Eleanor Warnock

Welcome Eleanor. Such a pleasure to have you today. I think you're the first guest we have who has spent more time than us thinking about AI and taste. you've written about it, you have interviewed quite a few people on the same topic, you talk about it on your on your Substack. Maybe maybe to start one one question I would have for you based on like the range of people you've talked to, is who do you find is finding the concept of taste kind of a cool concept to talk about?

And who finds it a little cringe? Like who finds it a little annoying to see Silicon Valley talk so much about taste?

that's a really good question. Who is finding it very cringe? I feel like people who are maybe more in the creative space, who are, yeah, more traditional creatives. So in writing, creative writing or

Mm-hmm.

art design, those people find it a little bit cringe that Silicon Valley is speaking about taste. but I also think that there's a big like a cup so two weeks ago there was South by Southwest in London and

South by Southwest is a festival that brings together creatives and people from the tech world. And being there and attending a lot of the sessions really reminded me that these two worlds don't talk to each other very much. And so actually,

Mm-hmm.

some of the conversations that we're having in technology about taste are the creative worlds, let's say, isn't even really conscious that that's going on. and there's a big disconnect there. so.

That's what I that's what I would say. I feel like sometimes

I'm curious what's

Yeah, go ahead.

Son Wha w did you have example of disconnect? Like what what what are some examples where one side just doesn't understand the other?

I think that an example of one side not understanding the other is when tech people, people who are building technology products, let's say, propose products that don't really understand what people in a certain industry enjoy. You know, let's like

Mm-hmm.

talk about writing, for example. someone reached out to me the other day who is trying to create an AI novel writing product.

Mm-hmm.

kind of to help people plot out a story and and come up with yeah part different parts of the plot, develop characters, that sort of stuff. And there are there's software already that exists out there that novelists have been using for a long time to write things. but kind of in when this person was pitching me, and again I I didn't really go too much into the details of this product, I thought, well, but that's the stuff that I like to do personally.

Mm-hmm. Mm-hmm.

So it just feels

like it's very like you're building a product and you think that I'm the target customer for your product, but I'm probably not the target customer for your product. The

Mm-hmm.

target customer for your product is probably someone who's like very like a very amateur writer that's like, I'm gonna I that needs a little bit more guardrails maybe to to create and to get yeah and to produce something. so I definitely feel that a lot. It's like I enjoy that stuff. So why do you want to take that away from me?

ask a question about this. So you see basically these these things being pitched to you that are just missing the point.

Mm.

Let's turn it around. So what how would it how would it need to look like in order to kind of s spark your interest or say, yeah, this is actually a good direction.

That's a good question. I think that things that would spark my interest are things that I guess here we're talking about technology products, right? We're talking about software.

Hm.

And we are so inundated with software and software tools now. And so a tool that really wants to is gonna spark my interest is something that is curious about me that

maybe takes a really strong stance. So for example, a tool that I have used is a tool, a writing tool called Ellipsis. And they say we don't allow any generative AI in the in the platform.

Hm.

They don't have any features that use generative AI. And I think that's a really interesting, they're not necessarily the best platform that I've used, new writing platform that I've used.

But I really appreciate and I'm interested in the fact that they are like, okay, we have this really strong stance, and this stance on AI, for example, comes from things that we've heard from our community. And so we're going to champion that. And that kind of strong stance on something that reflects conversations with the community of people that are using that tool is is gonna be interesting, I think, to a practitioner.

Can I so just to make sure I understand the f for now your your favorite experience has been some with zero AI almost, right? For writing. Is it because you like all the aspects of writing? the reason I'm asking because when I was working a bit in that space of AI for writing, interviewing a large number of substack writers, we noticed there was there was kind of two camps. You know, some people really liked coming up with ideas and they were open to having AI

Turning the ideas into writing just because they were not very confident about their writing. And then you had some people who on the other side actually really enjoyed wordsmithing, but they didn't care so much about coming up with original IDs. They just really, really loved like writing every single word themselves. And basically for people who only like half of the job, maybe AI can do the other half. would you agree with that statement, or do you think it's usually better not to have any or like

So it it really depends on the creative process of the person. And I really loved the work that you did on that because you revealed that. You revealed how with something like writing, and I think the same would said can we can say the same about other creative things. There's no one process that people like, you know. Some people, as you said, like they like to come up with the idea and then they need they want some help with drafting those ideas, right?

Other people, like they want to lean on AI more for that idea thing, and then they wanna spend more time crafting each word, right? And that's why I think that it's really difficult when people come, like for example, right now, you know, I work for every, which is we we talk about it as the only subscription. You need to stay at the edge of AI. And we are very obsessed with writing and AI, and everyone in the team has a different process for writing with AI.

Right. And that just reflects the different preferences of the individual people in the team. And kind of to come back to you, actually, I think the one of the best software writing experiences I've had actually is one with AI. I think I really enjoy using Claude for some of the things that I that I don't like about writing. Like, for

Mm-hmm.

example, for my newsletter, I always have Claude clean up the transcript. I have a transcript cleaning skill that I've created.

