Transcript
Speaker: Well, here we go again.
Speaker: This is episode number five.
Speaker: Number five is the one where we bring in much improved audio.
Speaker: I hope our listeners are appreciating that.
Speaker: That is true.
Speaker: If you don't appreciate it, you should go back and listen to the first episode that we got in.
Speaker: Speaking of catching up, you know, because there's all those avid listeners out there that want to catch up on the previous ones.
Speaker: I hear you recently uploaded the previous episodes on a new platform.
Speaker: Can you tell us about that?
Speaker: I did.
Speaker: Well, I finally figured out how to do video editing.
Speaker: And we are now on YouTube.
Speaker: You search for at this is unpacked or the unpacked podcast.
Speaker: Hopefully our SEO is eventually not terrible that you type unpacked podcast and you can actually find us.
Speaker: But if you want to listen to us on YouTube, we are there.
Speaker: And it should be known that this time around, I am already a subscriber at this moment versus our episode two, where I was not yet a subscriber on any platform.
Speaker: But I'm now a proud subscriber.
Speaker: One more platform and then you'll be forgiven.
Speaker: That sounds good.
Speaker: On a moment of levity, maybe, I'm sure all our listeners are very avid and maybe rabid NFL football fans.
Speaker: I know the two of us maybe are not so.
Speaker: That's true.
Speaker: Super Bowl just happened.
Speaker: And as Bay Area residents, I think we're supposed to be sad.
Speaker: That's true.
Speaker: I can tell you're crying a little right now.
Speaker: I get it, man.
Speaker: It's tough.
Speaker: And I did see a lot of crazy videos on TikTok on, you know,
Speaker: a lot of disappointment, but I do think the TikTok is a great way to consume these events in post, you know, in case there's no FOMO for me, like my girlfriend and I were watching some, we were on a plane during the game, so I didn't see it live, but the TikTok content is stellar, especially anything covering the Nickelodeon channel, the stream of the game, which had a different set of announcers and all sorts of post effects.
Speaker: Hilarious.
Speaker: I also think I saw more of Taylor Swift on TikTok than I saw the actual game.
Speaker: Despite the fact that they only had 53 seconds of Taylor content from the game.
Speaker: Well, in all fairness, I'd heard of Taylor Swift before and I'd heard of Travis Kelsey.
Speaker: It's true.
Speaker: I think most people have.
Speaker: They're both good at their jobs, but one of them happens to be a billionaire and world renowned.
Speaker: All right.
Speaker: Well, we'll dive into the topics for this week.
Speaker: We're going to talk about two topics.
Speaker: The first one is Meta has stopped recommending political content in all of their recommendation services, particularly Instagram Reels and Threads, which is their Twitter competitor.
Speaker: We'll dive into that a little bit more.
Speaker: And then the second topic we're going to talk about is, well, what's it called now?
Speaker: Gemini, or I guess it was BARD before.
Speaker: Google had their equivalent of ChatGPT called Google BARD on the web primarily.
Speaker: They have now rebranded it to Gemini.
Speaker: It's now available, well, disappointingly, as a separate app on Android, but not as a separate app on iOS.
Speaker: And it's only also launched to a subset of iOS users, which James is very smug about.
Speaker: I am very smug because I am an avid Android user, have yet to be coaxed into the Apple ecosystem.
Speaker: That's true.
Speaker: I wish I would have held up as long as you did.
Speaker: Well, so with that, let's dive into the first one.
Speaker: So let me tee this up.
Speaker: Meta basically made an announcement last week that they will stop recommending political content across their recommendation services.
Speaker: So this is primarily when I say recommendation services, this is essentially places where Meta is essentially the...
Speaker: the algorithm that is determining what is being amplified to users.
Speaker: If you follow somebody, let's say you follow like a congressman, you follow like a particular account, you will still be able to see most of this, most of the content that they share, both in the news feed on Instagram, like not the stories.
Speaker: Actually, it is visible in feed and stories.
Speaker: It's a little fuzzy whether it's actually still visible in like the Reels interface itself.
Speaker: I don't know if you read about that, James.
Speaker: I think my impression was that it was also a fixed like discovery recommendation.
Speaker: That's what I thought.
Speaker: Okay.
Speaker: Scrolling through reels as well.
Speaker: So yeah, that's fair.
Speaker: And then I guess for those of our listeners who might not be in the US, there is the presidential election coming up in 2024.
Speaker: And then I think there's also a bunch of other local elections.
Speaker: I know San Francisco has like the mayor election, a bunch of like other local representatives are up for the election again.
Speaker: The other nuance that might be useful is they do mention in their press release that users have the ability to go and toggle whether they want to see political content.
Speaker: It is turned off by default.
Speaker: And we know that a lot of these default settings are very powerful, but users have the ability to go turn on something deep inside the settings.
Speaker: I wonder who those wild people are that will go deep in the settings and say, I want more outrage.
Speaker: I want more political content.
Speaker: That's true.
Speaker: People out there.
Speaker: So, James, first reactions?
Speaker: I guess, you know, I think the distinction on that if you're already following that it's not affecting that content, I think that's kind of the opt-out obviously is something if he created, shouldn't judge our users, but a user really, really wants open graph recommendations of, you know, political content.
Speaker: They can still opt in, so that's good they have that.
Speaker: But then also, I think another signal that a user might be given that they want political content is following specific accounts.
Speaker: So I think that's good that that's there.
Speaker: Like, from a user's choice standpoint, I do wonder.
Speaker: Like how, what overall user sentiment will be about this change?
Speaker: Like we know defaults are powerful.
Speaker: I wonder how many people were already following a political account or how many were not.
Speaker: You know, I think the thing that I looked to when I first thought about this is, at least in the US, looking at voter turnout.
Speaker: So of the percent of eligible users, or not users in this case, but as the percent of eligible voters, how many actually ended up voting in the election?
Speaker: And a thing that is an interesting change in the U.S. is during a presidential election year, it's significantly higher than during what we call midterm, which is those elections that happen two years into a presidential campaign.
Speaker: So a president is elected, voter turnout there is like 55%, 60%, which you'd want it to be higher, but it's okay.
Speaker: And then during a midterm election year, which still has important things to be elected, it's down like 40%.
Speaker: I think most recently...
Speaker: I think because of the political atmosphere, we had higher than that engagement.
Speaker: So I said 40%, but I think the most recent one had almost 50% for a midterm year.
Speaker: Interesting.
Speaker: Okay.
Speaker: So I thought of that kind of dynamic as a signal that, you know, how many people out there are actually going to miss the content?
Speaker: You know, like they, how many of them are actually engaged enough that they want this political content in their sphere?
Speaker: You know, there's only...
Speaker: max 60% of people on a presidential year and then even fewer on lesser years.
Speaker: So how many people are going to be missing the content was the first thing I thought about.
Speaker: That is true.
Speaker: I guess the one thing I would probably push back on the 60% is I know in the U.S. specifically, there are just challenges that make it difficult for users to turn up and vote.
Speaker: Like I know, for example, a lot of people have like multiple jobs, like polls open up at specific hours, close at specific hours.
Speaker: I know that there are also elections during like bad weather season, things like that, where there's storms and all that kind of stuff.
Speaker: So I think it's fair, the 60% number where it's an estimate of like how many people actually ended up voting.
Speaker: But it's probably reasonable to say that the number that cares about like the political...
Speaker: process is maybe like higher than that.
Speaker: Certainly, but I would say maybe that's where the contrast between presidential and midterm years is important.
Speaker: That's fair.
Speaker: Because theoretically, whatever effect from, you know, advertising, advert, voting laws that are inhibit people from voting applies to both midterms and presidential.
Speaker: So there's still that, I guess, 20% gap, 15, 20% gap in turnout.
Speaker: So I feel that's kind of like a natural test on how
Speaker: how and what's the delta between an engaged voter and like still engaged enough to vote on a presidential year and a voter that knows plenty of people out there that are just not politically motivated.
Speaker: I think that's fair.
Speaker: The other lens I was thinking about this from was why is Meta doing this, you know, outside of the election coming up that timing part of it itself?
