Transcript
Speaker: Episode number seven.
Speaker: James, we're back.
Speaker: We're back and ever closer to double digits.
Speaker: No pressure for episode 10.
Speaker: I know.
Speaker: Even though we missed a week, we've actually kept momentum for most of the year so far.
Speaker: Not bad.
Speaker: Not bad.
Speaker: You know, I've been thinking our podcast is missing something very crucial.
Speaker: I want to hear your opinion.
Speaker: Well, I think what we need to involve more to our podcast are giant worms.
Speaker: What?
Speaker: Okay.
Speaker: I don't know if you've seen Dune, the new Dune movie.
Speaker: I have not, though.
Speaker: I've heard a lot about it.
Speaker: Well, I have not either.
Speaker: So that's a bombshell right there.
Speaker: But I rewatched the first one.
Speaker: And, you know, that movie has a lot of political intrigue, a lot of things going on.
Speaker: Interesting.
Speaker: a lot of hallucinogenic spices, but the worms are pretty dang cool in there.
Speaker: You know, like I couldn't imagine, you know, right.
Speaker: I'm imagining this riding along on a worm, you know, doing a podcast.
Speaker: I vaguely remember hearing the word spices and worms a lot for some reason.
Speaker: Yeah.
Speaker: The whole concept is the spice must flow.
Speaker: You know, it's like they don't have computers and so they can only do interstellar travel if these like navigators can use this spice.
Speaker: You know, it was interesting.
Speaker: Great vibe in that movie.
Speaker: All right, we've got two on the agenda.
Speaker: One, we are going to talk about Reddit that recently announced that they're filing for an IPO.
Speaker: And the second is some interesting, well, maybe controversial regulation that's happening around AI that India just announced a day or two back.
Speaker: So two topics and we'll dive right in.
Speaker: Let's do it.
Speaker: All right.
Speaker: So the first one, let me tee this up for the Reddit conversation.
Speaker: So most people are familiar with Reddit.
Speaker: It's a social media platform.
Speaker: It's probably one of the only major text-based social media platforms that's around today.
Speaker: It was founded in 2005.
Speaker: So it's been a while.
Speaker: They filed their S1, which is basically...
Speaker: public disclosure prospectus that a company files when they are planning to go public.
Speaker: That's also the document that they use to go to the bankers and start raising money for their IPOs.
Speaker: The nuance about S1 that I learned a while back was this is the first time that a private company is usually declaring a lot of information that has been private for all of their 20-year history now.
Speaker: And this is the first time all of this information becomes public.
Speaker: So it's always a huge treasure trove of information that suddenly becomes available to everybody in the world.
Speaker: In the S1, they've basically said that they are aiming for somewhere close to a $5 billion valuation, which is actually less than half of what their last round of fundraising in 2021 was valued at.
Speaker: It's been a generally slower tech IPO market the last couple of years.
Speaker: The most notable ones in the last year have been Arm and Instacart, which have gone okay.
Speaker: It's a pretty exciting company and it's great to see somebody that's been private for 20 years, been through a bunch of different cycles and finally come out the door.
Speaker: Yeah, I think they also have a kind of interesting history where they, I think they early on in their existence, they sold off to a media company and they had some changes with their founders and stuff like that.
Speaker: So there's some interesting drama or background there for folks interested.
Speaker: They've also had a lot of CEOs.
Speaker: I think in the last, I think they had this founder who left and there were like three or four other
Speaker: CEOs and then the founder came back a few years back.
Speaker: Yeah, exactly.
Speaker: And I think another piece of drama that was making the waves around the internet was the compensation that was announced or that was in this filing.
Speaker: We're not going to go into it, but we had some pretty high numbers, but I think it's, it's, these are all kind of based off of depends on how a company is evaluated.
Speaker: That makes sense.
Speaker: For those of our listeners that are maybe not as familiar with Reddit, because I know that's a relatively niche social network.
Speaker: What's your experience been like, James, with using Reddit?
Speaker: So I'm part of the, what's it, like 90%, 99% of people on the internet that don't contribute content and are just consuming the content.
Speaker: So Reddit is organized into what they term as subreddits that usually have
Speaker: centered around a certain topic something like that they could be they could be around like a sport or they could be around like fantasy book series or a tv show or maybe like a financial advice forum or you know career advice at the wide variety of topics and they talk about how they have all these communities on them my own experience i've kind of seen the over time shift as i've gone through different periods of my life of what
Speaker: Yeah, subreddits I find interesting.
Speaker: The thing that I've found really fascinating about Reddit is there are really obscure subreddits.
Speaker: You know, typically when you join a Facebook group, maybe it's a very broad Facebook group with broad interest.
Speaker: While those also exist on Reddit, you know, r slash music, for example.
Speaker: The most incredible thing for me has been found some very small and narrow subreddits.
Speaker: Like, for example, when we started recording the podcast,
Speaker: I found this subreddit called r slash podcasting, which is basically people that are trying to figure out what tools to use and how to get started and things like that.
Speaker: When I started writing Substack, I actually found an r slash Substack, which is basically everybody that's like trying to get this off of the ground.
Speaker: There's like r slash Taylor Swift, which is pretty big.
Speaker: It's a million users or something now.
Speaker: Wall Street Bets was a huge one that became a sensation a few months back.
Speaker: So that's been one of the really incredible things about Reddit, where it's almost like if you have something very specific that you're interested in, and maybe in real life, it's hard for you to find people that are also exactly interested in the same thing.
Speaker: And Reddit just has these so many forks of smaller worlds within Reddit.
Speaker: Yeah, I think so too.