And then I go in and I edit that and I, you know, shape the piece myself and I really enjoy that part of the process. And then Claude is the one that kind of gives me the canvas that I can work with and takes away that annoying part of the process for me, right? I can see, I can imagine that there would be writers who would actually even want to automate that transcript cleaning process to the very end, who would just say, I want you to do everything. But for me, I really like just like being in the words and and like playing around with it.

and then another tool that I've been using is a tool that's been being built in London called Marker. And they have this very cool feature where you can put like XXX or TKTK, right, into a document. Say you're saying, you know, Bruno's book, which was published in XXX, and it will fill in the date that the book was published. And so you don't have to jump over to Google and say, okay, what was what year was that book published in? And

That is a great that keeps you in the flow and that's a very cool experience. but most of the time I really like keeping AI out of the environment in which I'm writing because I feel that I need to be responsible for everything that goes out. And so I need to, if if AI is in the same environment as me with writing, there can be so many points of failure where I'm like, I didn't even realize that that like got put in there or

you know, this was edited in this way and I didn't read over it. And so from the point of view of someone who's worked in newsrooms where it's all about responsibility and accountability and accuracy, I really like to separate the two things out most of the time.

When you you've talked about Every, which is both known for writing and for software. do you see a big def difference in the way say software has been transformed by AI versus writing? And actually

For sure I have.

this is a very leading question, obviously, because we all know coders are using AI a lot. I I g I guess the the real question is w why do you think there is kind of such a difference between

How much coders are using AI versus writers?

Totally. And I would love for you also for you, Bruno, to also like feed into this because I have so many questions on this. I have my hypothesis, but I'm obviously not a software engineer, right? I think the first thing is that models have been created to make they're they're adapted to software engineering, right? Because that's gonna make AI labs a lot of money. And so they're adapted to that. And helping people write is not necessarily gonna make

labs a lot of money or helping helping kind of close that gap of like making something really, really, really good with writing is not necessarily gonna be the cash cow. So I think that's one thing. I also think that with software engineering, and again, this changes a little bit now that we're seeing yeah, we're seeing like models like Fable that can work for a really long time, that can can correct themselves, that can kind of make judgments is

I feel like with software engineering, like you start with your spec, you're like, I know what I'm going to make. I know what the objective is, right? And so it's easy for you to tell, communicate that to the model and say, This is what I'm I'm trying to make. And you can be very specific about it because and then you can very objectively evaluate the output of that, right? It either fulfills the objective or it doesn't fulfill the objective, right? Whereas with Yeah, yeah, yeah, please, please.

Can I try to push back on this actually, on this distinction? Because

In software, before it got so well accepted, there was pushback from say the more like conservative coders because they were saying, you know, even if you give some goal, you're gonna have a lot of micro decisions to do in the along the way. You know, you're gonna decide which database to use, gonna decide which framework to use, which which colors to put the buttons, and so on. And I think with time, coders have realized that AI is actually better than them for so many of these micro decisions. and so it

Basic even though there is actually quite a lot of choice to be made,

Mm.

I think sometimes they just start trusting the machine. Do you do you think there is a possibility or risk, depending on how you look at it, that writing is gonna be the same? Where like at some point even good writers are gonna start thinking, you know what, the machine is gonna find better words than me, so I have to just surrender.

I think that people already are trusting the machine, like in terms of writing. Look at how many people put things on LinkedIn that are very clearly AI written and they're trusting

Mm-hmm. Mm-hmm.

the machine. I think we can even trust I I believe that we can trust the machine maybe to like 90%, let's say 90% success. Let's use success. I don't want to use quality as a word, let's say success, when we are doing things like writing marketing copy and you have very

like defined guidelines. This is our brand, this is the way we talk about ourselves. These are the words we use. And so when you can like make that very, very defined, then I think you can you can be very successful and get a get a a strong output, let's say. but as maybe with coding and anything, it just breaks down when there's like with writing, you are the objective changes as you write, right?

And I think that's true for creative writing, that's true for essay writing. Sometimes it's true for journalism in some sense. When you put everything together and you're like, okay, actually, as I'm synthesizing this through writing, my ideas are changing and my view on this is changing. And that is the stuff that AI struggles with, right? and that's yeah, that's what AI struggles with.

That's very correct. I mean that's the way AI is trained typically these days is with what's known as reinforcement learning from verifiable feedback, meaning tasks that are kind of clear cut and well, like the the that's the the real kind of notion of success for a goal that has been set in advance, which is indeed quite quite different.

Can I can I float another direction or thesis in in this in this domain? So if you look at software engineering, besides the question of individual productivity and how you how you yourself work, I think there is in some sense almost a not a universal, but there is there are relative established patterns also how to work together, right? You have something like GitHub, you have Git as a protocol, you have certain ways of basically pushing new code to an existing into an existing code base.

It seems to me that the equivalent is not clear in other domains. Right? If you take something like, you know, a a media outlet, obviously there are processes in place, but it's not like they're not so universally recognized. Would you agree with that?

A hundred percent. I think that's really interesting. I hadn't thought of it in that way. And I think working at Every has exposed me to those really strong systems and processes around working that you have in software development, which is amazing. But we do not have that at all in terms of writing. yeah, like imagine if I think part of it is because it is a little bit of a personal thing. I think it's also because maybe I don't know, every

company has their own way of writing and does it really help everyone if like I as Google share my style guide for writing with the rest of the world? Maybe as a reference, but you know, maybe it does. Maybe there is an an argument that people should be stir sharing more about how they arrive at the at the writing that they do. and and then the other thing is like can you really reuse components of writing? Maybe.

I think you can't really reuse some components of writing. And so something that I wrote in my article about the British economy is if I take that and try and make it into a module or a skill or something like that doesn't really help people, right? but I think we are we can go much more in the direction of standardizing the processes of writing. So this week I was speaking to this really cool team in the UK, and they are doing a cycling publication.

About cycling and they've created a completely AI native CMS. And the difference between the CMS that, you know, we used at the Wall Street Journal and what they were using is night and day. They had a feed, they were bringing

Sorry, sorry, c

can you c sorry, can you say what an CMS is? Sorry, I'm just I see. Thanks.