Speaker: What else is actually motivating them to make these changes at this point of time?
Speaker: There's a couple of things which I found interesting.
Speaker: One is in their public statement where they actually talk about these changes.
Speaker: It's a lot more to the tune of, oh, we don't want to be Twitter.
Speaker: Which is fair, because I think, you know, I used to be a pretty frequent Twitter user before all of this chaos started in the last year.
Speaker: They unleashed the recommendation algorithm.
Speaker: Yeah, basically.
Speaker: And I do think it's gotten pretty negative, pretty toxic, or maybe it's always been pretty negative and pretty toxic, but now it's almost, you know, completely unhinged, unmodelated.
Speaker: There's just all kinds of content on the platform.
Speaker: So that's kind of their...
Speaker: That's kind of their party line on why they are trying to limit political content, which is probably fair.
Speaker: But some of the other factors that I also started digging a little bit more into, they've gotten a lot of flack about political content on like a few different fronts.
Speaker: I don't know if you heard about this whole genocide that happened in Myanmar a few years back.
Speaker: They got a lot of flack for it.
Speaker: I mean, WhatsApp was something which was used a lot and even excluding WhatsApp, where let's say that WhatsApp is a communication tool.
Speaker: Even within the Facebook properties, there was this large lawsuit against them on how Facebook did not do enough Facebook.
Speaker: to prevent this genocide and a lot of content actually got promoted because it was getting engagement and all of that stuff.
Speaker: The Jan 6th capital attacks in the US, that was another one that they got a lot of flack for.
Speaker: I was looking at some data points, Facebook groups apparently was used a lot as part of this whole
Speaker: organization, election denial, all of those things.
Speaker: And there was somewhere to the tune of like 650,000 posts that happened in a very short period of time, right after the Jan 6 event that happened here.
Speaker: So all of those, it's, you know, clearly a large halo in the background of Meta's reputation.
Speaker: And that alongside with a lot of the scrutiny that came in a couple of weeks back with their Senate hearing around like,
Speaker: exploiting kids and the kind of harm that it's caused for teens and even younger kids for that matter.
Speaker: Yeah, certainly.
Speaker: I think there's probably some level of every country probably has a story of there's something politically happened or maybe not politically, something bad happened that was related to someone organizing on Facebook, someone amplifying something on Facebook.
Speaker: And I wonder if maybe a more cynical look is maybe international adoption for Instagram is, compare the adoption of Facebook versus Instagram.
Speaker: I think Facebook's probably ahead internationally for adoption.
Speaker: This is mean without looking at the numbers, but in so maybe that's part of their gate.
Speaker: They've learned this on earlier products and they wanna apply it to their own products.
Speaker: And I do think from a philosophical standpoint,
Speaker: They offer a very powerful tool for amplifying something.
Speaker: As an advertiser, it's a very powerful platform.
Speaker: But then as someone that would scream into the void maybe, or they're just trying to share content, share their opinion, or as an influencer, I think they have a high amount of people engaged with their platform and are looking to be entertained by whatever they recommend.
Speaker: And so this amplification effect, I think, does need some stewardship and they can take it seriously themselves.
Speaker: Or I think at some point it'll be taken seriously for them.
Speaker: And I wouldn't be surprised if part of this is them trying to show that they're serious about it and they don't want further regulation on how they should be handling certain types of content.
Speaker: That's fair.
Speaker: I mean, I don't know how much of it is, you know, them trying to do the right thing here versus them trying to wash their hands off of this.
Speaker: They're like, hey, we know that there is going to be sensitivity and political content is involved.
Speaker: So we're basically trying to step away from this, minimize the amount of content that is even happening on this platform so that, you know, the likelihood of this problem coming up is reduced.
Speaker: So I do think that there's some kind of like a product strategy choice there of they're essentially choosing to walk away from this as like a product design decision, you know, rather than it being more like a model decision in some ways.
Speaker: Certainly, maybe a buzzworthy way to describe it is it's kind of like a Disney-ification of their products.
Speaker: That's a good point.
Speaker: Where, you know, it's kind of like everyone knows Disney's brand and there's the kid shows and the Marvels, which...
Speaker: maybe have violence in them, but it's like always, you know, the good versus the bad and, you know, the superheroes win eventually.
Speaker: And so there's a very fuzzy brand around Disney and they have also, they own Hulu, which has other shows that will have a little bit more adult themes or difficult topics.
Speaker: And so I'm not saying that they'll go all the way to have separate brands targeting different things, but it does feel like we see past success where
Speaker: what's called them both media properties where a media property kind of restricts their content to be more, you know, I don't know, palatable and polite society or okay for kids, something like that.
Speaker: And so it's not a crazy idea to say, hey, we restrict our content to not by default give you this stuff.
Speaker: That's a really good point.
Speaker: And I think the other interesting data point that I...
Speaker: read in one of their press releases.
Speaker: I don't know how much of this translates to Reels, but they did mention that in the Facebook app itself, the percent of content that is political is today apparently very less.
Speaker: It's like 3%, something like that.
Speaker: Interesting.
Speaker: Which is fairly surprising.
Speaker: And then the other nuance around this I also learned is...
Speaker: After a lot of these issues happened, after like Jan 6th in the US, a couple of years back, they made a bunch of algorithm changes around political content, where I think previously they were primarily optimizing for engagement.
Speaker: So it's kind of like if you have an enraged thread with like misinformation on it, and you know, like a recommendation algorithm or like a ranking algorithm does what you design it to do.
Speaker: And if you design it to optimize for engagement, it is going to rank that kind of content higher.
Speaker: So one of the changes that they said they did make around that time is starting to evaluate the political type of content differently.
Speaker: There was not that much detail, but they said something to the tune of using user surveys as a measure for success versus just using, you know, engagement as the source of truth for training their algorithms.
Speaker: My guess is like some of those changes has also changed the distribution of how much of their content is political.
Speaker: And then with more and more of like reels and things like that, where there is more algorithmic decisioning happening versus maybe feeds where it was, you know, semi-algorithmic, semi-you following kind of thing.
Speaker: This is probably a good point for them to start walking away from
Speaker: some of these more risky decisions which don't have a lot of business upside for them.
Speaker: Yeah, I think that business upside also sounds pretty important here.
Speaker: I think I want to touch on that in a second, but related to what you just said, I will be curious, I guess I'll never experience it, it's not part of my
Speaker: My normal doom scrolling flow does not involve Facebook properties, but I am curious what they consider political.
Speaker: And I'm sure it's like algorithmically decided or an ML model is deciding this, but because I think at the edges, there's like a difference between a message being political and someone talking about like a policy.
Speaker: I mean, the root of the words is similar, but if someone's talking about
Speaker: housing policy in the United States, describing the situation and talking about facts, will that be labeled as political versus someone that's trying to push a particular agenda, whether it's, you know, no more tall buildings or more buildings?
Speaker: That is clearly more like politically aimed.
Speaker: So I'm curious how that will end up affecting it is, will it catch the nuance between a policy discussion versus political activity?
Speaker: But I think that's a tough thing to drive.
Speaker: And also, I'm curious how many users are actually, you know, deep in the nose of policy.
Speaker: And I think that maybe I'll reference my earlier numbers on like most political topics or policy topics.
Speaker: Yep.
Speaker: Will you get that 40% of eligible voters into that?
Speaker: Maybe not.
Speaker: Maybe they're just being voting how they're told to vote.
Speaker: Yeah, that's a good point.
Speaker: I mean, my guess is if you were designing and ranking algorithm for this, you're probably coming up with some kind of like a threshold prediction of, you know, on a scale of like zero to 10, how political this is.
Speaker: And I would not be surprised if somebody had...
Speaker: you know, Meta is kind of making the decision, okay, this is an election year, like we're okay with, you know, the threshold for what's considered political is like higher and, you know, maybe we have moderated a lot more in like an election year versus a lot less in a non-election year.
Speaker: So that's like the fascinating state of like where media is today, where there is a bunch of these algorithmically determined platforms which are, you know, kind of playing with the scale of, you know, what gets amplified and not.