Speaker: And it was also exciting or interesting, or maybe just a part of the human experience is that all these places, they all have...
Speaker: volunteer moderators, like the people that are in charge of the content or like keeping their own little, I guess they might see it as their own little fee, like in order and moderating people's behavior is all done by other Reddit users, you know, people that maybe founded the subreddit or have experience moderating, which I think is in Reddit's benefit because they're not paying people to do that, but also kind of an interesting like bottoms up dynamic around the communities there.
Speaker: That's pretty fascinating that they've actually managed to get to this scale with just volunteer moderators, essentially.
Speaker: I think they have, based on their S1, they basically have 73 million daily active users and somewhere about like 267 million weekly active users.
Speaker: It's pretty damn impressive that they've actually managed to have a huge set of like subreddits with just volunteer moderators.
Speaker: Yeah, I think so.
Speaker: And those user accounts, what's also interesting, they kind of can speak to the niche aspect where they cover topics people can't find elsewhere.
Speaker: They cite this in their S1 in a couple different spots.
Speaker: They talk about the percent of their users which are not active on other major platforms.
Speaker: Interesting.
Speaker: So obviously none of these overlap, but they're talking about how like 32% are not active on Facebook.
Speaker: 37% are not active on Instagram.
Speaker: These are big places that a lot of people spend their time.
Speaker: And just throw in the last one I'll throw in is 41% are not active on TikTok.
Speaker: These are all places that I think in my day-to-day life, that's like where people go to kind of consume content idly or do something.
Speaker: But you can see there's a significant portion of their user base that is coming here, not exclusively, but not using, getting content here that they could otherwise get there or getting only content here they can only find here.
Speaker: That's true.
Speaker: I guess there's something really fascinating about Reddit where I know that they have had cases where, you know, people have complained that it's not moderated enough or there's been like toxic threads and things like that.
Speaker: But there's something still very like candid and natural about Reddit threads where even though it's pseudo anonymous, like most people don't use their own names, there's still something about Reddit threads and Reddit conversations, but it still feels like there's
Speaker: Real people that are brought together based on like very specific interests, which I think has been really hard to replicate in probably any other platform.
Speaker: Yeah, it is interesting how people that you see engaging there, talking to each other, they are humanized to a bit because of being around that shared interest.
Speaker: But they do call out on the anonymity part.
Speaker: They do call that out in their S1 as well, talking about how it's important to them.
Speaker: They think it provides a healthier, safer space for users to be able to be anonymous.
Speaker: They're not required to have a real name policy, Meta's properties do, which is kind of interesting, kind of more like towards the Twitters of the world being more like that.
Speaker: But obviously, I think Reddit's had less moderation controversies than I think Twitter has in recent years.
Speaker: of the different formats of social media that exists in the world of, you know, text to image to video.
Speaker: I think Twitter has gone pretty crazy over the last, you know, few months, maybe a few quarters.
Speaker: Threads, I know, came up as like a Twitter alternative, but it's kind of its own echo chamber where there's maybe not as many real conversations happening.
Speaker: Reddit, I would argue, is probably the most real and the last standing text-based social media platform that like exists today.
Speaker: Yeah.
Speaker: And I also think the part of that is that community aspect.
Speaker: I think a lot of the experience on Twitter that people might talk about, oh, I'm on NBA Twitter or like I'm on XYZ Twitter where like they're in, they, a lot of their content is like that because that's their interest.
Speaker: But it, so a lot of that is the, the way the content is structured and engagement is done.
Speaker: It's people like.
Speaker: posting stuff out there and seeing if people react.
Speaker: Whereas Reddit has a lot of more, someone makes a main conversation topic or main question or main piece of content and people follow up and talk about that one.
Speaker: And it's, I think I find it's a little bit more structured and a little bit more rigid than these other places, but also people can,
Speaker: stay focused on their topic a little bit more.
Speaker: And so I think there's less like wild minute by minute swings in what's going on, but it allows people to go deeper and maybe that's where the human element comes from.
Speaker: Yeah, that's a good point.
Speaker: So maybe we should talk about, you know, we've gone through the S1 filing.
Speaker: Maybe we should talk about what our outlook looks like around some of the specific things that they talked about in the S1.
Speaker: The first area that kind of stood out to me was they do have a lot of daily and weekly active users.
Speaker: which is great.
Speaker: And the growth rate has also been pretty impressive.
Speaker: You know, I've seen somewhere to the tune of more than 20% growth year over year in both their daily and weekly active users.
Speaker: I wonder how much of this, you know, how much of this growth continues.
Speaker: Like Reddit has always had this aura of being a relatively niche platform, you know, catered to certain types of people who liked the vibe of Reddit.
Speaker: It was like a lot of like engineers, for example.
Speaker: That's definitely something that I was thinking about of, you know, how much
Speaker: of this is actually scalable beyond this particular niche set of users that are attracted to the Reddit aesthetic?
Speaker: Yeah, it was actually inspired or...
Speaker: I felt positive feelings about Reddit because I think I had a simple stereotype of what the user would look like.
Speaker: And the problem with stereotypes is if the, or in this conversation would be that they would be limited to a certain type of user and ideally your growth would be unlimited to be everyone.
Speaker: And one of the things they reference about, they use some com score data to understand the demography a little bit of their user base.
Speaker: And they actually said that the, that the,
Speaker: that users were 50% male, which was surprising for me.
Speaker: I kind of, I feel like there's been several different situations and a lot of communities on Reddit do have a problem with being male dominated.
Speaker: And so I think that speaks to a growth in like user base that are like more equally split across.
Speaker: That's pretty impressive.
Speaker: I did not expect that.
Speaker: Yeah, exactly.