It's a content management system. So when we have an article, right?

When you have an article, it's like the back end of the of the website that shows the articles, right?

Mm-hmm.

You go in, you upload the article, you add the title, you add the tags, right? You fix the SEO, all of that stuff.

Thanks.

And think about WordPress. You've probably used WordPress before. WordPress is like the CMS that most even very sophisticated publications use. And these people had this amazing CMS where they were pulling in data from

cycling RSS feeds from YouTube videos of races and of cyclers who were giving interviews. And then the feed was suggesting what the news was supposed to be and then writing the draft. And so when they were going into the CMS, there was already a draft and they could edit that and they could add components to that. And that sort of a process, like that starts to look a lot more like something you could share on GitHub, something that I could use, right? Even if I was in a different domain. So

Maybe maybe we with like the kind of tools that we have now, we can start to approach something like that.

Follow up question. This also suggests to me that the idea where we have a let's say a newspaper, a monthly why not have something multi mo multidimensional, multi model, right? What you describe is already going in this direction. What would be the interesting or what would be an interesting ex direction to explore if you want to go beyond the formats that exist, but basically look into a mesh of things?

I mean, you can think of so much. You can think of a day when we're gonna be able to create entirely personal publications for people, right? Or a day when the journalists, I feel like journalism is also becoming much more about the individual and individual brands, when you could actually maybe ask questions through a simulated version of the journalist. So they could write a piece about taste and AI, and then you could have a interviewer have a chat with them.

about the piece and about what they thought about it.

Hm.

So there's so many, so many things that you can think about that are very exciting. But but I think also

No. Yeah.

we can't overengineer things because I think like VR

Yeah.

and and XR showed us that people don't necessarily want the overengineered solution, you know?

Can I can I ask about this use case?

So I think this is one of the most that was after ChatGPT came out, I would say in the first six months, I would say like one of the most common projects I was hearing about was people were people trying to automate newsletters. Like we were inundated of like AI generated newsletters. sometimes maybe for very like technical specific topics, I would see value, you know, as like an aggregator of like no facts and figures.

but generally speaking, I think there was a pretty universal feeling that this is just slop and completely unreadable. we've gone kind of a long way, I guess, since like a couple of years. but I think still most people would still say that this sort of like AI generated content or or the kind of you know press a button to generate an entire novel is generally still slop in the sense of really kind of a a taste problem. I I think that's often one of the

one of the example that comes to people's mind when they think of the taste problem of AI. Do you do you already have a kind of mental model of like what is w what makes it slop? Or like or first of all, do you agree that what comes out of these fully automated pipelines today is mediocre? And what what do you think we need to change fund fundamentally for this to become truly kind of interesting to read?

Totally.

So I interact with AI generated writing all the time. And

Mm-hmm.

I it's it's bad. It's I think the reason that you can tell it's flop is that when

Yeah.

you start to edit it, because I'm an editor, it falls down. So in

Mm-hmm.

journalism, I have this I kind of have this concept where I talk about bulletproof

Mm.

writing. So bulletproof writing is writing in which every word is like the right word. It is accurate, it is considered.

And AI writing is it's like a piece of glass, and when you hit it with a hammer, it falls apart because the words don't have that same kind of consideration behind them. And you're like, why are you using this word? It sounds fine when you read it, maybe the first time, other than some very ridiculously AI jargony sentences. but then once you kind of tap it, you say, like, why did you choose that word? Right. What's

Why in this context did you choose that word and then it falls apart?

and I mean, I guess it's just the mechanics of how generative AI is trained that makes it that way, right? for now.

I'm just

sorry, I'm just having a Eureka moment. It's incredible. What you just said. Yeah, what what you you what you what you just explained is potentially linked to an effect people have noticed a lot, which is one shotting something seems to work much better than iterating on something. It's actually quite true for software as well, right? But basically if you give a very long prompt, something I I I personally like a lot is spending an hour doing a very long voice memo or just running or like in the transport or whatever.

And then giving that to a frontier model and one shot something, whether it's software or blog post or whatever. And usually the first iteration is a wow moment. Like, my God, it looks so nice and consistent. But indeed, every time I try to ask to like, but it's almost perfect. Just change like the a couple of sentences in the intro, it all falls apart. I think that's that link links very directly to what you just said, right? But do you have

What do you mean I guess when when you said this is how models are trained? Like what what would explain the fact that it's good at one shot but not iteration?

I'm not really sure how I would explain

Y'all.

like one shots versus iteration. maybe one thing that I feel when I work with AI and for example, with writing, and I'm and I'm going back and forth and I'm iterating, is that I feel like it's very like linear in the way that it it's like t we're going down a path together, and we can't really go back on the path or like change our course very well. And

That's and especially as like context fills up, right? It gets harder and harder to kind of get off this path. And so that's I actually really agree with you. I do do a lot of one-shotting of stuff because you do get a much better result on something that's bigger, I find. And then I'll just have that like output that the AI has given me. And then I kind of like take notes or I like think about that on the side and then take some actions based on that. So like a really good example is like.

I have been writing a novel. I rewrote the first four chapters completely. And that meant that many things in the book had to change, obviously. Characters had to change, whatever. And I gave Opus 4.8 the entire novel, which is 80, 90,000 words almost. And I gave it a really long thing about like, here's why I changed this thing. This is like the meaning of the themes and like why I made this change, right? And I gave it the original.

