where I basically merge different voices and attach them to a new text and maybe a new tune. Is that different or at the end of the day the same? If I create new music out of something or if I just merge different music styles? Welcome back to another episode of Lutz & Jasper, where we talk about AI, latest trends.
And today we would love to focus a bit more on the creative part that AI is creating right now. But what the heck has happened last week and the week before? Let’s start with the most important question, Lutz. How’s your coffee?
It’s good. I’m recording today from Stockholm. And Stockholm has actually a good barista as well as a good local roaster, which is amazing. So I had already a coffee tasting this morning, five different coffees.
I’m ready to roll. Very good. I had the amazing Cherry Office coffee. I think 10 cups because I didn’t sleep that much tonight.
So we’re ready to roll, ready to rumble. And as you said, a lot of things happened this week. I think Google was pretty happy. AI.
Generative AI. It uses AI to bring AI. AI. Generative AI.
Generative AI. Lutz, how did you like the Google presentation? Super interesting. I mean, Google has been always the company who has driven AI upfront, right?
They were the ones first to bring in the transformers. And it’s a kind of an irony that somebody else makes now the big splash out of it, OpenAI and so on. So they really want to show the world they are here. Yes.
And there is this leaked paper as well. Yeah, we spoke about it, right? Yeah. So what is important to know is that Google launched Palm 2.
And Palm 2 has, you know, essentially is very similar to, from the token links, comparable to ChatGPT. Same context. We talked about token links and how important it is. Google launched something like Copilot.
I know a lot of my Google friends use ChatGPT to actually improve their code, right? So yeah, here we go. Now they have their own Copilot internally. They’re using it.
They used it internally and now they’re launching it. So that’s cool. Do you think they took some time because they actually know how difficult, I mean, you worked at Google, we should maybe be transparent here. But obviously you can’t disclose anything that is secret.
But do you think they took their time because they wanted to launch something properly so they can counter OpenAI or did it just take more time for them? Oh, wow. Yes. So I don’t work at Google anymore.
So I can say this. I actually believe. I believe that Google has an amazing bottom-up culture. However, they have not really done a very good job of launching a lot of new things.
They’re making a lot of money out of search, right? And now they have AI. And where does AI fit? And can we build on something new?
And can we launch it in health or can we launch it somewhere else? I think they didn’t do a good job there in the past. And the like OpenAI and other companies are leaving. But I think they are making a lot of money out of search.
other companies are a little bit stealing here the thunder yeah and there was a certain wtf what has just happened moment at google and their last presentation and say we did it in paris where the guy even didn’t have his phone with him and i mean like the presentation didn’t work i mean it was kind of chaos it it showed google being scared if you look at now io and they they do this so sweet they kind of showed okay you had autocomplete and now people write emails they kind of bring home the point of hey we google have been always at this place we are ready to roll we are ready to take this on that was a way more calm approach so i still think google and innovation is hard because it’s a big company but i do think there are and we talked about that the model participants And we talked about that the model is probably not the most important thing. And we talked about it, that innovation is very much happening on existing products. Google is extremely well positioned. And you’re right, they’re back on stage, but they are still a large and maybe a bit slower company because I think just this week, Anthropic launched their Claude model with 100,000 tokens.
So much, much more they can ingest. It takes a bit of time. I read 22 seconds to actually get something out of the model, but they came from 9,000 tokens to 100,000. So that’s pretty impressive.
So they are back on stage from Google’s side, but then they’re still this fast moving companies. Which is, by the way, I mean, one thing which comes from Google I.O. is you saw that they launched their model Palm 2 based on the data, which goes up to, I think this January or this February, right? So that means their training cycle is actually very much reduced, which comes down to the whole point.
And we talked about this before that open source still wins. How do you simplify your interface? Is the chat ability, I think you have two questions. Let me answer them separately.
How do you simplify your interface? Like if you’re having a startup, how do you simplify your interface? This is what we see everybody slamming an LLM on. And sometimes the interface.
Might become easier because now suddenly you can write. Yeah, we were basically debating why a closed or an open model. So open source model to the public would win. And our conclusion is that the open source is clearly winning.