Speaker: You'd expect 60, 70% mail, right?
Speaker: Probably more, yeah.
Speaker: But 50%.
Speaker: So I think that's speaking to their growing and their growing to be like a more wide spectrum of people, which I think is a healthy thing for, you know, not only for everyone having access to helpful information in, you know, online communities, but also as a company, it's good to have more people engaged.
Speaker: And if more people are being welcomed by it, I think that's a good sign.
Speaker: Yeah, that's a good point.
Speaker: I also wonder how much I don't know if you've noticed this on Google search recently when you actually search for specific questions, you know, you might search for like best movie theaters in San Francisco.
Speaker: And there's now a subsection of Google search, which shows you more Q&A type content.
Speaker: which I think predominantly has been Reddit and Quora so far.
Speaker: So I'm curious, like how much of their new traffic is also coming from, you know, a lot of this SEO traction that they're seeing lately.
Speaker: And they do call out this as, you know, one of the risks in their S1 where they've seen good traction with SEO and this kind of discovery from search platforms.
Speaker: But
Speaker: Not sure how much of this actually continues to last in the long term.
Speaker: Yeah, it sounds like they're doing smart SEO.
Speaker: I think it's tough for me to imagine that Google would want to make a product where they're like gathering this content themselves.
Speaker: So if anything, it kind of sounds more symbiotic to me, but that doesn't mean the risk isn't there.
Speaker: Yeah, but I do think that like with somebody like Google, they are incentivized to maybe serve their consumers, at least in the short term, till they figure out something that works out for them long term.
Speaker: So I do think that the SEO bump that they've gotten now has been probably a bit of a change in strategy for Google, which may or may not continue to last long term.
Speaker: True.
Speaker: I saw that in the numbers.
Speaker: For some reason, I couldn't explain it, but the Q4 2023 and their numbers was like a significantly... Interesting.
Speaker: Interesting.
Speaker: Not significant, but materially higher number of users than compared to other quarters.
Speaker: I don't know where that's coming from.
Speaker: Maybe they're just juicing it pre-IPO.
Speaker: Yeah.
Speaker: But it could have recent wins in an SEO landscape.
Speaker: That makes sense.
Speaker: Yep.
Speaker: The second aspect in their S1 that was interesting was they do talk about their majority of revenue coming from advertising.
Speaker: So they had somewhere about $800 million in annual revenue in 2023, up 21% year over year, which is, you know, obviously pretty meaningful for a platform of this size.
Speaker: What's your reactions to it?
Speaker: Were you surprised?
Speaker: Were you, was that expected?
Speaker: I think expected for me, I kind of expect no one's going to pay for Reddit.
Speaker: I mean, they do offer Reddit premium, doesn't have ads.
Speaker: So, but, but no one's going to, this is like a consumers are going to want this for quote unquote free.
Speaker: And so it's going to be ad supportive.
Speaker: They're, they're, they're like 98% of their revenue is ads.
Speaker: No surprise there.
Speaker: I think they talk about a little bit about what's their mix on customer base, 65% like big companies.
Speaker: And then they talk about, I forgot the mix of the type of ads because they think previously had been like more CPM as in impression style ads, like versus performance advertising.
Speaker: And they've shifted to more performance advertising.
Speaker: I actually came away with a little bit of a bullish take on their future because of reading into that.
Speaker: They had a lot of specific things in their S1 and they're talking about their advertising technology, which is something I've worked in.
Speaker: And the takeaway I had from it was like a little felt like they were describing it like it's early days.
Speaker: Like they were started a big reinvestment in 2018.
Speaker: And then they made mention to some, I would say, pretty like
Speaker: some technologies that they've started building that I would say people built in 2012, you know, like these are the catching up on ad front.
Speaker: And I actually see that as meaning to have a bigger runway for revenue growth, which gave, that's where I'm coming way bullish here where it's like, oh, I'm excited for them.
Speaker: Yeah, their revenues, you know, 800 million, but I think there's actually a long runway because their, their advertising technology isn't up to the performance level that you would expect from a company at this point.
Speaker: But that just means there's more upside for them.
Speaker: Yeah, that's a reasonable take.
Speaker: I mean, I think I also the nuance I also read was they were trying to go after advertisers that aligned more tightly with the kind of subreddits they have so that, you know, more of the ads that they show are also like a lot more relevant, given it's like heavily contextual.
Speaker: I think I partially agree with the bullish take.
Speaker: The other thing that also makes me bullish about this is with all of the Apple and Google privacy changes that have happened around advertising, I do think that Reddit has this unique advantage where it's heavily contextual advertising.
Speaker: They're probably one of the few platforms that can do contextual advertising.
Speaker: in a way that's, you know, it works for them versus in a lot of platforms like Instagram or TikTok, it's very hard to actually do contextual advertising.
Speaker: So that's also something which I think works well in their favor.
Speaker: Yeah, 100%.
Speaker: They even talk about specifically in the S1 how they are insulated from these because of their type of targeting.
Speaker: Interesting.
Speaker: It's a little bit like the flip side of what I said previously, where, oh, they're advertising technology so young.
Speaker: Well, the flip side is that it never got sophisticated enough where they're going past doing contextual stuff.
Speaker: But they do have this benefit of people are, you know, reading about bikes, show them ads from bikes.
Speaker: That's something that's kind of similar to showing people search ads.
Speaker: But I do think it will limit them long term because in the end, people aren't, you know, single track minded.
Speaker: You know, sometimes you want that bike ad, but sometimes you were, you know, you know, just just
Speaker: researching showtimes for a movie and then you're reading your bike things, you love bikes, but then they're like, oh, look, this movie theater just opened up.
