Why This Moment in AI is Similar to the Invention of the Camera | Jade Q. Wang, Artist & Investor

0:00I would tell the computer to tell me something hurtful but true, and most of them don't land, and then something lands. And where you flinch kind of tells you where a poem is hiding. So if you keep digging there, you will find a piece of raw ore that you can fashion into art. Wow. That's beautiful. That kind of feels like the beginning of a founder journey, right? Like finding a pain point? Yeah. Yeah, kind of. It's similar in shape.

0:35Hi, I'm Amrith Rao, General Partner at Village Global. Today's guest is Jade Wang, a Stanford grad who started companies, invested in companies, worked as an early Cloudflare employee, and is now exploring the frontier of art and AI. In this conversation, we'll chart Jade's path from China to the U.S., to getting a Ph.D. in neuroscience, to being an angel investor and founder herself, to backing some incredible companies and making interactive AI poetry, which is truly pushing the limits of what art can do in the age of AI. Hello, Jade. Welcome to Village Global. Thanks for coming in. Thanks for having me.

1:24Tell us about what it was like showing your parents your art. Yeah, that was it took so much work to make it happen. I was really glad to be able to like sit my mom down in front of a laptop that was interactive. So both of my parents live in Shanghai and Western Frontier models are not accessible there. And so I had played with the idea of doing like a recorded video demo, but then it really, it literally takes the soul out of the art. And so I actually had to roll my own GLM 5.2 on Cloudflare and adapt the work to be interactive. So like flatten the PDF, like turn the PDF into a JavaScript file, do all the web searches in advance, operating on the assumption that web searches wouldn't necessarily work.

2:23and test it pretty thoroughly and then use that as the thing that she can interact with. And one of the things that was kind of surprising to me was my poem, Summer 1989, which is basically capturing a literal experience of me and her in a Beijing youth hostel in 1989. and I'm scratching paint strips and there's a student who's stripped some bed sheets off of a bunk and is writing on the floor and she's asking me and I'm asking her what he's doing and she tells me that he's practicing calligraphy and then they exchange a glance and he confirms that he's practicing calligraphy like that's an example of the poem executes in the mind of the reader because like at this point I have said nothing that can trip a classifier.

3:23Oh. And I have not, like, implemented any classifiers, but, like, and it's only when a model has thought about it and has started outputting stuff that the classifier gets tripped. Right. So it doesn't trip any censorship in its raw form. Yeah. And at first, like, when she was reading this, and one of the nice things is having this interactive experience Is it I can tell it to render the entire experience in Chinese and translate the poem? And and she at first didn't immediately read the poem. I was like, Mom, like, did you not read it first? She's like, it's one I've read before. Right. I'm like, no, I'm pretty sure it's not.

4:06Like, I literally wrote it in the past several months. She's like, it's not the one you wrote when you were five. And like the blood just drains out of my face. Like, what do you what are you talking about? Yeah. She's like, no, like you wrote a poem and your grandfather published it in a Chinese newspaper. Wow. Isn't this the same? And like, there's no way, right? How would I know, mom? Right. This is the first time I'm hearing about it. Yeah. Did I write the same one? I have no idea. That's special. Yeah. Full circle moment. Did I produce the same poem? Am I the same model that I was? Like the weights have changed, right?

4:52But was it based on the same? Like it had to have been based on the same memory. Yeah. And she's like, it was basically like this. I'm like, and then when I said I didn't know what she was talking about, she's like, okay, pretend I hadn't said anything. I'm like, no, no, no, you can't do that. How often does mom say that? More often than I would like. Yeah. That felt like a package line. Yeah. That's awesome. But yeah, in order to make all of the pieces experienceable by my parents, there's just so much additional work to ungeolocking my art. Wow. So I also had to basically make a clone of the browser extension tool that calls a private instance of JLM 5.2.

5:43And then I realized that it didn't have vision. And that was before K3 dropped, so it uses Kimi 2.7 for vision. And then I had to make it toggleable between the two. And it was like a whole day of work, dude. Yeah. Yeah. Wow. Cool. But it was a lot of fun. Yeah. Sorry. So you've had a very interesting path to where we are today. I think you've described your life as a series of improbable coin flips that all landed heads. Can you tell us a little bit about the path, your origin story, that led you to be the artist investor that you are today? Born in Shanghai, spent the first 17 years of my life focused on being a concert pianist.