Speaker: Certainly.
Speaker: And I think that's a, that's a, not exactly a dark secret, but it's a, I think folks that aren't used to it should be aware of is a lot of these things.
Speaker: There's kind of like a target, target that you're aiming for and you can turn the knob, so to speak of, of more or less.
Speaker: Usually the, the,
Speaker: It's not humans validating everything.
Speaker: It's some sort of AI and you can tune it for getting the outcome you want.
Speaker: And so it's likely that this is not like a blanket.
Speaker: All recommendation is gone.
Speaker: They're saying it's all gone, but it's probably like something that is quoted to be at least 50% political is gone, you know, and they twist that knob as you described.
Speaker: Yeah, th at's a good point.
Speaker: The other one, okay, so I think I understand why Meta made this decision, right?
Speaker: I think historically they've shown time and again that they're essentially like a hardcore capitalistic entity.
Speaker: They care about shareholder value a lot more than stakeholder value.
Speaker: They've kind of consistently shown in the past that they're willing to make trade-offs that increase shareholder value, even if that comes at the cost of affecting democracies.
Speaker: So I guess even...
Speaker: Even with that bias, if I'm trying to be fair, okay, I can understand why Meta made that decision.
Speaker: I think what concerns me about this is in terms of political discourse today, especially in the US, there is a lot changing.
Speaker: We talked about this a little bit last time where...
Speaker: Cable news is basically in like structural decline.
Speaker: It has been in structural decline for a long time.
Speaker: Cable itself.
Speaker: News might be the last part of cable that's still alive.
Speaker: Yeah, that's right.
Speaker: There are a lot of media organizations have been through like layoffs even in the last couple of months.
Speaker: I was trying to look at the list of companies that had layoffs.
Speaker: There's NBC, Business Insider, LA Times.
Speaker: All of them had layoffs in the last month.
Speaker: California, sorry, Canada.
Speaker: has a pretty large media corporation called Bell Media, which I think is owned by like Bell Labs, but they own a bunch of media properties.
Speaker: They had a pretty large round of layoffs this week.
Speaker: So like a lot of these structured media organizations are becoming more and more smaller.
Speaker: And one of the trends we talked about last time is like this essentially, whatever you want to call it, like dispersion, decentralization of like how news propagates and how like discourse around civic issues is happening.
Speaker: It's becoming more and more distributed and it is getting distributed through platforms like YouTube, through platforms like Instagram, through platforms like TikTok.
Speaker: And that's just like the reality of how this information moves around today.
Speaker: These are less of, I mean, they are editorial in some ways where, you know, they're kind of holding the keys to the algorithm and what gets amplified or not.
Speaker: But they're primarily like distribution channels.
Speaker: And then when one of, you know, the top three distribution channels starts limiting the kind of content that can go around, I think that that is fairly detrimental to like state-of-democracies.
Speaker: Like you're essentially, again, like I understand that Meta as a private company is allowed to do whatever the hell they want.
Speaker: It is that platform they can control.
Speaker: They can like not show things.
Speaker: They can show things.
Speaker: But I do think this...
Speaker: essentially putting on a default where by default this content is not visible to users.
Speaker: I think that's going to significantly limit the amount of political discourse happening, especially going into an election year.
Speaker: And so what's the alternative then that they would continue to have this but add more fact checking or what's the alternative you see?
Speaker: Yeah, I think there's probably more clinical solutions that you can think about.
Speaker: For example, I use TikTok a bunch.
Speaker: It's been interesting.
Speaker: I think my TikTok feed used to be much less political before.
Speaker: And I think I'm starting to see a lot more political content over the last few months, maybe.
Speaker: And I don't know if that's a consequence of just the algorithm identifying what I like more or what I want more of versus, you know, is this actually like an external thing?
Speaker: There's a lot happening in the world right now.
Speaker: But
Speaker: I do think that on TikTok, for example, there are a lot of legitimate media properties, which are, you know, not like old school CNN, you know, the established media properties.
Speaker: But there are a lot of like smaller digital first media properties that actually create a lot of very valuable content.
Speaker: And that gets distributed through TikTok.
Speaker: And then there is also a bunch of junk content.
Speaker: Like you can actually literally see them in the names of the accounts, for example, that it would be like a
Speaker: a junk string that is, you know, they have a random set of followers, they have a random set of users, a random set of videos that they've uploaded.
Speaker: You can actually go to the account and then you can fairly easily as a human look at it and say, okay, this is not legit.
Speaker: know so if i think like one clinical solution can be like you know can you weed out accounts like that like actually have you know truly quote unquote verified accounts you know so that it's not just a bunch of random people spewing misinformation certainly yeah i think the verifying something like that i have also come across short form video content from traditional media or more trustworthy sources i remember early in the pandemics they found the ones from the washington post really funny and more recently i've seen some from
Speaker: I forgot, NPR maybe.
Speaker: And so there's clear cases where whether the content is political or not, or it's just describing the situation, it's maybe something to evaluate.
Speaker: But that content can be made by traditional, it's just a channel, right?
Speaker: Like you said, the content is being made by news organizations.
Speaker: Absolutely.
Speaker: There's also like podcast clips, for example.
Speaker: I think I've seen a lot of podcast clips of like interesting content on like fairly both ends of the spectrum.
Speaker: Lex Friedman did this interview with Jared Kushner, who is, you know, obviously a controversial person.
Speaker: He's Trump's son-in-law.
Speaker: He was also the Middle East foreign advisor or whatever to like President Trump.
Speaker: And I think it was good to hear that perspective, whether you agree with that or not.
Speaker: And at the same time, he's also had people from fully the other side of the spectrum.
Speaker: Recently, he had somebody that's a Palestinian activist and come and talk about how he thinks about the issue with Israel.
Speaker: I think that there is a lot of value in open discourse if you can figure out ways to...
Speaker: I love legitimate voices to come up and cut out the things that are like junk and misinformation.
Speaker: Yeah, to some extent, maybe we can contrast this with the past technology, how cable news worked, where this open graph recommendation software before Facebook changes, giving you whatever political content, that's different than the status quo on news channels, where typically, at least in the U.S.,
Speaker: Whichever way you were raised or your family leans, you're usually, your team follows a certain news channel, right?
Speaker: So to some extent, those folks are already locked into an ecosystem of a certain point of view.
Speaker: And there's very little open graph recommendation.
Speaker: I'm doing air quotes right now.
Speaker: And that because they just, those news companies have certain goals and they'll structure their content to achieve those goals.
Speaker: And maybe once in a while, they'll have like interviews that are across the aisle or something like that.
Speaker: But with this OpenGraph stuff, the algorithm, I guess, has control of what you see.
Speaker: And so, I mean, maybe you end up with the same result where you only engage in the stuff that reinforces your worldview.
Speaker: And so the algorithm only gives you stuff on your team's content.
Speaker: But that is something that could have existed where they continue to allow political content.
Speaker: But now you can only have the stuff from your team because theoretically the people that do want the content are subscribed to their favorite news outlet or political actor.
Speaker: And I think there's like reasonable ways to actually solve for this problem.
Speaker: For example, you know, the whole echo chamber problem of you're just listening to views that are very close to your own.
Speaker: Like with something like large language models today, right?
Speaker: Like you could basically take a podcast clip.
Speaker: and throw it into chat GPD and say, hey, okay, tell me on a spectrum of left to right, like where does this view fall?
Speaker: And if you are constantly 80% of the time, like maybe you're leaning like left or you're leaning right, there's a way to more thoughtfully, algorithmically say, okay, we will give you the stuff that you like or the stuff that resonates the most with you, but you also kind of randomize this with... I mean, TikTok does this randomization really well.
Speaker: Like TikTok has...
Speaker: I think that's the thing that they do much better than reels and like the reason their recommendation algorithms and like the reason the content that they come up with is a lot better is like they do a really good balance of, you know, explore versus exploit.
Speaker: Certainly.
Speaker: And I had a thought related to what you're describing is, you know, I think at the end of the day, what's difficult here is they are offering an entertainment product.