Speaker: Like, I think there's situations where that type of advertising doesn't have to be contextual.
Speaker: It's great that they can make a business off contextual, but long term, they're going to need to do a mix of it.
Speaker: That makes sense.
Speaker: And to that end, I think part of something I also thought was encouraging is that they're growing user base.
Speaker: They're also growing in login rates for their users.
Speaker: Logged in users are growing in more.
Speaker: And that kind of speaks to the things you alluded to with privacy changes.
Speaker: A lot of that can be insulated or you can, one big tool you have for mitigation there is having logged in users because they're ones that you can match with people with a
Speaker: If they do a transaction or something like that, they can be seed data for that kind of matching.
Speaker: But in general, I do think that they benefit from not being dependent on third-party data signals for developing good ad targeting.
Speaker: Yeah, that makes a lot of sense.
Speaker: I think the part that I'm skeptical about with their advertising opportunity...
Speaker: I don't think I've seen great examples of text-based social media platforms having great ads.
Speaker: Anytime recently, you know, maybe Facebook used to have it at some point of time, the original Facebook app.
Speaker: But the Twitter ads are like terrible.
Speaker: I literally do not remember seeing a good ad on Twitter.
Speaker: Same with, I guess, Suresh doesn't have ads right now.
Speaker: Even with Reddit, I think my experience has been...
Speaker: I don't recall seeing a memorable ad on Reddit for as long as I've used the product.
Speaker: And I'm sure there's some part of it that can be addressed through better targeting, maybe better understanding of the users.
Speaker: And I'm sure there is advertising levers to pull to improve that.
Speaker: But I think the part that maybe I'm a little bit skeptical about is I don't know if it ever gets to a place where, you know, Reddit ads are like as organic or feel as natural as, you know, something like TikTok ads do or something like Instagram ads do.
Speaker: I think I have a split opinion on this one.
Speaker: I think I agree with you on the, I've seen two ad units on Reddit.
Speaker: There's the one who's just going through the feed of posts.
Speaker: And those ones used to are typically, in my experience, have been like image dominated where they have a big image and they have some text.
Speaker: And as a human, my eyes are drawn to the image and then that's where I start.
Speaker: And then I realized, oh, damn, this is an ad and then I move on.
Speaker: And so those ones, I feel like you could drive some engagement or those have an ability to
Speaker: to be quote-unquote good ads because they are a little bit more complex than just text they're kind of more you know instagram ads in the sense that they they got you know a photo there and you can put like some action button to get people to do stuff the other type of ads i've seen in them is once you click through to the one of the posts and you have the comment threads yep it's the top comment it's like not exactly disguised but the top comment is actually an ad yep
Speaker: and it'll be purely text-based.
Speaker: I don't find those very engaging at all, and my eyes usually just skim past them.
Speaker: Interesting, okay.
Speaker: So I do agree on those comment-based ones.
Speaker: Yeah, I don't really see how those are affecting people because humans, I don't want to put in the effort to read through a whole ad, you know?
Speaker: But like an image is pretty engaging, and it's easy for us to understand.
Speaker: So I do feel like the feed-based ads work.
Speaker: work on me.
Speaker: That's interesting.
Speaker: Okay.
Speaker: I actually, I originally had the other side of the take where, you know, I think in a comment based thread,
Speaker: It's probably easier to show contextual ads because, you know, it is one context of a thread.
Speaker: There is a very clear context there where it makes sense to add something contextual versus.
Speaker: I do think that, I don't know if you felt this, like I personally as a user, one of the challenges that I've always felt with the Reddit home feed is there's just already so much context switch between like subsequent cards in the thread.
Speaker: Which I think is unlike other feeds.
Speaker: I mean, I think other feeds also change, but here it's suddenly I'm seeing like R slash healthy food and then I'm seeing like R slash sub stack and I'm seeing like R slash, you know, letter Kenny and some just random stuff like that where...
Speaker: I think there's already a pretty significant amount of context which happening.
Speaker: And I think the ads add to more of the chaos rather than fit nicely into, you know, an organic setting.
Speaker: Yeah, that's a challenging atmosphere.
Speaker: Just as a side note, they also have those ads if you go specifically to a subreddit.
Speaker: So even if you go specifically to a cycling subreddit, you could have that kind of engaging ad unit within the feed.
Speaker: But back on like the jarring topic changing and how do you get an ad that fits in there?
Speaker: I think I would argue back on, well,
Speaker: How do you experience that in TikTok?
Speaker: Because I do feel like when you're swiping through TikToks, you get the same experience where the topic can change wildly between videos and the ad units.
Speaker: That's a good question.
Speaker: Yeah.
Speaker: I don't know the answer to that, but somehow the TikTok ads do feel a lot more organic than the Reddit ads do.
Speaker: It might be as simple as videos are a little easier for humans like videos more.
Speaker: I do think with videos, there's a little bit of a warm up, you know, like most videos, you're like, OK, it takes a couple of seconds.
Speaker: It like seeps in.
Speaker: And then by the time you finish the video, like moving down.
Speaker: So I don't know, maybe there is a little bit of that transition that the video format is enabling that the feed one doesn't.
Speaker: Yeah, no, but I do think I, outside the feed, unless you get in a feed, it's a good context.
Speaker: I do kind of agree.
Speaker: Like besides getting users to scroll forever and just throw ads in there, like I do question, can you get, can you monetize more?
Speaker: I don't know if they can, but they can certainly improve their targeting, but can they monetize more is kind of the question.
Speaker: That's a good point.
Speaker: So let's dive into the third nuance that we want to talk about.
