Building and Investing in Consumer AI

Talia Goldberg is a partner at Bessemer Venture Partners where she focuses on consumer internet, software, and AI. She became the youngest elected partner in firm history. Bessemer is a VC firm with a long and storied history with investments in companies like Twilio, Shopify, Pinterest, LinkedIn, Yelp, Twitch, and many more.

In this episode, we cover a range of topics including:
- Why is consumer the biggest beneficiary of AI
- AI is generating massive consumer surplus
- Why is RLHF only a secondary moat
- Distribution tactics for consumer AI: virality via social proof, the Etsy effect, building for niche communities, standing on the shoulder of giants, "first order irrational, second order rational"
- Conversion tactics for consumer AI: entertainment as conversion, community-driven conversion, gambling psychology
- What does 2024 look like for AI startups and VCs

Talia's favorite book: Poor Charlie's Almanack (Author: Charlie Munger) 

--------
Where to find Prateek Joshi:

Newsletter: https://prateekjoshi.substack.com 
Website: https://prateekj.com 
LinkedIn: https://www.linkedin.com/in/prateek-joshi-91047b19 
Twitter: https://twitter.com/prateekvjoshi 

Transcript

Prateek Joshi (00:01.569)
Talia, thank you so much for joining me today.

Talia (00:04.726)
Thank you for having me. I'm excited to be here.

Prateek Joshi (00:07.601)
Let's start with the amazing article you wrote early this year about who the biggest beneficiary of the AI space race is. And it's not the computer chip providers, it's not the foundation model or the application layer, not even the incumbent platforms. You said it's the consumer. Why do you believe that's the case?

Talia (00:31.65)
Yeah, well, I think I wrote that article in March and I should look at Nvidia's stock price between the publishing date of that piece and today because they've certainly been a big beneficiary and I think will continue to be. But I do really firmly believe that it is the consumer and the individuals and us as

from all of the incredible progress that's happening in AI. And I call it a space race because it now really does feel in many ways like a space race. But for a long time, I've been having conversations with various investors and entrepreneurs, and there's a lot of different opinions on where there's gonna be the most value capture and different approaches. And the reality is that I think all different layers of the stack from...

the chip layer to the infrastructure layer to the application layer are going to have different, are going to capture massive value and, you know, in varying quantities. But the consumers stand by far and away the most to gain. And I think that's increasing, I've increasing conviction in that because we're seeing this dynamic of consumer surplus, which is basically the difference between the price that a consumer is willing to pay and the price that they actually pay.

And when I look at the value that AI is providing today, but also is really poised to provide over the next one year, five years decade, it is mind blowing. We have this augmented human intelligence, this ability to augment human intelligence, to drive massive progress, to lower the cost of healthcare, to give everyone a personalized tutor, to improve access to education, to increase productivity. And the benefits of that relative to the cost of it,

um, are just mind blowing. And I think we're already seeing that prices are plummeting. Um, we've seen, you know, ma orders of magnitude decreases in just the past two years of, um, APIs like open AI and, uh, the pro and, and the, at the same time, the quality is improving. So things are getting better. Prices are decreasing. Um, there's more and more competition. You have the open source ecosystem kind of innovating and improving. Um, and in general, I think that's really exciting for.

Talia (02:55.382)
humans for us as consumers and for us as a society.

Prateek Joshi (03:00.173)
Amazing. And in one of your writings, you mentioned how AI is generating massive consumer surplus. And you quoted an amazing Charlie Munger story, which I really liked, that you talk about, he talks about the textile business where more efficient machines will result in cost reductions. Great. But the benefit of cost reductions, they don't go to the person who buys the equipment.

So can you explain what this means in the context of AI and why does it matter?

Talia (03:36.854)
Yeah. So the example that Charlie Munger gave is an example that I think we've actually seen throughout a lot of technology and innovation and progress. And it's why in reality, I think a lot of technology has been, I think, deflationary, if anything. And so it may not even always show up in the GDP growth stats. And so you have this tension between, oh my gosh, we have this incredible increase in productivity and gains, and that's incredible. But on the other hand...

