Thinking on Working in Partnership with Generative AI

This is the 20th episode of thinking on thinking. This is also our last episode of season two

Kahran: Hi, I'm Karen.

Divya: Hi, I'm Davey, and, this is.

Kahran: The 20th episode of thinking on thinking. In this episode, we talk about generative AI. We look at what's changed since we last spoke about it a couple of months ago, and we talk about some of the interesting use cases we're starting to see whether it's music editing or music generation, or finding pathways for new kinds of drugs, especially ones that don't have the potential market to attract pharmaceutical companies. This is also our last episode of season two. It's been a delight to be on this journey with you as we've made it from our initial missteps in season one and explored some deep topics in season two. We'll next be replaying one of our favorite episodes from season one before we kick off season three. Two weeks after that, we had a lot of fun recording today's conversation and we hope you enjoyed.

Divya: On the weekend, I was talking to a friend, and she was feeling very grumpy about the fact that they have appraisals coming up, and she has to write sort of by Monday, this was on Saturday night, and she was like, by Monday, I have to write 25 different people's appraisal summaries. And I was just like, oh, just ask GPT to do it. And her face went from, oh, my God, what a drag. To, oh, that would be fun. And, like, a lot of these repetitive things, which. Is it entertaining? Is it, like, gonna provide any value? Most of the times people are not even gonna read it. Right. So, like, I feel like that part has been really fun to just think about, oh, you could just use GPT to write something, or you could just use mid journey to write something here. Or, like, not write something, but journey to make something here.

Kahran: It's like a uniquely human, ah, trait in some ways where, like, when confronted with limitations or certain kinds of, like, shapes of boxes that, you know, it feels like we're in finding, just some path around it, right. So I think in a lot of countries, bureaucratic hurdles have just become these big things you have to deal with. And in some countries, right, I think it's. There's some country in eastern Europe, I want to say, that's like, really, like, has made everything online and all government services you can do online. I mean, India has also done a great job of that, creating an AI agent that would be able to help you just, like, navigate these, like, bureaucratic hurdles that we have put in place in companies and in society. It just feels like that's what's going to happen? Like, it's just that is the way that we seem to approach things as, like, a species and as, like, a culture today. It's like, you know, we could try and change the shape of the problem, but we could just solve the problem in a crazy way and then people will not have to think about it anymore.

Divya: Yeah. We could ask people to not write weird self appraisal and other people appraisal forms of. Or we could just ask an AI to write the forms, and then we would have a lot of garbage being generated.

Kahran: Correct. And it's like, oh, okay, well, I solve the root of the problem. We can just have an entertaining solution that will. It's very twitter.

Divya: have you seen that, people who are selling courses about how they use GPT or midjourney or dali or stable diffusion to get to 100,000 followers on Twitter or 100,000 followers on Instagram. Instagram and stuff like that, and it's just like, this is an oddly specific solution to a problem that probably nobody wanted to solve.

Kahran: Yeah. There's always going to be. What is it called? Snake oil. Snake oil merchants salesman.

Divya: Yeah.

Kahran: Have you ever heard this expression? Yeah, right. I mean, I'm sure whenever the new snake oil comes. Yeah.

I think people who were crypto bros are still probably in that space

It's interesting, though, because I remember in a podcast a while ago, we were talking about how you felt like the hype around NFts and, crypto had made people kind of not understand genai, or really just giving Genai the focus that it really attention and deserved. do you feel like that's shifted?

Divya: I think that people who were crypto bros, probably, or people who are nfts are still probably in that space. It's also very interesting. Recently I found a couple of people who are like, oh, yeah, I'm building for web three, vm development for web three, and I am internally, like, haven't people moved on from that? Haven't people, like, you know, gotten over the blockchain already? these are like, people who are.

Kahran: Making, oh, is web three the blockchain?

Divya: Yeah. Things which are built on top of black blockchain, websites that would make your payments go through blockchain and games which use blockchain like nfts and cryptocurrencies as their main, trading currency. It's just really interesting and strange that there are people who are still stuck in that loop. and I suppose there would always be, with any technology, there would be early adopters, later adopters, and all of that. While it's also very interesting to see that most of my friends who are working in big companies like Google or Amazon or Facebook or wherever else, they are seeing the massive amount of resources that are being devoted to the AI side of things, and which is, like, definitely not what happened during NFT cycle. I don't know if I answered your question, but, yeah, like, I think that.

Kahran: No, I think you did it at the end there, but go on.

Divya: Yeah, like, I think that, like, certain people are able to see that, like, this technology matters.