Speaker: So we were talking about this before we started.
Speaker: They talk a lot about their data licensing deals where essentially they've licensed their data to a bunch of different companies.
Speaker: I think Google is the only one that they explicitly talk about having done this with.
Speaker: But I think that there's a bunch of other deals which they haven't listed who they've
Speaker: license their data to but it's essentially something to the tune of 200 million dollars over three years roughly is what they mentioned as part of their uh s1 which is like non-trivial so 200 million over three years that's 60 million 65 million a year on a base of 800 million dollars a year so that's like
Speaker: almost 10% of the revenue is coming from data licensing, which is really interesting.
Speaker: Like I was quite surprised that they were going pretty aggressively after these data licensing deals.
Speaker: Well, I have a bearish opinion on the data licensing in general, but something that I do have at the back of my mind is,
Speaker: feels more early days for this data licensing thing.
Speaker: We heard the news recently about the one where they're licensing to Google.
Speaker: I do feel like it's like a dirty secret for these large language models that they actually do use a lot of Reddit data for their training.
Speaker: So it's good that they're paying to use that content.
Speaker: Where I'm bearish on the selling to LLMs, like that does sound like good money, but I'm thinking about what I would go to Reddit for.
Speaker: And how if the LLMs have that data, that that would essentially replace my need to go to Reddit.
Speaker: So they're kind of cannibalizing themselves, their organic user interactions when they do that.
Speaker: It won't replace like the community aspect.
Speaker: I do feel like there's some subreddits out there.
Speaker: Maybe it's one that's based off of the Bay Area or like some locale that you might want to go back, just see what's
Speaker: people are saying and maybe there's other ones where it's oh the new episode of blog came out i want to hear what other people are hypothesizing about the next one or how they react so there's some amount that like will never go away like the for when people are going for the human the human interactions but to your example on like the q a and the google search results do feel if i'm ever like researching air fryer
Speaker: You know, I feel like you might go to Reddit to hear about people's experiences, but now you might just go through the LLM and guess what?
Speaker: That's like a very commercial intent that you're looking for an air fryer.
Speaker: And so the Reddit's missing out on this kind of commercial activity you might be going for.
Speaker: So I don't know exactly how that breaks down and like how much LLM answerable content is commercial monetizable or like heavily monetized.
Speaker: And if it's high, then, you know, I do feel like they're risking cannibalizing themselves.
Speaker: Yeah, that makes sense.
Speaker: I mean, I was also wondering why they would license their data versus potentially building their own search kind of experience, which I think someone like Quora, for example, has a bot where you could interact with the Quora bot and ask it questions and it gives you like summaries of answers.
Speaker: That seems like a pretty nice play for Reddit to keep users within their ecosystem versus, you know, actually licensing some of this data out for training and risk not having users come back to them.
Speaker: Yeah, I think I'm way on board with Quora's approach.
Speaker: That seems a lot more defensible.
Speaker: Maybe something that's behind the scenes is difficult.
Speaker: For the longest time, I think a lot of people love to make fun of Reddit's search abilities.
Speaker: If you're looking for something, it's easier to search Google and say Reddit than it is to use Reddit search abilities.
Speaker: So they must be missing maybe some technical chops there.
Speaker: Because I do kind of feel there's a lot of consumer experiences that Reddit themselves could build that would be powered by LLMs that
Speaker: would replace anyone's need to go through ChatGPT or Gemini to access the content.
Speaker: Like whenever you're doing some sort of research that involves Reddit and you just go straight to, wouldn't it be great to ask the Reddit chat bot and it could summarize multiple.
Speaker: So I don't know.
Speaker: It seems like quick, easy money before you IPO, which maybe is unfair, but that's all I do.
Speaker: Yeah, no, that's fair.
Speaker: I mean, I also don't think they've talked about the structure of their deals with some of these companies.
Speaker: I wonder if part of the deals is also, you know, providing some kind of like backlinks into Reddit, you know, maybe for example, for a question of, hey, which products are good?
Speaker: I don't know if users just want to take an LLM response for, hey, here's five products you should check out.
Speaker: Maybe they actually do want to see some of the source and like the trusted information.
Speaker: This is the Reddit there that it's coming from.
Speaker: I do think it still opens up the risk of cannibalizing their ads revenue, but...
Speaker: If at least they're getting links and preserving some of the traffic from coming back to them, it's probably like a good middle ground.
Speaker: That's a good point.
Speaker: I hadn't thought about that.
Speaker: We talked a week or two ago about RAG or retrieval augmented generation.
Speaker: And in my experience within Gemini and how it was using Google services, imagining that it farms out some sort of search to Reddit and shows the Reddit branding.
Speaker: It's like Reddit embedded into it.
Speaker: Maybe that's a way that, just like you said, it wouldn't cannibalize, but it keeps people engaged.
Speaker: That makes sense.
Speaker: And then this is my tinfoil hat theory is I wonder if part of the said or unsaid quid pro quo is also the new presence that Reddit is now getting as part of Google SEO.
Speaker: You know, that seems like a nice upside to have if you're giving access to Google for training your data.
Speaker: That would be a clear commercial exchange there.
Speaker: I like that.
Speaker: I like that theory.
Speaker: I like it.
Speaker: All right.
Speaker: That was a good discussion.
Speaker: With that, we'll transition to the next topic for the week.
Speaker: All righty, and I'll tee up our second topic of the day.
Speaker: It was news that came out recently about regulatory action that the Indian government is going to do.
Speaker: They are going to require that companies get permission before they release models or software that uses models.
Speaker: AI models or large language models before they use them and release them to the general public.
