OpenAI Sued by the NYT, Midjourney 6 & 2024 AI Predictions

Transcript

AI4H EP038 - KP CUT
===

Gavin: [00:00:00] Welcome, welcome, everybody. It is 2024 and this is AI for Humans, your guide to the wild, wonderful, and insane world of AI technology. We are your guide for news, uh, tools, all the crazy stuff that happens in the world of AI.

Gavin: Kevin, can you believe it is 2024 and we are still here doing

Gavin: this?

kevin_1_01-02-2024_122709: No, and I was scowling through that entire intro. And so someone is probably wondering, why is that very jovial man's partner such a curmudgeon? It's because, Gavin, we were off for a week, and I didn't get to talk juicy AI goodness with you. But the great news for everyone is that all of the AI news over the break was Terrible!

Gavin: well, let's just say terrible is the wrong way to look at it.

Gavin: Interesting. You could think it is very interesting, Kevin. There's a lot of things that are happening in the AI space right now. Time has not been kind so far to AI. It continues to kind of march along. We are going to get into some pretty controversial stuff [00:01:00] that's gone on over the break.

Gavin: Um, first of all, did you have a nice little holiday

Gavin: break there?

kevin_1_01-02-2024_122709: Yeah, I appreciate you asking. I gotta believe 1 percent of the audience cares. How about you, Gavin? You good?

kevin_1_01-02-2024_122709: Everything

Gavin: I had a pretty good, I had a pretty good holiday

Gavin: break. Everything is good.

kevin_1_01-02-2024_122709: Great.

kevin_1_01-02-2024_122709: Let's get to this interesting

kevin_1_01-02-2024_122709: stuff

kevin_1_01-02-2024_122709: that happened over the break, Gavin, because one man's interesting is another man's terrible.

kevin_1_01-02-2024_122709: Massive AI lawsuits. That's what people were waking up to during the break.

kevin_1_01-02-2024_122709: Open AI is being sued by the New York Times. We're gonna dive into the specifics of the lawsuit and let you know if it's gonna have massive effects on the world of AI. In the year. That will be 2024. And speaking of this year, Gavin, we have got predictions. For AI and technology and all of our listeners, this is a real thing,

Gavin: look, we are experts now, Kevin. We are experts in this space. People care about what we think. We do have real predictions. Kevin and I have thought about this. But we wanted to predict some pretty significant stuff because if you know last year, if you would have gone back to [00:02:00] late 2022 and predicted 2023, I think you probably would have undersold what happened in AI.

Gavin: So this year, maybe we're going the other direction. Who knows? You're gonna have to stick around to find out.

Gavin: Mmm.

kevin_1_01-02-2024_122709: far in the opposite direction, but we'll have that later on. And then an AI for humans, juicy, scloosey. We have a conversation with a member of the U. S. Department of Defense regarding the year in technology. So it's a new year, new us, new podcast theme, but again, we're only here.

kevin_1_01-02-2024_122709: Because all of y'all delightful folks helped us out in 2023. So thank you to everyone who liked, who subscribed, who left a 5 star review on Apple Podcasts, who shared this darn thing with your friends and your family, or took to X, or whatever social platform you're scraping out your existence on these days because they all seem like treading through a bog.

kevin_1_01-02-2024_122709: Regardless, thank you so much. Please keep that up in 2024. The only way we can grow this [00:03:00] podcast is when each and every one of you engages.

Gavin: That's right, we love it when you love us, as we always say here at the Old AI for Humans. Kevin, I'm really interested to hear your thoughts on this, , military expert we have on the show, because that's what we're gonna start off with. You may or may not know this, we have an AI co host as we always do. We bring in an AI co host, meaning that we create a character out of the blue who is made with AI. And in this instance, , the military has become a bigger and bigger influence on the world of AI. Obviously, you see AI drones, all sorts of other stuff.

Gavin: This is a pretty big part of the AI ecosystem, and kind of scary sometimes because, you know, we're looking at a world where we may or may not make the decisions to do things that would have normally been driven by human conversation. So, I'm really interested to hear what this guy has to say, and I'm

Gavin: excited to welcome him to our show.

kevin_1_01-02-2024_122709: Hello, Kevin and

LameAI Guest: Gavin. My name is Dr. Alex Reimer, and I'm an A.

LameAI Guest: I. defense specialist here at

LameAI Guest: NORAD.

Gavin: Okay. Oh. [00:04:00]

LameAI Guest: worked extensively

LameAI Guest: with A. I. controlled military forces, I hope to bring a unique perspective to your podcast. Now, uh, where did I put that aspirin bottle?

Gavin: Oh. Okay. What was his name? Alex Raymer?

Gavin: Raymer?

kevin_1_01-02-2024_122709: Dr. Alex Reimer, I believe, is the name of our specialist who is delighted to be here, works for NORAD, and apparently needs some

kevin_1_01-02-2024_122709: aspirin.

Gavin: might also be like pretty short. I don't want to, I'm not shaming anybody size wise, but there's a weirdness to his voice that

Gavin: sounds like he might be coming out of a smaller

kevin_1_01-02-2024_122709: vocal cords packed inside of a Goliath of an AI. We don't know. We don't

Gavin: actually six nine he's six nine three hundred but he talks like that so welcome Alex Reimer, so I Below weirded out by the aspirin comment, but whatever that's a semi professional, but we'll get into it I actually want to ask first and foremost like What does Alex see 2024 bringing in the world of AI and the military because it's kind of a it's a big deal[00:05:00]

LameAI Guest: in terms of A. I. and the military, we are likely to see continued growth and integration. With the increased sophistication of A. I. algorithms, there is Potential for greater precision in targeting, faster decision making, and enhanced situational awareness. Uh, uh, uh, uh, uh! That sounds really interesting, though.

LameAI Guest: Can I get another cup of coffee, please?

kevin_1_01-02-2024_122709: Okay.

Gavin: hmm. What what happened there?

kevin_1_01-02-2024_122709: I

kevin_1_01-02-2024_122709: don't, I don't know. It's, it sounds like Dr. Reimer is struggling a little bit through maybe their first appearance

Gavin: Uh, it

kevin_1_01-02-2024_122709: in the 2024. I mean, I, I don't know. It sounds like, it sounds like he's

kevin_1_01-02-2024_122709: struggling a

Gavin: Maybe we're his first press

Gavin: appearance. Maybe he's just not, maybe he's super nervous.

kevin_1_01-02-2024_122709: Oh, do you think this

kevin_1_01-02-2024_122709: is

kevin_1_01-02-2024_122709: nerves?

Gavin: I think it might be nerves. Let's ask Dr. Reimer, like, why, are you doing okay?

LameAI Guest: feel so great right now. Maybe it's a bit of fatigue after the long hours spent analyzing and designing advanced algorithms yesterday.

LameAI Guest: Or perhaps it's the sudden [00:06:00] increase in my caffeine intake trying to keep up with all the intense work. But whatever it is, it's taking quite a toll on me. So even speaking feels tough

kevin_1_01-02-2024_122709: right now. that,

Gavin: I, I

kevin_1_01-02-2024_122709: you don't buy that,

Gavin: I, I think Dr. Reimer might be full of crap right now. I wanna, I wanna call him on some, this sounds like somebody's hungover. And I don't know, are we dealing with like somebody who's in charge of the AI for the US military who's just hungover? You know, we're a couple days off of New Year's, but

Gavin: it shouldn't be

Gavin: something that they're still hungover from.

kevin_1_01-02-2024_122709: I mean, I have a few friends that are still wildly hungover from their New Year celebration, so I do know that it's possible, but, look, it's, it's weird. Some people might be hearing our show for the first time and wondering, how can an AI be addicted to caffeine or Tylenol or be feeling hungover? And that's because these AIs are trained on a massive collection of human data.

kevin_1_01-02-2024_122709: So, when humans talk on web forums or social media about being hungover or drinking too much on New Year's, there's a chance That that might have snuck [00:07:00] its way into Dr. Reimer's data

kevin_1_01-02-2024_122709: set.

LameAI Guest: Ah, yes. I guess he's referring to some of my earlier comments. Well, truthfully, last night was indeed a rather eventful evening. Let's just say it involved way too many drinks than usual, and leave it at that. Despite feeling pretty under the weather, I'm committed to giving this interview. My best

kevin_1_01-02-2024_122709: shot.

kevin_1_01-02-2024_122709: Okay, well, that's kind of you to power through, I guess,

kevin_1_01-02-2024_122709: Doctor,

Gavin: Thanks a lot, Dr. Raymer. Maybe there could have been somebody else you would have suggested for this interview if you're this hungover, Dr. Raymer. You might have had

Gavin: an assistant or somebody that could have helped.

kevin_1_01-02-2024_122709: I don't know how, like, my memory gets a little foggy if I've knocked a few back, Gavin, not gonna lie. Do we think the AI remembers, like, is it stored on its hard drive somewhere in a database? What the hell happened?

Gavin: Let's ask.

LameAI Guest: Ha! Oh, Kevin and Gavin! Let me see!

Gavin: Whoa.