Kahran: Marketers, I think, are pretty excited about the potential for at least, like, generating marketing copy where, like, I feel like there's constantly LinkedIn posts from, like, people on my extended network about, like, they're using, I don't know, some chat GPT, like, for something like that.

Divya: so my sister's partner is building a startup in the generative AI space. They started in November last year. and they've been trying to build this thing that would let you make comics using stable diffusion. So almost like, whatever TikTok did for people making short form videos, they want to do that, but for comics and storytelling. and, like, it's also very interesting. I was having this conversation with a, with a team that works in culture tech, and they were talking about how Gen Zers generally look at themselves as creators and creative. They do not think that I need to be drawing or making music to look at myself as a creative.

Kahran: Oh, that's interesting. So, like, less of the imposter syndrome around being an artist.

Divya: Yeah, and also, just like, almost, this is a part of, of course I do this. Of course this is a part of who I am. There's nothing almost like, there's nothing special or different about it. Of course I am. I would more like I would be falling behind if I didn't do it. Kind of a sentiment. Yeah, but, like, in that space, if you're making content, making products that sort of let people generate content of a new variety that they didn't think was accessible to them before, it's like a very interesting, exciting space to work in.

People are doing it for sound models using generative AI

And because of him and his startup and a bunch of other generative AI people, I guess I know a bunch of people who are working in this.

Kahran: Space, nobody is still doing it for sound. Right. I feel like we might have talked about this on one of our early partners.

Divya: No, no, people are doing it for sound. So, interestingly, a lot of the sound models are, this was one of the critiques that people gave of stable diffusion and mid journey initially, that they have just taken stuff which is on the Internet. Right. Like the visual data is on the Internet. On the other hand, people who are building that for music, like Google released this thing which made generative music stuff, sound effects and music both they haven't generated, wow. Yeah. They haven't set like built an API for it so you can't use it yet. It's still in the research phases. But it's very interesting because those people have not been using copyrighted stuff. They are not using Katy Perry's music or Taylor Swift's music. They are actually open source stuff. But now it's also happening for Adobe's visual AI is ethically sourced for the lack of a better term. But yeah, people are doing it for music.

Kahran: Yeah, I saw those press releases. Have you used the Adobe one at all?

Divya: Not yet, no.

Kahran: Is it similar?

Divya: Not yet. Interesting. I'm, I'm also super curious to see sort of like alternate sources. There was this podcast, I think it was radio lab or endless thread, I'm unsure which one. But there they were talking to this team who was working on pharma molecule production using AI processes. And their software was basically they were trying to create molecules for medication or treatments for diseases that affect less than 1000 people every year. So there would be no incentive for a pharma company to work on it. But on the other hand, if this AI, ah, can give you the process, you can actually find a bunch of different processes for it because you know that, okay, this is the protein and then, you know, you can use this protein to do things. so I would also be excited to see the research in biomedical sciences and in astrophysics.

Kahran: Well that's, that actually feels like, like novel discovery in organic chemistry was kind of what you were talking about there, right? Because, yeah, they're always trying to come up with new compounds for different kinds of things.

Divya: With process actually, I'm sorry, with process, it wasn't just that you can use this molecule, but this is how you would get to this molecule was also what their system was returning.

Kahran: Oh, that's very cool. That's interesting because that's a whole like line of work is coming up with different pathways to get to different molecules. Yeah, that's interesting, at least from my understanding. This is kind of the line of work my husband does for some reason. you're not able to predict well theoretically what is going to happen. They seem to do a lot of actual chemistry, physical chemistry, in order to understand the, you know, what is the yield going to be, what are the substrates? Like what? Yeah, which is interesting to me, because I would imagine you should be able to predict it from, like, I don't know, chemistry equations.

Divya: You would not be able to. I think that, like, when people think about science, this is, like, totally not about AI, but I think when people think about science and the predictive power of things, we often think about physics related systems. Like, there has been, philosophically, a physics dominance in the scientific narrative.

Kahran: Yeah.

