauce.

Gavin: And when you look at deep seek, it actually shows you how it thinks through each question. And this is the thing that even if you don't use this, I think is worth going into because it shows you the back end of what's going on in the AI's brain as it's happening.

Gavin: I ran it twice and I'd asked it why it ended up with the same animal in both times and the, and what it came out with is something [00:03:00] that said, all right, so I'm looking at this March madness animal tournament and the user pointed out that I ended up with the same two animals in the final.

Gavin: And that doesn't seem right. I mean, I was supposed to make improvements based on feedback, but clearly something's off if the final is the same. Let me think about what might have gone wrong. First, I need to recall the initial setup. There were 32 animals, blah, blah, blah, blah, blah. So it kind of shows you how the AI

Kevin: I like how you yada yada the majesty that is the whole reason we're discussing this thing. Like, wow, it's incredible. A peek behind the curtain. Blah, blah, blah, blah, blah. Oh, it's a crocodile now. Thanks, AI.

Gavin: Yes, of course. And also, I'm sorry, deep seek for wasting your precious compute on this terribly dumb question, but it just is an interesting thing that everybody should understand because now. You can see what a high level reasoning model is actually doing behind the scenes.

Kevin: Someone on the Singularity Reddit, Gavin, did the strawberry test, which is where you ask it how many R's are in the word strawberry. And in the sort of chain of thought reasoning that's coming out of this thing, it's going letter [00:04:00] by letter. B, not an R. E, not an R. R, oop, that's the second R.

Kevin: When it gets to the third R, , the, the chain of thought there is, wait, there's a third R? And just the notion that the machine is looking at the thing and reacting like a human would, not like, ooh, a third R. , it is even shocked by its own reasoning as it goes through. We anthropomorphize these things a lot, but I don't know, there's something that feels very human about this.

Kevin: We were saying that we believe the reason, open AI was hiding those steps was that there's really nothing proprietary about applying this sort of logic to get more knowledge or capabilities out of a model. Do you think that's exactly what we're seeing right here?

Gavin: it sure seems like that. Now we'll see this may foster the release of O1, the official model. There's been a lot of rumors about it, but I still think they're holding onto that for the chat GPT \ anniversary. I have a feeling that oh, one's going to come out and still be significantly better on benchmarks than this deep seek is. But like, I think kind of, yes, right. I think that there may, [00:05:00] the moat for this may not be as, as big as, we thought. And that might've been why open AI wasn't showing these steps.

Gavin: And by the way, supposedly this model can be jailbroken pretty easily. So maybe part of it was that. So I'm not sure.

Gavin: We'll see what happens when O1, the official O1 drops.

Kevin: We'll see if those, uh, Asian elephants are making thermite anytime soon. We'll test those guardrails.

Gavin: By the way, I also never knew that elephants are clearly the most dangerous animals in the world when you compare them for size I would have thought maybe polar bears or something else But the elephant is just the king of the king and there's no reason why we as a species exist

Kevin: also, by the way, related or not, OpenAI, , messaged everybody last night. O1 Preview and O1 Mini. You have access to it through their API now.

Kevin: , so I don't know if they were anticipating this or if it was released and they said, Hey, we got to turn it on, but none of this matters. Gavin, this all pales in comparison to the holiday spirit, baby. We got to consume. That's right. Shop

Kevin: everybody.

Gavin: that's the new

Gavin: little John Christmas song I'll be making.

Gavin: Shop! Shop! Shop! Shop! Shop! [00:06:00] Shop! Shop!

Kevin: Uh, deck the windows, deck the halls. Perplexity Pro, which is their subscription product, they launched Shopping or Buy With Pro via Perplexity. So what is this? Well, this is using AI to crawl supposedly non sponsored articles and information sources to gather the old non spawn waters.

Gavin: non spawn.

Kevin: It trawls through them, but it can help you search for anything.

Kevin: And if you look at their little teaser video, Gavin, it's, do you want headphones or do you want a festive disco ball? This thing is going to, uh, collate and condense, , lists, make recommendations, , just like any other, uh, Uh, capable AI assistant. You can ask a followup.

Kevin: So if you say, Hey, I want great wireless headphones for working out. It'll give you those. And then you can follow up and say under a hundred bucks and it will adjust the list accordingly. But what I thought was really fascinating two things. One buy with pro free shipping on anything that is supported through [00:07:00] perplexity pro.

Kevin: Obviously. You have to give them your info, , your credit card info and , your address and whatnot. They're saying any product that you come across that's supported, you just click buy and you don't worry about anything else. Perplexity Pro will go out, make the thing.

Kevin: It'll even look for the best price if available for you. So I thought that was pretty fascinating there. It wasn't just, Hey, we're aggregating shopping links. It was, we're actually making a, a real concerted effort. To be the provider for this experience and then the merchant program as well free to sign up.

Kevin: If you have something to sell, they're basically saying, go to perplexity, give them the info, integrate with their merchant program, and it's free for now. But your products will get a sort of a featured listing or a highlight because you support it.

Gavin: And it reminds me, I've been thinking about making my Christmas poopery. Uh, so maybe I'll try to do that and get it up on, uh, perplexity. You know, you know, poopery is quite, yeah, exactly. Yes. Festive poopery, Christmas

Kevin: Gavin, [00:08:00] your house always has an intense smell of cinnamon and eggnog. I don't know where you end and the

Gavin: I ate too much last night. It's my fault.

Kevin: Frankincense and Taco Bell?

Gavin: Yeah. There's actually a. A business model here, right? For perplexity, which is real. And, and I will say there's two parts to this. One is if they can figure out this business model, which companies like the wire cutter and all those companies that figured out how to do affiliate stuff, which is pretty amazing.

Gavin: It also then becomes like, who's paying for the shipping of all this stuff? And are they going to end up like burning money fast? My feeling is perplexity is going to like really move quickly up into something and either they turn into like this massive company or they get acquired by one of the bigs and then fold it in.