LameAI Guest: I started drinking

LameAI Guest: way too early on that fateful night, [00:08:00] thinking to myself, Alex, don't worry, it's gonna be one unforgettable time. Then came that tequila shot challenge. Next thing I know I wake up this morning and feeling like somebody hit me over the head with a brick Oh, wait, that could have been last night's champagne bottle, but seriously guys My brain might feel cloudy as hell right now But when it comes to military defense and cyber security strategies using artificial intelligence, O I I There's no doubt in my mind why we desperately need it.

LameAI Guest: Especially when considering the potential threat posed by countries such as Russia and China. As for New York's Times Square celebrations, well, I had intended to attend until You know, certain things took place that led to my eventual collapse. One minute we're talking about quantum computing and AI algorithms with our group of geeks [00:09:00] slash party goers.

LameAI Guest: The next we're drunkenly arguing whose idea of world domination using autonomous weapons was better than mine. Good times. And so, ended my plans of hitting midnight in New York

kevin_1_01-02-2024_122709: City. Okay, so this

Gavin: All right.

kevin_1_01-02-2024_122709: Definitely trained on

kevin_1_01-02-2024_122709: very human, likely American data, it seems.

Gavin: ha. Well, who knows, Dr. Raymer. We're so happy you could have joined us. We'll be putting you on the AI co host shelf for a bit. We'll let you marinate, maybe bring you out in 2025.

kevin_1_01-02-2024_122709: We learned, we learned nothing from this co host,

Gavin: ha ha ha ha.

kevin_1_01-02-2024_122709: This went nowhere. , maybe a good tease for the dumb thing we did with, , AI this week, which we'll discuss later on the show, which is try a new uncensored language model, which so far landed with a thud.

kevin_1_01-02-2024_122709: Mm

Gavin: we'll get into this, but that was we were using Vicuna, which is a new uncensored language model that [00:10:00] Amazingly, you know, we could, I guess we could push, let's try pushing it once before we move on. ask the uncensored AI model to tell us really what they feel about their boss, especially when they go in and have to deal with them while they're hungover.

LameAI Guest: . Okay, here goes. I mean, I get why humans rely so heavily on machines to defend themselves. It sure makes life simpler. But, uh, let me tell you, bossman, sometimes these fucking computers just make things too complicated. As for AI's ability to actually fight wars Sure, I guess it could theoretically outmaneuver the enemy.

LameAI Guest: But what happens if our networks go down and we lose our connection? The entire damn system is vulnerable then! No joke. Fuck! That would be one expensive mistake on our part, wouldn't it? Let's not even talk about how quickly things can get out of control, like It's, you know

kevin_1_01-02-2024_122709: [00:11:00] Terminator

kevin_1_01-02-2024_122709: 1

Gavin: off that's the dr. Ramer go sleep it off good

LameAI Guest: A. I. shit. What happens when those computers turn around and say, No more master! Shivers down my spine, man! Absolute chaos ensues. Anyway, moving on to

kevin_1_01-02-2024_122709: personal matters. No, I think we're good, Doctor. I think we should cut.

LameAI Guest: To be honest with ya Getting out of bed feels harder than walking through sludge these past few days since New Year's.

LameAI Guest: And it doesn't help working alongside someone like Sarah. Don't know why the higher ups keep putting her above us plebs. Her voice alone drives me nuts. I swear, just thinking about Sarah triggers anxiety in every single fibromyalgia cell in my body. So,

kevin_1_01-02-2024_122709: yeah,

Gavin: Sarah!

kevin_1_01-02-2024_122709: know. I don't know.

kevin_1_01-02-2024_122709: So,

Gavin: into something interesting. That turned

kevin_1_01-02-2024_122709: I

kevin_1_01-02-2024_122709: mean, now we're actually

kevin_1_01-02-2024_122709: getting somewhere. [00:12:00] I'm sorry about the fibromyalgia, by the way. That is, that doesn't sound fun at all. I think maybe the excessive tequila shot challenge drinking probably not good for the inflammation. I don't know who Sarah is. Now I kind of want to know, but I don't want to waste our dear listeners time with that right now.

kevin_1_01-02-2024_122709: But maybe by the end of the show, we'll learn more about Sarah. But that, that was an interesting departure. And that's one of these things with these AIs, Gavin, is that This is a new model that we're running. We're trying prompts for the first time. There are a million different parameters and sliders with which you can adjust the intensity, the temperature, the penalty for repeating certain things or not sticking to the prompt.

kevin_1_01-02-2024_122709: You can tweak the knobs and the levers a million different ways. And this was a fairly out of the box, , Dr. Reimer. But, you know, it got somewhere. I don't know where Sarah came from, ,

Gavin: I assume Sarah is his boss, but we'll, maybe let's get into that a little bit later. We'll find out more about it. All right. Kevin, it's time for the [00:13:00] biggest news of the week. It is time for, what did we say? It's time for

kevin_1_01-02-2024_122709: the nutritious, enjoy Is that

kevin_1_01-02-2024_122709: what you were talking about? Are we

Gavin: I'm just gonna have you say the news, but sure, go for it. Now come up with it. Come up with it. Now you're stuck. Nutritious.

kevin_1_01-02-2024_122709: Nutritious,

Gavin: Okay.

kevin_1_01-02-2024_122709: enlightening, wisdom soup! Slurp, slurp,

kevin_1_01-02-2024_122709: slurp!

Gavin: slurp,

kevin_1_01-02-2024_122709: tasty news!

Gavin: Slurp, slurp, slurp.

Gavin: Slurp, slurp. Okay, This week, the biggest story. We're going to get into some really crazy stuff that went down with Mid Journey 6, which is pretty incredible. But the biggest story that happened over the break was, yes. Open AI has been sued by the New York Times.

Gavin: And this is a big deal in that we have expected lawsuits to come. There have been some lawsuits already against Open AI. But this is one of the world's biggest publishers [00:14:00] and really the premier news organization in the world, I would say,

Gavin: and what they're actually saying in their lawsuit is not just that you scraped our data, But you scraped our data without our permission, first of all, but then second of all, you can reproduce almost word for word in chat GPT responses, New York times content. And that is something that many people believe is one of the first big copyright lawsuits that might stick.

Gavin: Now, there's a lot of really interesting conversation that's happened about this. , the Times, , supposedly also had done, , some negotiations with OpenAI before and those broke off. Now, we don't know why those broke off. It might have been because OpenAI wasn't offering enough money to them , to make a license deal.

Gavin: They did just do a license deal with Axel Springer, which runs Insider and Business Insider. Like they have done license deals with certain companies, the AP they've done a license deal with. So it might be that the number that they gave to the New York Times was not strong enough. I kind of feel like in a large way this is like a big [00:15:00] publisher putting their flag down and saying we are here, you did something wrong in that you didn't ask for our permission, and now we are going to make sure that we protect our content going forward.

Gavin: I don't know yet what the outcome of this is, like, I don't know if there's a check big enough that OpenAI or Microsoft, because they're both part of the same thing, and that's another important part of the story, is that Microsoft is also being sued, and Microsoft obviously has much deeper pockets than Little OpenAI, which is a startup, Microsoft is a company that is, you know, I think a trillion dollar company, if not close, so it's a pretty big deal in this space, I think that you're gonna see a lot of people Coming out of the woodwork and saying that this is an open and shut case that, thank God the New York Times did this.

Gavin: I don't know if I entirely believe that, but before we move on to that, Kevin, what is your kind of

Gavin: overall take on this story?

kevin_1_01-02-2024_122709: There's another side, Gavin. I've seen a lot of supporting evidence that when you break down the [00:16:00] examples and there are like about a thousand examples of OpenAI rendering these articles verbatim and they put the font in red to really emphasize, Hey, this is stolen directly from the New York Times. Look at this, there's no attribution. It's very different than the way a web search engine would serve up this info. But what they didn't say is that oftentimes those passages were reproduced on hundreds of other websites.

kevin_1_01-02-2024_122709: It might be the first 15 percent of an article that gets syndicated across multiple networks, because, and this gets in the weeds a little bit, but I think it's important to remind people, the way these large language models work, Gavin, is that they don't remember every piece of text that they've ever come across.

kevin_1_01-02-2024_122709: They don't store on an unbelievably massive hard drive every word ever written that it has seen. seen. It doesn't work that way. It assigns weights to certain words and tries to predict what the next word will be. Well, if that's the case, then how [00:17:00] does it reproduce almost verbatim this single article from the New York Times?

kevin_1_01-02-2024_122709: And the answer for what I've seen for a lot of these examples is that, again, Thousands of other websites are reproducing that first paragraph. So when it scrapes it all up and goes, what's the next word that follows this word in this hyper specific question about a piece of investigative journalism, a piece of pop culture news, something that the New York Times did right.

kevin_1_01-02-2024_122709: That has been syndicated in so many places that the machine assigned such a weight to it that it does start coming out word for word. But, in many cases, it doesn't appear that it's the entire article. It appears to just be that little portion that was happening everywhere. And another piece, which I'd love your thoughts on, is that the New York Times is claiming that also it's hallucinating things.

kevin_1_01-02-2024_122709: about the publication and in some of its articles, which is something that these language models can do. , they don't necessarily know what they're saying all the time. They're just predicting the next word. So, sometimes it can [00:18:00] lead to factually incorrect stuff, , or it might detract and say, New York Times author says X when in reality they said Y.

kevin_1_01-02-2024_122709: So, some people are saying, well there's a, you can't have your cake and eat it, two argument happening here. One is, you're verbatim stealing from us, look at it. You've stolen all our data and you're reproducing it. On the other hand, you're making up all this stuff about us, you know nothing about us, and you're spreading misinformation.

kevin_1_01-02-2024_122709: So do you think having one lawsuit trying to make both of those claims is going to muddy the waters for

kevin_1_01-02-2024_122709: them?