But if you look now back March, we had 11 large language model heading the market. In April, like is it six, seven? I kind of lost count there, right? Yeah.
to see Google I.O. is, I think, in two ways. It’s A, open source wins. That’s the same thing, this leaked document from Google.
At the end, essentially, and nobody knows 100% if it was leaked deliberately or not, but it says they went through everything that is important to have a defensible mode, so mode like USP and something that can last and create an unfair advantage. And they came to the conclusion that the open source community, the creativity, the way you work with AI needs transparency, needs also the community and fast development, and they can’t defend against that with anything that is closed. I think that’s one part. Which is, by the way, an interesting setup because if you look now on all the regulators around the world who are very keen to understand what AI is doing, and I will be very cautious commenting on this, but.
If you would have regulated cars before we have millions of cars on the roads, you might have stopped the car industry in the tracks. Because they can kill people, that’s for sure. Yes. Exactly, yeah.
So it will be interesting. However, Google is actually well positioned there, right? Because they are not launching in Europe and in Canada initially, but they said we will work with a regulator to launch. And I think also another interesting development, this week was that OpenAI, apparently set by the information and news outlet, is also open sourcing their language model to the public.
So they are also responding like Google to this whole trend of open models. Now for a VC, and then we should probably dig into our topic of today, but for a VC, the interesting development will be, how do we see control, transparency, reporting, and reporting happening? The problem here is that the regulation is not yet clear. So it’s a question, will this come out of MLOps?
That you kind of see this transparency. And we have this, by the way, we had this in the finance industries. I always run this as a case at Cornell with my students that I actually look at using a model to predict whether you underwrite a loan. Regulation, however, says it has to be understandable why I underwrite the loan.
So now we use a model, and then there are companies who help to explain the model so that somebody can underwrite it. We will see something very similar here. And we have the same in the insurance industry, as you know, also probably in the US, but definitely in Germany. So you have to explain all your models.
Yeah, that’s true. And that means in terms of dear founders, there is an opportunity, I’m going to do that. I’m going to do that. We’re definitely a platform will come into play.
How that looks like depends a little bit on how the regulation looks like. And that is still in the making. However, I’m super interested to see how this is building up. Yeah, I agree.
So I wanted to come to a topic that is maybe a bit older, but it’s still happening a lot. And we actually like it because it’s again back to the fact why AI is taking off so fast, because everyone out there listening can have fun with it. So there was this song, It’s called From Ghostwriter. It’s a very weird video of you, but I don’t use TikTok.
Sorry to say. Heart on my sleeve. And it was sung by Drake and The Weeknd. I’m not a fan there, but it was it wasn’t created by them.
So that was actually the first one that went viral. Millions of visitors, listeners. And then it came out that an AI created the song. But how is that possible?
It’s a lean, a lot of flags. I’m going to be able to play if I end up. And she knows. Yes.
How is that? I hate auto tune, but I’m old. So now how is it possible? Right.
I mean, we are talking technically. We can talk technically. Technically means if I have a neural network which encodes music, so it codes an image, takes an image and says in this image, there’s a dog that’s an encoder and then decode and saying, show me a dog. It uses all the weights in.
The neural network to create a dog. And arrange the pixels essentially, right? Say there’s a blue one, there’s a brown one. Yeah.
That dog doesn’t exist. It is how the neural network imagines the dog. Meaning the same thing. If I ask you picture a dog, you have a dog in your mind, but that dog doesn’t exist.
It is created in your mind. Now, music is it’s instead of pixels, which are all happening at the same time, it has waves and tunes and they are in a flow, which is more difficult to create. And that’s the reason why we don’t see yet, big time thing yet so many videos. Video generation, we have Runaway and other companies who are trying to do videos, but music is definitely a tool out there.
Now at the end of the day, if I understand correctly, you would just change the wavelength and the loudness level of loudness, et cetera. But it’s still more complex than just saying the pixel, on the top left, the pixel on the bottom right. Totally. Now, so that is the same way how I can create text as a, with tokens, with context, life is like a box of, and then we know that it’s chocolate.
I can do music and can say, what is the next expected tone to happen? So I can create music in the same setup. And obviously I can write now text in the same way, in the same setup. And I can create an artificial voice to read that text.