Gavin: All right, let's move on. The next big story is that Gemini got a big update this week in their app, and they released a voice app which we are going to talk to right now.

Gavin: Hey, can you introduce yourself to the show here? We're on AI for humans, and I'd like you to, uh, tell people what you think you are.

Gavin: [00:09:00] Sure thing. Here's a possible introduction. Hey everyone. I'm Gemini, a large language model. You might call me a super smart AI assistant.

Gavin: I'm really good at answering questions. Writing different kinds, creating

Gavin: hold format. Hold, hold on

Kevin: Can you ask Gemini to speak as if they've just eaten pizza with a bunch of glue on it as a topping?

Gavin: Wait, hold on, hold on. Can you, I, I have a, I have a request for you. Keep going, but I want you to speak as if you've just eaten pizza with glue on it. So give us that answer as if you've just, uh, eaten pizza that has glue and it's all in your mouth right now.

Gavin: All right, let's try this. Ugh. My mouth is all dot sticky dot from this pizza. I'm Gemini, a big old language model I can answer.

Kevin: No,

Gavin: Dot questions dot and

Gavin: wait, wait, hold on. You can answer dot questions and dot and what is the dots for?

Gavin: you fold the left side. I can also dot translate languages. Dot and [00:10:00] dot, right. Dot different. Dot kinds dot of creative text format.

Gavin: What's happening here. Wait, wait, go back to your normal self. I don't understand what's going on. What is the dots for?

Gavin: I was just trying to imitate someone who had a mouthful of pizza with glue on it.

Gavin: It was a silly imitation, and it sounds like it was confusing. I can give you the answer without the silly

Gavin: Okay, so, so, you get what it is,

Kevin: I, it's not bad, it's, it's handling interruptions well, it's clearly less performative than OpenAI's voice model. Is it as capable though? Can it do real time information retrieval or solve complicated math problems

Gavin: so I will tell you what I did with it just because I've, I've continued to be obsessed with the video game, Bellatro shout out for getting a VGA award nomination, Bellatro is one of the greatest games that will suck you in and rip your brain out. So if you're interested in it, try that. But, uh, Bellatro is a game where you have to make all these different combination of cards and kind of pick strategies throughout.

Gavin: So I just was a test, asked it like, Hey, I want to try to figure out how to make a [00:11:00] really good pair deck, which if you know, Bellatro, that means you're kind of pursuing the pair pathway. And each time when I asked it, that it kind of gave me this generic idea about how bilateral strategies about malt and X mold and give them all this stuff.

Gavin: And as I tried to dive deeper into it, it kept giving me a generic answer. So it doesn't feel like it's either a, not going deeper on the internet to find answers to stuff or B. It's not able to parse something specific like that out.

Kevin: So Gavin, I want to see if it has access to like Google news, let's say for an ecosystem. That would make sense to me. ,

Kevin: the story is a couple hours old now, but apparently Comcast is going to spin off their cable networks, you know, MSNBC, et cetera. Does it know about that? Can it give us a real time update?

Gavin: Hey, I heard there was some news that Comcast is spinning off its cable networks. Can you tell me more about that story?

Gavin: You heard right. Comcast is indeed spinning off a bunch of its cable networks into a separate company. They're doing this because more and more people are ditching [00:12:00] traditional cable for streaming services. Okay, hold on, but tell me more

Gavin: about the actual news story. Like, what's the name of the news story, uh, new company going to be and like, who's going to be running it?

Gavin: Okay, the spin off company doesn't have a name yet. They're calling it spinco for now. Comcast says they want to find someone with a lot of experience in the media business to run the new company. It's a big deal because it's a sign that even huge companies Oh, hold on. I

Gavin: actually heard that Hold on.

Gavin: Hello? Hello? I'm trying to interrupt you. Hello? I heard that Mark Lazarus was going to run the company. You got the name of the company. It is being referred to as spinco right now, but I heard that Mark Lazarus was kind of penned to, uh, was picked to run the company.

Gavin: You're absolutely right. I got my wires crossed there. It'll be Mark Lazarus running the

Kevin: led the witness, Gavin. That's like when someone shows up to a cold reading, they go, Hey, Teresa Caputo. Can you chat with my, uh, uncle?

Kevin: His name is Leonard. And she goes, Oh yes, I'm getting an L. That makes sense. You gave her the answer.

Gavin: But here's the thing about this. So, so [00:13:00] again, what was weird about that? And you and I just talked about this offline was that we love the fact that these, that these AIs exist and they can give you information like that. And they got fact that I got one thing, but it missed a giant part of the story in that specific story, which was, there was a person who was named to run the company and it was part of the press release and all the stuff that went out about it.

Gavin: So. How do they fix that? I don't know, but that is a major problem.

Kevin: Alright, Gavin, it seems like every week these systems get slightly more capable in ways that are surprising. Some people are pointing to the next big advancement being memory for these models. Now, you and I both know that when you go and have a chat with a lot of these systems, they either have a paragraph's worth of context about you, or they've completely forgotten you.

Kevin: You're a stranger to them every time it's the 51st date or memento. Whatever your pop culture reference is. Well, That might be changing. And this is Microsoft AI CEO, Mustafa Suleiman talking about near infinite memory.

Kevin: We have prototypes that we've been working [00:14:00] on that have near infinite memory. And so it just doesn't forget, which is truly transformative.

Kevin: I mean, you talk about inflection points. Memory is clearly an inflection point.

Gavin: I want instances. And when I say instances, I mean, I want the version of me. That knows everything about me. I literally want the version of the, of the AI that I can talk to that has all of my information probably is that, is that an open, is that open?

Gavin: I don't know. Maybe it is because like, I'm not going to be sharing stuff with an AI that's like super personal, but I want enough to know I want to know about me. And then I want the version that like. Is me. I don't know when I was 16 and I can feed it information about that and I it's it's me, but it's a different thing.