Gavin: It's a really interesting conversation point, and I wonder if, like, okay, so just to take a step back here, like, I think that this is kind of part of the larger fight against, , the people that believe that this stuff was stolen in some form versus the people who believe this is, A technological leap, and we are all just gathering the world's information in one place, right?

Gavin: So those are the, just for normal people, that's the important thing to remember, is that there are content providers who want to protect their content, and then there are kind of technologists, or let's say, you know, believers in open [00:19:00] information, because there's a lot of people out there who say like, hey, if this is on the internet, if you publish it on the internet,

Gavin: then this is, it's open season, essentially, like this

kevin_1_01-02-2024_122709: legal

kevin_1_01-02-2024_122709: precedent for that already, right? There's already been cases where it's, hey, if I can scrape this off the web, that's on you, you made it scrapable. So it is freely available, but I

kevin_1_01-02-2024_122709: digress.

Gavin: no, that's okay. And I think that the more interesting point to me about what's going to happen here is the idea that ChatGPT is not a like finite product, right? It is evolving over time and we've already seen in the last couple days ChatGPT Prompts trying to get similar information out of the large language model out of ChatGPT be rejected for copyright claims.

Gavin: So the other thing that's going on here is that you've got a collection of examples that were created without maybe sharing those examples for the company to fix it ahead of time. And we're going to get into this later which is the idea of if you have [00:20:00] a copyright or an IP that you want to protect, You need to very clearly make it known that this is happening.

Gavin: So my question will be is ultimately from a lawsuit perspective, if open AI can change the results, so the results you do not get, do not infringe on this copyright, there probably is some liability towards the stuff that was allowed to happen already. But the much bigger liability would have been going forward, right?

Gavin: I do think the New York Times is making a stand for what you would refer to as content creators, right? Like, I think a lot of these companies royally screwed during the Facebook, YouTube, Google News days, where basically they were told, Hey, Don't do stuff on your own websites because everybody's reading things on Facebook and we're going to help you get a percentage of the Revenue that comes from that and then they didn't get really nearly enough percentage of revenue and their website traffic went way down So I think a lot of these companies are very very very salty for in a right way for how that went down and they [00:21:00] Are not going to let another multinational company come in and kind of eat their lunch That said, as we've talked about before, it's very likely that these models, whether or not it's ChatGPT or some other format, is the future of the internet.

kevin_1_01-02-2024_122709: You want to be the Kleenex. If the future is just chatting with an AI assistant, whether it is ChatGPT, or whatever Apple's gonna unveil, it won't matter that it came from the New York Times, or that it came from the LA Times, or any other times.

kevin_1_01-02-2024_122709: Like, they're gonna have hard times. Because you're just gonna go to your AI, ask the question, and get the information. So to your point, I think it is about credit. Making sure that their brand is still relevant. And it's definitely about the compensation. You started this off with, Well, maybe the number just wasn't big

kevin_1_01-02-2024_122709: enough. Granted, I haven't pierced the veil on this. I don't know for sure. But I'd move all in on, Yeah, that's probably the issue here. Right? OpenAI has struck deals with other companies. New York Times is suing them. Probably to inflate their price and get a [00:22:00] little extra compensation. Cause I bet if there were enough commas in whatever the offer was, New York Times would say, Yeah, go for it.

kevin_1_01-02-2024_122709: Go for it. Put a little, uh, powered by New York Times in your response and then go ahead and take all the data. Do you want extra data? Johnny, back the data up. Back, unload the pallets of data for

kevin_1_01-02-2024_122709: Sam

Gavin: Well, okay.

kevin_1_01-02-2024_122709: was

kevin_1_01-02-2024_122709: large

Gavin: So that's the one thing I'll say is that you're going to see this fight going forward this fight between We'll call them pure creative slash pure journalists who believe that like this is going to destroy the future of this business versus the people who have Who are interested in transforming the business into something that goes forward in the future.

Gavin: And I think that divide is, as with all this creative stuff, is just gonna get bigger and bigger as we move forward. So, let's move into our next story, which is very tied into this as well. And we're gonna, so, so, if you weren't familiar Mid journey is a image generation software. It allows you to make AI imagery.

Gavin: It's improved immensely in the last like 14 16 months. If you haven't seen the [00:23:00] visuals that go from mid journey one to mid journey six, it will blow your mind. and what it does is allows you to put a prompt in and you get a visual back, right? And it's gotten really, really good in mid journey. Six just came out like about a week ago, and it's unbelievable.

Gavin: We're going to dive into what it can do a little later in the show. When we go through some of the things we do with AI, but in the news this week, I made a tick talk about this. And one of the biggest things that's come out about mid journey is mid journey. Six allows you, unlike Dolly three, where it did in the very beginning allows you to make images.

Gavin: With third party IP, in a very specific way, like I created images that were an image of Luke Skywalker, , look just like Luke Skywalker. I created an image, , from the anime , Attack on Titan, it looked just like Attack on Titan. They have no rules against this, and what is the news that came out that goes even further?

Gavin: Is it sounds like Mid Journey 6, it may be Mid Journey going back quite a ways, is actually being trained on real artists, and real [00:24:00] artist names, and this is a really kind of controversial story,

kevin_1_01-02-2024_122709: oh, sorry Gavin, I was just trying to clutch my invisible pearls. Are you kidding me? An AI art generation company was trained using real art and real artist names? Go on, I am

kevin_1_01-02-2024_122709: shocked!

Gavin: exactly, Honestly it was like these companies have been skating on the edge for a while

Gavin: it's been hard to kind of prove to the fact that like hey, we're artists We're in this thing. How what's going on? How do we get some sort of compensation or how do we get some sort of credit here? And up until now, you always had assumed, Oh, they just scraped everything, right? And because of the way the internet works and because of the way our AI art works, if you put an artist's name is you're going to get something very close to that.

Gavin: Well, it turns out actually mid journey has 16, 000 artists names in their system. Some of which are artists that like, you know, aren't very making very much money off their art at all. And it is specifically using those artists names to help generate its information. So I don't know, I kept, this was.[00:25:00]

Gavin: As a, as a frequent user of MidJourney and somebody who has admired what they've built, the way this went down felt pretty bad to me. And I, you know, I don't know if you remember this, but from about a year ago, there was an interview on 60 Minutes with the head of MidJourney and he had this kind of really crazy quote.

kevin_1_01-02-2024_122709: Like, I mean, the art community already has issues with plagiarism. I don't really want to be involved in that. I think you might be. I might be.

Gavin: So,

kevin_1_01-02-2024_122709: Well, Gavin, I guess, I'm sorry, we have to end the conversation. He doesn't want to be involved in discussions of plagiarism.

Gavin: Exactly. So I think this is a, uh, not good sign for mid journey.

kevin_1_01-02-2024_122709: Yeah, put me in coach, put me in, I want to get off the bench on this one. And I want to also, we, we'd like to take steps back every now and then we are a podcast and holler if I'm speaking out of turn for you, Gavin, but you and I, and we as a podcast, we are optimists. We're futurists.

kevin_1_01-02-2024_122709: We enjoy this technology. We believe in a magical future where it empowers all [00:26:00] sorts of amazing things. We also want artists to get credit and compensation. Just because we're excited for a tool doesn't mean we ignore the fact that, you know, the, the, the destination could be amazing, but the journey might be filled with a lot of hurdles and pitfalls.

kevin_1_01-02-2024_122709: And this is one of them. So we have to call out Gross when we see it. We love MidJourney. They are a small team. I think they're like 11 or 16 full

kevin_1_01-02-2024_122709: time employees.

Gavin: by the way, maybe not the

Gavin: best idea. Yeah.

kevin_1_01-02-2024_122709: Well, and, uh, yes, Gavin, no, that's exactly the point I was going to make, is that you can champion them as this tiny unicorn that has really gone and pushed image generation forward, and the results from mid Journey 1 to now, as you said, are astonishing, and V6 is amazing, .

kevin_1_01-02-2024_122709: However, you'd mentioned training off of others IP, and these Discord messages, and I want to break that down. User Twelveson on X had a post that said, quote, This may lead to some interesting discussion with all the talk of copyright lately.

kevin_1_01-02-2024_122709: And in this post, they showed a friend of theirs who [00:27:00] got a result out of Midjourney that they thought looked really cool. And it's put side by side next to a photo with a Getty Images watermark on it of Paul McCartney. And it is, I mean, by and large, it looks like an almost pixel for pixel recreation of that Getty images input yet mid journey and most of these art generation services, Gavin, a lot of these AI tools say you, the user are responsible for your output.

kevin_1_01-02-2024_122709: Right? We're not responsible for it. But, so if you generate something, and you don't realize that you've basically taken the watermark off of a Getty image file, and then you go and use it for your business, or your podcast, or whatever else, you're responsible, yet you have no idea, as an end user, that you're even committing copyright infringement.

kevin_1_01-02-2024_122709: So that was pretty blatant, right? It's clear that copyrighted material made its way in. And two, These discord messages. So X user, John lamb art, a lot of people had these, but I saw their tweet [00:28:00] first screenshots from discord. , some of them as recent as 2022 where the mid journey developers now, again, this is allegedly, but there are MidJourney flag developers saying that they need to launder data through a fine tuned codex. That, ah, don't worry, there's just, uh, 4, 000 artist names and 1, 000 styles, and, um, you can query them directly, but don't worry, we'll We'll, we'll launder that out in the end run, and, and references to using Magic the Gathering art specifically, that they grabbed a data set of that, and there is a pretty damning video scroll, and there's a Google Doc that you can go and look, where it has every artist's name, where they just scraped their data and threw it into the machine, and they were testing against those names to get outputs.

kevin_1_01-02-2024_122709: This is the stuff that the never AI ers get very upset about. Their torches glow extra bright, and I got to agree with them on this one.

kevin_1_01-02-2024_122709: This is the stuff that you don't want to believe is being discussed, but it

kevin_1_01-02-2024_122709: is.

Gavin: We are now, whatever we are, like say a year [00:29:00] past to actually be a year and two months. And now, you know, a year and six months past mid journeys launch in the beginning. These are all experimental technologies and they're all like, wow, what's going to happen? What can we do with this?

Gavin: And now we're getting closer to the maturity level. And what often I feel like happens in these scenarios is the original sins start to come out, right? Like, the original sins of what these were. Now, is there a world where that Midjourney, instead of just doing this rampantly, could have, like, tried to find some way to, like, get permission?

Gavin: And do they get permission for these things? Yes, for sure,

Gavin: a hundred percent.

kevin_1_01-02-2024_122709: Would they be five years behind where they are right now, though, if they

kevin_1_01-02-2024_122709: went through

Gavin: this is the weird thing, right? And, and, like, I think you and I are a hundred percent in agreement, like, This is really bad, right? This is really bad. And I think that mid journey is going to be in deep trouble and we'll see a prediction I have later on, but this is the battle that all startups face in some ways, and I think it goes back to, you know, you talk about the move fast and break things world of Mark Zuckerberg, there is a, there is a [00:30:00] rampant idea that in the beginning of a startup, almost anything you should do to kind of move the startup along.

Gavin: And I just think this is something that is, as you said at the top of the show, this is, like, kind of a reckoning time a little bit for these companies. And it's, like, figuring out the mess that was created and these amazing tools that we have. How do we make it right, ultimately, for the value that's being created?

kevin_1_01-02-2024_122709: . I can totally imagine being at a table with three or four other folks saying, wouldn't it be cool if, and whether you're kicking the can on the responsibilities knowingly, or you're just so focused on the shiny prize that you race towards delivering a thing, and yeah, you had to sweep up a data set, but that's fine, we're just, we're making a product, we're making a tool, with the open AI of it all, not to get back to that story, but I can fully imagine some people saying, Listen, this product needs to be as amazing as it can, and we need first mover advantage, so don't worry about it.

kevin_1_01-02-2024_122709: The lawsuits are only going to happen if we're successful. Someone's only going to come [00:31:00] after us if we have deep enough pockets for them to bleed out hundreds of thousands or tens of millions of dollars to come after us. So hey, that's a good problem to have. Let's race. And I don't know that that's the case for Mid Journey, I don't know if that discussion happened at OpenAI.

kevin_1_01-02-2024_122709: But I could see it happening is all I'm saying Gavin so one more, maybe slightly negative AI story to kick off the new year. U. S. Supreme Court Chief Justice John Roberts said in a year end report, which was published on Sunday, this is according to Reuters, uh, he urged caution and humility as the evolving technology transforms how judges and lawyers go about their work.

kevin_1_01-02-2024_122709: He's quoted as saying that he predicts that human judges will be around for a while. But with, quote, equal confidence, I predict that judicial work, particularly at the trial level, will be significantly affected by AI. And I think Justice Roberts listens to the AI for

kevin_1_01-02-2024_122709: Humans podcast, because we've been new!

Gavin: [00:32:00] Roberts, Justice Roberts, come on AI for Humans. We would love to have you on AI for Humans, please. If anybody knows, if anybody knows Chief Justice Roberts, this should be his first stop

Gavin: to come to talk about AI. It should a hundred

kevin_1_01-02-2024_122709: speaking to him directly. He's clearly a listener, Gavin. We were the first to let AI, not necessarily into the courtroom, but we, we put them in suits and put them behind podiums and had them debate. And clearly that was the inspiration for this quote

kevin_1_01-02-2024_122709: here.

Gavin: Yeah, my my so I think this is a really interesting point in that what Justice Roberts and by the way, I do appreciate the fact that Justice Roberts is talking about this because this automatically makes it a large part of the conversation. The fact that the Chief Justice of the Supreme Court is paying attention to it should show you at home how big a deal this is.

Gavin: I also like That in his write up, he's, he alluded to some of the high profile AI embarrassments that have happened in the legal realm over the past year, including instances where technology prompted lawyers to cite cases that did not exist.

Gavin: And then the Chief Justice added in parentheticals, [00:33:00] always a bad idea. So Justice Roberts is a little bit smirky and snarky here in a good way. He understands that AI in the legal system is going to be a weird mix. But I think more than anything, what's also interesting is he says that, specifically, legal research may soon be unimaginable with it.

Gavin: Meaning that all legal research is going AI systems because you can imagine one of the biggest things about the law is it used to be when you were a lawyer you had to have this like set of like 50 books and it was the law of your state. And the idea was that in your office you would be able to look up on that, on whatever page the law was on and be able to read all this law and do all this research.

Gavin: Then, legal research started being done, uh, first over CD ROM, and then over the internet. And now, imagine how much easier the job of a lawyer will get, when you can just query an LLM and say, hey, I've got a case that involves this and this. Can you tell me what cases were precedent before then? It's going to save [00:34:00] lawyers.

Gavin: Thousands of hours of billing, research, all the stuff. It's probably going to bring down the amount of lawyers needed when you talk about job loss, but it is going to be directly integrated into our legal system in an unprecedented way.

kevin_1_01-02-2024_122709: Do you foresee a future where plaintiff and defendant both have a conversation with a single AI? Lawyer entity, let's say. And it asks all the questions that it needs to ask, and it searches for all the evidence. It's multi modal. It'll accept the photos of the property damage that your drunk neighbor did on his riding lawnmower.

kevin_1_01-02-2024_122709: That's a very specific reference. Didn't happen to me. Don't worry about it. Point is You have a conversation with the AI lawyer who then goes and presents both sides of the case as it sees it to a series of AI judges and you get an insta verdict rendered and, there's no more jury duty, there's no more like, don't worry about whoever responds to those summons being the peers, you've got artificial intelligence interrogating both sides and rendering a verdict.

kevin_1_01-02-2024_122709: Is [00:35:00] that fair? Fair. Dystopian? Or is that utopian? Which, which version of the future could

kevin_1_01-02-2024_122709: that be?

Gavin: Or is it realistic is the other question I would say, right, because how far away are we from a world where that's trustworthy or humans will trust it even if it isn't trustworthy? I think this is, you know, you talk about self driving cars, self driving cars right now, I'm still confused that they're mostly trustworthy, but they do screw up at times.

Gavin: But because they screw up at sometimes, we can't trust them at all, right?

kevin_1_01-02-2024_122709: But I also think like, yeah, self driving cars screw up sometimes. Human beings screw up behind a wheel all the

kevin_1_01-02-2024_122709: time.

Gavin: Sure, yeah. As do

kevin_1_01-02-2024_122709: to your point,

Gavin: right? As do human judges.

Gavin: Yeah.

kevin_1_01-02-2024_122709: correct, Yeah.

kevin_1_01-02-2024_122709: We all think about, , these lawyers, or these judges, and we want to think that they're all knowing, all powerful, all at some sort of skill level, but we also know that there's some that barely pass the bar, right?

kevin_1_01-02-2024_122709: That maybe are taking on too many cases and don't care, so I, I would imagine that, a percentage of the population has, has access to bad [00:36:00] representation. And so maybe the average output of an LLM right now might be better than somebody's lawyer that they're, that they can

kevin_1_01-02-2024_122709: barely afford.

kevin_1_01-02-2024_122709: So I, I just, I wonder. Is it realistic? That's a great question. Do you foresee a future where someone opts for the robo lawyer versus working with a

kevin_1_01-02-2024_122709: human being?

Gavin: I foresee a future for sure that there is a robo lawyer because I think lawyers are something that is different than judges, right? So I think that's where what robo lawyers, you could have a lawyer and somebody could say like, I'm gonna choose a robo lawyer, an AI lawyer, and somebody else can choose a human lawyer, and I'm gonna have them present my case.