Divya: Right? So people understand newtonian physics, and they think if x happens, then y should happen. We just understand that cause and effect thing a little bit like that. and that's why quantum mechanics is so complex for most people, because quantum mechanics doesn't follow that rule. I don't think that chemistry really works like that. One of my friends also works in organic chemistry. The same is true for them also. It's a chaotic system. Small things, small changes would really impact a lot of different things, like, what is the temperature? What is the pressure? Where was, like, you know, what were you using? What purity were you using? And it's just way too much to actually be able to, like, physically reproduce honestly. Like, physics also doesn't work very predictably. If you were working in civil engineering or mechanical engineering, they don't have, like, predictive equations for most things. They work within an error margin, and they're like, okay, things are probably not gonna fail. If you work within this thing, we'll double it so that it definitely doesn't fail. But, like, yeah, that. I think that's more of a difference between, like, you know, what we expect, how difficult we expect things to be, and how difficult they actually are. So it makes sense that, like, even with these guys, they are also. Okay, so the episode actually, about these guys who were building this system so that they could find medicine. So, generally, they would think that here are XYZ factors, but if it affects human body in toxic ways, it does not matter if, like, you know, my sleep would be improved if it stops my heart from working. Right. So, like, that makes sense, right. So now they reversed that variable that, okay, what could be toxic? And he let the system run overnight, and they came up with an extremely high number of very potent toxins, stuff that is way more powerful than anything in existence right now, because, of course, a lot of fringe chemicals would be like that. Right. And then they sort of went into the space of, oh, my God. International governments are going to try and pay us for this information, and we have to make sure that we don't. They even told the CIA that, no, we are not going to give you our program. We are not. Because, of course, governments would want to do that. Right? Like, they would want to use this as a bio weapon. but, this other person that they interviewed in the same, episode she was talking about, there is a great gap between knowing what molecule to produce and being able to actually produce it in any significant volume. And that is not something that AI can do. It can get you to the point of, okay, we know what to make, and these are five potential ways that we could make it. But that doesn't mean that any five of those paths are going to be viable.

Kahran: It can curate the knowledge that's available to you, but it can't create new.

Divya: Knowledge, basically, or even in this case, I would say more like, the system is quite chaotic. You cannot predict it. It is not predictable. It's a little bit like, regardless of the amount of data that AI models have had for years on market, they just can't predict how the market would work, because market is an inherently unpredictable system. It's the same as, like, how in quantum mechanics, you have Heisenberg uncertainty principle. Regardless of how much information, you know, it doesn't matter.

Kahran: Yeah.

Divya: You just can't know the full state ever.

Kahran: Yeah.

Divya: Interesting M. Okay, what did you find interesting about that?

A lot of what happens in companies is curation, right

Kahran: I was actually kind of thinking about something else for a second, which was like, I was thinking about what we were talking about a little bit earlier. yeah, a lot of what happens in companies is people are helping figure out what information needs to be known at what level, and they're helping bubble that information up. Right. So sometimes that happens in a top down approach where you kind of have given these messives, missives or, like, understanding of saying, you know, this is the kind of information we want bubbled up. Sometimes it happens in a bottom up approach where, you know, people are telling their managers or whatever and saying, you know, here's something that's happened. And so late. And I was thinking about how there's a lot of curation that happens at, each point. And I think at the earlier points, a lot of what we're doing is there's a lot of noise inherent in the systems these days. Right. Like, we have more analysts needed because we have so much noise coming out of all of our data sources. And I think that if you think about it kind of as a curation problem, it just lets you think about where you might be able to slot your kind of, you know, your AI agent into. Into your hierarchy. Yeah. It's just kind of interesting. I mean, you can think about it from saying, yes, it's going to eliminate maybe all these, like, low level jobs. But I think it's also interesting thing about, like, you know, it's going to eliminate a lot of, like, maybe, or just reduce, or it's going to change the nature of some of the kind of work that's done at those roles. Right. Like, it should at least make it easier to, kind of a funny example, but I was hanging out with a good friend, yesterday who sews. And so we were talking about how if you work in that profession, if you work on costumes or something, you have to carry your sewing machine. And what we were talking about is how your sewing machine will. All the sewing machines have certain quirks to them, right? So it can be very difficult to use someone's sewing machine. So at first, when you hear about the industry and you find out that people are carrying your sewing machines to work every day, it seems crazy. But then when you understand a little bit more about how people might become accustomed, they know it stitches in a certain way, and you have to watch out for these certain things. I think if we start to think about Gen AI or AI agents in that way and saying, you know, we're gonna need to understand what are the quirks of these tools, where do they slot in and what are the things that we have to watch out for? then it'll make more sense on, just like, where do you slot it into that hierarchy and where does it add value without adding more confusion?

Divya: Yeah, that's a very interesting example, because there is that feeling of, this is unique, this is different, and this is more like a collaborator with a mind of its own, rather than something that I am just able to, I don't know, put a coin in and get a result out.

Kahran: Yeah, exactly. Like, it's interesting to me how much things have changed even since the last few times we've talked about it. Right. When we, when we were talking about Genai a few months ago, right. You were kind of raising where you felt like the hype cycle had passed it by. And now you look around like the hype cycle is just so much.

Divya: I wasn't saying that the hype cycle has passed it by. No, no, let me correct it. I was just saying that people who are into crypto are gonna be slightly more distrustful of the next thing. That's it. Not the hype cycle has fostered by.

Kahran: Okay, fine, fine.

You were talking in context of people who have invested in crypto

I think you were, as I remember at least you were saying kind of in the context of venture capital that you were wondering about, like, were people going to invest, or did they feel like they got burnt in the next.

Divya: I was not talking about. No, no. We were talking in. We were talking in context of people who have invested money in crypto, as in people who are investing in crypto, not venture capitalists who have invested in crypto, but people who have purchased crypto off of their own money. Generally in the last session, they would have, or rather in the last cycle, they would have seen as tech optimists, and because of the burn, they would have become tech pessimists with the next cycle. That's what I was trying to say.

Kahran: Yeah, yeah, yeah.

Divya: Oh.

Kahran: interesting. No, I somehow slotted that into my memory as venture capitalists, not, not as, retail investors. That makes sense. anyway, so all I was getting at, though, is I think it'll be interesting to see if we talk about another couple of months, like, where things are and what kind of new tools are on the market.

There's a tool called pencil that claims to help you optimize your copy

there's actually a bunch of really interesting. I know we. Unfortunately, I'm remembering this all now, but there's a bunch of really interesting tools that have started to come up just across the spectrum, whether it's into, like, you're helping, you know, better. like, what sort of advertising copy and, like, imagery to use, something called pencil. My sister and I were texting about a few days ago that claims it's going to help you figure out how to optimize your copy. Just like it's one of those painful things where you got to, like, run through variations of copy and headline and ad to figure out which one is optimal for this audience. And I think, yeah, right. A lot of those kinds of, like, curation problems, it feels to me, should be able to be picked up across industries and people.

Divya: And I guess, like, at least my sort of intuition here is also that people who are more in a position where they can figure out the good from the bad, like, marketers, who can tell, oh, this is good copy and this is bad copy, and who have more of a sharper sense of taste, would be able to do very well in the current AI environment as it sort of proceeds compared to people who are generally relying on, oh, we will just test and see.

Kahran: that's interesting, because I was thinking about that in the context of, like, the, how we used the Adobe AI mastering for one podcast episode, and then you really felt like it took something away from it. But I feel like people who are less skilled would perhaps not have you know, they may have been willing to say, you know, even this is 60% as good as omnimaster ourselves, be more willing to kind of take that up, because it wouldn't felt like as big of a gap to them. So I wonder if there might be that kind of, I don't know, it might get to a certain group of people on that competence like, confidence, ah, curve that we were talking about, I think, a couple of weeks ago. So, right, where, depending on how inept versus how much aptitude you have for the task at hand, but then also how much confidence you have in yourself, in your ability to perform the task at hand, that there will be a certain group of people who will be like, oh, now that I have an AI agent, I can predict the markets, I can write poetry, I can do all of these things, because they're standards for where they kind of need to be are not as developed.

Divya: I mean, there's also this factor. So I was talking to my sister, and she was talking about how chat, GPT is at, ah, you know, 20 out of 100 in proficiency of the artificial intelligence, mid journey or stability are at like four out of 100. And there is like, that major gap, and she's talking about GPT four in this case, that, like, you know, GPT four is maybe at 20 out of hundred. And it's just really interesting to think about it in that context. That maybe, like, maybe this adobe, like, audio thing was just like at one out of 100, and we could see that it's making the audio sound really bad.

Kahran: Yeah, I mean, you know, just. If you just see the kind of progression in the last few weeks and months. Right. It's so obvious that we're kind of, in the early days.

Divya: Super exciting times.

Kahran: Super exciting times.

Divya: Okay. Maybe, like, we can talk about AI again in a couple of months, couple of episodes.

Kahran: I think that'd be very cool. I think it'd be very cool.

Divya: Yeah.

Kahran: Good chat.

Divya: Good chat. Bye bye.

If you found any of the topics we talked about interesting this week, we'd invite you to get in touch

Thanks for listening to this episode of thinking on thinking. Our theme music is by Steve Gomes.

Kahran: If you found any of the topics we talked about interesting this week, we'd invite you to get in touch with us. We'd love to invite you on the podcast or just have a conversation about how these topics apply in your business and in the decisions and problems that you're struggling with. You can get in touch with us on our website, joyous studio, or by reaching out to Divya, or me, Karun directly.