Chapters and I gave it the new chapters and then the rest of the book. And I said, based on what I've told you about why I made this decision, what this changes about the story, what what I want this to achieve, give me a plan for how I need to change the rest of the book. And it gave me a really, really great, like, it was like all laid out. It was like, this is this change, this is why you need to make that change because of this theme.

You need to say have this character say here, here's a suggestion. You know, you had this thread that you established at the beginning. Maybe you could like wrap that thread later in the book. And again, I didn't really talk with it very much after I had that plan. I was like, this is awesome. Agree with some things, don't agree with other things. But but had I tried to engage with it and like go, I would like have gone down this path, this linear path of like, okay, well, I really don't think that you're right about me changing that one character. And

you like lose flexibility. Especially since I loaded it with an insane amount of context. So Yeah.

I have a f I have a follow-up on this. First, actually this is really an excellent picture that you gave or analogy on this. The idea that I and I would actually say you don't have to hit it with a with a hammer, you just have to nick it in the in

Yes, yeah.

one edge, and basically the whole picture is basically like a is cracking.

Yeah.

what you while you were saying this, what it reminds me of, on this Leban just stated and then maybe you comment on it, what it seems to me is that what it's missing is basically the the memory of the backtracing that we normally do, right?

The idea that if you're writing, you go to a point, it's a cool it's it's basically a dead end, you go back, you go another direction. But that that is basically in your muscle memory. It's in the back of your head of a creative project if you're writing a book or even if you're writing a scientific article, you know what you have tried, that it's not there. And it seems a bit to like that what you're describing is what is missing is exactly that kind of path-dependent knowledge.

I agree. I think there's something there, and that something transcends context. Like if you think about me writing, let's say the take the take the example of a journalistic article, I would come to the writing process of that with all the context that I could technically give an LLM. The interview transcript, right? The profile of the person I interviewed, the other articles that I read to help me understand the subject. But they w

There would be that like meta context that you're talking about that the LLM wouldn't necessarily understand, which is like, why did I pick this topic? You know, is there some sort of connection with this person? Again, are there other ways that I've thought about telling the story? And I was like, No, I thought about interviewing that person, but I'm not gonna interview I'm gonna interview this one that have shaped why I brought that collection of context together. So I think, yeah, for sure.

Really I I really like what you said earlier about you know kind of deciding as you go what's gonna be the topic, like you're changing the the goal as you go. I I imagine there's something in in my mind there's some cases where it makes a lot of sense, right? Like if you're writing a very complex I don't know a screenplay and there's just so many constraints in a movie, right? You in time of you know keeping the

The energy up, but creating intrigue. And there's all this constraint that, you know, you're gonna try something and it's not gonna work. So you're gonna need to go back three sa three scenes before to change the premise so that you can unlock something else, right? But so this is almost like puzzle solving, right? but you you mention other areas like journalism, right? Where I guess it's it's not that you're just fixing a puzzle, it's that you just generally change your opinion about what's interesting, right? Like you would decide, actually this this

I want to know more about this. I'm curious how you I guess what's your mental model about how you decide when you write something. Well what makes you kind of decide what you actually find interesting and w why wouldn't AI be able to somehow have a similar process, if you will?

Well, I mean, now we're kind of coming into the the taste s you know,

Yes.

part of the conversation, I think. I think it's it's a it's a sense of

It's a sense of either I'm curious about the topic, I find it interesting, I find I I feel like, you know, to talk about a sensation that I feel like I feel like my brain is lighting up, right? I'm like, okay, yeah, this is the thing. Or that's the anecdote, or that's the story, right? and I think that that is honed in my case, for example, with journalism and

And nonfiction writing that's been honed by working in newsrooms and and by working with people editing nonfiction essays. and then when it's more creative stuff, it's just my own personal like what am I interested in? What is what is what yeah, makes my brain light up, right? but I think that that interest is something that you have to train if storytelling is your job.

if you are in marketing or if you're in communications or any sort of profession like that, you train yourself to see this is the story, this is the hook that that makes people that causes an emotional reaction, essentially, because storytelling is about an emotional reaction.

Is there I I I know in the camp of people who are most skeptical of AI and robots in general, I I think there's a camp of people who would say, you know, they can machines cannot have taste because you know they're just clankers. They're just, you know, bits and bobs. They they don't have this sort of like, you know, embodied emotions and all of these things. In the same I I I think actually you had some guests or you interviewed people who have made

I think Esther Moisie Kirschbaum who had mentioned you can't fake test taste. Like there's something machines kind of inherently almost can't really can't really do. Would you kind of agree with this view? or do you think there's kind of no real reason, no limit for why machines couldn't kind of reproduce this this mechanism of like lighting up?

Yeah, I mean I think like you can things that are machine created can definitely provoke an emotional response in us, right? Let's start maybe from Yeah,

On the reader side, yes. Yeah.

yeah, it can, right? So I mean, think about the fact that people fall in love with chatbots. That is an emotional response to something that a machine

Yeah, yeah. Yes.

is creating, right? And I'll remember like when I was a teenager when I had AOL Instant Messenger.

meeting some boys online and talking with them and liking them. And in that case, it was a it was a human on the other end, but it could have been a machine, right? I didn't know. I didn't see

Yep. Yeah.

them, right? and so I do think that digital interactions and things that are made from by machines can definitely emoke a evoke a emotional response in us, right? I think I maybe I sent you the the the email today or I sent you the article. Today I was reading

I watched this video. It was like an eight-minute long video that's completely AI created. and it's about a man with a, let's say, a deformed face who lives in Paris and he's looking for love and he finds love. And this has gotten a lot of interest, and a lot of people have watched this

Mm-hmm.

completely AI generated video, which took because of the limits of of like prompting.