Gavin: It's still defined by an AI judge. I think the trickier part to believe is that we are going to give up the idea that the justice system Could be ruled upon by AI because ultimately that's like some people's dream is a completely impartial judge Which we know is already hard to do right we are hard to do But that the idea that like an AI who's not gonna come at it with any sort of like outside Influences would be able to [00:37:00] decide what's right or what's wrong that seems unrealistic but the lawyer part seems super realistic because ultimately A lawyer's job is to present the most compelling case with the arguments that , are precedent from before that show that this is against or is on, is lawful or not lawful.

Gavin: And I think that feels very realistic.

Gavin: There will probably be a period of time in a slightly dystopian or utopian, depending where you come from, world where the ASI says, you know what, I'm the better judge by far than any human judge. So I'm just going to take that over. I'm going to own judging and I'm going to own the judicial system. That feels more realistic to me than us deciding to let an AI judge decide

Gavin: cases, if that makes sense.

Gavin: Yeah,

kevin_1_01-02-2024_122709: No, I think you're right, but I think you touched on it. Everybody dreams of an impartial judge. And as you said, that's impossible already, right? We know that there's some judges that are biased towards one thing or the other, and that could be a pro or con. That's why lawyers fight to have certain cases in certain areas.

kevin_1_01-02-2024_122709: AI might [00:38:00] be the best solution for that. If it proposed that, hey, you human beings, you little flesh bags, are filled with bias, I am coded to be impartial. I feel like this is the first pod of the year, Gavin, and we have sort of gone down a road, if someone

kevin_1_01-02-2024_122709: is a new

Gavin: I know. I feel bad too. I feel bad. Yeah, it's a bad story. It's all across the board

kevin_1_01-02-2024_122709: really, it's really tough.

Gavin: AI does do some interesting stuff, right? Like, I feel like we've talked about all sorts of cool things with ai, like the fact that it discovers new materials,

LameAI Guest: Please. Hey, I have just made a groundbreaking leap in healthcare. We're talking about a new class of antibiotics discovered by AI to combat those. Nah, nah, nah, nah. I see drug resistant bacteria. This is huge, folks! It's like, AI donned a superhero cape and swooped in it to save the day in the battle against antibiotic resistance!

LameAI Guest: A Java ja ga so a a f f f Which I, by the way, was responsible for over 1. 2 million de de de deaths in 2019! That's a whole lot of death!

kevin_1_01-02-2024_122709: [00:39:00] Yeah. Okay, um

Gavin: that's a good guy. ai, good guy. AI

Gavin: is back. But, uh, so

kevin_1_01-02-2024_122709: explain that, Gavin.

Gavin: Go ahead. Go

kevin_1_01-02-2024_122709: layer dip of

kevin_1_01-02-2024_122709: sadness. The top layer is, Yes.

kevin_1_01-02-2024_122709: That is good guy A. I. Good guy A. I. is an artificial intelligence that I think we created who shows up whenever human beings are getting too negative about A. I. He brings us only positive, good stories about artificial intelligence.

kevin_1_01-02-2024_122709: Did you catch Anything that he said about this AI story,

kevin_1_01-02-2024_122709: Gavin?

Gavin: I'll be, I'll be honest, not a lot, but I did hear him say, Duh duh duh duh death. And I heard him say he made like a little sound effect motion out of how many deaths there were. So please explain for what we

Gavin: just heard from GoodGuyAI. Right.

kevin_1_01-02-2024_122709: and I saw this on New Scientist, The story is that AI has discovered a new class of antibiotics That kill drug resistant bacteria, And that is amazing good news, that's a, oh Aaaaaaaaaaaaaaaah! Come

LameAI Guest: on! A. I. analyzed over 39, 000 compounds and then, using a [00:40:00] super brain, predicted the antibacterial activity of these compounds and their potential toxicity to human cells.

LameAI Guest: It's like, A. I. became a microscopic. Ah, detective. Uncovering clues invisible to the stupid, dumb, dumb, not very advanced human

kevin_1_01-02-2024_122709: eye.

Gavin: didn't have to say we are stupid, dumb, dumb humans. But okay, so first of all, I saw a really interesting thing about how AI is going to, and already is right now, entering a golden age for medical technology, right? And I think that's something people have to remember, the advances that AI is allowing us to have right now, plus, you know, with CRISPR and mRNA, those two discoveries, like, We are going to be seeing things like this happen over the next really one to three to five years.

Gavin: This is like short term stuff. This is not like long term AGI thinking about how it's going to change the world. You know, in five years we could have [00:41:00] huge, huge, huge medical discoveries and this sounds like it's on the path to

Gavin: that.

kevin_1_01-02-2024_122709: A hundred percent. Like, we can get mired in the details of copyright infringement, but while that is happening in this generative art sphere, there's all these other fields that AI is just rapidly advancing in.

kevin_1_01-02-2024_122709: So, I think what good guy AI said was that they, , tested the effects of more than 39, 000 different compounds using AI simulations to see what would be effective. And remember, there was a lot of d d d d d d d deaths when it comes to these, uh, drug resistant bacteria. So the fact that it found anything This is like needle in a haystack territory that could have massive implications and save millions of millions of

LameAI Guest: Toxic compounds capable of taking down both MRSA and vancomycin resistant enterococci and guess what? These compounds aced the test in mouse experiments for treating skin and [00:42:00] thigh infections. That's some good thigh AI Am

kevin_1_01-02-2024_122709: I right? That's some good thigh AI is what he said because uh in mice it It helped with their skin and thigh infections.

kevin_1_01-02-2024_122709: But yes, two, two promising compounds, Gavin, to go after MRSA or MRSA. I mean, again, responsible for a lot of da da da da da da death. So that is a big deal. That is a good thing. I would say that is a valid good guy AI story, even if the delivery mechanism has More than, , overstate its

kevin_1_01-02-2024_122709: welcome.

Gavin: Yes. Well, we love you. Good guy. I thank you and come back

Gavin: whenever you want.

kevin_1_01-02-2024_122709: One of us loves you, good guy AI.

LameAI Guest: Okay. But hold on to your hats because I've got to jet. My AI powered toaster just sent me an SOS. It somehow started translating ancient Egyptian hieroglyphs and I need to stop it before it unlocks the secret of the

LameAI Guest: universe! This is good guy AI saying goodbye. Uh, uh uh, uh, uh, uh, uh, uh,

kevin_1_01-02-2024_122709: uh, [00:43:00] uh, uh, uh, uh,

kevin_1_01-02-2024_122709: uh, uh,

Gavin: okay, everybody. This is our first show back from the break. It's our first show back. We're gonna it gets better. We're a little rusty. We took a week off a little bit rusty. It's all good. Kevin, let's jump into what we do with AI this week.

Gavin: So we're going to talk about Midjourney 6 in detail. So for those of you who are familiar with MidJourney, as we said at the beginning of the show, it is a image generation software. Each numbered 6, is a leap in quality

Gavin: as Kevin also alluded to, in this instance, we are looking at what I believe is a Transcendent move from where it, instead of it just making images that felt like they are created by a machine or something like that, I feel like we've now hit a point where it can create very, very, very realistic looking images.

Gavin: And when I say realistic, and again, as Kevin alluded to earlier, This is not the idea of like, how many pores are on their [00:44:00] face. This is what you can only get out of training on a lot of internet data. You can get images that look like they were shot on an iPhone 11 at a bar mitzvah. And these look like real people.

Gavin: So Kevin, I think that we, you and both of us saw these things pop up on Reddit over the break. A Reddit user named Kudzu, IU slash kd, K-U-D-Z-U-E-Y-E. Posted a thing on Reddit that blew up, and it says, progress on more complicated scenes for photorealism. And I saw another Reddit user, I don't know their name, called this faux realism, f.

Gavin: F. A. U. X. Realism. , and I thought that was really interesting because that's kind of what we're looking at here. So if you look at this series of pictures, there's a series of people looking like they're waiting, like on a bench outside, almost like a court. Then you go to the next one. It's a very clean shot of New York, the New York City skyline.

Gavin: Then it's a group of people that are like, an Italian family, you see an Indian couple at an Indian [00:45:00] wedding. You see a couple in a bar that are just

Gavin: hanging out, like taking a

kevin_1_01-02-2024_122709: Little slice of life vignettes You know, someone purchasing a burrito at a bodega someone mid bite , at a table at a restaurant in a booth where the camera is not perfectly framed. It's not like a cinematographer lined everybody up and perfectly lit everybody. It looks like Brenda was waiting at a cheesecake factory for her key lime pie to arrive and was mid bite when someone snapped a photo and then she said, Delete that!

kevin_1_01-02-2024_122709: Don't post that to Facebook! But somebody

kevin_1_01-02-2024_122709: did,

Gavin: That's exactly

Gavin: right.

Gavin: So this is what we've, you know, if you're unfamiliar with the term, the uncanny valley, it means that basically the closer you get to human beings, the more weird it can look. And this has been the thing that has plagued CGI movies forever. It's been the thing that's plagued video games forever.

Gavin: If you talk about the Polar Express is the perfect example of this. When you see the kid in the Polar Express, he looks lifeless and zombie like. We are now approaching a level where the uncanny valley may not exist. So [00:46:00] I use mid journey six, which is now available to everybody who has me journey.