Prateek Joshi (03:54.571)
Hehehe

Talia (04:07.474)
I think technology does often become deflationary, at least in markets like this. And the example that Charlie Munger gave is that they had this loom business or this textile business where they came across a loom and basically there's a better loom and a better machine that could do things at half the price. And he said, oh shit, on the one hand you could have looked at that and been like, wow, there's this amazing new technology and this amazing new loom. But their response was...

Prateek Joshi (04:29.683)
Alright.

Talia (04:36.514)
gosh, I hope that doesn't work because if that works, I'm gonna have to close my entire business because the benefit of that lower price is just gonna go straight to the consumer because everyone's suddenly gonna have access to this machine or buy this machine and then they're gonna compete on offering lower prices to the customer and just erode away the margin. And that's an example of consumer surplus. And I think that's a cautionary tale in some ways about what you can see with

happening in AI and in various ecosystems where it does get cheaper and cheaper. And you have to be cheaper because there's competition. But if there's no moat and there's nothing that makes it meaningfully defensible, it's really hard to build a long term, durable and valuable business. And so as we think about that as an investor, as I think about that, it really brings into focus that. The value that create that.

one wants to create as a business and to capture real value over time is going to need to come not just from providing a really great product and technology, but having really strong moats in your business that allow you to be defensible over time. And those are traditional moats that we've seen forever, but I think that lesson is increasingly clear.

Prateek Joshi (05:56.057)
Yeah, I think there's a lot of parallels to be drawn from that story to what we are seeing here. And speaking of moats, obviously the topic is discussed a lot, especially in venture, when you look at a business and you want to see in the long term what can be the moat here, what's going to help them defend if they build something meaningful. And on the other hand, people have been talking a lot about

RLHF, Reinforcement Learning with Human Feedback. And it's been a very hot topic. Many companies are building solely on that basis, but you have said that RLHF is only a secondary mode, right? And by itself, it doesn't mean much. So first of all, why is that the case? And why is that, what can people do to, what can be the primary mode in this case?

Talia (06:52.854)
Yeah. So this is such an interesting topic and I wish I, you know, I'm, I'm technically not enough of an expert, but I wish I had, um, uh, an entrepreneur and one of our portfolio companies with me, um, because I think there's two points to this and my view on it is evolving, um, and changing in real time. But why is RLHF a secondary moat? I think, you know, to get to even have RLHF at

any meaningful scale where you're getting real feedback. Um, you have to have. A lot of users that are using your product consistently and engaging with it over time. And if you, I look at a company like mid journey, um, as a primary example of this, where they've been able to get this really interesting flywheel, um, going by getting more and more data as they've had more and more usage, but their ability to do that stemmed from building a really strong community.

and a distribution advantage and some type of network effect in the community. Of course, that also made their models better, which then led to even more users coming in and more feedback, which led to their models getting even better again. And so it's this amazing virtuous cycle, but it all starts with the first part, which is build something people want to use and love, and then use those seedlings to grow. And so I think it's what comes, it's the second part of cracking that first chicken and the egg, not enough in and of itself.

Although I will say, we're investors in one pretty successful now growth stage company that is trained their own AI models. And then I was recently chatting with another entrepreneur at another very well-known AI startup. And both of them were making an argument to me that RLHF is overrated. And that even though these companies have massive, massive

Talia (08:52.554)
of all their competitors, that actually they've started to train their models more on much smaller select data sets. And so the value and the power of actually having those users may at least in some use cases be lesser than I even anticipated. So it's possible that it's over time maybe going to be even a tertiary mode. I don't know, but it's really fascinating just to see how quickly the technology and the best practices are changing and what we're seeing from some of these startups that are really at the forefront that are now saying,

may not be the best and there are other ways to achieve similar outcomes that don't necessarily depend on massive scale.

Prateek Joshi (09:29.233)
Yeah, that's actually pretty interesting because one would think that, you know, I think you made a point where RLHF is a result of a distribution advantage or a network effect. Now you've got to build the primary thing in which case get a lot of users. Hopefully they love your product and then you can figure out how to use that to your advantage. All right. You wrote, I'm sorry. Yeah. Um, I, yeah, I think.

Talia (09:53.194)
But what are you seeing? What are you?

Prateek Joshi (09:58.805)
People are still, I've talked to a whole lot of founders and they're figuring out how do we utilize the interaction data and so that the model is not trained on, for example, just commerce data but also the interaction data. So I think it by itself is not gonna do much but it can be a nitro boost to an engine that's already working really well. So in this case, if you have a strong network effect, people are using it, they're loving it. You can use RLHF to extend the lead

Again, by itself, it does contribute something to the engine, but the exact amount is TBD because people are still experimenting. So that's something we're seeing play out in the market.

Talia (10:46.076)
Are there diminishing returns to scale there?

Prateek Joshi (10:48.849)
Yes, I think that the whole bunch of papers to that as well. But yes, there is after a certain point, the additional data point about user interaction doesn't add much to an advantage. So yeah, I mean, there's definitely diminishing returns. And also there's a nice paper where they talk about compute optimal LLMs, meaning for a given amount of data, how many model parameters will give you the best performance and after which it's just kind of meaningless in both directions.

It's pretty interesting how that is playing out. All right, so moving to another amazing article you wrote about distribution and conversion models for consumer startups, not related to AI, but I found it fascinating because many, many founders are building consumer AI startups. So I thought it would be interesting to kind of take that and look at it through an AI lens. So.

Let's start with the distribution tactics. The first one, we won't be able to cover all of them, but we'll cover a few interesting ones. So, virality via social proof. In consumer apps, you mentioned social proof is a well-observed behavior where people mirror the actions of others in an attempt to reflect correct behavior in that situation. So, how has this manifested itself in consumer AI companies?

Talia (12:15.402)
Yeah, what has been really interesting and exciting for me to see in AI, or this latest wave of AI has been that.

Talia (12:29.454)
creating a product that is so jaw-dropping, magical, and different can be distribution in and of itself. And I think that's what we saw with ChatGPT when it became the fastest growing application ever in history. And it's because it was effectively magic and jaw-dropping. And of course, some people saw it as just like a cool party trick, you know, and maybe, you know, don't use it as regularly as others. But it was had this just unbelievable.

partially first mover advantage and just magic to it that led to massive, massive word of mouth. And I haven't seen something like that into that scale or degree and also repeatedly, chat GPT is a premium example, but there are others that have similarly, I think if you're looking in companies like Jasper and 11 Labs and others, but just have this magical experience.

that you're like, holy shit, that's amazing, let me share it and share it with others and share my creations. That itself is distribution and it's unclear, how long that advantage lasts. At some point, you have to be able to have retention and capitalize on that magical moment. But it also means that first mover advantage is pretty important. And I think that's maybe one of the newer things that I've appreciated in the AI ecosystem is that first mover advantage can be a really, really powerful.

advantage to have early on in this ecosystem and getting ahead can matter a lot.

Prateek Joshi (14:02.797)
Amazing. The next distribution tactic is the Etsy effect, what you've written about, where it's a phenomenon where a seller promotes their own content or product from a marketplace, and that in turn brings more business. So how has this played out in consumer AI? And also maybe, what's an example of a startup or maybe a big company that is kind of doing this well?