And I can create the music. So theoretically I have all the pieces together to create music. And it feels a bit like what you’re describing is when I use mid journey is imagine like same as the dog, imagine a dog, imagine Drake and the weekend singing, but that would create like new things. But this one.
Feels a bit more, like merge the merge comment in mid journey, where I basically merge different voices and attach them to a new text and maybe a new tune. Is that different or at the end of the day, the same, if I create new music out of something, or if I just merge different music styles. That’s an excellent question because I actually believe this is where much of the debate goes wrong in the public. We kind of saying, oh, we have AI and AI is now doing everything.
This is not right. Like, I don’t know. I mean, how do you use, for example, chat GPT or Bert? Well, you can’t use Bert because you’re in Europe, but like, how do you use chat GPT?
I’m currently using it honestly, just because I find it interesting to summarize text and to see if the summary of the summary is right. I’m not using it to write text because I don’t like the writing style. You could prompt the writing style probably in a way that you like it. But if you ask the, the computer to write something for you, it will write very generic.
Yeah, that’s what I mean. In chat GPT for many folks is an intellectual sparring partner. It’s a cross entropy, right? It’s kind of like I say something, let me see how they reformulating what I say, or I have an idea, how can they complement it?
So Party Pooper Lutz says now this Drake song, it wasn’t AI. And now people are like, wait, what? And all haters come talk to him. Yeah, it was Drake having a Drake style.
And by the way, Drake very publicly talked about his own production capability. He’s not in his own songs. He has a team. So now somebody used instead of it, like Drake goes to his team, he breathes his team.
He has a certain style and his team creates content and then he plays it in. And I mean, Drake, I mean, look at how many songs he produces, right? It is like, it’s not my music, but so like, but it’s more the production chain of music, right? A hundred percent.
I mean, it’s when you’ve seen the video with Dr. Dre and Jimmy Iovine, Dr. Dre has produced Eminem, right? Eminem didn’t produce the music himself.
So now we have AI helping in the same level, Drake would utilize his team to help him. And the person who created the song, Ghostwriter, I think the brilliance here is not, there wasn’t an AI where somebody typed in, I want to have a song from Drake and the weekend and it’s going to be awesome. And the music came out. No, I think there was a lot of human decision making in terms of how do I get the text?
How do I put it together? But the beauty is some one person was able to do it. Yes. Intellectual sparring partner or like, it’s kind of, it’s kind of, a cross entropy.
You kind of get stuff together, you build it up, you mix it up and you create something new out of it. And I think that’s, that’s the interesting part where you’re probably relating to that. This is not really AI because somebody created the tune, probably manually. Somebody created the text maybe with a little bit of chat GPT, but then you would also tweak it so it rhymes well and works well with the music flow.
And then you might even think, oh, I’m going to have a song from Drake. And then you might even sing it yourself, but you just change the voice that you have, not really with AI, but let’s say some of it. So it sounds similar to Drake and the weekend. Whereas, and I tried it myself because we also use it for this podcast.
The intro music is created by AI, but I tried myself those AI tools that would create music. I still have to select the different instruments. I still have to select the speed, the tunes, G, do or whatever you call it in English. So how it sounds.
I would create random output a bit like Imagine from Midjourney, but it might still not sound great. So I would change it. So AI doesn’t, and I tried some different tools and dear listeners, if you have something better, please let us know. But so far it’s a lot of experimentation still where the human has to give feedback.
This sounds good. This doesn’t sound good. I want a different instrument. And I think, so just to be clear, yes, Ghostwriter used AI.
They probably, it probably used AI. For the text. And it probably used AI for the music, but it combined those and the style. It combined those and the training it decided to use and the focus it set.
And that said, like, I want something and the weekend in this, that is all human decision making. Yeah. And this is so important for us to understand once we are going into this generative AI set up. It is not that.
Suddenly everything will be taken over. Yeah. It is the combination of a human with the AI, which will be the new norm. And the great thing is it’s not even proprietary.
So when I, when we looked at our research for audio music, Microsoft has Wally, OpenAI has Jukebox. Then we have Autogen from Meta, Music LLM from Google, Stability AI, Dance and so forth, Deep Composer from Amazon. So everyone can use it like mid journey. They have to pay for some of those, but everybody can be more creative.