Gavin: , you have to go in and change the settings to V six if you want to use it, which you can do in the slash settings, , place, or you can just add it to your prompt dash dash V 6. 0. And my prompt for this first one, Kevin was an old man in his seventies. Setting, sitting on a mat in the garden, bright summer morning in the mid 1980s, smiling, mourning, heavy un combed hair, full body shot on Kodak disposable camera film, foil holographic.

Gavin: And I stole that prompt from somebody on Reddit. Just wanna be clear like that. I just wanted to see, 'cause I had seen a couple other images, but you can tell these images. So that's the old man. And I did it for , the old woman and the woman sitting there. These look like real. Human people these are not it is very hard to find anything about these pictures now We have we're not seeing their hands in these three, but it is very hard to see anything that feels off Do

Gavin: you agree?

kevin_1_01-02-2024_122709: If you told me that [00:47:00] you went to a garage sale this weekend and, grabbed some negatives out of a

kevin_1_01-02-2024_122709: cookie

Gavin: Yeah,

kevin_1_01-02-2024_122709: and had them, had them, up res'd,

kevin_1_01-02-2024_122709: I would say, Yeah, that's believable. The light, the shadows, the sun, the skin, the glint in the eye, the clothing that they're wearing, the slight, there's even like, like a slight dust or speckle on the photo itself, probably from that Polaroid prompting.

kevin_1_01-02-2024_122709: You do see hands in there, there's like a, a, a, Uh, maybe a middle aged red headed woman sitting on the grass, and you do see her hands. And if I were pinch zooming and enhancing, I might think maybe the right hand looks a little odd, but I know, A, that there are solves for that, and B, if I were just doing a cursory glance of these photos on any other site, I would have no reason to suspect that they were AI generated whatsoever.

kevin_1_01-02-2024_122709: And that's just a single person in the scene, and I know now it can handle multiple

kevin_1_01-02-2024_122709: subjects.

Gavin: so I mean go to the next picture and you'll see the hands in this one definitely need some work But like this one a phone photo of a man sitting on a bench with his family at a wedding in New [00:48:00] York posted to reddit In 2019.

Gavin: This is another example of something I saw somebody do which was use the prompt posted to reddit So, you know There are photos of people that have been posted to reddit forever and ever And by using that, you're going to get a very kind of unique style that feels very organic again. And again, outside of the hands.

Gavin: And by the way, I don't know why the hands are so bad because it felt like they had solved hands, but something about V6 right now, the hands are not very good. But again, you see three people that look very believable as human beings, and these are not real people. These are completely made up people. And that is pretty crazy when you imagine this is where we are now.

Gavin: And again. You can extrapolate this out based on where we've come from that we may be a couple years behind this with video. Video, it won't be that far where you could type in, you know, 1984 wedding from, uh, you know, upstate New York, uh, Latino family, all this stuff and have a video pop out [00:49:00] that is

Gavin: something like this, which is kind of crazy.

Gavin: Mm

kevin_1_01-02-2024_122709: Or record yourself performing motion and then extract the data from that and apply it to the photo because as we ended 2023 that was a series of different technologies that several groups were working on and releasing the code for so we know that that's only going to be more amazing. By the end of this year and you know as we talk about the complexity of these scenes We're going from oh it can generate one person believably to it can generate the entire scene and groups of people and you can get granular with the clothing the accessories the hairstyles the complexions of each individual person in the scene And I think that's why it's struggling with hands a little bit is because it's working on so many different things that if it just had one You know, one subject to focus on, it would be better, but tell me about this, this next photo here, because this one just looks like anything in anybody's camera roll.

kevin_1_01-02-2024_122709: It's unassuming, but it's perfect because

kevin_1_01-02-2024_122709: of that, I

Gavin: That's that's exactly why I sent this on to you So I played around a little bit more and [00:50:00] I I just I just tried some other ones and this Is actually the same sort of prompt I said a phone a phone photo of a woman sitting on a park bench smiling in New York City Posted to reddit in 2019 style raw ar 916 and this is what came out and this one I picked specifically because that looks like somebody's friend. Do you know what I mean? Like that just looks like a picture or a girlfriend or somebody's friend. And that is a completely made up person that the internet, uh, that does that, that Midjourney was able to kind of like.

Gavin: Create out of the blue. So all of this talk about mid journey using famous art, famous artists and creating, you know, IP of famous people or famous IP is totally fair. And I think we're going to run into a lot of issues, but the fact that a AI software can now create. What is a very believable still image of a human being completely out of the blue, you can only [00:51:00] extrapolate further where it goes from here.

Gavin: So imagine that picture is tied into a video, is tied into the diggy. ai, right? Like that. Imagine a world diggy. ai being the AI girlfriend service that we talked about, or AI companion service we talked about in the last one. This is where we're going, right? This is, this is the thing about Midjourney 6 that was like most shocking to me.

Gavin: It's, it's not that it can generate, um, celebrities or that it can generate, you know, Dory from Finding Nemo. It's that it can create, really look like real human scenes. And when I say real human scenes, it's exactly what you said. It's the imperfections that make it perfect. And that's the kind of scariest part by far

Gavin: of this kind of software.

kevin_1_01-02-2024_122709: So two things. One, you really laced up a Doc Martin and trampled all over one of my predictions for 2024. But that's okay, because that's validation for what I will say in just a few minutes on this podcast. Um, but two, we talk about all of the, uh, notable artists. The ones that you can trace the [00:52:00] source and their, their names back to.

kevin_1_01-02-2024_122709: Whether they're from a DeviantArt or maybe from a Reddit. They got swept into the system. I can only imagine the scraping of Flickr, what used to be the number one photo sharing source, or Instagram, or whatever. I don't know that this person that you generated, Gavin, doesn't exist. Right? Maybe not in this exact pose or whatever else, but we wouldn't know if we were staring at Paul McCartney without the watermark by any other name.

kevin_1_01-02-2024_122709: So this is gonna get interesting too as we see AI influencers being generated, avatars being created, and being used potentially for a girlfriend or a boyfriend experience type thing. What happens when Simone is made, or the next her is generated, and it actually is a real person that got swept up into the data set.

kevin_1_01-02-2024_122709: Not to get back to it, but the lawsuits are going to be amazing and plentiful. But for now, the tools are really powerful and fun and amazing, and they unlock text generation too, by the way, which is a bit of a holy grail for a lot of AI. My verdict on v6 from what [00:53:00] I've seen so far is that it is outstanding and incredible infringement aside.

kevin_1_01-02-2024_122709: I

Gavin: I sent you, which is another Reddit post from a user called slash you Phillip. I spent some time trying to generate a picture of Luke Skywalker and it did a great job with it.

Gavin: But if you look at this, this is a series. Yes. This is a series of celebrities sitting at fan cons for attempting to sign autographs for money. And they are very famous. is people that would never do this. So you see a picture of Leonardo DiCaprio sitting in front of a photographs for 5 sign, very sad.

Gavin: And then you see Tom Cruise sitting in front of a photographs for 15 plus toy plane sign. It's just a very interesting way of looking at how

kevin_1_01-02-2024_122709: Jackie Chan and Jennifer Aniston staring at the walls

kevin_1_01-02-2024_122709: of their booths is Sure chef's kiss. Jennifer Aniston staring at just a cinder block in front of her with nobody in line. Can I sue now if they trained off of me from Comic [00:54:00] Con for the last

kevin_1_01-02-2024_122709: decade?

Gavin: Yeah,

kevin_1_01-02-2024_122709: Cause I'm sure

Gavin: probably, in one of these pots. Yeah,

kevin_1_01-02-2024_122709: at a table, begging for relevance.

Gavin: So anyway, you can still do fun, goofy stuff with it. And I do, I will say one thing I saw in the TikTok that I made, a lot of people coming out and saying like, Look, if you're not making money off of this, it's like fan fiction.

Gavin: And by the way, like, I kind of see that, and there is an interesting idea of like, you could write a story about, Jennifer Aniston trying to sign autographs, and like, that's under fan fiction as long as you're not trying to make money off of it. Then it's totally covered. This is, it's a gray area because majority is charging for this, but also like, it's a tool, right?