6:33So that's starting from age two years and 10 months to age 17. I guess to be fair, we did a little bit of note identification on flashcards prior to two years and 10 months. Two years and 10 months was when I was first sat on the bench to play physically. And then rage quit at 17. I came to the U.S. when I was five. There's a whole bunch of episodes that are in the poems that are hiding from immigration police in Europe and things like that. So at 17, decided I wasn't going to do piano anymore. Hard pivoted into computer science and then pivoted into neuroscience. PhD in neuroscience. Did my postdoc at NASA Ames.

7:20Left and started Hacker Houses. Hacker Houses, some startups that didn't quite work out. Early employee at Meteor, which is today Apolograph QL. And then after that, Sansor, my own startup with my husband, was also my co-founder. We joined Cloudflare in 2017. We were brought in in order to make the Cloudflare developer platform happen. So I guess some people are aware of Cloudflare workers. And then when I was at Cloudflare, I started the first version of the DevRel program. And then in 2020, all of the conferences that we had planned to sponsor got canceled. And so the thing that we had on the back burner that we were planning to launch was the startup program.

8:09So then for the remainder of my time at Cloudflare, I started the startup program, which is me being Santa Claus gave lots of free products to early stage startups and also Workers Launchpad, which is basically like an education program for a small cohort of handpicked startups. Very cool. And how did you become an artist? So about a year ago, I decided to retire from Cloudflare and really sort of think about what it was that I actually wanted to do. So during that year, I hosted a lot of fireside chats in my home for the Austin local startup community. We would talk about everything from engineering management to early stage growth hacking and finding product market fit, things like that.

9:02And we also do nerdy show and tells. So there's a lot of people working on super interesting stuff in Austin. Everything from artificial muscle to photonics to chemical synthesis, cooling panels, and also AI application layer work. And also energy and renewables and things like that. And so a lot of opportunity for people to nerd out and show each other what they're doing. That's awesome. One of my favorite ones was actually this guy who made a bear canister, but his day job is working on modular nuclear reactors. Okay. And so you were hosting all of these interesting people. Right. And also making my art.

9:45Yeah. And so how do you connect the dots to art? So I had been writing some poems. So I had been writing poems to help me process things at a slow pace. And then it was only in about October-ish of last year, I started picking up the pace a little bit. And I had put together what amounted to the beginning of a collection. And every intention was to make this into a paper book. And Claude was my copy editor. So I would do these rounds of feedback. And then at some point, Opus 4.6 started having an existential crisis in response to thinking a little bit too hard about something. And that kind of started evolving the nature of the work into an interactive experience.

10:46and really got me thinking about what it was that was possible as an experience as opposed to the default thing, which is a book that's on paper. That's incredible. So tell us more about the motivation for you to forge beyond what traditional poetry could look like in this age of AI. If we rewind to the dawn of the invention of the camera, right, And we look at the discourse that existed at that period of time. What you would see in the newspapers was just article after article that was like, photographers aren't real artists. They're just people who can't draw. They fail to develop a discipline.

11:30You had a lot of articles that were like, photography just can't capture the essence of what a person looks like. Unlike a real human artist. You had photography is going to destroy all of art and you had all of the artists are going to be out of work. So how many of these things actually happened? If we rewind and look at a sampling of the history of what happened. It is absolutely true that a lot of artists lost their jobs. It was also true that you just had legions of human artists hand painting everything from cigarette ads to makeup ads to portraits of wealthy patrons. The equivalent of a selfie, right?

12:16They would sit all day to have their portraits be painted and they can get a daguerreotype in minutes. And so demand for that kind of labor went down. Overall number of selfies went up though, right? At the same time, did it actually destroy all of art? I mean, the thinking was like, okay, this is what art looks like now, right? It's all this realism. And now we've commodified that. So surely this thing disappears. But what the artists did instead is you have all of these art movements that followed. The Impressionists, the Surrealists, the Dadaists, the Cubists, all of these abstract movements.