Mm-hmm.

like for AI video was really, really hard for the person who made it to make it. and like people have had an emotional response to this. I watched it and I thought it was a little bit clunky. But I can see how people would have an emotional response, right?

Yeah.

Mm-hmm. Yeah.

I I guess I was so I'm I'm totally with you on the on the reader side, the person receiving the you know the content. I I guess what I'm was what I was alluding to is some some people on the creative side have this maybe say this romantic view of the creative process where they believe you know the artist must have himself some emotion. You know, there need to be some authentic taste on the creator side.

to to produce something of value. And here I'm using value in a different sense, it's just you know popular and you know getting likes on YouTube. Obviously you can, you know, hack the algorithms and so on, right? But do do you think there's something to that view that like really it's something kind of fundamentally different into a you know a human using their own emotion as part of the creative process that will be difficult for machines to reproduce or

Or do you think there's no real limit?

I think it's really hard because think about let's I personally want and I'm interested in art and let's let's talk about yeah let's talk about art art that's made by people for a reason. because I'm I'm interested in the human experience, right? But think about the example of forged art, right?

You might see a painting and you might think it's a great painting, and then someone tells you that it's a forgery, and you know, all of a sudden the value of that completely disappears, right? And there's actually not really a logic for why the value should disappear, right? If you, if you think that it was a beautiful painting, again, you had that emotional response, then why should it not be valuable? Right. and so it's really difficult to say, like,

I think AI can in the future might be able to create, let's say, a novel that was stylistically as impressive as a great novelist, right? and it's kind of up to society to figure out whether or not they want to assign value to that, right? I think there are people who will and there are people who won't, right? in my case, I think I would be interested. I'd be like, okay, curious. I'm curious about this.

But at the same time, I would be more interested in reading something by someone who wrote it based on their own experience. But that's just my like value. those are my values.

Can I push back a bit on this?

Yeah.

I mean we so far we basically we have been mapping it as there's either s it's zero one, right? There is the AI creates everything and then there's the human basically. But much more realistic and close term obviously is and that that's just the history of how what happened in the past is that humans use technology to produce art. And do you think there is something fundamentally different about AI being used in that process, or do you think it's a continuation in a very long story that we are going down?

Focus on art, not in not on any other domains.

I definitely think it's another chapter in a very long story. and I think that it is it's it's almost like, is this another Photoshop, right? Is this another tool that you use in order to create? And I yeah, I agree with you. I think that the creative industry definitely see things they see things as very binary. It's either it was totally created by AI or it wasn't at all.

And I think that the future, I mean, again, I use AI when I'm writing. It helps me to when I'm doing creative writing. I it helps me, for example, I use it when I'm writing dialogue. And I'll say, are the motivations of these characters I these are their motivations. Is that coming across to the reader? You know, you can give it a reader persona, you can get feedback from it using a using reader personas. That's amazing. That's so cool, right? and that's a tool.

And I don't see anything wrong with that. there's you know, writers who there was a Japanese writer who f came out and said she won an award, a major award for her book, and she came out and said that five percent of the book was written by AI. And it caused this

Mm-hmm.

huge furor because people were like, What? How can you give her an award if five percent

Yeah.

of the book is written by AI? It's like, well, sh she had the idea, she brought it together, you know. And then as a joke.

She did a project where she had AI write 95% of the thing and she wrote 5%

Mm-hmm.

of the thing, which I think is really fun, right? and I think you see artists that are exploring this distance between them and the AI. Is it five percent or is it 95%? and I think there are a lot of, especially on the visual side of things, artists who are pushing it really to the extreme, you know.

Hm.

There's

the I think they're French, a French artist collective who created a portrait of a fictional person and it's signed by the algorithm, right? And

Mm-hmm.

this sold at auction. it's very cool. It's very cool. But that's

That's very cool. Mm-hmm.

like it's a bit gimmicky. Like when you do that the first time and people are like, interesting, okay. And it's interesting because they are the first people to do it. If you did try to do that again, it

I don't think from an artistic perspective it would be very interesting, right? and so I feel like most people are gonna settle on the like closer to the fifty fifty or like closer to the, you know, it's never gonna be that extreme.

I guess that for for me that's feels like a distinction that's quite important the the difference between imitative art. I mean obviously a lot of creation can be nice by just combining two existing good ideas or like, you know, taking a picture and say, hey, make it look like Studio Ghibli.

Yes.

and i in my mind it's not the same tier

as like truly novel and kind of grand art, if you will. Like it's not like you invented like the first cathedral or like make this, you know, completely new, grandiose kind of project, right? Do you do I I I guess I'm curious, have you seen people demonstrating already the ability of AI to go towards like true novelty? Or do you do you feel we're still in a regime where it's very it's very much in a a tool of kind of imitation?

I mean, I think a really interesting person to look at with this is an artist who's been around for a long time and working with AI for a long time. their name, they are they, is Su Gwen Cheng, and they use robotic arms that are trained on their drawing style, and they draw together with the robotic arms. And

Sugwen says that the AI is a collaborator, is a co-collaborator on the their work. And I it is extremely like intellectually interesting, intellectually rigorous work. and that I think I think that they are one of the artists where like that that's what it potentially could look like, you know, really interesting.

like mechanically interesting, intellectually interesting. and I haven't really seen that in writing yet.