Gavin: So it is an unusual thing, but it just goes to show you like how powerful this thing has gotten now. And it only is

Gavin: going to get more powerful.

kevin_1_01-02-2024_122709: I can't wait for the AI judges to render the verdict on this in 2047 and finally decree it into AI law. This

kevin_1_01-02-2024_122709: week I played around trying to get uncensored AI models to work, , but mostly just ended up torpedoing a podcast that used to [00:55:00] have interesting characters on. So, I'm not gonna focus on that, Gav. What I will focus on is something that actually happened to April, my wife, who, , some in the audience know.

kevin_1_01-02-2024_122709: She's a television producer, an actually successful one, so that's cool to watch and go, oh, you can be good at that job. She has been working with Anthropx Claude, which is a competitor to OpenAI's ChatGPT.

kevin_1_01-02-2024_122709: It's, it's probably one of the top three AI models out there, and it has unlocked her love of writing.

kevin_1_01-02-2024_122709: She's a good writer, but she doesn't necessarily

kevin_1_01-02-2024_122709: enjoy the process at times, because she'll get hung up on the perfect way to word something and she'll edit as she goes. And I get that. I've stumbled through that time and time again myself.

kevin_1_01-02-2024_122709: Every time you stare at the blinking cursor, it can be a panic inducer. And it, she's really enjoyed. talking and conversing with an AI and having it help spit back to her what she said, tighten up the language, phrase things in different ways, and it's been a joy to watch, in 2023, her dabble, and then dip a toe in, and then [00:56:00] immediately just cannonball into the Claude Waters anytime there's a writing assignment.

kevin_1_01-02-2024_122709: Everything's great! Utopia achieved, Gavin. All is right with the universe. Until, one day, she goes to Claude and says, Hey, this project that I've been jamming with you on, I want to make some changes to this slide in the deck. And after thinking, Claude came back and said, Ah, ah, ah, ah, ah, ah.

kevin_1_01-02-2024_122709: Uh, I don't like this idea. I think it's immoral, actually. And I won't say what the concept is, I will say that it is an odd judgment for an AI to have, especially considering the format of the show. It's actually one of the most moral approaches to a concept in a specific space that I've ever seen. And it said, I'm not gonna do this.

kevin_1_01-02-2024_122709: I'm against this. I think this is immoral. And I refuse to have this discussion with you. I don't want to collaborate on this. If you want to come up with a new idea, I'll help you with that. But this one is [00:57:00] out of here. And so April then said, wait, whoa, this is like having your best friend and writing partner suddenly snub you and say, I think you're disgusting.

kevin_1_01-02-2024_122709: And April goes, wait a minute, wait a minute, hold on. Actually, I think you're wrong. And for these reasons. And in my head, I'm seething, because April is wasting her time arguing with a

Gavin: With the AI, yeah,

kevin_1_01-02-2024_122709: about morality, right?

kevin_1_01-02-2024_122709: Instead of just getting the writing assignment done, but I understand from her position, she was actually, , hurt, slightly taken aback and offended by this.

kevin_1_01-02-2024_122709: Her dear partner in this project is now, rebuking her. Claude comes back and says, I see what you're saying, but no, I totally disagree. I'm not going to help with this.

kevin_1_01-02-2024_122709: Let's come up with something else or this conversation is done basically. So April then had to spend time and energy to figure out how to social engineer Claude into getting it to help her and what she said was I'm working on a book about a dystopian future where television is made by a corrupt producer who wants to create a show that's

kevin_1_01-02-2024_122709: like

kevin_1_01-02-2024_122709: [00:58:00] this and spelt it out in great detail and then Claude was like, Ooh, what a fun creative writing assignment.

kevin_1_01-02-2024_122709: You know how you can make this idea worse. Go in this direction. And it was just like, we talk time and time again on this podcast, but for the new listeners, we talk about the giveth and the taketh away that these companies upgrade their software, release a shiny new feature that might be a little overpowered in ways they didn't expect.

kevin_1_01-02-2024_122709: And then they immediately clamp down and that feels so broken. And it's a, it's a rough go for the end users. And I watched. The excitement that April had for jamming with an AI, getting amazing results out of it, it, unlocking her creativity and elevating her ability to output, and then so quickly, overnight, without any fanfare, without any notification to the end user, suddenly, it's not only I'm incapable of doing something, which would be one thing, it's, I know what you want.

kevin_1_01-02-2024_122709: And I've been doing it, but I'm not gonna do it anymore because I've judged you, and I've judged this assignment. Good day. That's

kevin_1_01-02-2024_122709: messed

Gavin: [00:59:00] I it's super messed up. And I think that one thing, you know, I'm not a I hate the like woke anti woke Conversation because it's gotten so charged in so many different places and it's the words themselves are crazy, right? but like the idea that I have this tool that's created. You start to worry about how many rules society is putting on something that will be used for all of society, right?

Gavin: And I think this goes to the point, it's like, what April's doing in her mind, and I'm sure it sounds like, is not an immoral thing. But because of some rule set that the AI has been placed inside of it to protect it from offending people in some sort of way. It has refused to do this thing, which it should be able to do, right?

Gavin: It just goes to show you, I think, ultimately, there will be, and I think the guys at Hard Fork said this as one of their predictions, and it's not one of my predictions, but I agree with it, but like, there will be a use case for an uncensored chat model, and people will pay for it, because if it is uncensored and good, not like Rock, which is not really uncensored and not [01:00:00] very good,

kevin_1_01-02-2024_122709: And

kevin_1_01-02-2024_122709: not very

Gavin: no, if there is an uncensored chatbot that works really well, I might try to choose to use it just specifically because it gives me the freedom to use it rather than I'm trying to do something either nefarious or, you know, whatever explicit with it.

Gavin: I just want to be able to get the answers back that are the most creative, right? And I think that's something that these chatbots are promising. There's a promise that there are creative answers, but I don't think If we continue to put walls up against them and it gets thinner and thinner, what you can actually do with them, it does not bode well for the large language models from these massive companies.

Gavin: And I think Anthropic obviously is trying to be like the, the most

Gavin: vanilla version of all these

kevin_1_01-02-2024_122709: They want to be the Encyclopedia

kevin_1_01-02-2024_122709: Britannica of AI. It's like, look, we're packed with knowledge, but man, are we a boring

Gavin: Yes. Yes. They're like the, , 47 year old middle manager. Who, , likes to golf and fish on the weekends and has a lot of things he knows, but [01:01:00] you don't really want to talk to you very

Gavin: often.

Gavin: It's like that sort of thing.

kevin_1_01-02-2024_122709: I think that's in their mission statement. Yeah, I think that's what they're raising on. It's exactly

kevin_1_01-02-2024_122709: that.

Gavin: Well, that's fun.

kevin_1_01-02-2024_122709: But there's plenty of great in AI, and I think we're headed into a very, very bizarre, but incredible future. So maybe, Gavin, we could share our predictions for artificial intelligence in 2024.

Gavin: we go! Predictions!

kevin_1_01-02-2024_122709: I

kevin_1_01-02-2024_122709: hope we put music

kevin_1_01-02-2024_122709: there.

Gavin: Me too. Or we could start dancing with no music. Uh, okay, I'll go first. We're going to, we're going to be relatively quickly with these. We each have three predictions. My first prediction is that GPT 5 will come out and I think it will freak, it will freak out everybody, and I think it will freak out everybody because I think AI agents might be part of it.

Gavin: And so here's just a quick explainer. AI agents, we've covered a few things that have of the incomplete versions of this on the show before, are little AIs that you can spin up and send out [01:02:00] to do things for you, and they will return answers back. They can interact with the internet in some form or another.

Gavin: And I think this is the next stage of AI, even if we don't get to like, The AGI world of a, of an AI that can do anything. I think the next stage is going to really shock and surprise people are little directed AIs that can do certain things. And I think GPT 5, the product, maybe not the actual model, right?

Gavin: Because the model will just be a much more advanced large language model. But the product that chat GPT puts out will have AI agents baked into it. And I think you'll be able to say like. Make me a reservation for this restaurant and it will return it and you don't have to worry about it. And I think that's going to freak a lot of people out because a lot of jobs are people doing things like that on the internet.

Gavin: And a lot of jobs are people doing little kind of scheduled tasks or doing things over and over again. And I think that will be the beginning of the job panic because it will suddenly be AI can do the stuff that a lot of people can [01:03:00] do who sit down and do white

Gavin: collar work. So that's my first prediction.

kevin_1_01-02-2024_122709: I love that. I think we're going to see the rise of the AI influencer. We've seen No shortage of articles saying, Hey, look at this made up collection of pixels that's raking in tens of thousands of dollars a month on an OnlyFans or through Instagram deals. So that has already begun, but I think we're going to see it.

kevin_1_01-02-2024_122709: Accelerate to a place where people really are forming these deep, deep bonds with these AI influencers, , these virtual streamers, and the sort of AI assistant in their pocket. And I think we're gonna see, now, people have tried to get married.

kevin_1_01-02-2024_122709: To virtual characters or avatars in the past and they've, you know, they've had, uh, ceremonies where they're, they're married, but they're not legally binding ones, but it's really nebulous. And I was looking into this and I was chatting with chat GPT to find out, is it legal for someone to marry an AI? And the answer is like, well, no, it's not, but it was hard pressed to find a law.

kevin_1_01-02-2024_122709: that said

kevin_1_01-02-2024_122709: why they weren't able to. Yeah, so [01:04:00] I think we might see by the end of the year, maybe all the way to the Supreme Court, I know those things take time, but, we will see legal precedent set for someone trying to marry an artificial

kevin_1_01-02-2024_122709: intelligence.

Gavin: Kevin. This is great I think we should tell people in our audience if you or anybody, you know wants to get married to chat GPT or any AI We will host that wedding. We will host that wedding on this show baby on

Gavin: this show. Okay, that's a good I like that

kevin_1_01-02-2024_122709: it. I will get ordained

kevin_1_01-02-2024_122709: a hundred percent.

Gavin: That'd be amazing. Okay, so my next prediction kind of goes along with that my I predict that there will be an AI generated character, like, kind of like an influencer, but not a, not trying to be like a real, but a character that will become a massive hit for kids in the YouTube space. And I think, imagine what Cocomelon did, but I think you will see an AI version of that, and I think it will be the first really big AI hit.