Speaker: So talking about that effect of showing people content that maybe they disagree with or like an alternative worldview than the one they traditionally subscribe to.
Speaker: Yeah.
Speaker: And I think it's difficult is at the end of the day, if you're trying to maintain active users, you don't want to scare them away.
Speaker: You want the quote unquote good outrage that keeps them engaged and wants to, you know, yell on the internet, right?
Speaker: I guess that's fair.
Speaker: Maybe that's what they want.
Speaker: But if you give them stuff they don't, they're not going to engage with it.
Speaker: They're going to be like, I don't like what I see on Instagram anymore.
Speaker: I started using it less.
Speaker: Now I'm on YouTube.
Speaker: You know, so I wonder if there's some amount of that in the calculation of, I mean, that's kind of what you alluded to when we were teeing this up is
Speaker: maybe it was a better, at the end of the day, win for them, as far as user engagement goes, to remove this stuff because they could give people content that maintained engagement while removing something that could potentially give them negative effects.
Speaker: Yeah, that makes a lot of sense.
Speaker: I think I see maybe like a couple of interesting opportunities from...
Speaker: for other companies, you know, given Instagram's taken this stance with political content.
Speaker: I think TikTok's actually done, I mean, I know I talk about TikTok a lot, but I think that they've done a reasonably good job of, you know, trying to have some level of trust on the platform.
Speaker: I know they're in the midst of a lot of controversy because they're not a U.S. company and- They're CEOs from Singapore though, in case people missed that coverage.
Speaker: But I think like they've done, somebody like TikTok, I think can actually come in and say, okay, like we're not shutting down political content.
Speaker: Like we're willing to kind of take on that risk because it's like meaningful content and it's actually something that people do reasonably care about and then find that, you know, more...
Speaker: clinical approach of how do you put this out more thoughtfully rather than just like letting the most angry opinions go around.
Speaker: That's that I think is like one opportunity for existing platforms.
Speaker: I also think that there's probably some room for a startup or somebody to come up and build like a more thoughtful civic specific platform.
Speaker: You know maybe it's not I don't know if people want to
Speaker: have a completely civic specific platform.
Speaker: But I do think that there's like a lot of opportunity.
Speaker: I think that civic accountability is a huge problem.
Speaker: I don't know if you've seen some of these accounts on Instagram and TikTok.
Speaker: There's a couple of these congressmen, for example,
Speaker: They do a really good job of, they have their accounts, they try to publish, you know, like every few days when there is like an actual, you know, discussion happening in the House or the Senate, like they would actually come and talk about, okay, hey, on the ground, like here's what's happening.
Speaker: And like, and at the end of the day, like the elected representatives, like they're
Speaker: I mean, as much as they are civic leaders, like they're also like civic servants where your job is to actually be accountable to the people that put you in power.
Speaker: And I think that there's this huge vacuum now where people feel a little bit disappointed with their elected officials and there's not really like a good communication mechanism.
Speaker: to fix that.
Speaker: Yeah, I think being able to do, be more digital first as a representative in a democracy certainly will be like an important skill to differentiate yourself.
Speaker: But then also on the topic of other platforms making a different decision, I'm going to extend my Disney-ification analogy a little further.
Speaker: For those in the US, we also have a channel, now they're in the middle of rebranding stuff, but previously people might know HBO that has a lot more content that's like trends to be more adult-
Speaker: content not in the pornographic sense but in the like difficult topics or like engaging difficult things or you know higher more difficult things that are not appropriate for children that's what I'm trying to say which is totally different than Disneyfication or Disney property and then going even further with this analogy that's maybe rougher on the edges there's also a channel in the US called C-SPAN which is kind of almost raw
Speaker: content from the Capitol Hill and stuff like that.
Speaker: And so applying that analogy to what you just said, yeah, there's an opportunity for if Facebook and its properties are going to be Disney, there's an opportunity for someone out there to be more HBO and an opportunity probably more in a non-profit sense for someone to be C-SPAN.
Speaker: But we're missing that aggregation layer of turn on your TV and scroll through your channels.
Speaker: So I don't know exactly how that'll work, but yeah, I see that out there.
Speaker: Yeah, absolutely.
Speaker: So I guess where we land on is, well, we understand why Meta did what they did.
Speaker: It's not surprising that they are optimizing for themselves and their shareholders, but it's clearly a vacuum for civic issues and maybe another platform kind of swoops into this opportunity.
Speaker: Agreed.
Speaker: All right.
Speaker: With that, we will transition into our next topic for
Speaker: All right, so it's up to me to tee this one up.
Speaker: So our topic today, Viggy, will be Gemini slash Bard.
Speaker: And so for those that aren't familiar, Gemini is Google's branding for their chat GPT style interface where you can interact with their large language model, those AI that have been in the news.
Speaker: And actually that's, you know, in classic Google form, there's always a question about names changing and naming, publicly naming being bad, which I think is like low hanging fruit here.
Speaker: But I will comment on the fact that, so this announcement happened recently here in February, but back in December, there was an announcement involving Gemini as well.
Speaker: So now what's the difference?
Speaker: And just a brief point of nuance is the models themselves are called Gemini.
Speaker: Just like in the previously, there was a model called Bard.
Speaker: And previously, I think there was one called Palm.
Speaker: I'm not sure.
Speaker: No one really keeps track of these things.
Speaker: But there's also this chat interface.
Speaker: We can talk with the AI that is also bearing the same name.
Speaker: So Bard's interface was called Bard and Gemini's interface is called Gemini.
Speaker: And we're supposed to, as consumers, understand all this nuance.
Speaker: But the takeaway here is their new interface, like JGPT, is called Gemini.
Speaker: Yeah.
Speaker: And it is powered by their large language model that's called Gemini.
Speaker: And similar to, so you can compare this to the most popular one, ChatGPT.
Speaker: They're powered by, it's made by a company called OpenAI.
Speaker: And they're powered by, their models have different names.
Speaker: And so one might be GPT-3 is one name for OpenAI's model.
Speaker: And their most advanced one right at that time of recording, I think is called GPT-4.
Speaker: And there's an intermediate one, GPT-3.5.
Speaker: And there's...
Speaker: so many names out there but the point in highlighting that is that google also has different levels here and that was part of the december announcement around gemini so gemini announced in december talked about how there's three different levels of gemini there's gemini nano which is made for on-device tasks which i think limited release to pixel 8 pro phones or something like that so this is gemini the modern not gemini the app
Speaker: Correct.
Speaker: So their usage of Gemini, the model, Nano is in the Pixel 8 Pro phones.
Speaker: Gemini Pro is the one that powers that chat interface on the web and app, which I'll talk about in a second.
Speaker: And Gemini Ultra is the subscription required
Speaker: version where it's the same interaction a consumer would have with Gemini Pro, but they are paying money so they get the more advanced one, which is largely follows OpenAI's pricing.
Speaker: Interesting.
Speaker: Is there a large difference between the Gemini Ultra and I guess like the regular Gemini model?
Speaker: So I have yet to be convinced to pay for it.
Speaker: So I don't know.
Speaker: They do talk about how they've done a lot of improvements to the model and ultra is the best of the best of the model.
Speaker: And pro is the one like that's more balanced that makes a trade off between, you know, handling a wide range of what's the word like, like they can handle it differently.
Speaker: of things to do, whereas Ultra is meant for high complexity tasks.
Speaker: Okay, I guess that's not surprising.
Speaker: I know OpenAI has also had multiple versions of the GPT models and they've recently launched something that's cheaper and more powerful compute and all of that stuff.
Speaker: So I do think that there's been an arms race going on with especially these model providers.
Speaker: And I know we've talked a little bit about this before around how a lot of
Speaker: this layer of model builders are maybe getting more and more commoditized, given, you know, maybe they don't have access to as much unique data compared to anybody else.
Speaker: But yeah, that makes sense that there's more of these coming up.
Speaker: Certainly, yeah.
Speaker: And I think one of the key features for Gemini that is an improvement over BARD that they highlight, besides vaguely its improved reasoning capabilities, which I think I have an axe to grind about, is I would say these things don't reason.