Yeah, they're the only artists that I can think of where I was like, wow, okay, yeah, totally. But when I see their work, what's interesting about it is that they always talk about the AI as a co-collaborator. So it's like me and the AI. And I always think of AI as a tool. So it's I did this with. I did this with Photoshop. I did this with a pen. I did this with AI. and I think the level at what AI can do right now in terms of text output.

is not ready to be a co-collaborator because it can't bring me that idea or can't push me into some sort of totally new thing like you were talking about, totally new level spectrum of creation that it deserves to be a co-creator. However, with like Sugwen, I think there's something totally new and novel about the idea of a mechanical arm that's been programmed on you, that's interacting with you, right? That you can't predict.

That for me is totally not

Love it.

Maybe sorry, maybe shifting a bit into a different direction. Just a thought, and this is also prompted by a previous interview we had. as far as I know, you spent a decent amount of time in Japan. there is a lot of talk about the current frontier models basically being a bit I mean, being a bit being biased towards a certain cultural sphere. there is evidence for it, but I would like I mean, for you who have lived in a who have lived in a different context

What's your perception of this? Do you feel this? When and where?

I am so glad that you asked me that question because I really wanted to talk about this aspect, especially as it relates

Mm-hmm.

to taste. Because I think even the fact

Yes.

Please.

that we're talking about taste is such a Western concept. So when I was

Mm-hmm.

prepping for this conversation, I was like, what is the word for taste in Japanese? And it doesn't really exist. Someone with good taste, you say sensei, they have good sense, but that's

Mm-hmm.

not really.

exactly what we're talking about with taste. And I think, yeah, taste is such a Western concept and it is so centered around the idea of fine art and a dichotomy between the high and the low and

Klass. Klass, hmm.

class. And exactly like think about like Pierre Bourdieu, right?

Mm-hmm.

The whole idea that our preferences, which is taste, is all

Mm-hmm.

defined by

social class, the way that we were brought up, the way that we were conditioned. And so when we talk about taste, I think it's so funny that we talk about it so much in tech now, because we're it's just people trying to like get social status at a s at a certain point, right?

Mm-hmm. Yeah.

it's saying, can I create something that will give me social status, basically? and I think that the way that Japan thinks about taste

is completely different. And again, I'm sure that other countries in Asia, I'm sure

Mm.

that countries in Africa have a totally different conception of aesthetics and taste. And maybe to give you like one thing about Japan, Japan is a country where everyday aesthetics have a very strong value. just the way that you make your area, your home, your garden beautiful in their own way, aesthetic in their own way.

and through those aesthetics of the everyday and the mundane, you can foster curiosity about the world around you, about community, about other humans, about nature. and that that doesn't exist in like our taste, you know,

Yeah.

conversation. So

It's so so interesting. Silicon Valley obviously would be the heart of with all the love we have for the people we know there and all the you know all the positives that come out of it. it's really it's really the city of God complex, right? It's it's a place where people do not hesitate to say I'm gonna build a product and ten billion people are gonna use it tomorrow because I want to own the entire planet.

and so that if you combine this with the kind of the baggage that comes with the word taste, there's obviously something to feel a little a little uncomfortable about. and I I guess what I wonder there is the I would definitely put me in a put myself in a camp of people who think there's something behind the idea of taste that's valuable. and yet I have all this hesitation about, you know, what is the right way to go about it. I wonder

I don't know, let's say a Silicon Valley startup working on taste versus a Japanese startup working on the same problem. Like what would they do differently?

that's such an interesting question. I would say first that I think a lot of the so right before we again we started talking, I was looking at this company in the US that's trying to train the models essentially for better taste on design decisions specifically, right? And gathering the judgments of many creatives and many designers to to make the to make the model better, to make better decisions. and

They use taste all the time to talk about this. And I'm like, that's not really taste. In my view, that's like making a better aesthetic choice, maybe higher quality, right? but it's not taste. Because I also feel like in that sense, they're they're working towards a commercial goal, and there are actually some like kind of object objective things about

Aesthetics that need to be present in order for it to be commercially successful. Like the font needs to match this thing and sizing needs to be like this, et cetera, et cetera. Right. And so I think that in the US, there is a lot of this confusion between taste and just like something that's commercially valuable or like quality. Just like let's make something that's let's make the output of the models higher quality for this vertical of things. Great. That's amazing, right? But like let's make the UX more addictive.

Let's make the UX more addictive. Let's make the let let's keep

people scrolling forever. Yeah.

Okay.

I mean and maybe that's fine. Like it would be cool if you know I could there was a tool that I could use to make my blog really beautifully designed for a cheap price. That would be really cool, right? Without having to pay for an expensive designer. but I think if it was like a Japanese team that was maybe working on this, I think A, like there would just be much more thought about.

What does this mean for community? What does this mean for

Mm.

everyone in society? Right? that would definitely be more at the heart of this. And

So you would say

instead of instead of making products more addictive, they would be they would care more about making the CT nicer for people, or this sort of this sort of use case.

Yeah, exactly. Or what are the aesthetics

or like how does this how does this make our lives more beautiful or that sort of stuff, you know? I think that would be more at the heart of the project.

Just

I just noted something when you when you emphasize the collective, maybe this is wrong or not, but I recently read that someone basically said, you Silicon Valley makes money, but the money flows back into companies. There is not really a new cathedral being built by the money or a new research institution, new university being built. What would it take to not in Silicon Valley but more broadly?

kind of grand scale, that we're there is something created basically that helps the overall world or the overall perspective on aesthetics, how we live, etc.