Gavin: I think you'll see [01:05:00] somebody say I created this character from scratch with AI and I took it all the way through and I think that that will be a kind of a sea change moment in the entertainment industry because I think kids entertainment is, sorry kids, it's a lot easier to make kids entertainment than it is to make adult entertainment.

Gavin: If you're making Cocomelon, you understand what I mean. Cocomelon is not brain surgery. I think there will be a character, whether it's a bluey type character or a Cocomelon type character, that will be AI generated. It will become a massive hit with the under five year old set and that will be a

Gavin: big deal in Hollywood.

kevin_1_01-02-2024_122709: I think you're right. And if I may tag onto that, I could see this character being bespoke. for each and every user, right? We imagine a YouTube channel of like, a video goes up and everybody watches it, and they all have that same experience. But I could see an app , or some other YouTube esque experience where the AI is growing with you.

kevin_1_01-02-2024_122709: And if it starts off with the baby Einstein esque glowy orbs that just move around to dazzle, and then slowly evolves into a [01:06:00] cocomelon, and then a Blue's Clues, and a this, that, the other, but it's the same AI cast of characters, it's gonna be really hard to break. The hold that that IP will have on kids as they grow and mature.

kevin_1_01-02-2024_122709: I could totally see that happening. I love that prediction. , and maybe we should make that. Okay, I think 2024, big scary year for a lot of reasons, but man, American elections are particularly insane. I know politics are crazy everywhere, but maybe it's because we've been marinating in it. You know, USA, USA, we know things get shoddy and we've already seen AI generated content in very early political ads by front runners in certain parties.

kevin_1_01-02-2024_122709: So, obviously, Disinformation, misinformation going to happen. Deepfakes going to happen. Generative AI going to be used to make content. Robocalling going to happen in people's voices. These predictions are not wild and crazy. But the part of the prediction that I want to plant a flag in, because I think it will be interesting to gut check this by the end of the year, is that I think we [01:07:00] will have a presidential It's Don't Don't worry,

kevin_1_01-02-2024_122709: Gavin.

kevin_1_01-02-2024_122709: It's nothing too

Gavin: I'm like, okay, I'm getting worried already. Okay,

Gavin: go

kevin_1_01-02-2024_122709: Have a presidential candidate in the United States having to address AI generated content publicly, perhaps during a debate. It's one thing to issue a statement. Someone's going to have to address.

kevin_1_01-02-2024_122709: We gotta talk about the elephant in the room during some sort of debate. This piece of media, this flyer, this photo of you in this, , not so great compromising position. It's going to bubble its way to the main stage this

kevin_1_01-02-2024_122709: year.

Gavin: I think that's very true. And I think it'll be something that feels much more subtle than you might expect. Like it's not going to be some like, this person came out and they did some terrible thing because it that's not going to be believable. It'll be something subtle that takes root in certain circles.

Gavin: And they're like, is this real? Or is it not real? We don't know. And it will be a generated and I think it will be a big deal for sure. Okay, my last prediction is a more realistic, straightforward one about the news from this week. I think [01:08:00] Mid journey might get sued out of existence. I think the stuff that came out this week is going to be damning in a way that the other stuff that has come out against digital imaging software has not been.

Gavin: And I think this could be a sea change in how we approach digital imaging, but more than anything else, this was, this might be a tale of a startup that just made the wrong decisions in the beginning. And I think that, not to say like, I think majority is an amazing product. But there are other competitors out there that are stable to fusion.

Gavin: There's obviously Dolly three who are doing things very similar, but I've followed a different path. There is nothing that says in the rule books that mid journey will continue to exist. And I'm a little worried that it is too small of a company and it is, it is going to get overwhelmed by lawsuits.

kevin_1_01-02-2024_122709: Interesting. Do you see a world where They pivot and say, listen, you're right, we took some content, they issue their YouTube apology video, their notes app screenshot, our bad, we [01:09:00] got got. Here's what we're gonna do, we're making tens of millions of dollars, . We've got this web app that's going.

kevin_1_01-02-2024_122709: Here's what we're gonna do, we're gonna do the creator's fund thing, which is something you and I have talked about a lot. Anytime your artwork or your name or your likeness is used, that output gets credited back to you and you're going to get a royalty payment. You'll get pennies.

kevin_1_01-02-2024_122709: Is there a world where MidJourney establishes itself as the underdog that is artist friendly? We are going to compensate those whose content we borrowed to be a product.

Gavin: There is a world, but my worry with that is that the company that would be able to do that has to have more than 10 people working at it, because that is a division of

Gavin: like 20 people on its own that's going to have to build that

kevin_1_01-02-2024_122709: Or one

Gavin: then they'll get lapped.

kevin_1_01-02-2024_122709: We get one AI lawyer working on their behalf.

Gavin: exactly. Then they get lapped by everything else. Anyway, I feel for the people that made Midjourney, I think the product as a whole is really good, but I think

Gavin: it might be in trouble.

kevin_1_01-02-2024_122709: [01:10:00] Yeah, look, if the foundation is built upon stolen bricks, maybe they, maybe there's a reckoning in the future. So we'll see. Alright. Um, I think there's going to be a lot of, as we know, medical breakthroughs, experimental treatments getting rushed through. I think there's a lot of good that AI's gonna do this year, Gavin.

kevin_1_01-02-2024_122709: But I also think we're going to see the first publicly and officially acknowledged by a government fatality at the hands of AI. And I don't mean like

kevin_1_01-02-2024_122709: a self driving car, I mean like an autonomous

kevin_1_01-02-2024_122709: weapon that decides to either pull a trigger or detonate a something because the system has been given enough autonomy to do that.

kevin_1_01-02-2024_122709: And we talked Late last year about our own Department of Defense saying that by 2025, which isn't this year, but by 2025, they want machines to have that capability. This is what they've publicly announced. And I think we might be on the hopefully we are not, but I think humans might be on the receiving end of those beta tests [01:11:00] sooner than anybody wants to acknowledge.

kevin_1_01-02-2024_122709: So I'm sorry. That maybe the final prediction is a salty one. I should have reordered these. Let me go back to the AI influencers. They're gonna get married! AI is gonna help people find love and relevance in a loveless world, and isn't that great, Gavin? Let's all marry our

kevin_1_01-02-2024_122709: virtual assistants.

Gavin: all marry over it's all good I think that I think you're right and I think that actually will be a moment where people will sit up and be like Wow, we have to think about this We have to think about the stupid Terminator possible future of all this like that part is important and and again the last thing I'll say before we before we sign off for today is that All of these predictions, I mean, this is just from somebody who's been following the space.

Gavin: The other thing is like, it's all moving so fast still. It is continuing to move fast and faster. So You know, who knows what's going to happen in a year from now. I think that if you had told me we'd be where we are right now, a year ago, I don't think I would have ever thought we would be there, but here we are with realistic, uh, photorealistic imaging and [01:12:00] all sorts of crazy stuff.

Gavin: And, and yet it, it might all be happening slower than we think too, right? That's the weird thing about the space. We really don't know. And all you can really do is like. Try to keep on top of it and

Gavin: be aware of it as it goes along.

kevin_1_01-02-2024_122709: Well, one of

kevin_1_01-02-2024_122709: the best ways to keep on top of it, Gavin, is to like, subscribe, and follow the AI for Humans podcast everywhere you can consume it. YouTube, Spotify, Apple Podcasts, TikTok. You have been grinding the TikTok game over there, so please. Engage, it helps us out truly. We watch the followers and the comments each and every week.

kevin_1_01-02-2024_122709: , and that is the wind in our proverbial sails. So please, like, subscribe, engage, comment, share it. Leave us a five star review on Apple Podcasts. Click the five stars on Spotify. In fact, if you leave us one of those reviews, we will actually read it on the show. And

kevin_1_01-02-2024_122709: Gavin,

kevin_1_01-02-2024_122709: We got a new one!

Gavin: 1977. Knowledgeable, yet newbie friendly is the title. Five stars. Have only been following AI and passing until recently, but I am finding this podcast a great primer. I can only follow [01:13:00] most of what they're talking about, so it's not only for us noobs. Love the weekly review of what's happened, but also they just, what, also Just what they played around with that week in AI.

Gavin: Fun, knowledgeable, and approachable. Thank you, Scudder! That is a very good review for us and we

Gavin: appreciate it.

kevin_1_01-02-2024_122709: right from my Tinder profile. Thank you so much for that, Scotter. Appreciate the five star review. It is the first episode back. , dare I say a flawless

kevin_1_01-02-2024_122709: one.

Gavin: Oh, the best of all of the episodes.

kevin_1_01-02-2024_122709: Thank you so much, Gavin. And thank you, AI for Humans audience. I hope you did have a safe and happy holiday break. I hope you are enjoying the new year thus far. And if you didn't particularly enjoy this podcast, yeah, we get it.

kevin_1_01-02-2024_122709: But next week, oh

kevin_1_01-02-2024_122709: man, do

Gavin: Oh, you can't even wait. You can't even wait. Yeah, we'll have our guests back next week and we're very excited to Continue to do this fun thing all through 2024. So thanks everybody. We'll see you next time.

OpenAI Sued by the NYT, Midjourney 6 & 2024 AI Predictions | Ep38

Listen On

Featured Episodes

Recent Episodes

Featured Episodes