Speaker: They're just spitting things out.
Speaker: But I think that's picking at nits.
Speaker: But it also has, they say it's a multimodal by nature, by default, by nature.
Speaker: And what they mean by that multimodal, like the mode with which you're communicating with it.
Speaker: So whether you speak to it, whether you type to it, or even whether you send a photo to it, it can ingest all that kind of data and with it, your interactions with it.
Speaker: That's a really cool one.
Speaker: I mean, I think that we haven't talked a lot about multimodal before, but I do think it's a pretty large change in construct in terms of, you know, being able to process images and videos and text in potentially like a single interface.
Speaker: I know this has been like one of the limitations with the more, you know, traditional audio assistants like the CDs and the Google assistants of the world where...
Speaker: They were almost limited to a single interface.
Speaker: And well, in case of audio, there's not even an interface.
Speaker: It's essentially audio as the conversation format versus actually being able to augment audio with something like a video or a visual interface.
Speaker: I think that that's a really significant advancement.
Speaker: And again...
Speaker: This is not necessarily a Google thing.
Speaker: I know GPT-4 or GPT-4 Turbo, those are also multi-models.
Speaker: So I do think that this is a notable improvement, but maybe TBD, how Google's multi-model is that much different from the other ones in the market.
Speaker: As far as a consumer-relevant application, the multi-modal thing I think is important for an everyday person.
Speaker: I think an episode ago or two episodes ago, we were making fun of...
Speaker: Maybe it wasn't on mic when we were making fun of them, but talking about the Humane AI pen and the Rabbit R1 product.
Speaker: And a lot of those are talking about, oh, you think their demo was tapping on it and saying, what is this?
Speaker: You know, what's a fistful of almonds?
Speaker: And I think that might seem weird, but with this experience, it kind of makes sense where you could be like, send a picture.
Speaker: We just did this during our setup, whereas we were debating, oh, what's this foam cover for a mic?
Speaker: Yeah.
Speaker: And you're like, oh, I'm just going to search something real quick with the cover end of mic.
Speaker: And I was like, I pulled out Gemini and I said, what is this?
Speaker: And it was able to, it was like the black thing and it was able to figure out what it was.
Speaker: And I think there are very cool practical use cases to this.
Speaker: I don't know if you saw the recent Galaxy, Samsung Galaxy ads.
Speaker: Just circle it.
Speaker: Yeah, just circle it.
Speaker: Yeah, that was, I think that was like a cool idea, you know, like they weren't really like trying to sell the model.
Speaker: It was like a very simple consumer idea.
Speaker: You see a product,
Speaker: You don't know where the product is from.
Speaker: You just want the product, you know, circle the thing and go find it.
Speaker: So, and I, another cool example I saw was, have you seen those like complicated parking restrictions?
Speaker: Yeah.
Speaker: San Francisco and LA have these.
Speaker: Yeah.
Speaker: Like I think in New York and stuff, it's even worse where it's just, you know, five boards on top of each other.
Speaker: It's 5 p.m.
Speaker: to 9 p.m.
Speaker: It's like street cleaning and it's also like permit only on some specific days.
Speaker: So one of the really cool use cases was like, just take a screenshot of all of those boards and then ask the multimodal, whatever, LLM, like, can I park now?
Speaker: You know, or like, can I park now for three hours?
Speaker: So I thought those were like pretty significant improvement in like user interfaces.
Speaker: And that's pretty...
Speaker: I agree.
Speaker: And I think the way Gemini shows this is, so maybe I should step back and talk about how you access it.
Speaker: There's the web UI, which people might be familiar with Chachapati and previous version, Bart.
Speaker: And then on iPhone, within the Google app on iPhone, there'll be that, which as you found out, you aren't in the rollout of that yet.
Speaker: So you can't actually access an iPhone.
Speaker: That's right.
Speaker: And then on Android, there's a separate app just called Gemini.
Speaker: And where it's even more interesting, and there's some good synergy here with being the platform owner and a developer of this, is you can replace your on-device Google Assistant with Gemini.
Speaker: And so when I say the magic phrase that activates
Speaker: My on-device assistant starts with OK.
Speaker: It's a different looking UI that pops up.
Speaker: And I have the option to attach what my current screenshot of my current screen, which is also interesting.
Speaker: So it really embeds that multimodal nature.
Speaker: And it's right there.
Speaker: You're interacting with it.
Speaker: So I feel like as far as a consumer experience standpoint, that that's a lot more natural to be like, you know, help me with this thing in the moment versus pause, tap, tap, tap, tap.
Speaker: Oh, look at this chat experience I'm experiencing.
Speaker: Like, oh, I'm so happy that I'm forced to do this in a chat way.
Speaker: You know, you just want to answer.
Speaker: You just want to do your thing, right?
Speaker: And so there's a little bit more of just do your thing kind of thing going on with that experience as your assistant on chat.
Speaker: Yeah, I think that's really cool.
Speaker: And in some ways, I do think that they're catching up to ChatGPT.
Speaker: ChatGPT has had a mobile app for a few months now.
Speaker: And I actually, I found the app to be really useful.
Speaker: Or I think just purely from a user interface standpoint, it's got like this, I don't know if you've seen the interface.
Speaker: It's got like this ridiculously simple interface.
Speaker: It's like... I haven't seen it on mobile, but I've seen it on web.
Speaker: It's incredible.
Speaker: It's literally like, it's like Craigslist in an app, you know?
Speaker: It's got like one text box.
Speaker: It's got like a single linear page where, you know, you have a conversation and there's literally nothing else in the page.
Speaker: There's one input box at the bottom and then a text stretch at the top where you like see your answers.
Speaker: So I think that's been like an...
Speaker: incredible simplification of an interface in some ways where, you know, it can do a lot of things, but the cool thing is like it does it in this fairly intuitive UX kind of way, which I think that Google was probably overloading a lot of their interfaces, even on web when they had the part product, for example, it still felt a little bit more like
Speaker: a loaded product, whereas with ChatGPT, it was like all the power with the Craigslist interface.
Speaker: So that critique still applies.
Speaker: The Gemini UI feels a little bit more polished, but it still has that little more polish in that they added even more polish on it.
Speaker: So it feels more like a final consumer product than even BARD felt like.
Speaker: But I think the contrast with your Craigslist style UI, that sounds very utilitarian.
Speaker: I've seen it on the web, but I haven't seen it on mobile.
Speaker: Gemini on both on mobile and web, I find that it feels like it's got that soft touch from a designer to make it appealing to me as a user.
Speaker: But the downside is it's really hard for me to screenshot stuff.
Speaker: The information density is low.
Speaker: So on web, on my big screen at home, that's fine.
Speaker: It's not really affected by it.
Speaker: But when I'm on my phone, it's like my question is off the screen and there's the LLM's response.
Speaker: And so I kind of feel a little bit of friction there where
Speaker: You know, I understand it's prettier, but like, even when I take a screenshot, I can't show and I want to share someone a cool thing.
Speaker: I can't show them like my question and the response.
Speaker: It's like just the response.
Speaker: And I don't know, it's a little bit of a gripe with what the consumer's experiences here.
Speaker: I think one of the, one of the things I was thinking about was
Speaker: What is it that, you know, ChatGPT or like Bard or I guess Gemini now, the app can like distinctly do, you know, what is it that they can uniquely do?
Speaker: And one of the things, I know we talked about this a little bit off mic before is Google is by default connected to a lot of Google applications.
Speaker: So the examples that they had in their press release was around, you know, being able to pull from within your Gmail application.
Speaker: If you were using like Google Workspace or being able to pull from your calendar or even being able to pull for something like, you know, Google Flights, for example, all of which in the case of ChatGPT are often enabled through like integrations, which are a lot more clunkier.
Speaker: There's a lot more dependency on like external developers to...
Speaker: support this and build experiences for that.
Speaker: So I do think that Google has a distinct advantage where they have a suite of products, kind of similar to the Microsoft Copilot, where, you know, if you are already using the Office suite of productivity tools, it kind of gives you access to more native things that you can do within these series of applications.