Talia (14:29.49)
Yeah, I mean, I think we're seeing already in the AI, like character and open AI both leveraging this character, this distribution tactic. And the Etsy effect, the reason I called it that is that I think of Etsy as one of the early pioneers and examples of this where you might see an interesting craft, like an interesting woven bracelet that you thought was beautiful.

and you meet the craftsperson and they tell you, check out my storefront on Etsy, or you can purchase it on Etsy. And then you go to Etsy and you're looking at the woven bracelet. And suddenly, though, you see that there's like a hundred other woven bracelets that look just like it from, you know, many, many other creators and the likes there. And so on the one hand, it benefits that seller, that merchant. But on the other hand, you've suddenly now been introduced to this

Prateek Joshi (15:14.347)
Alright.

Talia (15:27.61)
has so much more there. And I think we've seen this with companies like Character and OpenAI because they've basically said, hey, create something, create a bot or create a character of some type. And let me create the critique character and share it with others and distribute it. And then they come in and they use it and they think it's really fun, but then they realize there's so much more that you can do there. There's so many more different types of applications or recipes.

and the like that emerge. And that just leads to this really, really nice distribution, one distribution because you're incentivized as the creator to get out the word and you wanna share it with your friends, but then this network effect because as more and more people join, you have more and more things that you can do there and more and more reasons to stay and more and more behaviors.

Prateek Joshi (16:18.001)
Yeah, if somebody were to create a Pratik character and sell it, there'll be a lot of returns because of the insufferable nature of Pratik as many of my dear friends have told me. But no, it's fun. I think that's a very good point you mentioned where I think it's a tactic that obviously Etsy is very famous and a big fan of just what it stands for. And I feel like it's a very, it's a great way to get your...

sellers who promote you, whereas obviously they're winning business, but you also get to win. The next distribution tactic is, you talked about, is building for specific niche communities. And startups can create world-class experiences to have that wow effect for that specific vertical, and then hopefully it spurs liquidity and growth, and then hopefully you expand. But on the other hand, consumer

companies, they're so devoted to being horizontal. How would you advise a consumer AI founder to leverage this distribution tactic?

Talia (17:25.806)
Um, so I don't know if I, you know, on the one hand, this is a distribution tactic because if you focus somewhere and you tackle a really, really niche market, you can much more easily get to all the people in that market or reach them and know who they are and build for them. Um, but, uh, you know, this is also true for vertical software and we're seeing, you know, this also play out really nicely with AI and companies that are building vertical software products, leveraging AI and on our portfolio companies, like even up.

in the personal injury law space or a bridge in the healthcare space. Like they're tackling these very specific segments. And so if an even ups case, personal injury law, it's not even just saying, let me build for lawyers or the legal industry. It's like, let me build for this very, very specific use case. That is demand letters and demand letters as a service in many ways and building for this very specific type of law. And when you see that, it's like a slam dunk for their needs.

And it's totally incredible and built perfectly with all the small little features and tiny things that only someone in that industry and in that specific sub niche of an industry really wants and cares about. And that's amazing because then you have other companies that come along and are creating these broader, even legal horizontal, but horizontal for legal software. And it just doesn't even compare and doesn't even compete and hold a flame to it. And I think similarly in consumer.

Prateek Joshi (18:37.715)
Yeah.

Talia (18:55.894)
You can see somewhat similar dynamics. I mean, Amazon started as a bookstore, right? It was selling literally just books online. And slowly but surely, you can have the potential to expand and expand from there. And so another example in our portfolio of this in the software ecosystem is Service Titan, right? They started very specifically selling software to plumbers and electricians. And then over time, they've added other field service categories. And...

water treatment and landscaping and different commercial types of services, not just residential. And slowly but surely, it's like a layer cake, but you can start to build out other layers beyond you.

Prateek Joshi (19:37.693)
Yeah, I think that... Yeah.

Talia (19:38.622)
And I love that everything big starts small. I think a lot of great companies look small or like toys early on. And in fact, I think market size is one of the easiest things for investors to get wrong. And many of the very best companies also create their own markets and create totally net new markets. And so you look at something like Airbnb and it, you know, it was a going on Craigslist and looking to stay at a stranger's home. Like that was a tiny market. And now look at what this market has become.