Speaker: This is a hot topic we talked about a couple of weeks ago about companies signing agreements about they're going to try their best to limit the impact of large language models.
Speaker: And this feels like a follow on to this.
Speaker: But this is India's reaction to how to regulate these companies.
Speaker: What's your takeaway when you're reading into this?
Speaker: Just a couple of nuances as we start.
Speaker: One is this particular announcement covers both AI and generative AI.
Speaker: So when you say AI, that also includes, you know, predictive models, I'm assuming the way that they framed it.
Speaker: And it includes both the models themselves and the applications, which is pretty overarching.
Speaker: It's literally like anything AI touches from the sound of it is, you know, what is in purview from this announcement.
Speaker: That's one nuance.
Speaker: The second nuance, and I guess this one's more of a caveat, is this is like fairly new.
Speaker: It's like a day or so old.
Speaker: There is no public announcement about this.
Speaker: There's a couple of Twitter threads by Indian officials, but it's mostly gone out as a private letter to a bunch of these companies.
Speaker: So whatever we're talking about is kind of based on what public information is available as of today.
Speaker: I was looking up on Twitter a little bit.
Speaker: I mean, a lot of people are fairly surprised and critical of the extent to which this one spans over.
Speaker: They do also talk about a couple of other specific nuances as part of this announcement, like potentially having a consent pop-up to clarify when something is not factually true information.
Speaker: It's worthwhile to peel a couple of layers and think about why this is happening right now in particular.
Speaker: The other one is, this is an election year, so there's a general election that's happening.
Speaker: India is the largest democracy in the world, basically, by population.
Speaker: I would say that India has generally...
Speaker: had a lot of political polarization in the last few years, especially the ruling party.
Speaker: I don't necessarily want to go too much into the politics of it, but it's essentially, you know, they are pretty hard line in terms of being religious.
Speaker: And that's been something that's somehow gotten a lot of popularity.
Speaker: So there's a little bit of that.
Speaker: And then there has been
Speaker: more curbs on free speech over the last few years where a lot of journalists that are critical of the government have been shut down pretty consistently.
Speaker: The other nuance that's also helpful with the Indian regulatory environment is especially when it comes to tech, a lot of the regulatory action has been pretty brute force.
Speaker: I don't know if you saw this last week.
Speaker: Basically, there were a bunch of apps that Google shut down in Play Store.
Speaker: Did not see that, no.
Speaker: Okay, so it's basically because of some payment terms, you know, where, you know, Google wants more money, the...
Speaker: owners don't want to pay as much money and then Google basically enforced their Play Store rules and said we're shutting you down and then the government basically went fairly ballistic on them and they're like no way in hell you're doing this and then Google backed off so it's a pretty large enough market it's you know more than one and a half billion people in the world it's like one sixth one one sixth of the world population something like that so it's a pretty large country with a regulatory landscape which I would say is skeptical of tech in a lot of ways
Speaker: And some part of it I do appreciate.
Speaker: I mean, I do think that in the US, for example, it's been incredibly hard to regulate large tech companies.
Speaker: Yeah, I think it kind of rhymes some of the things you said.
Speaker: I feel like other countries have gone through this polarization effect and some of it's been polarization, but also one side is anti-tech and one side is pro-tech.
Speaker: I think that happens in the U.S. And then there's also other areas that are going through a period of more direct regulation on tech in the European Union.
Speaker: And all this is centered around like an emerging piece of technology, which I think is
Speaker: is difficult.
Speaker: I think if anything, I would want this technology to be allowed to emerge a little bit more and figure out like what the dynamics are.
Speaker: It's tough when you take it all in context.
Speaker: Yeah, well, yeah.
Speaker: You know, this could be used as a tool.
Speaker: We talked a couple weeks ago about the AI using to fake Joe Biden's voice in a phone call.
Speaker: Absolutely.
Speaker: Like you could, you don't want these kind of.
Speaker: And this has happened a lot in Asia, especially one of the biggest controversies that Facebook got in the middle of was in Myanmar, where there was essentially a genocide that happened because a lot of misinformation was allowed to spread on WhatsApp.
Speaker: Everybody that I know in India uses WhatsApp and WhatsApp forwards are like a very natural part of the.
Speaker: communication system.
Speaker: So it's actually very easy for this to get propagated fairly quickly.
Speaker: So I think that some of the concerns there are justified.
Speaker: I do think that the extent of this regulation and the way that they've done it does seem pretty overarching.
Speaker: But based on some of the communications that they've put out, the risk of election integrity does seem to be a factor, especially at a time where the current political party has been able to consolidate more and more control over the last few years.
Speaker: It seems, feels like half the topic is interesting dynamics in the political landscape in India right now and half of it is AI.
Speaker: As I was thinking about it, something I was curious about was like who, who would be impacted by this and how would they be impacted?
Speaker: We talked about the nuance between
Speaker: models and applications which use models, which was one area.
Speaker: But another area was, I think they were phrasing it as, this would only apply to significant platforms and not to startups, whether it was Indian startups or not, maybe we'll assume it's Indian startups.
Speaker: But I thought that was some immediate points to start with, like how would things be distorted by this regulation?
Speaker: And curious if you had starting thoughts on those.
Speaker: Yeah, that's a really good point.
Speaker: I do think that they clarified that a lot of this regulation applies to quote unquote significant platforms.
Speaker: Like I don't think they've defined what significant platforms are, but they did clarify that this applies to these large platforms and not necessarily to the smaller startups, which I do think is important in India.
Speaker: Historically in India, there have been
Speaker: much fewer homegrown core tech startups.
Speaker: There have definitely been more applications like Ola, for example, is the equivalent of Uber.