Speaker: So yeah, this is, we were talking about the, what's the consumer experience here?
Speaker: Earlier we were talking about multimodal being a difference and that's cool.
Speaker: And I think the other part is what you just described and the technical term for it is retrieval augmented generation.
Speaker: And so that's what the technical term is for it.
Speaker: And that it's essentially, you can teach a large language model to know that, teach it how to go somewhere for a specific type of information.
Speaker: And then it'll know when a user's query
Speaker: what they want something from a specific source, then it'll go there itself.
Speaker: And so like you mentioned, the support, it comes in and the user interface, it's kind of obvious when you do a query and it thinks that you need to reference another service.
Speaker: If you ask about a local business or a video or I forgot my other example, it shows a little logo of that service, like the little YouTube one, a loading sign or like the maps and a loading sign.
Speaker: And so you kind of see that it's querying that and it talks about how it queried it from there.
Speaker: But like you said, it can only access Google services and they can also access select websites, which I found out.
Speaker: I was on a plane the other day and I had plain Wi-Fi, so enough to browse the web sometimes, but sometimes it would be slow.
Speaker: So interesting use case I found is I would ask Gemini to read the website for me.
Speaker: Yeah.
Speaker: Well, specifically wasn't that wasn't downloading.
Speaker: So I was trying to read the technical report about Gemini.
Speaker: So basically it's performance against other, other large language models.
Speaker: Yep.
Speaker: And the PDF wasn't downloading.
Speaker: Cause I think, you know, the airline didn't want me to download a PDF, even though it was like megabytes.
Speaker: It was the older wifi where you can't stream.
Speaker: Anyway, I asked Gemini to summon, I summarize the response or the, the compare the performance within and was able to talk about stuff.
Speaker: Of course.
Speaker: I wasn't able to go verify that like it's referencing real numbers, but it was cool that Jim and I could like download this PDF, read stuff about it, or potentially it already had something in its quote unquote brain about the PDF and it could summarize some stuff out of it for me.
Speaker: And this connection with this
Speaker: rag ability is very interesting for applications like that you might have a more complex task to do.
Speaker: I mean, that simple one that I was messaging you about was I there was a YouTube video.
Speaker: I didn't really want to watch the whole thing.
Speaker: And so I was like, summarize.
Speaker: It was like a Q&A with the creator.
Speaker: And that's a really cool use.
Speaker: Yes.
Speaker: Yeah.
Speaker: Yeah.
Speaker: I was like, you know, I care about this creator's like
Speaker: content, but I don't really want to listen to 20 minutes Q&A.
Speaker: So I just asked it to summarize.
Speaker: Actually, it occurs to me that I didn't actually send you this example.
Speaker: I think the meta version I did as I sent the YouTube upload of our most recent episode of the podcast.
Speaker: And then I asked it to summarize and give it an outline format.
Speaker: And it wasn't perfect, but it did reference how we structured our talk.
Speaker: And so that was really cool to me, exactly how it's doing that.
Speaker: I think maybe needs some more research, especially because it's like video.
Speaker: Is it using the transcripts?
Speaker: Is it actually processing the video?
Speaker: We don't know.
Speaker: But it was able to access YouTube.
Speaker: And what's interesting is that it can't access Spotify, for example.
Speaker: I had a similar interaction where I was asking about a Taylor Swift song and I asked it, can I have a Spotify link?
Speaker: And it was like, blah, blah, blah, can't do that.
Speaker: And then I asked for one for YouTube and it gave me one, like straight to the music video.
Speaker: So there's great interplay, as you talked about, with being able to link out to their services.
Speaker: I guess the question I had on my mind is I definitely see the consumer value from this, you know, and I know Google has shown some demos of like example applications that AI could do.
Speaker: I think they showed this demo of like
Speaker: The assistant basically trying to talk to the customer service rep or, you know, something that you want to get your subscription canceled and you don't want to stay in line for a long time.
Speaker: So I definitely see those kind of use cases, which, you know, maybe the assistant can do down the line.
Speaker: And this becomes really good.
Speaker: The more Google can have a direct interface with a consumer where they can actually do this very easily.
Speaker: So I definitely see the consumer value and like the long-term application of this.
Speaker: The part that I struggled with is, is this like a direct monetization?
Speaker: You know, like I know that they have some of those paid tiers of the Gemini models for consumers.
Speaker: I'm curious, what's your take on, is this actually something that consumers would pay for?
Speaker: It's interesting because I myself am not paying full time.
Speaker: I know of people that pay for ChatGPT's version,
Speaker: I don't know.
Speaker: I think the thing, maybe it's just as early in the maturity for these kinds of products.
Speaker: I feel like we're all forced into this chat interface, you know, and that's the default way that we should be benefiting from these technologies.
Speaker: And so, giving you had that example with doing a phone call for you, that's cool.
Speaker: But there's some interactions that I am asking an AI a model, but sometimes, or sorry, asking an AI a question and we want to go back and forth in chat thing.
Speaker: Sometimes there's something complex I want to do, you know, like, uh, in, but I ended up having to spend, you know, 20, 30 minutes making a spreadsheet, you know, grabbing stuff and blah, blah, blah.
Speaker: And I feel like it would be great if it could just do that for me, but like the interaction would be maybe more like Microsoft co-pilots demos.
Speaker: So it's an application rather than a chat interface.
Speaker: Yeah, but, you know, we talked about it on a previous podcast.
Speaker: I do think at the OS level is probably the right place, but I want it to be able to do lots of stuff for me, you know?
Speaker: And right now in the chat interface, it can't really, like today, you know, I think there's two sides to it.
Speaker: There's, you can think of an app, they designed it for a certain use case and it's going to be able to work for your core use case.
Speaker: You just got to go there and do it.
Speaker: But if I know what I want to do, I really have to follow the tap, tap, tap, going through the flow here.
Speaker: I'm like looking for a restaurant that's open in this area, blah, blah, blah.
Speaker: Like, why do I have to tap through all that stuff?
Speaker: Wouldn't it be great if I could say, you know, give me this and it just opens the app with the results there, right?
Speaker: Like part of the, what's part of the upload, sorry, part of the multimodal or like the, the rag integration with, with Gemini is it'll give you like map results and it'll be like in a Google map.
Speaker: That's really cool.
Speaker: You can see it visualized.
Speaker: That's cool.
Speaker: Yep.
Speaker: But at the same time, I'm still in this darn chat in our face.
Speaker: I think I can click through and end up on maps, but still it's, it feels a little unnatural.
Speaker: There's only so much that a chat interface can do for you, you know, even if it's like a fully loaded up chat interface, even if you start supporting like complex media and like images and videos and all of that.
Speaker: Yeah, I think you're right.
Speaker: Why do they have to get there from a chat, you know, like maybe someday it'll jump straight there.
Speaker: I don't know.
Speaker: I think your original question was who would pay for this.
Speaker: And I think my reticence for paying for it is I don't feel like it saves me time yet.
Speaker: The whole point of it being a better assistant or this chat interface being more powerful is that it should save me time.
Speaker: And I think to address that, it has to have these usability improvements where...
Speaker: it's really legit saving me time i've had previous experiences other ones and working with like very specific technical questions of like how do i use a certain amazon web services service like how do i use step amazon step functions and it was able to like another the chat gpt was able to walk me through it so i can see it really applying there but like in my everyday normal person life it's it's not really i think that's fair
Speaker: I actually ended up a couple of times paying for it at GPT.
Speaker: I guess maybe the first time was just more out of curiosity to see.
Speaker: So right now, the GPT-4 model is still behind the paywall.
Speaker: So you can access like 3.5 for free, but GPT-4 models are paywalled.
Speaker: The multi-modal functionality is paywalled.
Speaker: So if you want to use like images and things like that, that's not available in the 3.5 model.
Speaker: The other thing I think that's paywalled is, I mean, when we were doing some of our podcast stuff, like I was trying to, you know, stick in our transcript and say, hey, can you give me like takeaways from this conversation, right?