That's a great question. And I think that it would be really, really hard. I think especially in the United States where incentives are so skewed towards capital and power accruing in the hands of a few people, that's that's really difficult. And I also just think that that culture, that's not a that's not a how would I say, that's not a priority for the United States. I mean, take healthcare. Sorry, I'm like spewing off a little bit on politics, but you know, the idea

Yeah.

that

There are many people in the United States who don't think that universal healthcare is a basic right of people. That is a fundamental, like in the US, that is fundamentally about you take care of yourself. And if you cannot take care of yourself, then you are weak and we should not protect you. And so I think that's you can't really like reverse engineer that fundamental thing that America's built on. And

America is great because of that. America is like, let's build stuff. I need to hustle. I need to, you know, pull myself up by my bootstraps. but I don't know. I think maybe one thing that I that could change, and maybe this sounds a little idealistic, is like sometimes I wonder like, where are like the different cultures, women, where are different people in when we build these products? And maybe if we just had some different

perspectives around the room, we could create something that was more useful or more reflective of the of society. And that's why I'm I'm glad

Mm-hmm.

that you asked this question about Japan. because I'm curious about that. I'm curious about you know how what are the what are the the gaps that these models have because they haven't taken those other perspectives into account, you know?

We've talked a lot about the US and Asia. obviously the US is right now winning the race on GPUs and like brute force capital injection into kind of AI intelligence. Asia is kind of winning on making things cheap and affordable, which maybe is actually a path towards, you know, releasing some of that pressure that AI companies have to to monetize, right? Why

Maybe ch changing gears a bit. I I know one of your many areas of expertise is the actual the European tech industry. and I I'm quite curious where you see I guess the best opportunity for Europe in terms of changing what we do with AI. And obviously we can talk about taste if you think that's one of the application domains, but I'm I'm curious also more generally.

Yeah, I think that in terms of AI in Europe, I think there's obviously Europe can be really, really quick adapters of AI. We've already seen that European individu like individuals in Europe and European companies have been really quick on the uptake. So I think that actually even if you are using American models or even Chinese models, there's a difference between

Like building the model and then also getting it to work in a big organization, you know? And so there is a there is an opportunity if Europe wants to, you know, do that for adoption, I guess. There's also I think that Europe is differentiated as well by its values. you know, we just talked a little bit about the values of the United States, and obviously, like, you know,

Healthcare, yes. Yeah.

that's a a generalization, but I think that

Europe is a place where there's more of a maybe more towards the collective side of things of we want to make sure that we're thinking about the effects of technology on our populations. and so that could that could protect its populations, Europe's populations, and that could actually be an edge. and but I don't really know.

in some sense, if there was something really bad that happened with AI, maybe like regulation in Europe could help. But again, I'm

Mm-hmm.

not really sure. and then I think there's also something around taste, actually. You know, taste is social status, you know, I think.

Mm-hmm.

And think about like some of the biggest con companies in Europe. They're luxury companies like LVMash, right? So

Yeah. Yeah, yeah.

those places are

Everyone around the world wants their sense of taste and their sense of style and their products. And so what if there's something that it what if they could apply that hype and and that kind of sense to an AI product or something to do with AI? Like that could be really

Mm-hmm.

powerful, right? And again, like there are people that are making knockoffs of, you know, Louis Vuitton bags all over the world and knockoffs

Mm-hmm.

of like Hermes bags and stuff.

But people still want to buy the real things, right?

Mm-hmm.

and so I wonder if there's something around that sort of taste that Europe could create.

Is there a risk though that we we I I guess if you build a a taste company that only sells a Louis Vuitton, is there a risk that we are back to the kind of critique we made earlier of AI being kind of a you know, a class kind of marker and so on?

Yes, completely. I think we would totally be back to that being like AI as a taste marker. Completely. It we it would be a

If you like PI as a T S worth.

corruption of AI in just another way, right? And just another way to serve these kind of ta class markers that we have.

So can you think of an even more positive take on this? Something that doesn't require

And even

Necessarily the

class system.

Could Europe be more creative? I I I guess what I'm what I'm leading to is, you know, the Renaissance, you know, type of ID that some some people would have loved to see emerge from, you know, big revolution like AI. What what would it take for Europe to somehow take AI and have a new renaissance?

Such a good question. Mm.

I think maybe if it's it depends on what kind of renaissance we're talking about. If it's like a scientific renaissance, then maybe we're actually already starting, right? there's amazing research that already exists across the continent. That's something that Europe has as an edge, I think, on the United States as well. and so maybe like we're gonna see that, right? but if it's gonna be an artistic renaissance, I think then we need to probably

Have ways for artists to be supported by either very wealthy people or I don't think it necessarily should be the state, but you would need people with money and resources to support artists to play around with AI tools, basically. right? Because the people who were creating in the Renaissance, they were financed by really rich people, right? And the church. So, right.

and so you would need to give people like the time and leisure to be able to do that. and play around with that. So maybe it's that. Rich people, come on, give us some money, help us play around with AI tools

and create create some more, you know, creative ways.

Yes. I think that's yeah.

Yeah. And like I can see someone in yeah, like when will when will someone do that? That would be really cool.

Can I add on this? There is a there is a tradition or a thinking in economics basically that one of the secrets of the success of Europe was actually the competition and the density, right? That you have the Netherlands or the Dutch Republic, you have the United Kingdom, you have France, all basically competing on ideas but also on technology, obviously, and obviously, unfortunately, also on may w in military terms. But let's put the money aside.

What would be the way of creating this kind of creative let's call it competition for now, for lack of better word. Dynamism, yes. The idea

dynamism and yeah, all the things that come with it, yeah.

but the idea of basically being ambitious and trying to outdo on a positive way others.