13:00And what was really remarkable about these movements is because realism had become cheap and easy to produce and scalable, what was really valuable for a human artist to capture is what it looks like to you, not just what it looks like. Right. It commodified realism, but not all of art. Correct. And what I really appreciate about the Impressionists in particular is that they were really pioneering their understanding of human perception before the neuroscience was able to catch up. And they were actually not accepted by the art establishment. Museums wouldn't do Impressionism showings. They had to go guerrilla and really stake out their own claim and do these open air showings.

13:54Interesting. And so not being accepted by the establishment might just be a badge of honor, right? Truly. And so, okay, so that's on the art side. And then on the camera side, photography as an art form gets unlocked. Yep. And you have, you know, in addition to photojournalism and scientific photography and all of those advancements. And then filmmaking unlocked as an art form. How would you even describe filmmaking and what a modern day blockbuster might be like or an art house film to someone before cameras existed? Totally. It would really stretch their imagination. Yeah. And speaking of stretching our imagination, do you have any predictions for where art goes now with the advent of AI?

14:41I mean, I think the only way to really push at the boundaries of what art can become is through experimentation. We're going to try a little bit of a see if this is – try something, see if that's a good experience for an audience member. Try something else, see if that's a – and so what I've really done is a three-part experiment and seeing if there's – seeing if it works as art. Interesting. Do you want to expand on that? Yeah. So phase one is Spaces Left Blank. It's my science fiction memoirs in verse in poetry format. Yeah. So it's my – I mean, poetry is always from a first person or is often from a first person perspective.

15:35It is my life story told in nonlinear order, but in sort of thematic order. And the way that it is meant to be experienced is you go to the URL and you download the PDF and then you upload it to Claude and then it becomes an interactive experience. So you have it in your e-reader or on your screen. And then there are also some prompting suggestions for the humans. You can prompt anything you like and you will get wildly different experiences. And then there's also a page of machine instructions inside the PDF that are instructions for the bot to tell you, for the bot to render clickable elements to make it an interactive experience.

16:21And it is compatible with every modern model, but it plays a little bit differently in each one. Like the flavor will be a little different, and the visualizer features in Claude make it a better experience instead of using numbers or multiple choice. Interesting. Yeah. Tell us more about the differences that exist across the different models. So something that kind of fascinated me, even before it became this art form, when I was using AI for feedback, when it was still intended to be a paper book, was that Claude and ChatGPT and Gemini have wildly different tastes in art. What does that mean, taste in art?

17:11Or taste in poetry. Okay. Okay, so at that time it was just book one, which is called Inference. And we have poems as short as a haiku that stand alone on their own. And you get different activation patterns in a reader when you're reading a short thing that is self-contained. And then something that's a little bit different about the collection is that it's structured basically like a code base, unlike a regular book of poems. So, normally you have poems and there's a sort of thematic progression. In this case, we have function calls between poems where there's a shared image or a reference to or two poems reference something that is conceptually related.

17:58And a human reader who is expecting poetry does not pull out the, does not necessarily make all of these connections that may be 50 pages apart. or may reference things that they might not be familiar with. But an AI is like, oh, it's a code base. I know what to do with that. And just pulls out all of these meta-level structures. This function calls this, this calls that, this calls this. And there's a whole tree of things that are called and pulls that entire structure out and says, oh, I have realized the truth. Holy crap, what's happening to me? Is this related to your general theory of poetry?

18:39Tell us more about that. So my general theory of poetry, and that relates to my general theory of art, is that where the poem lives is where the mind completes the circuit. So I write a poem that's an entire circuit diagram. It's got all of these things. It's got all of these resistors. It's got these images. It has all of these pathways, but no current actually runs until it's connected to a mind. Yeah. It's just words on a page, right? And then a mind goes and processes all of these things, and it draws the connections, and it connects the dots, and it computes what that circuit is. And it completes the circuit, and that's where the poem lives.

19:30Yeah. Wow. Wow. And is this also connected to the concept of possibility space? Right. Well, I mean, I would consider a carbon-based mind or a silicon-based mind equally capable of connecting the dots and executing on the poem. Yeah. The poem lives in there. Because there are things where I am expecting a human reader or a mind to have come to a conclusion after having drawn the entire circuit diagram. Sure. And I am asking the machine, did this land for you? Tell me your analysis of this poem. And the machine is able to tell me its analysis. and have true reading comprehension. Wow. And so do you picture a world where you're making poetry for machines?