So I think the question is more now what actually happens with the industry with the help of these extra tools versus will this be replaced with this whole industry will be replaced? Well, I’m never, never like a big fan of replacement. It’s a dream. And we, I think by using those tools, production cost goes down.
The ability, you don’t need a big team anymore. Yeah. You don’t need a big producer anymore. It becomes easier.
And we can look at two examples probably in the industry to see what’s going to happen. One is in media, text journalism, right? It used to be that it was a profession, journalism. It was hard to create text.
It was hard to bring it together. Once you had the article, then it was hard to actually create, create the printed thing that you needed. Yeah. Printing press.
That was. Design it, put it together, like plotted. That’s on one page. Yes.
And then it was hard to actually distribute it. You need to have access to all the shops. Now there was a great unbundling. So first of all, everybody started to be able to operate their own computer and now writing was easy.
Then the internet came and now we could actually distribute it easy. And then everybody got a mobile phone and the mobile revolution came. Now everybody could be their own journalist. And you saw new business models spinning up there and the whole area and coming together.
So that is what. And, and sorry, you forgot one part. The research was even also happening over the internet because before the journalists, they would use the network, they would have to go to cafes, et cetera. And suddenly now I get all the information as well.
Nothing is proprietary anymore. And that has changed hugely the way how we consume media, how we consume text, right? That had like went from, we buy, the newspaper in the morning to we read the feed in Facebook from our friends to with the Kardashian is recommending me as an influencer. What I should read to TikTok is AI driven.
And to me, I mean, there has to be huge shift. So now let’s look at the music industry. Well, it used to be hard to create music and you needed a team, a studio, an orchestra, crown singer and so on and so forth. And now you can actually saying, well, let’s play it like Drake.
And I just want a violin solo here in it. So you can make it way easier. The future will be indie. And to my point, you just go there to the store and had to buy it or even before you had to go to cafes, listening to jazz music.
Now you can listen on Spotify on the internet, but then you can even remix it. You can change the style. You can change as to your point. I can play with it because before I would listen to a song and then I would get inspired.
But now I can listen to so much music. I can mix it up. And I can change it in a way that I like maybe more modern style autotune, this bad autotune that I don’t like. But the experimentation is much faster and the inspiration, the input and the output both is much more, yeah, much wider, much more creative, which leads to your point of more indie music.
Yes. Future is going to be indie. Now, the funny thing is when you talk about this, this mixing, we have seen this, we’ve seen that story before, right? Yes.
People started sampling music. Yeah. And the industry’s reaction to this was, oh my God, don’t sample our music. Don’t do it.
And then there were people were starting fighting and saying, no, this has to be protected. And how do we pay our dear artists? Over time, obviously DJ sampled music. Suddenly this became a new style.
People loved it. And sampling music became, a thing. Yeah. And if you, same thing here.
And a good example, just watch hip hop evolution. The whole rap music in the eighties, nineties, probably even before was with jazz samples. That’s why I love this type of music, but that was free. That was the main reason they got it for free.
They could produce music so they would earn some money. And that created rap at the end of the day in hip hop. Now this is, but the interesting part is so overall AI is not replacing it. It just lowers the barrier.
Mm-hmm. Yeah. And that is why participants suggested that participants suggested that participants suggested that participants So that’s all unclear. But overall production cost becomes lower.
Overall, when production cost comes lower, let’s remember, we only have two ears, we only can listen to one song at a time, it’s a zero sum game. Meaning, if everybody now starts to create more music, and by the way, Drake, isn’t he? Like, I mean, he creates music, like other people, like, like he creates a lot of songs, because he’s trying to occupy that space. Now, if AI makes it way easier for me, I’m in direct competition with Drake.
Well, not me, because I don’t have the music taste. But I haven’t, like the person who’s creating this is in direct competition with Drake. Meaning, the content. The price per content will go down.
Yeah, will be way more to be produced and there will be way more to be heard. Now so they I said early on the future is indie is a second possible trend. And the second possible trend is the future will be music aggregators, because if you have many songs to choose from, you will see essentially what happened in the media industry, where suddenly everybody’s creating content. And everybody could consume everywhere the content, but that was then a calling for the aggregator, which after the great unbundling, you had the great bundling going via Facebook whatsoever.