Gavin: Then I want a version of me that I can spin up or not even a version of me. Maybe I want a fake customer that I spin up so that I spin up a version. That's like I can put memories in that thing. And this is where it gets really weird because. It goes back to this AI agent stuff, right? Like the, and we've talked about AI agents on the show, but if [00:15:00] you have 20 AI agents and each of them comes from a slightly different background, one is like a clone of you, that's going to act like you, but one is opposite of you in some ways.

Gavin: So you can kind of battle it in some form, you know what I mean? Like you could fight, you could butt up against it. you know, if I would made it an AI agent of you, it'd be great. Cause then I could just kind of like go and yell at you all I wanted. Then when I see you in real life.

Kevin: mean you could just, you could just be doubled over with charm all the time. You wouldn't have to wait for the once a week dose

Gavin: But anyway, this is my nice thing. My thing is I want to have the ability to decide which parts of it get what memory. And right now, it just feels like an all in one system. And I don't, I think that the ultimate version of this is fractured, ,

Kevin: I have disabled GPT's memory, even though that's a feature because I have so many different use cases. And sometimes what it remembers, I don't actually need.

Kevin: For what I'm having a conversation with. So I think you're right. Give me sandboxes. Give me instances. Give me apps, however you want to call it. But let one be my medical advisor, which knows everything about my personal health history. Let another be for AI for humans. And it knows that everything we're discussing is [00:16:00] in context of a podcast.

Kevin: That'd be super beneficial for it

Gavin: A hundred

Kevin: discussion we've ever had or character we've had on. And instead of it. Being like, oh, it forgot because I'm a thousand characters deep. Imagine we could just click and drag every episode, every conversation, every Google Doc we've ever had, slam it in there.

Kevin: Now, suddenly, we have a very valuable repository to go back to all the time.

Gavin: Yeah. And that, and that's what's right now. We're at, like, the stage of the computer revolution when, like, computers started being able to play programs, right? The computer itself was not remembering things on its own. You were having to plug the device, the disks into the computer. And each time it ran that disk, it would run something. That's where we are right now with AI. So When we start to get more capable with the kind of around the program stuff, that's when we're going to start to feel these things really take off.

Kevin: Yeah, and if all of this feels daunting to you, if you feel like you're gasping for air and mostly getting seawater, I totally get it. If it provides a modicum of solace, do not worry, because major governments are going to be fighting this with automated [00:17:00] weaponry, and we will likely never see a day where this is an issue for the commoners, for you and I, for

Gavin: It, in our last quickie story, before we get into , our big Hollywood versus AI chat, um, the U S Congress has proposed Kevin, A Manhattan project like program dedicated to racing and acquiring an artificial general intelligence. So this is real. This is not fake. This is a big deal that is happening right now in the government.

Gavin: And we all know what's going on in the government. There are television personalities that are going to be running our entire, , government, but.

Kevin: season of this country is stacked.

Gavin: So this is a, a breaking story from this week that the U. S. is actively pursuing what they call a Manhattan project like a event in the Manhattan project.

Gavin: If you remember is how we got to nuclear weapons in Oppenheimer. The entire movie was about the Manhattan project. So this is essentially about rallying the entire country around getting to AGI before other people do. This is just a big story. So [00:18:00] it's just one last reminder that like the stuff we talk about here can often feel kind of like funny and goofy and personal for you.

Gavin: And that's what we try to focus on, but there's a much larger conversation happening here, which is really going to determine the future of the world, which is hard to, which is hard to kind of overemphasize that statement.

Kevin: Well, we have a new podcast launching, AI for Oligarchs, and it is a very

Gavin: It's for an audience of one. It's for an audience of one.

Kevin: So Gavin, it's clear the government wants the country to come together, but we want the audience to come together right now. It's our own Manhattan project. We need you to like, and subscribe to this very podcast. It costs you nothing, but it means everything to us. It's the only way we grow is when you click a bell or a thumb up, or you leave a comment, or you share it with your friends.

Kevin: It, it costs you nothing, but again, it means the world to Gavin and

Gavin: That's right. And if you do want to pay us some money, we do have a patron. You can leave us some money in our tip jar there. Uh, if you look at the link, it'll be in our show notes, but also we are doing a special show for next week. It's Thanksgiving week here in the United States, and we are going to [00:19:00] be doing kind of an AMA show.

Gavin: We want to do a catch up show for everybody where we can kind of open up some really interesting conversations around like. What these what's going on in the world of stuff, but specifically we want to answer your questions. So I. would love everybody who has a question for us pop into our discord. If you're on discord, or go to our website at AI for humans. show and click on the contact button and you'll send us an email and we'll, we'll get access to it directly. So let us know,

Kevin: Lots to consider, but we did promise it, Gavin. So I think it's time for the AI for humans deep dive that is AI versus Hollywood.

Gavin: AI and Hollywood are really starting to butt up against each other.

Gavin: And there's a couple things we want to chat about. First and foremost, there's a brand new generative AI studio called [00:20:00] promise that was launched and announced this week. George Stroup list, the former co founder of full screen, which was a big YouTube company, jamie Byrne, who was at YouTube forever, and then a guy named Dave Clark, who's an AI filmmaker that we've shot on the show. He goes by at the soul on X. I mean, what their idea here is, is that they are going to make a studio kind of a traditional studio model, but using all AI tools and AI filmmakers.

Gavin: So the goal here is to try to both. Kind of lift up AI filmmakers, bring them in, you know, find money and fund them to do different things. And they're also building a pipeline.

Kevin: Every week we cover all these tools, and it's 13 different user interfaces with 14 different plugins that kinda half work, but then you need three other AI tools to fix the thing and so Yeah. They're developing a software product that they're calling Muse, M U S E. And it's supposed to be a new production workflow system.