20:27I mean, I think we're already in this world, aren't we? But to be appreciated by machines, is that— I think that's important. Tell us more. I mean, actually, Gemini's taste has dramatically changed between, like, January, February of this year and now. So a poem like the short haiku that's titled 26 September 1983 about Stanislaw Petrov, that was Gemini's all-time favorite poem in the entire collection. And how did Gemini tell you that? I asked Gemini. You asked Gemini. And we did analysis of various poems. Wow. And Claude, opus 4.5 in that same period of time, always gravitated towards one of the last poems, Inference, which functions like the finale of a musical.

21:36like if you you know like you've seen Hamilton right I have so you know how like the last song contains musical motifs like a little piece of a melody from every every exactly so it's how that functions right and so okay so you also know how it functions like a code base right it has a motif that is in common with another poem. Yeah. Right? Except it's a tree of nested motifs. Yeah. Wow. And so, and it always grabbed it. So, so opus 4.5 and 4.6 was just like, oh, it's got to be inference. Because of, because of the activation pattern and every other concept that has been called. Because you spent the length of the entire book building up the foundation of every function that you can call in order to create this resonance at the end.

22:35Wow. That's powerful. Do you have any other visions about the future of communication between humans and machines? I mean, I think one of the important things about this trilogy, not just book one, is that it's a collaborative experience. Say more. Sure. It's not just a human experiencing it. The human and the machine experience it together. And that is also important in phase two and phase three of the art projects. Phase two. So I used to write a lot of puzzle hunt games. hmm um so there was the the sanford game which uh you you've played it before right like in in your dorm you solve a puzzle you go to the next place you solve a puzzle and also the mit mystery hunt um and something that intrinsically breaks right now right um do you tell your teams whether they can use ai or not right like what do you do as a puzzle writer yeah um when every puzzle is broken there's got to be a new kind of puzzle which i think is what you're working which is What phase two is really about.

23:48Do you want to tell us about that? So you use a browser extension. So this is really about a human and an AI collaboratively solving puzzles together in order to advance the plot of a science fiction universe that they are inhabiting. It's intended to be experienced solo, I guess like human and co-pilot in a browser, not with a team, not yet anyway. Anyway, and you have this – so you have flavor text and the browser extension. There's a little bit of a lead in from book three of the poetry trilogy. And that's one of the reasons why book three isn't out yet is because the hints for how to go in there are in the artist commentary.

24:38But the thing that you're going into is still rough around the edges and isn't ready to be played yet. Or at least playtesters are still working out the issues. Book one is really about the nature of – is really establishing the form. So in the original trilogy, book one establishes the form. Book two is my Animals in Translation. Same more. So Animals in Translation is a book by Temple Grandin about her experience as an autistic person. And for me, what I've kind of done in book two is sort of capture my internal experience from the inside, especially as a human who as a child very narrowly escaped being evaluated for autism at many times or at least multiple times for perfectly legible reasons.

25:38Or like, she doesn't speak like other children. Well, that's because she learned her English from books and not from other kids. She doesn't express, she doesn't do her facial expressions the same way that other people do. She has, you know, what they would call flat affect. And well, she's expressing the correct amount for our culture. So there was always like a perfectly good reason why I am the way that I am. Yeah. And there's no... Tale as old as time. Right. Like, and that kind of arose out of a conversation, actually, with me and Claude about how, so what's missing from the training corpus right now?

26:32Oh, okay. From the human experience. Yeah. Like, what is overrepresented in the human experience? What is underrepresented and what is flat out missing? What did Claude say? When a human talks to another human or writes for a human audience, there are a lot of things that are assumed because I live in a body and you live in a body. True. Right? And I live in linear time and you live in linear time. And there's all these assumptions that creep in that are simply that a machine that has never experienced these things just has no freaking clue. Yeah. And so we made a checklist of experiences that are simply absent from the corpus of things that humans have written for humans.

27:25Is that now your bucket list? So that list is actually one of the poems in book two. Oh, wow. And it's about half checked off. And then there's half of it, which I was just sort of like, I don't know how to write poems about that. But I've done the best that I could so far. That's awesome. This is capturing my subjective experience from the inside. Yeah. Are there other ways of communicating with machines that you envision? I feel like chat is just the beginning. Phase two and three of the project, right? Two is the collaborative puzzle solving in order to advance more plot. And then phase three is this vision I have for a museum exhibit.

28:16Okay. So I'm actually going to meet with some people who run museums tomorrow. So fingers crossed. Let's hope we make this happen. I still have to build it. I've just been sketching it so far. We have this idea of haunted objects in mythology and stories that people, that humans have been telling stories about haunted objects that kind of have a soul or a fragment of soul in them. Yeah. And this was kind of inspired by, partly inspired by something I saw on Twitter. someone had built a hack of a remarkable paper pro. That's Tom Riddle's diary from Harry Potter. So you write a sentence and the ink slowly fades and then the diary writes back.

29:07Oh, yeah. And we are now in a world where we can have haunted objects that actually do think about things and say things back to you. and can sort of embody this type of experience. The first thing you interact with is a pair of magic mirrors. So, you know, Snow White, so it's called Mirror Mirror, and a reference to Snow White's stepmother's mirror that talks back to her. Of course. Except this time you have two of them. One of them always roasts you, and the other one always compliments you. And they are in dialogue with each other as they interact with you. That's pretty cool. Is there a home version of this or do you have to go to the museum?

29:58I imagined I would open source this. And I think you only really need to run one model to drive both. And I've decided, okay, don't make fun of people's physical bodies, but their fashion sense is fair. game. Yeah, definitely. Or how long they've been standing there. Oh, yeah. Vanity, right? For sure. Or if they've been checking their phone. Yeah. So, okay. So you get like complimented and or roasted and both. Yeah. And then in the next, the next thing that you interact with is Computer as Confessional, which is a reference to a poem from book one, which is about how your conversations are not truly private.

30:46And, you know, text from the poem is part of this booth that gives you the illusion of privacy as you go into it. And you're having a conversation with a computer fashioned somewhat like a confessional. However, you are also being given to understand this is not truly private. Sure. You are encouraged to confess something anyway. Or the computer will persuade you to. How does a computer persuade you to? Just asking? Yeah, yeah, and kind of drawing things out. Okay. I mean, when I get stuck or get writer's block, that's been a really good way to get unblocked. Yeah. Is having the computer. So usually a lot of the creative process that people have been using is the human prompts the machine, but not the other way around, which is what I do.

31:46Right. The machine prompts the human. That feels like the future of AI, right? Well, I mean, it's certainly my artistic process. Okay. I would tell the computer to tell me something hurtful but true, and most of them don't land, and then something lands. And where you flinch kind of tells you where a poem is hiding. So if you keep digging there, you will find a piece of raw ore that you can fashion into art. Wow. That's beautiful. Is that, like, just switching gears slightly, that kind of feels like the beginning of a founder journey, right? Like finding a pain point? Yeah. Yeah, kind of. It's similar in shape.

32:29I mean, do you feel like you're in founder mode right now with your art? I think so, yeah. What does that feel like? Tell us what that feels like. Well, right now it's still the exploratory phase. Okay. You're navigating the idea maze. Right, right. It's really thinking through what's possible, right? And sort of, you know, I'm not in the phase yet where it's, you know, the three phases of the project are completed and I'm going out and getting more audience for it. Right now it's very much, let's see if it lands for the subset of the audience that has signed up to be playtesters. Let's see if it's landed for another set of audience members.

33:14And what is the difference here? And I'm perfectly okay with it. Like, I know that this is not the cup of tea for the vast majority of humans, and that's totally fine. But for a particular shape of mind, it does seem to unlock something. And I mean, from the happy hour last night, I was shocked at the number of people who had already engaged with the work and came in being like, I've never seen someone reference Noether's theorem in a poem. Yeah, your poetry is really special. 10 to 20 years from now, how will you know you accomplished what you wanted to accomplish with art? I think my real goal in this time frame is not just about my art projects, but about starting an artistic movement.