So maybe we should all invest now into companies like Spotify, Storytel, whatsoever, by the way. I think your point is also we only have two ears, because if I if I have all this indie music, and some sometimes that happens to me on on certain music streaming services. I just, it’s just getting boring. So I’m coming back to my brands or artists, and I just want to listen to something that I know something familiar, something I can hear in the background.
I remember, let’s call it, let’s come back to my rap music, or hip hop music. Timbaland was a big one, because he had this certain type of music style. And then he was talking with this voice. I’m not imitating that now, in the background.
And that’s why you would listen to it, because Timbaland is part of that. And same probably happens nowadays with Drake. And same will probably happen. happen also in the future because people are a bit lazy right i want to listen to nice music i don’t want to experience new music styles i just want to listen to what i know now that is actually an interesting one i’m like so either it becomes future is indie if the future is indie then that means that the brand is an important part so they will be music endorsed by drake or music like drake drake is is not the person anymore drake is the emotional component of this music and you see very many people complaining oh yes now we have ai that means that we are missing out the emotion no there is still the human who kind of says i want a certain things like with image generation i want a chicken made out of fried chicken and in the sunset that is you that is an idea which i put down right i think it’s also art at the end of the day i want to relate to the picture i want to relate to the story and if i come to music i also wanted to relate some people at least to the artist the people who have been to a concert i love concerts like live concerts it’s a total different atmosphere and if this is ai created music and you put an avatar on the screen dancing in front of me it just doesn’t feel right maybe that changes in 50 years for my children my children are already born so maybe it’s more 20 years but this is it’s not just brand that would be my interpretation for the trend so it could be in the music but somebody still has to perform i want to relate to some person i’m listening to maybe i even want to read about them on newspaper or follow them on tick tock i can’t follow ghostwriter on tick tock right now i can probably but it’s not maybe interesting to follow him yeah but again then okay maybe like the out tent authenticity of this is the important part of the story but it could be that we create a virtual drake right you have already instagram followers and thousands that are completely virtual and there’s a lot of beauty to it right if you you build up a person and that person has a certain personality and people follow that personality and follow the the brand follow the idea that this person doesn’t exist in reality might not be so important anymore and so there’s that’s a brand part the indie part there’s obviously the other part which we saw like the swiss rock right i mean there is a swiss music station which is testing completely automated music generation and if you listen to amazon music then you will see that amazon music is trying to push a lot of plain not branded songs and because it reduces their cost into your music stream right and i think there is a whole other area which is these aggregation areas so indie versus aggregation both are these two areas which will be enabled because ai is reducing our barrier but very important ai is not taking over it is the human together with the ai and one example i want to bring here is actually chess at some point in time kasparov lost his life and he was able to play the game of chess and he was able to play the game of chess against a computer yeah and kasparov was the last person who had at least a draw between him and the computer from that time on nobody else will win against the computer but the best chess playing teams are not human and they are not ai they’re the combination of a human with a computer these are the best ones and same thing with music you will well it is like as it is as we said early on it will be cross entropy it will be an intellectual sparring partner it will be bringing things together in a nice mix which is the future this is a creative podcast i like how you manage to combine a creative topic with chess but i don’t play chess so i think to cover the last bit of it because we’re running a bit out of time is i guess the copywriting discussion and it’s very early so we should only tackle it but everybody who’s now in this creative space might be a bit worried about copyright infringement getty images got sued because i’m sorry getty images sued stability ai because they allegedly copied 12 million images to train their eye model and and that’s a that’s a that’s obviously a big one it could happen with music as well and it happens all the time however as we just alluded on i’m getting inspired by other styles so that’s not good copyright infringement per se do you have any thoughts on that yeah i i had to chuckle a little bit lately as an and i you know dear listener what we always do is before we go and sit here and chat we obviously brief ourselves so we read articles we have we’re trying to get an overview and if you look at the briefing overview you see a lot of news outlets talking about oh copyright infringement this is bad what are all the musicians doing they need to make days end and they need to make money are we going to support them and i scratch my head and kind of think like isn’t if the future is indie meaning the market becomes bigger yeah and i agree totally because if you think about it who who are we protecting here are we protecting the rich musicians who have all the power for the lawyers and then start suing all the small indie ones or we’re protecting the small indie ones who get copied by the large ones i think it’s it’s yeah it feels more like for the first one the the big ones and if if the world is is getting more indie it might be even more just because everybody is able to create music it’s actually a funny thing i think the music industry hasn’t learned a thing they are reacting as they reacted at the sampling time this is kind of close down tell spotify to take all the songs down touch their muscle i would say spotify and deezer would be very highly recommended to focus on the indie part because that means the aggregation play they do is actually the value they’re creating for the market and the other thing which i think you will see is and we are seeing this for the whole ai discussion over and over The moment, the winner will be us, the consumer, because we get more choice.
We get more opportunities to listen to interesting songs. So the whole copyright discussion, yeah, Getty Images sued. It’s funny because essentially they sued because of the Getty Image logo. Yeah, it was found.
Yeah, it’s true. No, but I think, yeah, looking at it, I 100% agree with you. And that’s what we already discussed with our predictions. There will be more open models, everybody using this, more innovation happening thanks to AI because we’re augmenting human creativity, human output.
So we see now the same happening in music. Hopefully we have another episode also on video content creation. Currently Runway is doing a great job there from the US side, but it doesn’t look like we should be afraid that this whole AI, creative music will lead to more boring music in the future. Let me add one more to the video.
I posted on my LinkedIn profile one marketing video, which was all AI created. It’s actually super bad. It’s a very bad video. The point is it’s completely created by AI, which makes it so appealing to talk about this.
Video creation is still, it’s still difficult. You can create an image, but if you want to recreate this image because it is AI, the image will look different. So putting the images into a video stream is yet a technical challenge, which will get solved pretty soon. We will see that the world will become more indie.
You do not need Disney. I’m sorry, like they just lost so many users on the Disney Plus subscription, but like you will see that this will become more complicated. Because instead of owning Marvel or creating yet another Star Wars movie, people can now, all the fans of Star Wars can easily, and they’re doing this already online, easily create their own movies. If you own the Marvel comics, if you have a Marvel comic, you scan it in and you kind of say, create a movie out of this with real people.
And you don’t need highly paid actors anymore because the computer can create it. And I think that’s, that’s the interesting part. Because then the human touch comes with the story, the tension, the plot changes, all these kinds of things. And AI still cannot do it.
But then the human can create and iterate all on this and put the human touch in, which makes things exciting. And by the way, I’m still a Disney subscriber because I love Marvel. I love Marvel as well. Don’t get me wrong.
But like, and I really love the Mandalorian, but hey, was it so, was it so innovative? No, no. The computer could have created this as well. But I think, maybe not the story.
I would disagree there. Actually, this is a good one. Okay. The story, the computer can generate not only one story, it can create a thousand different stories.
That’s true. Yeah. The question is the agency of us as a human to say that is this storyline, which I believe sparks the highest emotional reaction or, is what I want to say at the moment. Meaning when you, and this is the same for music, for writing, for imagery.
If I say I want the chicken made out of fried chicken, I get from Midjourney four images. If I do stable diffusion, I tell them how many images I want to get, whatever. I need to make a decision of what is the right level. Think about our graphics for our podcast.
You didn’t say I want a graph, a logo for the podcast. And it came. No, we had an exchange about it. You said like this logo, or that logo.
And I just, I just felt like I’m back to when I started my own company where crowdsourcing was, was a trend. Now with all this AI creating, creating output for humans and users, but we need still the human feedback. So we come back to our feedback loop. Maybe a nice business model idea would be that I create a platform of all different users out there.
And I just show products, stories, images generated by AI to get some user feedback. So then I can throw them into the market with a bit more security that they’re actually working. So AI is actually creating something that where we need humans again at a huge scale. Maybe that’s the future job.
It’s not prompt engineer. It’s AI feedback engineer user. Well, I mean, yes. And it’s again, AI will supercharge us.
Human. I’m very positive about this. The tricky part, which we haven’t really addressed and that we should probably address at some point in time is how do we rescale, reskill all people who are getting not supercharged. We do that in another podcast.
Yes. So great seeing you Jasper. Likewise. Thank you very much, Lux.
And thanks for joining us. you