Speaker: It's a fairly large company.
Speaker: So there have been a lot of these tech enabled applications.
Speaker: But in terms of just purely software core tech kind of products, the ones that have gotten really huge are probably like Inmobi, which is a supply side platform in the advertising world.
Speaker: there's been a couple of CRM companies that have gotten pretty large.
Speaker: So I do think that there has always generally been this sentiment of, you know, India is kind of the place where you build for the local market.
Speaker: There is a very strong startup culture across the Indian ecosystem.
Speaker: It's like one of the youngest countries in the world in terms of like population balance.
Speaker: So I do think that
Speaker: I do think that the government understands that.
Speaker: I think that they are trying to play that balance of, you know, we'll stand up to some of these large companies while still enabling like the smaller companies to like continue thriving.
Speaker: The part that's, I think, challenging to grapple is, you know, especially when it extends both across models and applications.
Speaker: It's just incredibly hard.
Speaker: There's a pretty large supply chain of, you know, there's like open AI that's making models, which is the base model.
Speaker: If you're a startup, you're probably taking this model and, you know, fine tuning the model for your own purposes, based on your own data, based on your use cases.
Speaker: And then there are folks that are picking up this and like building applications, whether it's like the startups themselves or, you know, there could be like other second, third degree consumers of these applications.
Speaker: So,
Speaker: I just don't think that a very strict licensing regime where all of this is impacted, I think it's going to significantly slow down the level of innovation that is happening in the ecosystem.
Speaker: Yeah, I think I agree with the being good that they're protecting the startup ecosystem.
Speaker: I think other countries, and just trying to figure out where they include the country in the European Union, have also made those kind of distinctions where they're applying the new regulations to these different words every time, but to large companies and not applying these more rigorous expectations on smaller companies.
Speaker: The supply chain, I think, is correct.
Speaker: That's a tough one.
Speaker: Where should this actually be happening?
Speaker: Right.
Speaker: The thing that I would be critical of, though, is like when there have been very bad things that have happened, let's take the Biden robocall as an example, right?
Speaker: The actual deepfake that was created was actually created using an application that was created by a startup.
Speaker: So it was a startup that has an AI-wise creation, whatever software, and this deepfake was created using that application layer, which technically falls in the bucket of, you know,
Speaker: A startup, not a significant platform that, you know, based on this ruling might not be regulated, but it's usually like these ones which are less regulated, have less compliance, where there's more risk for reuse versus, you know, if you were like...
Speaker: If you were like chat GPT or even like a Google board, you generally have a large enough compliance and like legal and like, you know, quality assurance teams to actually make sure that there are more and more controls before you start seeing objectionable content.
Speaker: That's a good point.
Speaker: I think I had kind of jumped straight to, oh, where will the pain be felt?
Speaker: The consumer is consuming content that's generated by LLM that may not be factual.
Speaker: And so that's definitely done by a big company.
Speaker: So they, by default, should be having these compliance departments.
Speaker: But the problem does lie with...
Speaker: It's not just when the consumption happens.
Speaker: Like how many people are using ChatGPD?
Speaker: I mean, I know in the US, I don't know the numbers in the US, but it's like 10, 20 million, something like that.
Speaker: Maybe those are the paid numbers, but maybe the non-paid numbers are higher.
Speaker: I just don't know if we've gotten to a level of...
Speaker: maturity where, you know, these are the predominant search forms are probably used by a minority of people.
Speaker: So in some ways, you know, I think the Indian government is sort of laying the foundation for this is how we're thinking about this particular space as it grows up.
Speaker: But I do think in terms of where a lot of these systems are today, the risk probably lies in less of the like the Google Gemini results, right?
Speaker: Like, sure, it was offensive.
Speaker: Sure, it would not produce images of white people.
Speaker: And it's not great, of course, but it doesn't seem as objectionable as, for example, that startup that actually enabled creating the Biden robocall, which is what should have been regulated rather than, you know,
Speaker: Google Bard, not a pretty good example.
Speaker: It's true that the implicit power and value coming out of these LLMs are based in their name is to be able to generate content, right?
Speaker: That can be entertaining to a human or fit the need expressed in the message you want.
Speaker: And to be able to just generate something for someone to consume is actually something that a company of any size can do.
Speaker: If Microsoft was to build a marketing tool that used generative AI, you could
Speaker: bet that it would have all sorts of compliance stuff built into it so it wouldn't have misinformation.
Speaker: But like you're saying, a startup could, it's fairly simple to take a generative AI tool to generate content and a startup's going to be trying to move fast and satisfy customers.
Speaker: So
Speaker: They won't be regulated because arbitrarily they're small, but they can still leverage that same ability.
Speaker: It's not like forbidden technology that only big companies have access to.
Speaker: It's pretty democratized.
Speaker: Yeah, absolutely.
Speaker: The other nuance is India has also had a complicated history with licensing in the past.
Speaker: This was like before 1990s, like back in, you know, post-independence, pre-1990.
Speaker: There was this whole phase where India was in a huge licensing era where if you had to start a business, if you had to do... There's like a lot of movies around this.
Speaker: Oh, licensing, like having a license to operate?
Speaker: Yeah, basically, right?
Speaker: Like you needed the government to sign off on anything.
Speaker: Just resulted in a lot of corruption.
Speaker: So if you like knew the right people, if you were willing to grease the right hands, you can actually get these licenses.
Speaker: And usually what that means is in the U.S. system, it's called recapture.
Speaker: If you're a large company, you have more lawyers, you have more money, you have more ways to work the system versus if you were a much smaller company.