But see, the thing with it now is I feel like we don't necessarily yes, we have national identity. And again, I don't really have any data to to back this up, but this is my initial reaction to what you're saying. Is like when you are a creative, I feel like you're kind of competing with everyone in the world. It's almost like the competition landscape has been flattened, right? Like if you are at the top of your field and you are an artist that is exhibiting a

It's global.

Yeah.

world

class museums, you're competing with everyone else in the world. And I'm not really sure you your work might be very like defined by your nationality and be very tied to that. But I think it might I think that that maybe strips away some of the like, let's do it for this country kind of thing, you know?

That's also not what I meant. What I meant is that basically the being based in a locality, but also having access to many variants around it. Right? I mean,

Mm.

just to give you an example, if you're in the middle of Europe and you take the train in three hours, you can basically cover a lot of different countries and also very within even within countries, very different kinds of you know, backstories, history, etc. Very diverse. But somehow we seem not to be able to kind of leverage this in as we did in the past. Maybe I'm maybe it's just my my personal view.

I I I've read similar explanation, Philipp. So for me like the one thing that also s seemed to be very powerful at the time was that you could find a home. Like if you you know, if you were a researcher in Italy and for whatever reason your university was rejecting you because your ideas were too out of the box, you would just find another country in Europe that would give you a home. maybe this is the kind of dynamic we don't have, or or maybe there's just no

N nothing no equivalent, no.

Or people

or people move to the US.

Yeah, or people move to the US. You're right. I think that's

Yes. Actually that's a that's a big part of it, yeah.

Yeah, people move to the US. I think you're right about that, actually. Yeah. And then just like bureaucracy, right? It's difficult. Even it's crazy that even within Europe it's it's difficult. Like if you were moving from France to Germany to be able to get settled, to do all the paperwork, you know. I think it would be a pain in the butt. Yeah. Yeah.

Yes, I know. I very painfully know. The other direction, but

Totally. So but I mean I

Like that would be great if there was some sort of like a hacker creative house that in the middle of, I don't know, Poland where everyone from around Europe came and created.

This is sorry, this is a di very different topic, but I it just crossed my mind. One thing where I have seen this, and it's very unfortunate we have it, that but it's the it's basically the basically the military military technology and the around drones and the the movement basically, the hackathons that are taking place with them obviously mostly mostly focused on the Ukraine but also surrounding countries. That might be unfortunately the one thing where I feel like this is actually very close. It's the urgency maybe also that creates that. I don't know.

It it's true, but you also have to remember that there are very large defense companies that are also in France, in Germany that are competing against each other, right? So there's also that weird

For sure. For sure. Yeah.

tension in Europe. It's like, let's work together and reindustrialize Europe, but I don't want you to get too much market share.

Ha ha

Yes. Indeed.

You know.

Maybe

we've taken colour of your time. I I'd love to not finish on defense and go back to the to the topic of taste if we could.

Yes, please. Let's not do that.

so one of the reasons we we're doing this podcast is because we we feel so many people are focused on AI for productivity. we'd love to see more AI for creativity. you've obviously looked a lot into that space. I I guess what I'm curious maybe would be to have your your clothing thought on

what what is the most important thing people would need to do to get AI to unlock more creativity in general in the world? What what's what's the priority for people building AI or people using AI in your mind?

We are

end we are ending on a very simple question. Yes.

Yes.

Ending with a simple question. I'll I'll

try and give a simplish answer. I think the first thing is I want more tech and creative people to come together.

Mm-hmm.

and I want people to even if they think that they have ethical objections, and of course the ethics around AI is very complicated, right? We didn't even talk about IP, copyright, bias, all of this stuff, right? Putting those things aside, I think that

More creative people need to just sit down and use the tools and not be afraid because that's what's given me so much more agency is going headfirst into trying to use these tools because I see the limits, right? I have those you talked about Bruno, like that moment where you're like, this is amazing. But you also have moments where you're like, You are stupid, Claude, and I'm yelling at Claude, you know.

Yeah, yeah, yeah.

And I think I

I have so much confidence now that my job is definitely not going away because I use the tools every day. Right. And I think that another thing that companies need to do is they need to give people time to experiment. Give like companies should be giving their employees at every we have a week twice a year where we don't have any other work but to vibe code and experiment. And if I were the CEO of a company now, that's what I would be doing.

I would say you all have a week. You're gonna come into the office. I'm gonna feed you all of that good stuff, but you're

Mm-hmm.

gonna hackathon your way to whatever you want, right? Build whatever you want.

Yeah.

And it doesn't necessarily have to be with the company, like build something to help your childcare schedule or whatever. But because it has to be something, people will only have that aha moment when they build something that's actually useful for them, right? and so.

I think that could help with the creativity thing as well, if people were given the space to play around with stuff too. I mean, I wish I had more time to like play around with Mid Journey, the stuff that Mid Journey is doing, other than the weird MRI scan thing that they watched today,

Yeah, yeah. Released today. Yeah.

which was absolutely insane.

Yeah.

their like product is amazing. It's so cool. And I have designer friends who are saying how cool it is and how how much they enjoy playing around with it. So I wish I had more time to do that.

Find cool use cases and just make cool stuff. I feel like I was I was playing a little bit with it the other day and I felt like I was back in the beginning of my like internet days when I had my GeoCities website and I was like getting little graphics and like and like making my little my little world, right? And that's so fun. That's really creative. So just giving people time. More of that.

More of that. Yes, hundred percent.

More of that.

Amen.

And rich people giving money to to to play around. as we said earlier. Love it. thank you so much for your time. I don't know. it's wonderful. Thank you so much.

Thank you guys. This is amazing. Thank you.

Thank you very much.