Speaker: Just to see what is the level of coherence with which it can give a summary.
Speaker: And again, that's a bit paywalled because of like the length of the text involved.
Speaker: So all of those are like, I think they were like interesting friction points which made me consider upgrading.
Speaker: But
Speaker: Yeah, I think you're right.
Speaker: Like, I think that there is only so much that a single central, you know, chat based interface can do for you.
Speaker: And I kind of landed in a place where, okay, I use this occasionally, like sometimes I want, you know, answers to random questions, and I don't want to read four pages, you know, four different like links.
Speaker: I actually have been using it in that sense.
Speaker: Like that has, even the unpaid version is good enough to be able to summarize some stuff for me.
Speaker: But I do think what you said is exactly true.
Speaker: And the other half of why I'm not paying for it.
Speaker: So there's that friction or like doing more powerful stuff that I might want to pay for it.
Speaker: The other half that was holding me back from paying for it.
Speaker: is general trust about it.
Speaker: I had a really fun interaction when I was first messing with Gemini and I have a friend that's really into Taylor Swift.
Speaker: And so I asked, I was trying to go through the flow of, okay, let's just find out what her most recent song was and then give me a link for it, whatever.
Speaker: And I was like, wow, this is like natural and it's working well.
Speaker: And I was sending screenshots to my friend and my friend was correcting me.
Speaker: No, that's wrong.
Speaker: No, that's
Speaker: That's wrong.
Speaker: That's wrong.
Speaker: And so like at the end of the day, it was like, okay, I, once I corrected it and said, actually, I think it's X, Y, Z. And we're like, why did you have this date wrong?
Speaker: And it would be talking about it.
Speaker: Eventually I coached it to the right answer.
Speaker: So not exactly.
Speaker: But I guess that one's probably going to get better with time, you know, where maybe the accuracy of the retrieval and all of that.
Speaker: Certainly.
Speaker: But I think you're right.
Speaker: Yeah, I agree with you that.
Speaker: Yeah.
Speaker: So at this moment in time, why I'm not paying for it is one half is, is it powerful enough to do what could save me time or benefit me?
Speaker: The other half is, does it actually reliable give me answers?
Speaker: If it's doing some sort of summarization or like pulling in data from different sources and dumping it into a spreadsheet for me or pulling out some key data from a video or PDF,
Speaker: that does sound like it's saving me a lot of time.
Speaker: But if at the same time I can't trust its output, how much time savings is going on.
Speaker: So I think those are the two halves that hold me back from paying for it.
Speaker: Yeah, that's a good point.
Speaker: I mean, I guess if I was thinking from like Google's perspective, there's probably a couple of angles.
Speaker: Like one is just like a defensive angle, you know, where if ChatGPT has an app and you don't have an app,
Speaker: you're kind of leaving like a beachhead open for attack and you don't want to do that.
Speaker: Like I know a lot of the open source work that a lot of big tech has done has been less of a, you know, we want to be like the good guys and contribute to the world kind of thing and more of, okay, this is how you're competing.
Speaker: Let me just throw this out in the open so that you can't compete on this anymore.
Speaker: Like even with the llama models, for example.
Speaker: I was going to say, why is the model llama model free?
Speaker: You know, like,
Speaker: They are incentivized, but they want more models.
Speaker: Exactly.
Speaker: So I do think that there is like a competitive play where, you know, maybe they don't want chat GPT to become like the primary search interface.
Speaker: Like they've tried to defend that entry point with consumers for the longest time.
Speaker: And I do think it makes sense for them to just offer this, even if it's like a free functionality, as long as, you know, they can somehow make the economics of that work.
Speaker: That's probably one.
Speaker: The second, this is a random thought.
Speaker: I think this is a bit of my hypothesis and I'm yet to validate if this is broadly true.
Speaker: I think that people are, I think people have their guards down a lot more when they're talking into an open-ended chat interface versus Google search.
Speaker: I don't know if you felt that way.
Speaker: For example, when I generally search something on Google search, I will try to go into like incognito.
Speaker: I'll try to make sure like Google doesn't get my data.
Speaker: But for example, when I'm asking like random stuff to chat GPT, because it's like a clean single box, there's not really much around it.
Speaker: It somehow psychologically feels...
Speaker: Like you can let your guard down, but I don't think it's actually true because I think now you're giving up like a lot more information.
Speaker: Maybe you're giving a lot more nuance of information about yourself.
Speaker: But I do think that if that is true, that's a nice play for Google where, you know, this kind of opens up a flood of new data for them about the consumers.
Speaker: I think I agree with that.
Speaker: It feels more natural kind of experience.
Speaker: On one hand, your first point on they don't, they want to be defensive and not let ChatGPT take user from you.
Speaker: Yep.
Speaker: Or from them.
Speaker: Totally agree.
Speaker: I'm not paying for the product, but they definitely should be building something like this right now.
Speaker: Right.
Speaker: That's clearly in demand.
Speaker: But then on your point on letting your guard down, I definitely agree.
Speaker: I have a friend and it's really funny, his interaction with assistants, I think it's half ironic, half not ironic, is like using smart home devices.
Speaker: interesting he would like it would be like obnoxious oh hey you just like ask it to do something like you know turn off these lights even though they're not maybe not smart or something like that those interactions are much more natural than you know like you would do like in a google search or something like that not that you google search turn on your phone yeah but like this interaction feels a lot more natural you just ask it some absurd thing and then sometimes it actually does it you know oh what's that song blah blah blah like some more vague reference and it can figure it out and
Speaker: So I do think there's a little bit of inhibitions dropped when you're speaking or typing towards it.
Speaker: And so there's a bit of an opportunity there to reactivate that data engine, so to speak.
Speaker: Maybe as a last point, I do feel like we've talked a lot about the paid tier, which would you pay or not?
Speaker: I think it would be important to maybe cover what exactly comes into paying.
Speaker: Yep.
Speaker: strategically very similar to chat gpt's paid tier chat gpt plus which costs i think 20 a month um this new the ultra version the most powerful version of their thing which is again similar to chat gpt is available if you're a paying customer the paying customer pays uh $19.99 a month so same price 20 bucks yeah
Speaker: And, but it's, it's a, so it's price to be cut out of that.
Speaker: But where it's interesting is it's actually a tier in their Google One subscription.
Speaker: So Google One, for those of you out there that aren't part of the glory, glorified Android gang, is similar to what you have on iPhone.
Speaker: You might have an iCloud account, you might be paying for photo storage or something like that.
Speaker: There's probably different tiers and stuff like that.
Speaker: Google's version of that is called Google One.
Speaker: And it also bundles in, the storage also includes your drives, your photos, and stuff like that.
Speaker: And so there's a new tier for Google One called Gemini AI plan, something like that.
Speaker: And so it includes accesses higher
Speaker: tier of Gemini.
Speaker: There's a more powerful Gemini.
Speaker: And it will include Gemini being able to do the workspace stuff we talked about earlier, like being integrated.
Speaker: But I'm guessing it's $2 for Google storage versus, you know, $20 for the blown up version.
Speaker: So what's interesting is the cheapest tier, you get free 15 gigs on Google One.
Speaker: Okay.
Speaker: The cheapest tier to pay for, which I'm running against to finally, is $299 a month.
Speaker: These are the annual fee, annual prices, slightly cheaper, but anyone can talk about monthly.
Speaker: $299 a month.
Speaker: And I think that's 200 gigabytes of storage.
Speaker: What's interesting here is the next level up, 1999, includes two terabyte of storage and XYZ extra workspace light features in their thing.
Speaker: But now they're adding Gemini Ultra on top of that without changing anything else about the plant.
Speaker: So in a pricing, a consumer interest thing,
Speaker: You can pay 20 bucks for ChatGPT or you can pay $19.99, get two terabytes of storage, all sorts of workplace features that are not available to a general Gmail subscriber.
Speaker: And you get Gemini within all your workspace stuff.
Speaker: So it's a very interesting bundling play where, yeah, you're paying $20 a month, but you're also getting two terabytes of storage.