Prateek Joshi (19:44.671)
Yeah.

Prateek Joshi (19:51.357)
Right.

Prateek Joshi (20:05.323)
Hehehe

Talia (20:07.946)
and the ability to really create markets. And I actually personally get really excited about companies that are creating new categories and new markets and have that potential. And where it's something that's small and niche and maybe doesn't exist, isn't even a market really today, but has the potential to be that. And I think startups are really well suited to do that. And it's awesome because you also have very, no incumbents really, right? Whereas in other markets, there's often incumbents, whether they're strong or weak.

Prateek Joshi (20:32.894)
Right.

Talia (20:36.819)
If you're in a new category, there's no one to compete with.

Prateek Joshi (20:40.709)
Yeah, and I'm strongly in the camp of going after a sub niche of a sub niche and completely creating a very magical experience. Having those super fans, I think that seed will help. And obviously, as you said, the best ones create their own markets, they expand their own markets and the world changes and reorganizes around every point. So it's very interesting. All right. The next topic is, again, this is in AI.

the talk of incumbents or incumbent platforms is, it's a very big topic because there are some people who say, hey, all of the value or benefits are basically they have a lot of money, they have a lot of resources and they own distribution. So that's fine, that's factually accurate. So how can startups stand on the shoulder of these giants and find opportunities to exploit that distribution, right? So how should startups think about that?

Talia (21:37.358)
I think if you look throughout history, whenever there are these big new platforms, there's often a window as they're emerging before they've.

Talia (21:48.798)
before they've sucked out all the oxygen from the room, that you can grow with them and exploit different attributes of those applications. And so for example, in the very early days of Google or earlier days of Google, SEO optimization was incredibly effective for many companies like Yelp that basically built their business totally off the backs of Google.

Prateek Joshi (21:51.702)
Hehehe

Talia (22:16.634)
And you could say, well, they're still SEO today. And you're right. But if you tried to build Yelp today, it wouldn't work. Google ranks their own listings higher. They're much smarter and more sophisticated about what that looks like. When you search for a restaurant, they're going to show you the Google reviews, not the Yelp reviews. But they had this window in time to build off the backs of those giants. And I think as you see these newer

platforms and distribution channels emerging, whether that's even with OpenAI and the marketplace that they've launched and otherwise, there's probably going to be some really, really interesting opportunities to get access to customers or to leapfrog ahead that may not be sustainable forever, but give you a real opportunity to accelerate at a much lower cost and by leveraging some of these giants. And then lastly, as it relates to incumbents, there are some interesting...

examples that are emerging. There's a company, for example, called OpenEvidence. That's a new startup in the medical space that recently announced a pretty big partnership with Elsevier, who has access to incredible amounts of data and information as an information and analytics and publishing business, one of the most important in the healthcare ecosystem. And you can imagine the benefits of actually being in a partnership and having a relationship where you get unfair

access to that information more than any other startup and where they are also benefiting by having really, really strong technologists building a cutting edge product and distributing it to doctors, to health systems and the like. And so I also believe there are some really interesting win-win opportunities like open evidence that will continue to emerge.

Prateek Joshi (24:00.817)
Right. I could keep talking about distribution tactics, but also want to touch about the conversion tactics as well. And again, this is a part where, obviously for the listeners, you kind of got a little bit of attention from the potential user, and now you need to convert them. Obviously, end of the day, you're a business, you're going to monetize somehow. Okay. So, first one, you mentioned this phrase,

entertainment as conversion, which is a very effective conversion tactic to capture the attention of consumers by providing useful entertaining content. And there's a good parallel to be drawn here to the strength of RLHF, meaning if you do it well, then users will stick with you for longer, they'll engage with you more, you can monetize better. So coming back to entertainment as conversion, how does this work in consumer AI and who has done this well?

Talia (25:03.978)
I think there's, first off, I'm really excited to see this tactic or just how different companies can use entertainment to engage users and optimize different customer behavior with AI because we are at the very, very early days of a massive change in content and digital content and what that looks like and having new forms of content, hyper-personalized.

media, rich media, image, not just text, but the music that we're listening to. Just today, I saw this company that launched that effectively. It was almost like generating personalized Spotify playlists effectively. And you just think about how dramatically I think the ecosystem is going to change and the types of even celebrity, what celebrity even means is going to dramatically

Prateek Joshi (25:47.308)
Haha

Talia (26:01.858)
different companies take advantage of these newer modalities and models to create experiences that are unlike others and lend themselves to really deep engagement. Most of the internet and a lot of the time spent on the internet right now is actually lean back. It's browsing on TikTok and the like. So it's quite possible we'll see the generated versions of that over time.

Prateek Joshi (26:27.295)
Yeah.

Talia (26:31.858)
infinite scroll for AI.

Prateek Joshi (26:33.901)
Yeah, and the cost of producing this content is also going down. If you look back, if you had to do this 10 years ago, it's just a lot of human effort, but now we can use AI to augment a lot of that. And so basically, if you are the company supplying the content, the cost to produce per unit of content is going down, which is a very good position to be in. The next conversion tactic, we touched upon it a little bit, but

the community driven conversion where you engage, you create a highly engaged community and it can drive high conversion. And so when it comes to consumer AI, many people, at least some of the big ones, they have a one-to-one relationship with, for example, ChatGPD. Like Mid Journey, they did the Discord server. So maybe a little bit of community, but mostly I use the app by myself and I move on. I just use it for a task. So how can...

AI companies like take the community driven angle and create an engaged community versus a one-to-one relationship or a company engaging individual users by themselves.

Talia (27:45.59)
I'm personally really interested to see more companies doing this. And I have a thesis that we're going to see a really interesting social network or of sorts that is built where the way that they crack the chicken and the egg problem is leveraging AI and AI bots. And it used to be that bots were considered bad. But when you can imagine, I think of Reddit as a perfect example where Reddit

You could imagine that a whole Reddit forum could be generated with all different user IDs. It doesn't matter if it's a human on the other end or an AI, but you can basically fill the empty room leveraging generated content and AI models and then use that to start to crack the chicken and the egg and layer in more humans into that experience. So we're mixed.

And I think that's a really interesting dynamic and perhaps it's going to weaken network effects over time if it's much easier. Um, these still want to community network effects to do that. I haven't yet seen many companies taking, um, this approach, but I'm excited to. And if you're doing that, uh, please let me know. But I think this whole notion of bots being bad, um, maybe obsolete in a number of years.

Prateek Joshi (28:48.869)
Yeah.

Prateek Joshi (29:04.049)
Yeah, I have one last tactic, conversion tactic that would be interesting. And obviously for listeners, if you're listening, I highly recommend reading this post along with all of our other posts. They're like super interesting, rich with data. All right, the last conversion tactic is the gambling psychology. And a funny thing, even if you tell people this is how

it works, this is how you're getting tricked, you still go to Twitter and you do the thing because you want to know, oh what if I find something new, maybe there's a new notification, like so there's this social media companies that do it very well. So again, maybe two part question. One, can you explain the gambling psychology in consumer startups in general? And two, how can that play out for an AI company?

Talia (29:58.31)
Game and psychology is in my mind a subset of gamification and humans are incentive driven machines and animals. And for better and frankly also for worse, I think we're very much victims of these very subtle but incentive driven game mechanics that are in products, and by the way, not just consumer products, but even business products, software products, how we make decisions every day.

We're not rational beings, we're irrational in many ways. And I think understanding that in human behavior as you think about products is really powerful for you as a business to help with conversion. Now, hopefully you're aligning incentives where you're using it for good, not for bad. But I do think that when you see these different rewards and different dopamine hits and different...