Speaker: So something like this, I think inherently has, even though it's intended to actually slow down some of these large companies, it's more likely going to slow down everybody else rather than, you know, what the intended outcome of the approach is.
Speaker: That's an interesting background that I don't have.
Speaker: So in 1991, basically, India opened up, like, this whole move towards opening up, like, foreign investment.
Speaker: It was, like, a pretty significant moment where pre-1991, India was, like, very licensing heavy and very, you know, not open to foreign investment.
Speaker: And that changed pretty significantly since then.
Speaker: And so I could also see that how people that are...
Speaker: building companies around these technologies would also might have been around to see that aspect and it would be drawing ire just because they are worried that that's how things are going to go and that this is going to be a return towards you know maybe for some reason in tech you have all this kind of licensing things to go through and yeah and then they would probably be worried about that yeah
Speaker: Yeah, I do think that there's an open question.
Speaker: And I think this is not just in India, but probably across the world for what is the right mechanism to regulate, you know, these kind of risks.
Speaker: Like it is a real risk.
Speaker: It is a real risk that there is misinformation.
Speaker: It is a real risk that there could be compromise to election integrity.
Speaker: This is still something I struggle with.
Speaker: You know, who do you regulate?
Speaker: Do you regulate the model players?
Speaker: Do you regulate the applications?
Speaker: Is it like somewhere in between?
Speaker: I'm curious if you have any takes on it.
Speaker: I think I look at it like, what's the harm you're trying to mitigate?
Speaker: And I know going into this, my initial reaction was, well, it's all at the consumer level.
Speaker: When people are consuming the content, you should be able to know when it's generated by AI and should label it as such or block it if it's enough.
Speaker: And so that's where my first...
Speaker: thought goes but I think I've been making a couple good points here that's slowly convincing me about well on the production end it's also a problem it could be anyone anyone could produce this stuff so I guess I still in this I'm still in my home base is probably still regulate the area where the people consume the content but I I'm developing my opinion
Speaker: Yeah, that's fair.
Speaker: I mean, I think it's like an okay starting point, but I do think that a lot of this, you know, maybe the distribution platforms can build their own like AI detection capabilities.
Speaker: Maybe they should like proactively scan the content and figure out AI or not kind of labels.
Speaker: I do think that there's this huge middle ground, like we've talked about in the previous episode on like sales email generation use cases.
Speaker: If you're writing a sales email and that's like AI generated, that's probably okay.
Speaker: And it doesn't have to be like labeled to be AI generated.
Speaker: But when it comes to, yeah, I think it's just like really hard to implement or enforce at the distribution layer for this reason.
Speaker: Certainly like in other non-content consuming aspects is where
Speaker: you wouldn't be able to do it.
Speaker: One thing I'm thinking about is in the school setting, I think that's probably the biggest threat.
Speaker: Students learn by doing the academic activity they need to do, writing a paper, reading articles and stuff like that.
Speaker: But these tools, today a student could already take advantage of these tools and they won't ever actually learn the concepts behind it, right?
Speaker: I think there's a broader question around that, but how this relates to you is, your point is that wouldn't be on my end of the spectrum.
Speaker: It would be
Speaker: more around the tool itself.
Speaker: Like, is it okay for you to make a tool that does this?
Speaker: Yeah, that's fair.
Speaker: I mean, I do think that there's probably like a more clinical approach to how you could regulate this.
Speaker: I do think that
Speaker: there should be responsibility and liability on each of these layers.
Speaker: Like, for example, if you are open AI and if you're selling models, you're probably responsible to make sure whoever is using your models is, you know, not using it for bad purposes, whether it's, you know, Microsoft that's integrating this as part of the Azure offering or if you're selling like open AI API access to a client directly.
Speaker: I do think that there is, you know, one layer of liability which needs to exist there.
Speaker: And then I do think the second one that might also be worthwhile is the application builders themselves, like the tool that was allowing AI deepfakes for the Biden robocall.
Speaker: I think that there should be some chain of responsibility there.
Speaker: It could be at the, I don't think it should live at the distribution level.
Speaker: That seems one step too far in terms of how we control this.
Speaker: But at least making sure that there could be licensing where somebody, when they're using the model, needs to maybe produce an ID and actually talk about, actually give their credentials so that it's not random actors that are generating some of these fake content and passing it around onto them.
Speaker: Maybe some amount of logging around.
Speaker: I think in certain industries, you probably have, you keep track of someone accesses certain resources, like some sensitive document.
Speaker: And maybe there needs to be something around that of keeping track of what you've generated in the past two years and who got what.
Speaker: And so you kind of have this chain of responsibility of if there's something bad, you should have notified it.
Speaker: But if something bad does happen, you should have record of where it came from.
Speaker: There's a little bit of that.
Speaker: Seems like a good, you know, crypto use case if you actually have the supply chain of where this content was generated from.
Speaker: That's actually a very good point.
Speaker: You can probably come up with some sort of hash of the content and write it to some shared chain.
Speaker: Yeah, it's a good crypto use case.
Speaker: That's all LLMs need, right?
Speaker: More integration with crypto and maybe some NFTs out of it too, because AI could generate the NFTs and then write it to the chain.
Speaker: And then you could be a creator on Reddit and sell your NFTs.
Speaker: Oh man, what a time to be alive.
Speaker: Well, that brings us full circle.
Speaker: That's all we had for today.
Speaker: You can follow us wherever you prefer.
Speaker: We are on Spotify, Apple Podcasts, and YouTube.
Speaker: If you have any feedback, you can email us at thisisunpacked at substack.com.
Speaker: If you have a friend that you think would like our content, please do share with them.
Speaker: And we'll see you again next week.
Speaker: See you.