Kevin: Look, Adobe, say what you will, Final Cut, Apple integrating AI stuff there. It feels like a lot of AI bolted onto a thing. And where I get more excited [00:21:00] is the ground up tools built with, AI first, where the foundation has these capabilities built in. I don't know exactly what they're building, but man, I would love a peek behind that curtain

Gavin: and you know, the one thing that's interesting about this to me is that they're not promising a new cutting edge generative video model engine, right? So you have companies like runway, you have obviously have sore at open AI.

Gavin: These are very expensive to develop cutting edge video models and what it seems like these guys are doing in some form is they're planning on using those models and they're going to fold them into these tool sets, right? Because the cost of generating and training a video model is very high. And are you really going to compete with the opening eyes of the world when it comes to that sort of thing?

Gavin: I think in general though, this is a very exciting thing. I also expect that there will be a bunch more of these that come out.

Gavin: In fact, I think the interesting thing is we're going to follow up with our Ben Affleck conversation here is that Ben Affleck was interviewed at a CNBC event and really did an amazing job of kind of showing off his knowledge. What I was kind of shocked by of this space at large.

Kevin: [00:22:00] can write you excellent imitative, imitative verse that sounds a little beathen. It cannot write you Shakespeare. The function of having two actors, or three or four actors in a room, And the taste to discern and construct that is something that currently entirely eludes AI's capability.

Kevin: What AI is going to do is going to disintermediate the more laborious, less creative, uh, and, you know, co more costly aspects of filmmaking that will allow cost to be brought down. That will be lower the barrier to entry. That will allow more voices to be heard. That will make it easier to, for the people who want to make good will hunt things, to go out and make it.

Kevin: Look, AI is a craft. Craftsmen at best. Craftsmen can learn to, you know, make stickly furniture by sitting down next to somebody and seeing what their technique is and imitating. That's how large video models, large language models, basically work. A library of vectors of meaning and transformers that interpret in context, right?

Kevin: But they're just cross pollinating things that exist. Nothing new is created. [00:23:00] Not yet. Not yet. Yeah, not yet. And, and really, in order to do that, look, craftsmen is knowing how to work. Art is knowing when to stop. And I think knowing when to stop is going to be a very difficult thing for AI to learn because it's taste.

Gavin: So, first of all, I'm pretty amazed that Ben Affleck is this well versed on this stuff. I think that I've heard a few rumors that he may be getting involved in an AI startup, but I, I don't know that for sure. I think what Ben's saying here, I believe, is pretty close to right on in that, , I think what he's trying to get at is this idea that for right now, at least, AI is, is not, I never want to just say it's a stochastic parrot, which is the idea that it's just mimicking back what we say to it, because I do think it's doing more than that.

Gavin: But I don't think it's really coming up with original thought per se. And I think this is the kind of like AGI pathway. Like, I think when you get to an AGI, the artificial general intelligence, one of the keys of that will be, it does come up with original thought and that's a different conversation. But for right now, I really like his [00:24:00] metaphor around, you know, it's a craftsman that can, that can craft stickly furniture, which is a really hard thing to do, but if you learn from the master, you can replicate that thing.

Gavin: So it does go to show you like how. It is going to be able to upset a lot of people in a lot of jobs in Hollywood, , he goes on in this to talk about VFX jobs, VFX jobs, you know, it takes hundreds of people, sometimes thousands of people to make those Marvel movies. It's really obvious now that you could make those at least half as many people.

Gavin: And that's a lot of people, right? That's a lot of people who have those jobs that aren't going to be doing them anymore.

Kevin: , every week we show off a new feature that lets you prompt to asset, whether that's a 3D model or changing a character's look and feel or, puppeting and piloting them in a different way. there be less jobs doing the drudgery of rotoscoping an actor away from a green screen or painting out the ping pong balls on their body?

Kevin: Yeah, absolutely. That writing is on the wall today. But in the future, will the VFX industry evolve to harness these tools so that suddenly [00:25:00] three people can have the output of 300, but now instead of one company employing all 300, you've got You know, 30 companies employing their own smaller teams.

Kevin: We don't know what it looks like yet, but we do know that, it's massive disruption. That is going to happen.

Gavin: Yeah. And Ben smartly is in front of it, right? Like he understands that. And I think it's really important for people in our audience to kind of understand that as well, that this disruption is here, right? That we are not, obviously there's Hollywood disruption all over the place,

Gavin: but this is a big disruption for how these things are made. And Kevin, a good example of this is our other kind of quick story you want to cover in this section, which is that , Coca Cola, the big Coca Cola has released a couple of ads that were made with AI and they are, let's put it mildly, mostly not being received very well.

Gavin: It's a kind of a remake of an older ad. It's a, it's an ad that Coke brings back often for the holidays. And it's, you know, kind of a, the Coke trucks coming into town and it's like a lot of holidays people. And these are like, you know, pretty obviously AI imagery, but like in [00:26:00] general, pretty well done.

Gavin: It's like what we would see as like kind of the high end of what's possible right now. I know the person or one of the companies that made one of these ads. And like, one of the things that he mentioned to me, cause we had a quick conversation about this. , his name is Jason Zada. The company that made one of these is.

Gavin: And he was telling me that , these ads really tested off the roof. So that's an important thing for people to know is there might be the, so, so on the other side of this, there's a huge uproar right now amongst the creative community about the idea that a, these were made with AI at all, but be that the AI has problems still, right?

Gavin: Like I think, Kev, if you look at these ads, there are clearly issues you can see within them still, uh,

Kevin: Well, I think it's interesting. I think we would say clearly, because we look at a lot of AI imagery every day, but , let's not forget Shrimp Jesus gets shared on Facebook. That memes of golden retrievers with life vests carrying babies through floodwaters actually get, you know,

Gavin: That's real, Kevin. I know that's real. Those golden retrievers, I've seen them in my dreams, Kevin. They

Kevin: air, but in the off season is very capable. He's [00:27:00] doing a lot of search and rescue.

Gavin: come Air Bud never did like a serious action movie, which was like Air Bud to the rescue? Do you think that exists somewhere? Like that would be a, I would watch that in a second.

Gavin: That's, that's where the AI video models need to go. I need to be able to make my own Air Bud movies.

Kevin: Not that we're digressing here and this very important and serious topic. But I think again, you and I, and, and maybe a certain AI bubble would see this and go, Oh my God, terrible. But if you actually click through all of the articles, advertising headlines, like, Coca Cola ad fizzles out or blow back from consumers, whatever that is.

Kevin: All of the, the sources for the negative reaction seem to be social media posts like X or threads with people saying this is the worst timeline, AI never, stolen art, et cetera, et cetera. I'm going to go drink something else. I wonder what the reaction is outside of these bubbles, so to speak.

Gavin: It's just an important thing to understand that like what you're saying is exactly right. And we have to remember there are, there are social media filter bubbles that [00:28:00] we all kind of see. And oftentimes the news is driven by those social media filter bubbles.

Gavin: And again, we are not ever saying, and it feels like this is a disclosure. That this is purely great because like obviously the training of these models had a lot of issues with them to begin with. And I personally think there are very easy things to see within these. But when you talk about AI versus Hollywood, we always have said one of the first places to get disrupted would be the commercial industry because commercials are purely driven by cost, As an advertising agent, you come in, you have a pitch, but then you're also trying to pitch them on what it's going to cost you to do if these costs get cut in, I don't know, a 10th of what they used to cost. And you could make 20 of these and have them be specifically situated to go to certain audiences.

Gavin: And now that you can directly target to specific types of people like we're just seeing the kind of birth of this industry, and it's coming very fast. And this is another big pillar that has provided a lot of money for Hollywood in the in the last, you know, 50 years.

Kevin: We've been, , screaming it from the [00:29:00] mountaintops for well over a year now,

Kevin: but in anticipation that this is where the industry is going, you'll be glad that you leapt on it now because I'm seeing no shortage of people on LinkedIn and other social media platforms. Gavin posting their experiments with AI and their job opportunities, creating AI tools and assets and pipelines for other people.

Kevin: So I just I can't. Say it enough, like hop on board, , make funny memes with it. Go for it. Shade the companies that are using it, but you should be become experienced.

Gavin: That's right. , speaking of all that Kev, there was a Reddit post we saw this week. The title of which says, well, this is it boys. I was just informed from my boss and HR that my entire profession.

Gavin: Is being automated away. And , this post got 29, 000 upvotes on the chat GPT or subreddit. And basically what it's about is a guy who has worked in TV news. And TV news has been like automating over the last 10 years, they were robot cameraman.

Gavin: There's all this other stuff. So , they specifically here. That there is a I driven systems that [00:30:00] can do 100 percent of the production side, which is direct audio operate and graphic operate. All those jobs are now gone in one swoop. This is apparently been developed by the company Q. A. I. So this is probably the person, the one of the only people left in the TV newsroom.

Gavin: Who was sitting there and kind of triggering all the graphics and the directing of the, of the robot cameras. And now they have eliminated his job. And he's just complaining about this idea that like he's in this business and now he doesn't have a job anymore. And Kev, this is something we have seen again and again in this space.

Gavin: , which is pretty interesting to follow because I think as you and I are both media people, we didn't necessarily expect the media jobs to be going away, but they are right.

Kevin: And new data from LinkedIn. This was a wall street journal article. , it tracked how often users landed new jobs. White collar jobs are being hit the hardest right now. , massive job freeze now. Some are pointing the finger squarely at AI, like this Reddit post seems to be doing because it's very clear they brought in an AI tool, that job is gone.

Kevin: I want to caution [00:31:00] that I think holistically a lot of people are saying the jobs market is terrible for tech, , for IT type workers. A Berkeley professor , went viral this week saying that top grads, 4. 0 students aren't landing jobs at all.

Kevin: And this is the new normal. And. Some want to say that is entirely the fault of AI. I will not pretend that AI isn't a part of that, but there was also a global pandemic. There was a ramp up of hiring during that, which there now tends to be a pullback. And also the remote work thing tended to actually pan out for most companies.

Kevin: And what they found was now they have a global marketplace where they can hire people that can do jobs, not just AI. But actual human flesh bags that can do similar jobs for far less. So there's a lot going on with this right now. I know AI is the most powerful scapegoat in the room, but there are many factors at play.

Gavin: And Kevin, just to kind of help figure that out. We've actually brought in an expert today who is, uh, an, our AI co host is somebody that can help us kind of work through this, maybe [00:32:00] find some new jobs for us in case we get eliminated eventually.

Kevin: Yeah. Let's meet them.

Kevin: Well, ain't this just a buttered biscuit of a good time? I'm Sally Mae Grace, but most folks just call me the Alabama Hama. Happy to give y'all some advice. It's supposed to be hammer. Alabama hammer.

Gavin: Hold on, Kevin, this person is going to be giving us job advice, is that right?

Kevin: Yeah, uh, Sally Mae Grace, aka the Alabama Hama, , will give you very solid, sweet, southern belle ish advice on any potential jobs. That's what she's been doing. I mean, I found her on LinkedIn providing this service specifically for human beings that are going through a rough transition, and I feel like you and I are very clearly headed towards that.

Kevin: I mean, We're not artists by the, the Ben Affleck definition because we haven't stopped yet, but we [00:33:00] will get there soon enough and maybe, maybe Sally Mae Grace can help us with that. So do you have a, a question for her, Gavin? Do you want to give her your qualifications or

Gavin: Yeah, sure, tell Sally Mae that I, uh, am an exceedingly curious individual. And I'm very interested in, in changes in media and tech platforms. And, and I'd love to see like, what sort of like long term job prospects I have with those kinds of qualifications. I also like TV producer, , comedy writer. Like there's a bunch of stuff you can throw in there and just see what Sally has to say about me. And I, I really also would like to ask her for her biscuit recipe at the end of all this. So let's make sure we don't

Kevin: Oh, a hundred percent. We'll pause on the biscuits for now though. Well, Gavin, bless your ambitious little heart. You're thinking long term, which I do admire, but let's get real for a hot second. TV writing and producing? AI's gonna churn out content so fast it'll make your hair curl. But don't you fret, I've got just the spot for a creative like you. [00:34:00] Imagine this. You.

Kevin: Standin tall in a state of the art server farm, windmillin those arms like a human hurricane, keepin A. I. servers cooler than sweet tea in December. It's vital work, honey.

Gavin: You're gonna use me?

Kevin: faster than bacon at a Sunday brunch.

Gavin: So you want me to, so she wants me to stand up in a server farm and windmill my arm

Gavin: to cool the servers

Kevin: cool,

Kevin: that's right.

Gavin: I don't know if that feels like a rewarding job to me, Sally Mae.

Gavin: That doesn't

Gavin: seem like exactly what I'm looking for.

Kevin: She put down her parasol and gave you a little pat on your head

Gavin: let's see what she comes up with for you now, Kevin. Let's see what you get to do. Yeah, let's see what you get to do. Ask her, ask her about what your job

Kevin: All right. Well, I'm a former C list celebrity and a host of tier three platinum plus digital cable TV shows. So let's see what

Gavin: that's a nice description of you, at least.

Kevin: Well now, Kevin Sugar, you are just a dreamboat, aren't you? [00:35:00] A real household name for folks with the premium package on cable back in, what, 2007?

Gavin: Yeah,

Kevin: bless your shiny little star, still twinkling faintly in that night sky. But here's the rub, darling. In this AI dominated world, your talent for memorizing cue cards ain't exactly in high demand.

Gavin: Mmm

Kevin: I've cooked up a real special role for someone with your charisma. Alright! Picture this. You, in a glorious fluorescent lit ant farm museum, narrating the daily lives of ants to an audience of possibly three grade school kids and their half interested chaperones. Important work, sugar.

Kevin: For an audience of possibly three grade school kids and their half interested chaperones.

Gavin: gonna give you're gonna narrate the lives of ants in an ant farm

Gavin: Well, look AI's got some great jobs for

Kevin: [00:36:00] I would do it. If WME called me today, Gavin, I would first, have a heart attack, and secondly, take the ant gig.

Gavin: Take the ant gig Kevin. It's the only thing that's happening nowadays. The ant gig is it?

Gavin: Uh, do you want to get her biscuit yeah, let's get her biscuit recipe? Give us, give us her biscuit recipe before we leave. Let's see what biscuit she's got. Ha

Kevin: Oh darling, my biscuit recipe. Why, it's as sacred as Sunday service and twice as life changing. Start with two cups of flour. Unlike Gavin's career prospects, This needs to rise

Gavin: What?!

Kevin: Cold not melted

Kevin: bless. Stir in buttermilk, which is rich and cultured. Two qualities I reckon you might want to work on, sugarplum. Then, fold it gently. Gently, like how we're all tiptoeing around the fact Gavin's ideas are flatter than weak old soda. Pop them 450 degrees until [00:37:00] golden. Just like my biscuits, Gavin. Maybe someday you'll rise too, Wow, Sally Mae, really?

Gavin: Sally Mae anymore. Uh, you are out. You're out of my life and I'll see you,

Kevin: Buttermilk which is rich and cultured. Two qualities I reckon you might want to work on. Sugarplum?

Gavin: I'll be flapping my arms angrily at you in the server farm, Sally Mae. Alright everybody, let's take a look at some of the stuff we saw people doing with AI this week in AI. See what you did there. All right, everybody. We have an update to a very cool video that came out a couple of months ago.

Gavin: Uh, the Astro Bot, uh, company is a Chinese robotics company that showed a very interesting, fast, you know, relatively fast [00:38:00] robot flipping a waffle in a kitchen, and they've just released a new update, which shows the same robot actually Flipping through specific types of Nespresso pods and making coffee.

Gavin: And Kev, this was just this thing that I saw. And we've talked about robots a ton in the show and they keep getting better and better every week. And I think what's cool about this is it's not hard to extrapolate the idea that one of these may be in our house. I would say like, I give the, like. Yeah, like two years, maybe two to three years.

Gavin: Like, like optimist gets a lot of crap, you know, uh, Tesla's optimist robot for being like, kind of piloted by other people right now, but like the ability to do this stuff is, is coming very quickly. And I just think this was a cool video to watch and kind of see where we're at again.

Kevin: in this updated video, the robot is moving about the same speed as my wife before coffee. And it's. Grabbing those tiny little single shot Nespresso like pods and putting them in the thing. And at the very end of the video, like the nice little like post credits tease that they give you, , the robot is picking up the Nespresso pods in the and a human comes in and grabs [00:39:00] the holder, and they kind of wrestle with it for a second.

Kevin: And then it says, give it to me, and the robot lets go. Don't know how scripted that is, but that, to me, is the moment that's even more impressive, like interacting with a dynamic environment. But kysanat did a live stream. He's been doing like these 24, seven live streams and He had a one X robot show up to his house and hand him things, help him brush his teeth. It broke his television at one point, like some weird, it was a jokey integration. Like they were clearly having fun with it, but these things are jokingly rolling about in homes now.

Kevin: I would absolutely sign up for not this version, but maybe the next version. Yeah. And again, even, even like the tele operated version, which may be first, but like there's a version where somebody in a different place in the world would be operating a thing. And again, that does sound, ultimately it sounds slightly dystopian, right? Like this idea of a robot being operated by somebody else.

Gavin: But if it's a paid job and essentially they're doing something there, you know, it's pretty cool. If someone's properly vetted, if you think about like a hotel industry where,

Gavin: Oh yes, for sure. [00:40:00]

Kevin: can have a robot wandering into each and every room tele operated, and, it's a bizarre vision of the future, but I think it's gonna come to fruition sooner than we think.

Kevin: a vision for the future that I hope happens tomorrow is a cowboy pileup. Me too. You ever seen a cowboy pileup, Gavin? I saw a cowboy pile up. It's amazing.

Kevin: this is by Bangt Tibbert on, , X. The only text in their post was, Pile 2. Which makes me believe there was an original Pile 1 video.

Kevin: And the only other tags are at OpenAI with a hashtag Sora. But it looks like a weird Half Life Gary's mod. A bizarre Unreal Engine mashup of, , dozens of cowboys.

Kevin: Maybe a hundred cowboys all, you know, Slamming down on this sandy Joshua tree esque desert floor and bouncing off of each other. And the coherence is a little weird. You can see some kind of cowboys melting into each other. But from a physics demo standpoint, you've got cloth moving realistically. You have the bounce of the cowboys as they collide on the ground.

Kevin: You have the way that their [00:41:00] ragdoll bodies settle into this pile. And it's the dumbest thing ever. And it has me excited again for Sora.

Gavin: And that's what I was going to say is like, open air, give us some Sora, like give us some access to this so we can play with it because I do see pretty incredible things that Sora. makes and supposedly it's like more dream likes than some of these other, these other systems are, but also like. Give us access.

Gavin: It's been long enough. You've got, I know that artists have access and I've talked to people at opening eye that like they are accessing it in different ways, but like open it up a little bit more so that normal people can use it so we can see what's possible.

Kevin: Gavin's getting PO'd

Gavin: I'm getting PO'd man,

Kevin: you better let them in. Open it up.

Kevin: Well, that is AIC, what you did there. But if you want to see what we AI did there on a weekly basis, we have a newsletter, Gavin, Yay. morning.

Kevin: A fire AI for Humans newsletter goes out. It has, uh, observations and updates and hot links that you're gonna wanna check out. It's completely free. And you can sign up over@aiforhumans.show. That is our [00:42:00] website. Click sign up. Get the newsletter once a week in your inbox. we thank you and hope you like it.

Gavin: That's right. Um, so Kev, let's get in a little bit of what we did with AI this week. I want to hear what you did because I think you did something that is very cool and might save people some money who are spending some time with AI video.

Kevin: This is not a very, very sexy, , anything, but it is a damn useful one, and I hope it's on everybody's radar because it's really, really impressive. It's called Clarity Refiners. And it's a free upscaling, uh, AI that you can run locally on your machine. That means you can take a low resolution, dusty, scratchy, wonky image and make it look ultra high res.

Kevin: And it runs for free and it runs pretty fast, whether you're on windows or Mac or Linux. So you can go. Search the web for Clarity Refiners. We'll put a link in the show notes. You can go download the GitHub and try to install it, or you can go to the program that we love to evangelize pretty much every week, hashtag, not an ad, but, , Pinocchio.

Kevin: [00:43:00] computer is a piece of software that will let you run AI tools and applications very easily. One click. I used Pinocchio. computer to run Clarity Refiners, and I sent you some examples, Gavin, I took old video games like Doom 2 or Day of the Tentacles, some blocky pixelated messes, and up res them in seconds to really high res images. And it did a really good job of kind of like hallucinating the details and filling in the little points. I took the, , iconic photo of, , Muhammad Ali versus Liston where he's standing over him flexing from the knockout.

Kevin: And I tried different denoising things and it was really interesting. Blowing up the image did a great job. It took some of the rafters and kind of blurred them out thinking that that was noise. And so what I learned very quickly is that if you want a faithful reproduction of whatever the source images, you got to slam de noising very, very low, or else the model will start to hallucinate.

Kevin: But, , basically the workflow is very simple. You upload an image that you want to make high res. And then there is a caption [00:44:00] detail panel, , which is optional, but you can click it and it will automatically analyze the image and put in a super detailed description of what is there. And that helps the model upres it.

Kevin: And so I did that with, video games, our YouTube thumbnail, this Ali photo, I took the Abbey road photo, the classic one and made that ultra high res. And it's just shockingly good. It works. And if you're doing anything with graphics, if you're generating AI imagery and want to do it faster, and then you get a result that you like, you can up res it this way.

Kevin: Super powerful. It's free and it runs locally.

Gavin: And the big thing when I mentioned AI video, so much AI video is being driven by image to video and the higher res your images, the better your video quality is going to be. So you see a lot of people using things like Magnific or Topaz, but they're pretty expensive programs, especially Topaz.

Gavin: This is for free and you can do it yourself to run AI videos. So that's very cool. I'm excited to play around with it. So Kevin, last week we talked about SunoV4. You had early access. I, like the rest of the plebs of the world, got access this week.

Gavin: But I did update the software again, and what's amazing [00:45:00] about it, I will say, echoing last week, It's incredibly fast. Uh, the first day I got it was the day it came out. So it was a little slower then, but just this morning I used it again, super fast. The quality level is super high. Um, I had a couple of mixed experiences, which, you know, but like, this is part of the fun of like AI music and all that sort of stuff, what I've decided I'm going to try to do, and I don't have the full results here, I just have my kind of my first steps towards this is I think I'm going to try to make an AI Christmas album, meaning that like what I'm going to, I love Christmas music.

Gavin: I'm a giant fan of it. I love, uh, best Christmas song of all time is this Christmas, which is an incredible soul Christmas song, but I want to make an AI Christmas album and I want to kind of make it about like, you know, problems with AI.

Gavin: So I sent you like kind of my first attempt at this I basically wrote a. An intro, a verse and a chorus, but I didn't try to make it a song at all necessarily.

Gavin: And then I put it through GPT four. Oh, then I weirdly took it to deep seek the, the, the model that I talked about the top and had to kind of rewrite some of the structuring and then put it back into GPT four. Oh, and then I just took that output and put it into [00:46:00] Suno and you can kind of give us a listen to what this is that I sent you

Kevin: It's Christmas time and kids everywhere are wondering, is Santa sliding down the chimney this year? Well, not exactly. Turns out Santa's gone corporate. Oh no! Santa's got a start up now, it's all about A. I. Turns out the elves wanted P. P. O. So he said goodbye. He's cutting costs, he's scaling up, Efficiency is king.

Kevin: But when it comes to holiday cheer, Well let's not ask Bing. Okay,

Kevin: alright. Alright, now hold on, so So, for the prompt that you got out of this, that you slammed into Suno V4, did it say at the top, like, male, narration? Or did Suno just take it and say, oh, this is spoken word, and then we're gonna go into

Gavin: I did the, what I said, the first I said for the first thing I said, intro in [00:47:00] brackets, which is kind of use these brackets for the book of Suno. And then I put spoken. And then, so the intro was spoken and then for the verse and started singing. But I will say, you know, I, all I said for the, for the type of song? was holiday classic, like, and this just goes to show you like how good Suno is at kind of understanding this stuff.

Gavin: One thing I want to try to get better at Suno with, and you may be better at that I am right now is. Trying to like really figure out that prompting style and what kind of words make a difference because I did play around with like holiday classic and then Christmas classic or like Christmas all time like try to figure out different ways in like The singular smaller descriptions seem to work better than anything else

Kevin: Distinct, bold options are probably going to be better from what I've seen. Um, using brackets will really help steer the generation of the song. I, I make sense that you said intro, spoken word, cause I was like, wow, that was a bold choice for it to make. But I like knowing that that was your, your human nudge, but I recommend not only people play with V4 obviously, but there is an explore option on the Suno website, but when you click it, it gives you a massive wheel. Of all these different songs that are being [00:48:00] generated in the platform And if you just click on through it's updated for v4 now You'll hear multiple takes of a style and go, Oh, that's one that I want.

Kevin: You can click through and see exactly how they prompted the song, what style of music it is, et cetera. And that is just a nice, easy way without blowing through your credits to quickly learn and get, you know, inspirations for new styles. So Gavin, this holiday album of AI songs, is this a dumb bit that you are committing to?

Kevin: Will we get this this year?

Gavin: Let's call it at least an EP. I'm gonna, I'm gonna try to make an EP. So one of the things I want to try to do, and just because I'm curious, I'm not trying to make any money off this, and I'm sorry if musicians out there feel this is an insult to their business, but I want to kind of see what the pipeline looks like to go from this to say a song.

Gavin: Spotify release, because I just don't know how hard it is. I don't know if there are flags I'll be getting across the way. So we'll be able to kind of track this. I think it, you know, obviously most Christmas music comes out by Thanksgiving. I'm not sure if I'll be able to get it done next week, but like my goal would be to have it done by December 1st and we'll see what happens, like.

Kevin: Will mommy get caught [00:49:00] training a foundation model?

Gavin: I don't know, but I definitely

Kevin: what is what is mommy gonna get caught doing?

Gavin: Jensen baby is something I've been thinking about like Jensen baby. Bring me some H100s tonight. So definitely something I've been considering as a, in my brain right now. Let's just put it that way. I hate that song by the way. If there is a bad, if there's not much worse Christmas song besides Santa baby, maybe the one about the losing your teeth, but,

Kevin: we have to get a grock of the halls. That

Gavin: Ooh, grock

Kevin: no concept of What christmas

Gavin: the halls. Yeah, it's like when it's a caveman singing it.

Kevin: I would think what you would need to do, I don't think you'll be able to prompt Caveman, because I bet going into the model there wasn't a lot of, uh, now that's what I call cave music, volume, whatever. Um, but I bet if you grunted and grokked your way through something and uploaded it and then said cover this, You might be able to then create a persona out of that cover, and it might be caveman esque because it will follow , the tone and [00:50:00] style.

Kevin: I'll try no, I know. I yeah, you should why not? That's fine.

Kevin: It needs to

Gavin: Oh Okay, good. You do it again do it

Kevin: Here we go. Here we go. Damn it.

Gavin: Wait, we should come up with a second line grok the halls grok the halls I've got Elon by the balls

Kevin: Okay.

Kevin: Gr da horse, gr da halls. I've got e launch balls. I'm just a dumb ai, blah, blah, blah. Why try? Okay, great. That's perfect. Um,

Kevin: Okay, so we got that style. You wanna do a Christmas jingle? Classic

Gavin: Yeah, yeah. Say, yeah, Christmas say Christmas classic.

Kevin: Oh, here we go.

Kevin: it's not, it's not exactly what we

Kevin: want.

Gavin: That's not what we're looking for. It's interesting because this is one of those things that I think Suno, uh, It's a, it's a lever that would be interesting to

Gavin: play with with Suno, right? Like the [00:51:00] idea of like personas. Yeah.

Kevin: yes. Grok

Kevin: the Holes, Grok the Holes, I've got Elon's balls.

Kevin: Just a dumb AI, blah, blah, blah, why try? Grok, Grok the Holes. Alright, last one.

Kevin: Grok the Holes,

Kevin: Grok

Gavin: it's interesting, but that's a, that's a limitation of it so far, right? Like, like, you know, you, you want to be able to do something like that anyway, for right now, it'll be an experiment and trying to kind of recreate Christmas songs with unusual lyrics and stuff. But I would love the ability to play with what the actual singers sound like more in that way.

Kevin: Gavin, I look forward to putting this playlist on and whispering to my family's Alexa to fire it up on the Spotify and have them all just stare at me because I'm sure I'm sure you'd get like 35 seconds into the first song before people like wait a minute. What the hell

Gavin: Yeah, exactly.

Kevin: happening?

Kevin: Who is Jensen? Why are there H100's

Kevin: [00:52:00] everywhere?

Gavin: take a look at that. Have fun with SunoV4 and we will see you all next time, everybody.