Prateek Joshi (30:44.339)
All right.

Talia (30:55.674)
the ability to have chance or to spin a wheel and win an award, you see consumer behavior changing, even if rationally, you know that the odds of success are low. You continue to do that behavior over and over again. And so those are things that we see constantly in particular in the gaming ecosystem. But it's actually pervasive. And throughout, I think, most businesses, I mean, if you if you walk through a supermarket like you are, you will see that you are.

uh, walking through a maze that has been, you know, perfectly designed to try and get you to behave in a certain way. When you go through Ikea, like you're following a path that is, you know, very laid out, um, to have you behave in a certain way, um, and follow a certain behavior and, uh, and act in a certain, in a certain mindset. And the businesses that use those lessons from behavioral psychology tend to have massive advantages.

Prateek Joshi (31:53.561)
Yeah. I think it's too tempting to pass up. So speaking of rationality, one thing I wanted to bring up was this tactic I think attributed to Chris Paik about first order irrational, but second order rational. And it basically means that a young company startup can disrupt the market by doing something that is completely irrational with the hope that it will eventually lead to something. So...

For an AI startup, what are a couple of examples of like irrational things they can do to make a dent?

Talia (32:29.378)
Yeah, well, you know, I think we're seeing a lot of behavior that hopefully there's rational. I'm confident that the founders and teams behind them have plans and believe that by doing something that's seemingly irrational and seems too good to be true, actually they're they have a path to building something that is going to create massive value. And I think very

Clearly we're seeing some of this already with pricing, with access to different like inference models. So for example, with Mistral and Mixtral, you know, you've seen the prices, you know, over the past just three days go from, or one week go from, I forget what it was, to zero. You know, almost one startup to the next to the next, you know, undercutting each other and suddenly it's zero. Like you can't get lower. I mean, maybe they'll start paying people to use it. I don't know. But other than that,

Prateek Joshi (33:07.397)
Yeah.

Talia (33:19.926)
But the belief I imagine is that you're building this product that's now going to be integrated into different applications and you're going to be able to deliver a different value to those customers and those consumers over time. And so it's something that on the surface may seem irrational, but may prove to be very rational and a really, really great hook to bring in and show someone the value of your product. It's effectively a trade-off between spending your dollars on marketing and ads.

and finding other ways of doing that. And there's many different approaches. It's always dangerous to give away a dollar for 90 cents, right? So I think you have to be really thoughtful when you're doing that about making sure that that's not actually what you're doing. But we have seen plenty of examples where that has worked out and something seemed almost too good to be true. But.

Prateek Joshi (33:57.865)
Yeah.

Prateek Joshi (34:05.328)
Right, right.

Talia (34:14.946)
the businesses were able to monetize in other ways and through different ancillary services.

Prateek Joshi (34:20.609)
Amazing. I have one last question before we go to the rapid fire round. And this is mostly about your outlook looking forward. So what does 2024 look like for both AI startups as well as investors who are focused on AI?

Talia (34:42.002)
Um, I'm really excited for 2024. Uh, I think that.

I've been in the venture industry for about 10 years and I have not felt this level of

enthusiasm and excitement for what can be and real progress in a way that is meaningfully moving the needle and not just in a way that maybe it feels incremental, whereas I think there's been a lot of interesting activity and hype over the past 10 years, but some of that felt somewhat incremental in terms of where the businesses were going and what was being built.

there's this energy that I've never felt before that's snowballing and it's not, it doesn't feel like it's stopping at all and leading to real progress and real innovation that is just so beyond what I've seen in my career. And so I think that's gonna continue. I'm quite confident that that's gonna continue. And I'm excited for that to continue over 2024.

Prateek Joshi (35:55.977)
Amazing. With that, we're at the rapid fire round. I'll ask a series of questions and would love to hear your answers in 15 seconds or less. You ready? Question number one. What's your favorite book?

Talia (36:04.79)
Hey, let's do it.

Talia (36:09.818)
Um, my favorite book, which is very timely because we talked about it a lot in this conversation and, um, it was recently republished by Stripe is Poor Charlie's Almanac, which is a compilation of speeches and work by Charlie Munger.

Prateek Joshi (36:24.013)
Amazing. Next question. What has been an important but overlooked AI trend in the last 12 months?

Talia (36:35.886)
I think one of the killer use cases for AI has proven to be in loneliness. And I, uh, I think there's a lot of lonely people in the world. And I think it's an under talked about phenomenon that a lot of what's driving a lot of usage across the ecosystem today is, um, really human nature driven and not necessarily productivity driven.

Prateek Joshi (37:00.161)
That's amazing. I think all these episodes, nobody has given like that answer. So I think that's like, that's very interesting. All right, next question. What's the one thing about consumer AI that most people don't get?

Talia (37:17.866)
It's ubiquitous already. Consumers interact with AI daily, personalized recommendations, voice assistants, autocorrect, and don't always recognize this as AI applications. And I think that leads to a lack of understanding about how pervasive AI already is in everyday life. And it's not just these sci-fi scenarios, but it's already deeply embedded in the mundane.

Prateek Joshi (37:44.253)
Yeah. All right, next question. What separates great AI products from the good ones?

Talia (37:50.478)
Shipping fast.

Prateek Joshi (37:52.549)
Love it. I think recently the tweet also like speed is the first boat. I love that tweet. All right, next question. As an investor, what have you changed your mind on recently?

Talia (38:04.15)
Ooh, that's a good one. I change my mind all the time, by the way. I have reserved the right to change my mind. I recently changed my mind on the costs that the, whether AI is going to be, it's going to be cheaper or more expensive to build the next great company in AI. I've actually fluctuated between us can be much cheaper. And we're seeing, you know, a lot of ability to do things with fewer people and automate more.

And then flipping to actually know there's like more competition, lower barriers, century, it's going to be more and more expensive. And so I've been changing my mind quite a bit on this, um, that topic.

Prateek Joshi (38:42.205)
Right, next question. What's your wildest AI prediction for the next 12 months? It's like almost impossible, but not quite impossible.

Talia (38:51.635)
Ugh.

Talia (38:55.534)
I think it's possible that we start to see an AI that can learn skills separate from knowledge in the next year, which would be wildly impactful if we can figure that out and we might over the next year, it'd be amazing.

Prateek Joshi (39:05.501)
Yeah, right.

Prateek Joshi (39:11.338)
Yeah, that actually could be very, very interesting. It's interesting. Yeah, I'm curious too. Now, like we'll see how that plays out. All right, final question. What's my wildest prediction? Oh boy, I think we'll make a ton of progress in biology. I'm a huge fan of what it can do in like digital biology. So.

Talia (39:20.386)
What's yours?

Talia (39:24.662)
I love that question.

Prateek Joshi (39:39.265)
I think all of the things that right now are either very, very compute intensive or, for example, even like designing molecules or sequencing DNA sequencing, whatever we can do, it's going to be like a thousand X faster, better, cheaper. And again, I think that's my prediction of whatever it takes, like a couple of days or hours, it's going to be like seconds by the end of the year. It's my like, it's just very hard to do, but like, I'm really bullish on that. All right. Final question.

What's your number one advice to founders who are starting out today?

Talia (40:15.095)
Move fast.

Prateek Joshi (40:16.461)
I think that's the, yeah. So if you're pitching Talia, anyone listening, just know speed is the killer feature. So just adhere to that. So Talia, this is, thank you so much for sharing your insights here. Loved your view on so many things. You've been doing this a while, so you've seen a couple of cycles. So it's fun to talk all things started off. So thanks again.

Talia (40:43.406)
Well, thank you so much. I couldn't be more excited for the future.

Prateek Joshi (40:47.554)
Perfect.