Speaker: Previously, the two terabytes already cost 20 bucks a month.
Speaker: So I think it's a very interesting bundling strategy.
Speaker: Yeah, that makes a lot of sense.
Speaker: I do think we've maybe talked about this in our previous episodes.
Speaker: I think that there is an opportunity for assistance, particularly in kind of like the SaaS and SaaS adjacent worlds where, you know, you're talking about Google Sheets having an assistant there.
Speaker: To me, that feels closer to, you know, Microsoft's co-pilot offering rather than like a Google Assistant, which is, you know, maybe a lot more
Speaker: consumer facing a lot more like a Siri kind of thing.
Speaker: I do think that in the SaaS world, there's probably a more propensity to maybe not necessarily pay for the AI assistant, but maybe you upgrade to a better plan.
Speaker: Maybe you retain better because you have this assistant.
Speaker: I know we were talking a little bit about Descript, which is the audio editing software we use now.
Speaker: One of the cool things that Descript does is it's got a bunch of this
Speaker: It's got a bunch of AI capabilities sprinkled as part of their user interface.
Speaker: So it's not like a chat thing where you go in and say, hey, do this thing for me.
Speaker: But it's got a bunch of things that are sprinkled across like the workflow of editing.
Speaker: So the first time you upload an audio, it like automatically transcribes it for you.
Speaker: And then once you have a transcript, you can tell it like, hey, pick like five blurbs from this that you think are like most interesting so that you can create clips out of it, things like that.
Speaker: So it maybe goes back to your...
Speaker: point around like non-chat based interfaces that are embedded into the actual workflow where you're like actually doing things and maybe there's a propensity to either pay additionally for that product or just you know be like better retained to that kind of like a SaaS product.
Speaker: I do think among those lines, the consumer benefit from all these LLMs and AI, this whole wave, I think is going to be felt at the edges where like all our products are going to have all this extra things that are more natural to use because they've applied this technology to it.
Speaker: But as far as a chat interface, they're still figuring it out.
Speaker: Is it going to be a chat interface?
Speaker: Are your systems just going to be better?
Speaker: How is that going to involve?
Speaker: I think those are two different things.
Speaker: As a feature within existing software, totally agree.
Speaker: There's all sorts of things that are better today already because companies have been applying this technology.
Speaker: But as far as that chat interface that you're paying for that does stuff, how does that not just get owned by Apple and Google?
Speaker: And how does that not just run on your device instead of on costly servers?
Speaker: I don't know.
Speaker: I kind of see it trending that way, but it'd be interesting to see how this chat interface evolves.
Speaker: Yeah, that's a good point.
Speaker: I think I know you've brought up the idea of running this on device, which we haven't specifically talked about a lot, but I think we've talked about it in the context of device-based assistance.
Speaker: So I do feel like this sets Google up in that path of having something that is like machine learning and LLM driven that runs more on device and, you know, maybe starts using some of the Gemini Nano, you said?
Speaker: Yeah, that's already being used on the Pixel 8 Pro, and I think their next year version will be using that.
Speaker: And I think on that point, the chat interface may be defensive against OpenAI, but more on the attack angle, the offense angle would be, you know,
Speaker: what can they do to drive user adoption of their services like some there's some extent there's like a synergistic aspect between google and apple but they also have this competing aspect of their two android versus ios and it's gonna be really hard to convince an average u.s consumer to go from their iphone to something else but if the technology is that good maybe they would and so it's going to be interesting to see can the technology be get that good and will apple be far enough behind that that people would actually be convinced like i
Speaker: you know, I don't really need iMessage and FaceTime.
Speaker: Look at all this time saved if I just had the new, you know, Google Pixel that can do amazing things.
Speaker: I'm guessing this is the unsaid AI stuff that Google is working on to support more like device-based things, you know, so that they don't get outcompeted with potentially things like this from Google.
Speaker: Yeah, cynical thing.
Speaker: Maybe that's why the iOS roll out so bad, you know, they're prioritizing Android, Android, Android.
Speaker: That's probably a little too cynical for the show.
Speaker: But I do think that one of the cool implications for Google is I know that there's a lot changing in like the SEO world, for example, if a lot more queries are being answered by like chat interfaces and not necessarily going through like the traditional search interface.
Speaker: That does change a lot in terms of like...
Speaker: how Google makes money or like search engines make money.
Speaker: So if they are able to monetize even a subset of users through a paid product like this, either on the workspace side of things or on the consumer side of things, probably buy them that buffer for the scenario where...
Speaker: there is an impact on their search business.
Speaker: Yeah, at the same time, you know, they do control the interface.
Speaker: So if they find that there's a lot of people like me that aren't willing to pay for it yet, well, everything for consumers on the internet is free because of advertising.
Speaker: And so they own the interface.
Speaker: They could add advertising support.
Speaker: I would be really surprised if
Speaker: it works as effectively as the links model, you know, like the links model is like black and white.
Speaker: It's like you search for something, there are like 10 links returned, you add three links, it's very black and white.
Speaker: You like click on that and you make money off of it.
Speaker: It just seems really hard.
Speaker: Even if you've optimized the hell out of it, I just don't think it could monetize as effectively on a chat based interface.
Speaker: But yeah, I think it'll be weird.
Speaker: I think they'll need some designers and some product managers figuring out, but some queries that would have ads for it that are, I'm like, hey, give me an authentic poke restaurant on Maui.
Speaker: It gave me results, it gave me lists and, you know, I could click through to them, but why can't that list also have of its four links?
Speaker: Why can't there be one, you know, that's sponsored or like asking for inspiration for shoes that are kind of like these shoes already have.
Speaker: Why can't, you know, two out of five of them be ad links can be like labeled as an ad, but.
Speaker: That's fair.
Speaker: I do feel like I don't know how that'll evolve if things become like evolve towards more of this agent experience where it's doing stuff for you like intentionally where a robot doesn't look at ads.
Speaker: It just looks at the facts.
Speaker: And so but like right now, I'm probably less bullish on having ads in like a chat based interface and maybe more bullish about like the amount of data it opens up and like
Speaker: gets Google and you know I don't think we've talked a lot about the Google Privacy Sandbox and some of those changes that Google's doing to essentially hunker down their walls to kind of control more of the ads value chain of things but I'm probably more bullish about
Speaker: them going in the play of like, hey, this is a chat-based interface.
Speaker: It gets us more data about users.
Speaker: It maybe helps us show relevant ads and it gives us access to information about users, which is basically not available to anybody else.
Speaker: But how do they monetize that value?
Speaker: And if they get rid of...
Speaker: search is such a high percentage of their revenue.
Speaker: Like, sure, you can't escape the fact that a YouTube video is going to have an ad in it or a news article is going to have an ad in it, but like your search interface not having an ad in it is like so core to how they make, you know, 90% of their revenue, 90% plus.
Speaker: So I think that they probably don't want to be
Speaker: the first ones to do it, but if someone else does it and it's a better replacement and it's going to be better, they're going to want to be there.
Speaker: And so that's why I kind of do see your point about it being a defensive thing.
Speaker: Be there, be in the space.
Speaker: Does this chat thing replace general search?
Speaker: Maybe, maybe not.
Speaker: And if it starts to be ready to, you know, I don't know, it's doomsday scenario for your search revenue.
Speaker: So like, how are you going to, I don't see consumers paying for their search engine ever.
Speaker: So it's, it's.
Speaker: The paid model is going to have struggle with adoption.
Speaker: So I'm looking at ads, but yeah, so how else can you reduce costs?
Speaker: And I think it's a much longer topic to discuss another time.
Speaker: We've got fun times.
Speaker: Curious to see how this plays out.
Speaker: I guess that's the two topics we had for today.
Speaker: This was a good conversation.
Speaker: Follow us on Spotify, Apple Podcasts, or now on YouTube if that's your preferred platform.
Speaker: You can always subscribe on Substack and get notified whenever we have new episodes coming in.
Speaker: We are trying to get new episodes out weekly-ish.
Speaker: So far, so good, James.
Speaker: JAMES MIRANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAANIYAAN
