Transcript
Speaker: Well, here we are, episode number six.
Speaker: We're excited to do this again.
Speaker: I think we've managed to keep a good amount of consistency so far.
Speaker: So far, so good for the year.
Speaker: I'm proud of that.
Speaker: I don't know if it was either of our New Year's resolutions, but if it was, not bad.
Speaker: That is true.
Speaker: We are recording this from very cloudy and rainy San Francisco.
Speaker: I don't think anybody is used to the rain in San Francisco, so every time it rains, everyone loses their mind.
Speaker: Definitely.
Speaker: There was even thunder earlier.
Speaker: That was crazy.
Speaker: This is so weird.
Speaker: Like when I grew up in India and India always had tropical rains.
Speaker: Oh, yeah.
Speaker: So when it rained, it just like rained.
Speaker: And here, I think California has been in a state of drought for many years, and now it's just been raining over the last couple of years.
Speaker: Yeah, that's, I guess, good for us, yeah, or good for people in California.
Speaker: But before we get too serious, I know you just benefited from it.
Speaker: I would like you to tell me, Viggy, why did we have here in the U.S. the day off on Monday?
Speaker: That is a good question.
Speaker: So this year in 2024 was February 19th.
Speaker: I know the theoretical answer.
Speaker: I don't know what is the actual reason for the holiday.
Speaker: I think it's called President Day.
Speaker: Okay.
Speaker: Right?
Speaker: That's about as much as I know about why we had the day off on Monday.
Speaker: And I'm not complaining.
Speaker: Yeah, that's right.
Speaker: It was originally for George Washington's birthday, but then shenanigans, they moved around, and yeah, that's for all presidents.
Speaker: Well, we're going into 2024 election year, and I don't know how I feel about President's Day anymore, but the holiday is nice.
Speaker: I agree.
Speaker: Maybe that should be on one of the upcoming candidates' presidential platform is another three-day weekend, because in the...
Speaker: In the US, we have a dearth of a three-day weekend between February and May.
Speaker: So March and April are looking for three-day weekends out there.
Speaker: That's true.
Speaker: All right.
Speaker: So with that, we will dive into two topics this week.
Speaker: The first one, we didn't get enough hot button topics last week already from the political content.
Speaker: So we will do a follow up of that, specifically zooming a little bit more into a recent commitment by some large tech companies and social platforms about their policies around election interference and what they're planning to do going into 2024, which is a big election year, not just in the US, but across the world.
Speaker: And the second topic we have is around Twilio, which has been through an interesting journey over the past few years.
Speaker: They've had some troubles over the past year, a couple of years.
Speaker: They had a recent earnings last week where their stock dropped a little bit after the earnings was announced.
Speaker: And there's also been some activist pressure.
Speaker: We'll go all into that in a little bit.
Speaker: Cool, let's do it.
Speaker: All right, so let's jump into the first one.
Speaker: Last week,
Speaker: A bunch of large tech companies.
Speaker: This is about 20 companies.
Speaker: It includes the who's who of big tech.
Speaker: So this is Meta, Microsoft, Amazon, Google, IBM, and some of the AI platform companies as well.
Speaker: OpenAI, Anthropic,
Speaker: And the social platforms, TikTok, Twitter, which I guess goes by X now, all of them signed an accord, which essentially is a pledge to fight election interference going into 2024.
Speaker: A lot of their language is focused around what they call deceptive AI election content.
Speaker: So this is AI generated content.
Speaker: It could be text-based.
Speaker: It could be video-based ones, which have come under a lot of scrutiny lately, especially deepfake type content, which we've seen a lot of in the last few months.
Speaker: Again, so these are voluntary commitments.
Speaker: So it's not really any kind of a legal requirement.
Speaker: You know, there's no real enforcement on whether they actually stick to this or not.
Speaker: Broadly, there's a few different commitments that are part of this accord.
Speaker: Without going into the laundry list of everything that's in there, maybe lay out the few big ones that they've talked about.
Speaker: One is prevention.
Speaker: So essentially stopping bad actors from being able to generate some of this misinformation or deepfake type content.
Speaker: Second is what they loosely call as covenants, which is...
Speaker: essentially refers to what is the origin of certain content, whether something is AI generated or not.
Speaker: And they don't really commit to explicitly labeling AI generated content, but they do talk about facilitating ways in which, you know, uploaders of this content can disclose whether something is AI generated or not.
Speaker: And then the third broad bucket is they talk about transparency in terms of letting people know what is happening, providing access to researchers.
Speaker: And funnily, they use a word called resilience.
Speaker: Stay strong.
Speaker: Stay strong.
Speaker: I don't know.
Speaker: Which kind of seems if you actually read some of their press releases, it sounds a little bit like, hey, we can't solve this whole problem.
Speaker: There is going to be bad things happening.
Speaker: We're going to try our best.
Speaker: And one of the ways to do that is through resilience, quote unquote.
Speaker: which essentially refers to, you know, actually creating more literacy around with users to recognize some of this content and all that stuff, right?
Speaker: It does kind of sound like a workshopped word or maybe the lobbyists at a company and the lawyers at a company got in the same room and they discussed, well, what could we actually commit to?
Speaker: That is a good question.
Speaker: So let me ask you, this is a funny one.
Speaker: Imagine you were one of these large companies and imagine you had to give out a press release
Speaker: for this accord, what would you say?
Speaker: Well, I probably would say something about how advanced technology we have for a leader in applying this or using this and that we have a history of doing right by people and that we will continue to do right by people with these new technologies.
Speaker: That would be roughly what I would say.
Speaker: All right, so let me read this out.
Speaker: And you tell me who wrote this, right?
Speaker: We are committed to upholding democracy and the Munich Accord complements our existing efforts to build and deploy new AI techniques that are reliable, secure and safe.
Speaker: We believe this is an important part of our collective work to advance safeguards against deceptive activity and protect the integrity of elections.
Speaker: Okay, well, I have a first pick in my head.
Speaker: I'm gonna guess Facebook.
Speaker: All right, no, you can take a second one.
Speaker: Well, my second one would be, and it's a tie between Google or Microsoft, because they both have leading AI chat.
Speaker: Okay, that's like warm, you know, so this was Amazon.
Speaker: What?
Speaker: Okay.
Speaker: This is, again, I guess the broad commentary is this is a loose outline for what they're committing to do.
Speaker: There is not really much of enforcement.
Speaker: There's not really too much accountability.
Speaker: They're essentially agreeing to this loose framework of what they're going to do going into this year.
Speaker: Which is an interesting, I guess it's progress.
Speaker: Yeah.
Speaker: I'm curious, was it done under the realms or because of any motivation from a particular political body?
Speaker: Like it's in Munich.
Speaker: So was it like a meetup involving people that are in the EU?
Speaker: Was it like a worldwide NGO?
Speaker: Was it like a UN meeting on something?
Speaker: I'm curious, like why did they do it now?
Speaker: Yeah, that's a good question.
Speaker: I think it's a couple of things.
Speaker: One is 2024 is a big election year across the globe.
Speaker: In the US, the presidential elections are, I guess, the final election happens in November of this year.
Speaker: I was looking up which other countries have elections.
Speaker: There is a bunch of them.
Speaker: So in the Europe, for example, Austria, UK, and a bunch of other countries are going into elections this year.
Speaker: There's also an election to the EU parliament, which is happening in 2024.
Speaker: India, which is the world's largest democracy, has gendered elections this year.
Speaker: Indonesia has elections this year.
Speaker: Taiwan just had an election sometime last month.
Speaker: So it's a huge year where there are a lot of these major elections in the midst of a fairly volatile political climate.
Speaker: And that has definitely increased the amount of pressure that is on a lot of these platforms to, you know, act responsible or at least make sure that they are in a position where they can defend their actions, even if they're not the best Samaritans.
Speaker: That's number one.
Speaker: The second big factor is there have been a couple of bad things sort of in the realm of like AI content which have happened recently.
Speaker: I don't know how much you've followed this.
Speaker: So
Speaker: A month or so back, the election primaries were happening in New Hampshire in the US.
Speaker: Okay, yeah, I remember those.
Speaker: Right.
Speaker: I don't know if you heard about this.
Speaker: So basically, there was a robocall that was in the voice of President Biden.
Speaker: And it essentially said it would call people and it called people on behalf of the local Democratic chair or whatever of that state.
Speaker: And they would basically ask the voters not to show up for the elections.
Speaker: It was a very vague thing where it said something to the tune of, oh, these primaries don't really matter.
Speaker: Don't waste your time showing up today.
Speaker: What I need you to show up is in November and not today.
Speaker: So this is like very clearly election interference.
Speaker: I don't think there's a clear verdict on who was behind it, but this was something which, you know, kind of became like a pretty serious issue that got a lot of media attention.
Speaker: It turns out that there was, you know, an AI voice generation kind of startup that whose product was used to create some of this content.
Speaker: But essentially, I think it's a combination of
Speaker: the volume of elections coming in this year, as well as some of these sensitive issues which have happened in the last few months, which have really accelerated a lot of these conversations.
Speaker: Do you feel like there's something like a new inflection point happening?
Speaker: I guess I could see because of the elections happening and past election interference, there'd be some zooming and some highlighting of this being a problem.
Speaker: But do you think the technology has managed to be chained since we've had a lot of the elections that coincided?
Speaker: I definitely think so.
Speaker: I mean, I think we've probably seen this a lot with generative AI, the difficulty in creating content at scale.
Speaker: That's dropped a lot, like the cost of creating content, the cost of creating content at scale has dropped a lot, which is likely going to cause a large implosion in the amount of misinformation and the amount of any kind of content that is available on the internet.
Speaker: That's definitely one.
Speaker: And then I think with...
Speaker: some of the LLM technology and like more of the generative technologies, especially around videos.
Speaker: Like some of these deep fakes are very, very, like they're very believable.
Speaker: So I listened to the audio, which was the Biden robocall and
Speaker: I would reasonably believe that that was like Biden's voice, you know, and that is definitely a problem.
Speaker: Like, I do think that there are non-nefarious use cases for this.
Speaker: I don't want to specifically name startups just because we're critiquing a bunch of things.
Speaker: But there are a couple of startups that do things around like marketing communications.
Speaker: For example, if you're a B2B salesperson and you want to send, you know, customized emails,
Speaker: video recordings instead of sending emails to a potential client.
Speaker: There are legitimate use cases like that where these kind of tools could be helpful.
Speaker: But I do think that the power of these tools, the effectiveness of these tools has increased pretty drastically over the past year where these are starting to become more and more serious issues.
Speaker: To double down on what you just said, I think to
Speaker: if we called robocalling like an automated, you know, marketing campaign via phone, okay, it's a little fuzzier name, but also if this was an intentional campaign by one of the political parties to want to do something, say it was, they wanted to automate phone reach outs and they would want it to be in their candidate's voice.
Speaker: I do feel like that's a good use of this technology and that's one of the improvements.
Speaker: Like you said, it really reduced the cost to make the content.
Speaker: But it matters that
Speaker: the candidate himself was trying to do something.
Speaker: Where we're talking is the misinformation risk comes from a nefarious actor or opponent having the poor judgment to do something like this where they pretend to be a candidate.
Speaker: And I think that's where the misinformation comes.
Speaker: Yeah, absolutely.
Speaker: And I think the other version of this risk is I was reading Microsoft's press release around this commitment.
Speaker: So one of the things that they talk about is they don't specifically call out the root cause of the Biden robocall.
Speaker: There have also been a couple of similar issues with the UK prime minister, for example.
Speaker: What they do call out is that there have been other instances of this kind of content that's created in the past, which is often created by like nation state kind of actors.
Speaker: So this is not just necessarily one person's political opponent trying to do something nefarious.
Speaker: It's also just, you know, it's
Speaker: You are a country that has a lot of access to technology and you want a savvy way to do geopolitical warfare without actually having boots on ground.
Speaker: After cybersecurity, this is probably one of your most effective ways of going after that.
Speaker: It would be interesting, how do you combat that in a democracy with open freedom of speech and where there's not, I guess there's been scandals in the past, but it's not like everything is monitored everywhere in the US.
Speaker: So how would a government agency in the US be able to monitor that though there's a robo call campaign going on?
Speaker: Well, how do you distinguish between a robocall campaign that's nefarious versus a regular marketing campaign?
Speaker: You know, it's not like they can listen to everything, especially as we've moved on to increased security for consumers when they access their banking applications, for example.
Speaker: It's a lot harder to snitch network traffic and know what's going on there.
Speaker: So it's really hard to do that.
Speaker: So is there like a global approach they could take or?
Speaker: Maybe this is the origin of these companies using the very workshopped word resilience, where they're trying to put the onus on the average person to be able to recognize this stuff.
Speaker: Yeah, absolutely.
Speaker: I think the part that I struggled with in terms of how to solve this problem, if you look at the list of these 20 companies,
Speaker: I would actually break them out differently in terms of where they fall in the AI value chain.
Speaker: You know, like there is OpenAI Anthropic, for example, who are primarily model providers.
Speaker: Like, yes, OpenAI also has ChatGPT, which is an application, but they are primarily like a model provider company, right?
Speaker: There's the Amazons and the Microsofts of the world, which are probably more at like the
Speaker: Compute, you know, they provide like Azure and AWS, which helps run these models, train these models and all of that.
Speaker: But they're still basically in the infrastructure layer of things.
Speaker: And then there are more applications, I would say, which are...
Speaker: They're closer to the consumer.
Speaker: This could be something like a Google search or a Google bot or even chat GPT for that matter.
Speaker: That's kind of one on the applications layer, which is used for creating this content.
Speaker: And then if you go one more step above that, there is the distribution platforms.
Speaker: This is like the TikToks and the Facebooks and Twitter.
Speaker: So the world or phone calls.
Speaker: Or phone calls.
Speaker: We'll shut up through the second segment where we talk about Twilio, but I'm not going to talk about this subject, but an automated phone call is also a distributor.
Speaker: Right.
Speaker: And even like emails for that matter, things like that.
Speaker: So I guess the part that I struggled with is, you know, where do you hold, like, who do you hold liable when something bad happens, right?
Speaker: Is this the infrastructure player?
Speaker: Is this, you know, application player?
Speaker: Is this the distribution platform or is it just purely the person that, you know, created the content?
Speaker: You ask the average person, all their ire would go to the distribution platforms.
Speaker: The average consumer, like where they saw it.
Speaker: It's Facebook's fault.
Speaker: I saw this, or it's, I can't believe, you know, YouTube would host a video like this and they'd want to bring it down.
Speaker: And, you know, like theoretically the thing that's protecting these platforms from this is...
Speaker: Section 240?
Speaker: 230.
Speaker: 230.
Speaker: Yeah.
Speaker: Got it.
Speaker: And where they're not responsible for the content they host.
Speaker: And so I wonder if this is, maybe it's a softer thing where they want to be on top of it because they also don't want to be the cause of spreading misinformation and it would put them at risk of further regulation if they're not keeping their own house clean, so to speak.
Speaker: Yeah, I think I agree with that take.
Speaker: Distribution platforms are definitely the ones that play a pretty large role in amplifying this content.
Speaker: But I guess the challenge is there's different types of distribution platforms, right?
Speaker: I guess social platforms is definitely...
Speaker: one of the more common ones and maybe the largest volume of where distribution is happening.
Speaker: We have seen some conflicting moves from different platforms where, you know, we talked about this last week, Meta is moving away from actually recommending political content, which
Speaker: I do think it's a tough trade-off of, you know, when you start heavily regulating some of these things and when you hold a single layer responsible for all of the problems, we'll start seeing more and more of these platforms back away from this, which I do think, you know, limits the amount of discourse that's happening.
Speaker: Maybe a quick analysis point on that one is what sounds interesting about the meta approach is I think we didn't totally resolve
Speaker: if the open graph recommendation applies to just organic content or it also counts for ads and if we continue with the thought that it doesn't apply to ads so if you're paying you can reach people that don't follow you yep following that thought process and combining it with what we're talking about today what i find interesting is
Speaker: like a common technique that folks use for combating bots is requiring, once is adding some amount of payment method needing to be on the account.
Speaker: So either a payment have to have a credit card attached to the account, that's, that filters out a lot of scammers.
Speaker: And then having to have a small amount of charge on the account, like a, like a usage fee, but it was really negligible.
Speaker: Yep.
Speaker: And so being able to regularly charge a credit card also reduces fraud and bought traffic.
Speaker: And then I think the final form is, well, you have to pay for your usage and paying for your usage.
Speaker: It just makes it expensive to run a scam at that point.
Speaker: I guess this was the rationale behind Twitter's whole like verified, or this was at least the externally expressed rationale behind the Twitter check mark and why they were launching verified.
Speaker: Exactly.
Speaker: And for that, you know, at the end of the day, the first scamming, there's the upside and the cost calculation is you can make a spreadsheet for this.
Speaker: And I,
Speaker: would say maybe Twitter's version wasn't a very robust spreadsheet because clearly a lot of people were able to take advantage of it for that $8 fee.
Speaker: And I think it's also a question of who is this nefarious actor, right?
Speaker: If it is like a state-sponsored actor, for example, I don't think they would have a problem submitting a credit card.
Speaker: This is actually what happened in the Biden robocall situation.
Speaker: So...
Speaker: I guess these guys are in the news, so maybe it's fair to call them out.
Speaker: So the startup that actually enabled this is called Eleven Labs.
Speaker: So they basically helped create this Biden robocall deepfake audio.
Speaker: Essentially, their stated policy on their website is they do allow users to make voice copies of politicians and things like that.
Speaker: But only when used in the context of like parodies, for example, if you make something funny and if it's like
Speaker: obvious that it's funny you're allowed to use it which I don't know it does seem like fairly iffy policy it rhymes to me like as a not a legal expert kind of rhymes with me that's probably pretty close to like they're trying to restrict themselves to a certain type of fair use right where it's like
Speaker: Reusing something for analysis is fair use, but also using something to make a joke or make fun of it is a classic example, protected speech that you're allowed to do regardless of what you're doing.
Speaker: So I could see how that's what they limited themselves to, but seems pretty easy to take advantage of evidently.
Speaker: That's true.
Speaker: Yeah.
Speaker: I mean, I guess like payments is maybe an OK starting point, but I doubt if it actually solves the problem of some of these well-funded, well-organized actors who are trying to intentionally cause some of this misinformation to spread.
Speaker: I do wonder though, how much of the onus should also lie on some of the more underlying like platform providers.
Speaker: I do think that anybody that is using a large model or like needs to train a model, things like that, I don't think you have a better choice than to use, you know, Azure or AWS and things like that.
Speaker: And I think most of their policies today are designed very loosely of, you know, don't be a terrorist.
Speaker: But I don't think it actually accounts for some of these more nuanced situations.
Speaker: Yeah, I'm struggling and I think I'm only going to arrive at something that's imperfect.
Speaker: But I think what would help to predict what the average person would think is try to come up with a metaphor.
Speaker: A very rudimentary one would be someone gets hit by a car.
Speaker: The driver's at fault, evidently.
Speaker: Is the manufacturer at fault with the car?
Speaker: There could be arguments that makes the manufacturer at fault.
Speaker: Now, what if I told you that the automated driving mode of the car was enabled?
Speaker: Like now is the company, whether it's the same or not, of that software, are they at fault?
Speaker: Or maybe extrapolating even more like their self-driving cars in San Francisco.
Speaker: Like at what point is who's at fault for any sort of interaction there?
Speaker: And I think it's a really difficult, like a lot of limitations of that metaphor, but there are different actors where the person driving a car.
Speaker: Yep.
Speaker: is maybe the distributors, right?
Speaker: Because it's the one that's using the final thing.
Speaker: The person that manufactured the car is maybe OpenAI.
Speaker: So the fact that Ford made a car and someone ran them over with it, it's hard to say it's Ford's fault.
Speaker: So if this bad metaphor, if someone used an OpenAI model to generate a Biden soundalike or another world leader soundalike, I struggle to think of OpenAI as at fault here.
Speaker: I think that's fair.
Speaker: The thing that I'll probably push back on is, let's take the example of this startup, right?
Speaker: Like 11 Labs, which I'm guessing they used some model from like an OpenAI or Anthropic or whatever to actually generate this audio replica.
Speaker: I do think that there needs to be some amount of responsibility on Microsoft Azure or like
Speaker: you know, Amazon AWS or OpenAI or Anthropic to actually make sure that there is some chain of liability that's happening where, okay, if OpenAI is like not talking to the startup, that's fine.
Speaker: Like maybe they pass that responsibility on to, you know, Microsoft because maybe that's how the model is being accessed.
Speaker: But if Microsoft is actually selling to someone that is doing a questionable application that's promoting all of this, I do think that there should be some liability at that
Speaker: applications layer of creating content, which in your case is, I don't know what's a good extension to that analogy, but let's say somebody bought a Ford car and added, I don't know, something to like get the car to run 2x faster.
Speaker: right and then sold it to a consumer you know that seems like something that the reseller or the you know the one person before the consumer that's not ford should be held responsible for it's true because at the end of the day what was theoretically possible in the past is you could hire a voice actor that could approximate the voice right and so what we're automating there is that thing so that's kind of like you're going from
Speaker: a car working regularly to going to X faster.
Speaker: And so you're going from holding the voice counter voice actor accountable to holding the, the people that make the automated voice accountable.
Speaker: Definitely.
Speaker: And, and I guess the point there would be like, well,
Speaker: I could see how you'd want to hold the people that generate the audio content or video content, regardless of where it's posted, you would want to hold them accountable.
Speaker: So to what level do you expect them to do that?
Speaker: You know, like how do they just have to detect it and not do it?
Speaker: That's pretty hard.
Speaker: Or what's the minimum they would need to do to be able to?
Speaker: Yeah, definitely.
Speaker: I do think that if somebody is actually trying to create, you know, Biden robocall voice that is asking people not to
Speaker: vote in elections, like I think there should have been a trust and safety framework at the applications, at like the final applications layer before the nefarious actor gets like that piece of content in their hand.
Speaker: So I do think that this startup in this case should have taken up more of that responsibility.
Speaker: But I do agree with you that it has to be a shared liability framework where
Speaker: At the end of the day, it's the part.
Speaker: And I do think that there is clear regulatory movement on the...
Speaker: end nefarious actor like for example creating deep fakes for political uh reasons is like now criminalizes intense states in the u.s i don't think there's still a federal law around this but for example if in these particular states you are the person that's creating this content and uploading it to your tiktok or whatever as the end person who's doing this you can be held accountable for it
Speaker: not TikTok, not the tool that you used to download this content, but you as the final, you know, person that picked up this content and distributed it.
Speaker: Right.
Speaker: So I do think that we are already at that end of the spectrum.
Speaker: I do think that there is need to move a couple of layers in.
Speaker: That's true.
Speaker: If I was to make an analogy with previous things that happened in the US, there's an example of how
Speaker: widely you're supposed to have free speech in the U.S. but an example of where it's limited is when it can cause harm to others and a famous example to cite is you're not legally allowed to yell fire in a theater because it causes panic right and so in this case if you're the person like causing a panic because of misinformation that kind of tracks with that thing I'm going to cheat a little bit and use some of my notes as this leading question or maybe set this leading thought so
Speaker: I think it sounds kind of hard for you to know categorically that someone just generated some content that's going to be misused.
Speaker: So is there any way that you could, as this startup or as Microsoft, or maybe I should say as the provider of generating content, could communicate to the distribution layers, this is LLM, or this is LLM generated?
Speaker: Is there something that they can do there?
Speaker: Yeah.
Speaker: Yeah, I think that's a good point.
Speaker: I do think that as part of these voluntary commitments that they signed, a lot of these players have essentially said that they would create mechanisms where somebody that uploads a video can self-declare saying that this is AI-generated content.
Speaker: I do think that there are reasons why it doesn't make sense to, you know, enforce broadly that every AI content needs to be, you know, labeled as such.
Speaker: Because it is difficult, right?
Speaker: For example, let's say you are drafting like an email that's going out, you know, it's a sales email that's going out to like a prospective client.
Speaker: And do you want to label that as, hey, this is AI generator?
Speaker: Like what if ChatGPT generated a version and you edited like 20% of it?
Speaker: Like I do think that that is...
Speaker: hard to enforce as like a broad policy.
Speaker: Google's talked about this in their SEO sense where they've explicitly said they're not going to penalize a piece of content just because it is AI generated.
Speaker: So I do see the arguments of the other side where, you know, I don't think it's reasonable to expect all AI generated content to be labeled so.
Speaker: But yeah, I do think that
Speaker: there do need to be like better systems for, you know, governance and more clarity around who created the content and where it came from.
Speaker: Something I was, I was looking for, like, I do agree with, with that perspective, but something I was looking for in, in, in posing that, that question was, so for Microsoft or OpenAI to be to communicate to Facebook or TikTok, that, that any arbitrary piece of content was machine generated.
Speaker: Yep.
Speaker: I think
Speaker: This wouldn't obviously protect against people running the models themselves.
Speaker: But if you use one of these platforms to generate it, something that you could actually do, you know, scammers beware, is you could kind of opaquely watermark what you're generating to say this was generated by XYZ, you know.
Speaker: it's, it's pretty crazy that you could actually detect an arbitrary text that might totally read normal to you as a human.
Speaker: Actually, there's a, you know, programmatic way to detect, you know, what's likelihood this is generated by an LLM.
Speaker: So there's that.
Speaker: I think even with sound and video, these are much more robust, like file types.
Speaker: You could hide a lot more data in that.
Speaker: And I think there's been lots of
Speaker: past examples of people hiding data within photos and videos and stuff like that.
Speaker: And so I kind of see that like an easy mechanism for these companies to promote resilience would be, even if a user doesn't self-declare that something's LLM generated, it'd be fairly easy for these distributors in your example to check for any sort of watermarking and for these companies that generate the content to...
Speaker: intentionally add a watermark to any generated content.
Speaker: That makes a lot of sense, yeah.
Speaker: I'll maybe finish with a couple of random ideas I had on things that could actually help with this problem.
Speaker: One is, this has been an interesting one.
Speaker: I think one of the powerful things about LLMs and some of these models is you can use them for all kinds of creative use cases.
Speaker: You asked the question of, if you are one of these startups that is
Speaker: actually have this capability, which, you know, could be used for nefarious purposes.
Speaker: Like, how do you kind of predict whether this could be used for nefarious purposes?
Speaker: Like, one solution to that is you could literally feed this content into an LLM and say, hey, give me, you know, do a bunch of prompt engineering.
Speaker: It's come up with the 20 most nefarious things I can do using this.
Speaker: And then, you know, come up with some kind of risks score for, you know, based on this, like, here's the level of risk that I expect from this content.
Speaker: And so I do think that there's a little bit of that application of the generative models for the trust and safety layer, which I think is really, could be really interesting and could be interesting for, especially for startups that are opening up themselves to a lot of this risk by, you know, through unintended consequences of their products.
Speaker: That's one.
Speaker: The second one, something along the lines of what you said, alongside watermarking, again, I think it's possible to use some of these LLMs themselves to
Speaker: essentially, you know, reverse and figure, you know, ask the LLM, do you think this is generated by an LLM?
Speaker: And so I think there's some of those circular use cases where I think there's an opportunity for more AI detection tools.
Speaker: Like I was kind of thinking about this, like an ad blocker, for example, you know, like, I know we're not like blocking AI content here, but
Speaker: There's probably an equivalent of that as a consumer product where, you know, if you want more transparency, if you want something that like just reads something and, you know, maybe it's a Chrome extension and it like tells you whether something is a generative or not.
Speaker: I think that could be like a really cool consumer product opportunity.
Speaker: Yeah, for the listeners out there that might've thought initially, well, that's a dumb idea.
Speaker: I want to back Viggy up on that example.
Speaker: People that know more about LLMs than me have talked about, talked to, I've talked with them and they've talked about how crazy it is that one of the most effective ways to detect LLM content is to ask an LLM what is LLM content.
Speaker: And it's, it sounds ridiculous, but it actually is true.
Speaker: All right.
Speaker: So with that, that was a good conversation and we'll transition into the next topic.
Speaker: Alrighty, jumping over to Twilio and Twilio segment.
Speaker: For our listeners out there, they recently had a earnings call and kind of also recently in the news, they had some layoffs as well.
Speaker: And we thought it was an interesting case of a company's business model kind of shifting and things like that.
Speaker: And so maybe to start, I should talk about what Twilio actually is.
Speaker: from like on a personal note whenever i've as a developer messed around with a project that involved needing to automate texting or phone calling i their marketing worked really well on me so i would always use twilio as my default so for to be more precise
Speaker: Twilio is known for offering the ability to programmatically interact with, let's say, telephones, you know, send texts, receive texts, do phone calls, receive phone calls, stuff like that.
Speaker: That's sexy.
Speaker: It's a very, very sexy business that's, you know, bound to blow up to billions of dollars.
Speaker: Also very relevant to our previous topic.
Speaker: And over time, they've added some new things to their business line.
Speaker: A very obvious one that clearly worked was they also added ability to do like email based stuff.
Speaker: That was an acquisition.
Speaker: It was an acquisition.
Speaker: Yeah.
Speaker: Okay.
Speaker: This was SendGrid?
Speaker: SendGrid was the name.
Speaker: Yes, exactly.
Speaker: And so you probably know some other providers of automated email sending, and maybe you're aware of some other providers of automated texting and phone calls.
Speaker: You know, I think a big one out there that as long as you only have to send a text out, like a message blast,
Speaker: This is actually an AWS product that you could do like that.
Speaker: There's another one that's called SNS, like with simple notification service, which also can do texts, but a notable differentiator here where Twilio is something more is that you can receive texts with Twilio.
Speaker: You can also do phone calls.
Speaker: And so they are a market leader in offering these kinds of communications products.
Speaker: And then if I understand this correctly, they're primarily like utility for businesses in some ways, like, you know, if you are Uber and you wanted to send text messages to your customers for, you know, whatever, one time password verification or if you're right is your things like that.
Speaker: Twilio is essentially kind of like the utility layer that lets you do that.
Speaker: That's how I'd see it too.
Speaker: Yeah.
Speaker: You know, other random consumer things you might run into like your bank, one password or a email or text verification from like your dinner reservation.
Speaker: Plenty of things like that.
Speaker: And so that's what they're mainly known for.
Speaker: And so rewinding the clock a little bit to around 2020, that was what Twilio looked like.
Speaker: And they announced they were going to acquire a company called Segment.
Speaker: So let me take a second to talk about Segment.
Speaker: Segment itself is a different kind of business.
Speaker: It offered development kits for developers across all major consumer platforms, you know, web, web.
Speaker: Android, iOS, to be able to send information to the server about the user interactions.
Speaker: So this could be helpful for basic analytics if you'd want, if you're not already using one of the popular analytics providers out there.
Speaker: It might be helpful to maybe just give an example for some of our listeners.
Speaker: I know we are familiar with customer CDPs because we work a lot on ad tech.
Speaker: Yeah.
Speaker: What's a good example of, you know, what kind of a consumer interaction would trigger this data transfer?
Speaker: So maybe an example, any sort of event that could happen on an app or a website.
Speaker: Say you're a big brand, sells shoes, and someone adds something to a cart, and then they never check out.
Speaker: So they're kind of like the data transfer layer in a lot of ways.
Speaker: Let's say you're Nike.
Speaker: You want to know when somebody has added a shoe to the cart, but maybe not checked out.
Speaker: So Nike's app basically sends a ping to segment and says, you know, user X,
Speaker: did event underscore checkout.
Speaker: And then, you know, we can actually use this signal to maybe, you know, retarget them with the same product and get them to buy the shoes and eventually get them to actually convert and make a purchase.
Speaker: Exactly.
Speaker: So, so I think there's a couple of things to unpack there is there's that action of the thing I was talking about the development kit is like being able to
Speaker: easily send this piece of information that event happened for a specific user did X or Y or Z. And so you would be able to report back to segment that someone added a specific shoe to the cart.
Speaker: Then they went to the checkout page and then mysteriously there was a missing event of actually completing the order.
Speaker: And that might make an interesting segment, hence their name.
Speaker: for doing something against later.
Speaker: And so that's something you mentioned retargeting.
Speaker: It would be like running an ad to advertise that shoe to try to convince them to purchase.
Speaker: Or it could be like exclude them from future things because you think that they're not interested in that.
Speaker: But the word you used earlier, CDP, customer data platform.
Speaker: There's also a word that's pretty much the same thing, DMP, data management platform.
Speaker: We're kind of like advertising, like broad cross website advertising to do these kinds of things.
Speaker: They would want to build a profile, so to speak, of their customers across all these different areas so that they can do any sort of advertising they'd want.
Speaker: That makes sense.
Speaker: Yep.
Speaker: And then I also, if I understand this correctly,
Speaker: One of the reasons some of these larger companies use CDPs is also when you want to have a singular view of a customer.
Speaker: Like, for example, let's say you've emailed somebody, you've also texted somebody, and then somebody also shows up in your store.
Speaker: In a lot of ways, these customer data platforms essentially act as that.
Speaker: central data store where you have, you know, a lot of information about every single user.
Speaker: Exactly.
Speaker: I think that was a major marketing thing.
Speaker: Like omni-channel.
Speaker: Be the one place.
Speaker: Like across everywhere, see what your users are doing, your prospective customers are doing.
Speaker: And that's the vision, I think, that the Twilio saw with acquiring segment was they already had some amount of interaction for
Speaker: people out there or to connect with real humans because real humans receive phone calls and texts and brands will want to reach them, but then also brands want to reach them digitally.
Speaker: And so they thought, I think they saw it as a natural kind of feedback loops that build into each other where someone's going to want to do automated texting to give you like your shipments on its way kind of update, but at the same time also want to track whether you got all the way there or something like that.
Speaker: I think that was the initial vision.
Speaker: And for what it's worth, I do think that this unlocks a lot of interesting use cases for marketers, especially if you want to create personas around user behavior or if you want to infer certain things about a particular user based on different touch points the brand has had with a particular consumer.
Speaker: I do think that in a lot of these problems,
Speaker: The biggest problem is like access to data in one place.
Speaker: And I do think CDP actually solves a lot of that data access in one place.
Speaker: I agree.
Speaker: And it definitely did.
Speaker: I think there is a future or another alternate opportunity here that we'll get into in a second.
Speaker: that Segment has all this data coming in and they can be the place people come to for their own data needs, but then for sharing the data with each other.
Speaker: The concept of the clean room of how you can kind of match your data, but without sharing personal information.
Speaker: And if they're like the place where people exchange this data, it's also an interesting opportunity.
Speaker: Makes sense.
Speaker: So Twilio started with more like phone communications, extended email communications, and then
Speaker: purchased segment so they can potentially expand into the customer data's platform in 2020.
Speaker: Exactly.
Speaker: And I think a good one point of analysis I think that's interesting with this expansion is comparing their traditional business with this new potential business when they're doing the acquisition is a big part of their model is when you send a text in an automated way with Twilio,
Speaker: Twilio does that and they charge you on a pay-per-use basis but they also you know they they're not like magically the sole provider of all these things right they have other people they're integrated with and so they also have costs on a pay-per-use basis they still have a healthy margin on these products but still there's a little bit of the opportunities different for their pay-per-use community core communication services versus this kind of
Speaker: wide open advertising opportunity where their ability, they can be the source of data and there's theoretically infinite upside of how they could reuse the data to generate even more money.
Speaker: Whereas in their traditional space, every time someone sends a text or a phone call, they're getting charged, but also Twilio has to
Speaker: you know, pay for that use of themselves.
Speaker: So there is this kind of other like bright opportunity here is not only is it synergistic, it's also a good step for the company profitability wise, where this is a more profitable space.
Speaker: So it has a higher upside and the company will be more profitable over time.
Speaker: It does seem like a reasonable hypothesis at the time of, you know, this acquisition where they did think that they could extend their existing capability like CDP.
Speaker: Maybe segment brings in more of the not just the utility layer of it and maybe it helps them expand beyond the utility layer and actually connect with the core of the marketing data.
Speaker: So I guess at the time of acquisition 2020, that seems like a pretty reasonable path to pursue.
Speaker: Yeah, I agree.
Speaker: And I think I'm going to, I don't know what the journalistic word is.
Speaker: Burying the lead is the right thing to say here, but I'm intentionally avoiding a certain topic right now.
Speaker: And I want to talk about something else that's happened since then.
Speaker: So, I mean, skipping ahead to now, there was also the reporting was a bit dramatic about things aren't looking so hot, especially involving segment and Twilio and segment.
Speaker: And they had an activist investor.
Speaker: I think they've had a couple of activist investors come in.
Speaker: Is that right?
Speaker: I remember a change in CEO from the founder to this other thing.
Speaker: And there's also some other services that they offer that they're slowly shutting down.
Speaker: I think they have an authenticator app or a multi-factor authenticator app and there's other things they're shutting down.
Speaker: I don't remember a change in activist investors, but it would make sense because they
Speaker: based on where they are now, they had this clear, you know, solid business running and there's this other thing dragging them down.
Speaker: And that makes sense.
Speaker: And, and, and so I think something to analyze going through time is they've slowly since 2020, they had gangbusters year with segment, whatever the pandemic hit.
Speaker: There's crazy online activity, at least in the US, in advertising and e-commerce and everything.
Speaker: There's lots of demand, lots of money being made.
Speaker: And then since then, things have kind of petered out.
Speaker: And something that's also changed over time is Twilio's reporting of the revenues has slowly shifted.
Speaker: They had it split between their communication section and their kind of, which is the stuff you'd expect from that, and their...
Speaker: their software things is how they started very broadly.
Speaker: And it's come down to today where they have
Speaker: their segment offerings which is the customer data platform plus like how you would their product which does retargeting via texting though that's in one section of their revenue and the other section is pretty much everything else which is i think where the drama is coming like what are they going to do a segment as well they've really streamlined the reporting such that there's two things here and they're reporting out performance on both and they're talking about how badly segment is going
Speaker: That makes sense.
Speaker: So I guess if I was reading this right, I was looking at some of their valuations through the pandemic as well.
Speaker: At the peak of the pandemic, they were basically somewhere about 5x.
Speaker: I see their high was about $400 sometime at the peak of the pandemic, the stock price.
Speaker: And the stock price now is about like $56.
Speaker: So at that point of time,
Speaker: segment probably seemed like a good buy.
Speaker: It looks like that's probably not how it's played out over the last two, three years.
Speaker: Yeah.
Speaker: I mean, maybe, maybe we should be a little less forgiving.
Speaker: They were probably already planning a buying segment before this pandemic thing hit.
Speaker: That's fair.
Speaker: So the, the, the,
Speaker: transaction went through in 2020.
Speaker: So that probably leading into it 2019, they're working on it.
Speaker: And then things blew up in the pandemic.
Speaker: They had, I think two tailwinds pushing them forward for success, right?
Speaker: They have all the advertising and craziness that was happening because of COVID.
Speaker: But then they also had all this social distancing stuff that was happening that I kind of theorize increased demand for kind of like, oh, we'll text you when your stuff's ready kind of thing, you know, increased usage of their core things.
Speaker: Like both of their businesses were like,
Speaker: buoyed a little bit or they jumped ahead in their, you know, their penetration in the market or like taking, fulfilling the market's needs because there's demands on both sides for increased advertising for discovery because everyone's trying to get discovered on the internet during the pandemic.
Speaker: And then, you know, everyone's trying to social distance.
Speaker: And then, so coming back to the thing I was bringing lead on earlier, I think the big thing that's changed and really
Speaker: I think is evident in the fact that they're kind of poo-pooing the future of their CDP of Twilio segment and talking about how they don't know what the right thing to do is here.
Speaker: Well, they're obviously never projecting uncertainty, but I mean, like they have a clear vision of continuing to be like a leader in communication software.
Speaker: And then they've isolated their CDP as a separate product line.
Speaker: And my theory here is what's also changed since then is I don't know how many listeners followed some of the drama that was following Apple and Facebook's privacy and advertising stuff.
Speaker: It affected more than Facebook, but...
Speaker: But Facebook was heavily impacted when Apple made the change, the app tracking transparency change, where now for a unique advertising ID that you used to be able to have as an iOS developer from every user of your app, that would be the same across all apps.
Speaker: So you could easily know, follow, you know, Viggy across the internet and be able to track that journey.
Speaker: This might sound familiar to how we characterized.
Speaker: And every app by default had access to this.
Speaker: Exactly.
Speaker: Until, yeah.
Speaker: And so things started to crack down in 2020, where now users had to opt in to allow that.
Speaker: And then there was scary messaging saying, are you sure you want apps to track you around the internet?
Speaker: Like fairly aggressive messaging, which the primary reason you would click yes is because, you know, you just wanted to get rid of the pop-up.
Speaker: Exactly.
Speaker: So tricking the user would be like how you got the opt-in, but otherwise most people were opting out.
Speaker: And so at the time, this affected pretty much, I think, all advertising on mobile.
Speaker: And Facebook was very dependent on that.
Speaker: And they, you go look at Facebook's stock price and you'll see, you can pretty much tell when that was released because of where the prospects went.
Speaker: Kudos to Facebook.
Speaker: They recovered from this.
Speaker: They built a whole way to mitigate this and still be effective at advertising, personalized advertising without this piece of data.
Speaker: And so that's a good success story.
Speaker: There's plenty of people out there such as Snapchat that haven't quite figured that out yet.
Speaker: And I think what's the learning here for me, at least, is Segment is another example of a company that either hasn't figured that out yet or their product is no longer valuable because they offer less value now because what they were doing before is literally impossible.
Speaker: Yeah, that's a fair point.
Speaker: I'm curious, how do you think about, I definitely see the argument for, you know, especially in the retargeting world, why the lack of the Apple advertising identifiers and, you know, with Google's Chrome cookies changes coming as well this year.
Speaker: that does change how much first party data is available.
Speaker: But one of the other sides of the argument I've heard, especially in the advertising world is this is kind of the time everybody is starting to hunker down on all kinds of first party data that is available to them.
Speaker: Yes, objectively, there is going to be much less first party data available.
Speaker: But everybody who has access to any kind of first party data is essentially trying to have, you know,
Speaker: something central where they have this data stored.
Speaker: How do you kind of triage those two opposing forces essentially?
Speaker: I think the way I see it is that fundamentally Segment is providing less value now to the people that are using it.
Speaker: For example, especially for logged out users,
Speaker: You won't be able to re reconcile people's different actions across platforms, iOS versus web.
Speaker: A lot of, a lot of Apple's efforts have clamped down on other alternative ways that you could identify users.
Speaker: And so those are becoming less effective.
Speaker: And so while people are hunkering down on their first party data,
Speaker: the value in the first-party data is with the person that owns the data and not necessarily the platform that powers it.
Speaker: And I think that's what changed for Segment is they were the platform that powered everyone's ability to do this.
Speaker: And they had this whole opportunity ahead of them of how could they be the company that aggregates across these different third-party data sources.
Speaker: You might think for some reason, maybe Nike wants to use Kellogg's data.
Speaker: You know, that's like a wild connection, but they don't have that opportunity anymore.
Speaker: They're much more of a utility business, ironically, where now Twilio will find themselves owning two utility businesses where previously one had this glorious future that could expand their revenue and profitability.
Speaker: But now it'll be, they'll, all the
Speaker: All the margin that came from advertising was, has been shifted from kind of being whoever can, can demonstrate their value the most to if you own the data, you get a lot of it.
Speaker: And if you own major consumer platforms, you get a lot of it.
Speaker: And so that's, what's how things have changed.
Speaker: Everyone in the middle, all these ad tech companies that you might see throwing a fit about other changes.
Speaker: are kind of getting squeezed where they're not really providing as much value anymore because you can't do the things that you could do previously.
Speaker: And so if you actually own the data, you're valuable.
Speaker: If you're a place where a lot of users go, you're valuable.
Speaker: But if you're the thing that people use to get there, like anyone could build that theoretically or.
Speaker: I think that's fair.
Speaker: So that's how I see the differences is they don't actually own the data.
Speaker: Yeah, and I remember a bunch of these data platforms and even the DMPs, like the data management platforms, used to have, or at least I guess they still do have the data marketplaces essentially where you could go buy segments.
Speaker: Could be like, hey, I want an auto-intender segment, which is essentially, you know, they've somehow used all of these signals to fix.
Speaker: figure out who are people that could potentially be interested in buying a car.
Speaker: So I think that makes sense.
Speaker: Like now that there is purely first party data and like lesser volume of data, it's probably hard to infer some of those signals from a much lesser, you know, availability of data.
Speaker: I think I do agree with your take on the
Speaker: convergence towards you know utility and this business becoming more and more of a utility business i guess the question i would push on is uh how do you triage the trend on the communication side where they have sort of been a utility business for the most part versus on the segment side where you know maybe they were less of a utility before but are
Speaker: more of a utility now.
Speaker: Like what makes the communications part of their business, like a much better opportunity long-term versus the segment part of the business?
Speaker: I think this is the important distinction because like you said, what are, like I painted a not so rosy picture of their future as a communications company.
Speaker: and when I talked about their opportunity they saw when they or they potentially saw when they when they when they bought segment and I think my sense is what continues to the reason their company is still valuable is the customer base they have so as a leader of these communication services the that is actually valuable it's certainly it's kind of shifted to okay maybe now it's not like a high growth business it's kind of like a stable you know like they're not gonna they're not gonna go away
Speaker: maybe they're like IBM, but like on a much smaller scale where they're not gonna go away, but they're gonna grow much more slowly.
Speaker: Everyone uses them, okay, great, but it's not necessarily differentiates where I think the segment business line
Speaker: wouldn't be a leader in the number of customers they have.
Speaker: I think I was joking as we were reading this or going through the rundown earlier was that the main competitor for segment would be today, now that they're not aggregating cost data, there's something called Google Tag Manager.
Speaker: And it's something that helps you manage the different pixels you load on your website.
Speaker: And that's kind of what segment becomes in this space without this, this extra special sauce.
Speaker: And the kicker is Google tag manager is free.
Speaker: So if you just make it easier for brands to manage the signals they send between their different properties to a central place, there's products out there and they're free that you're competing with.
Speaker: And so I think the fact that
Speaker: Your competitor is free and you're not a leading service in this space.
Speaker: This kind of a really difficult position to be in for the segment part of the business.
Speaker: Whereas Twilio part, I used my evidently, evidently it's an anecdote, but they're the one I think of when I'm like, I want to send some automated text messages to my friends, you know?
Speaker: That's a really good one.
Speaker: I also wonder if there is a difference in, you know,
Speaker: who their primary customers is in some ways with Twilio like I think they do have a reputation of being a very developer friendly product versus selling a CDP is a very different go-to-market approach essentially like you're not selling to developers anymore you're selling to marketing team you're selling to a CMO of a large company it's often
Speaker: a lot more of a top-down decision than, you know, something that you can start bottom-up adoption with.
Speaker: So I would imagine, I see why it is a very different business to succeed in than the communications part of what they do.
Speaker: I agree that even if they did help sales folks today, the ones that are trained up on selling the communication stuff wouldn't be able to easily transition to sell the CDP stuff.
Speaker: So it would be an investment they'd have to make and one that would be hard to do.
Speaker: I think a good analog would be those in the tech community might fill in with Stripe, which does credit card processing.
Speaker: They're famous for having really good
Speaker: APIs and relations with developers and have a lot of goodwill with developers.
Speaker: I would say if you have a mental model for Stripe and how that fits in, I would say that think of Twilio similar to Stripe in that, oh, this is the place you go to, to do X as an API.
Speaker: Whereas I don't, I guess I don't have an equivalent of the boring, maybe it's like Chase.
Speaker: Chase credit card processing or something like that.
Speaker: I've heard IDM.
Speaker: Oh yeah.
Speaker: It's like pretty enterprise heavy.
Speaker: Yep.
Speaker: That's a good point.
Speaker: Yeah.
Speaker: Maybe that's a good contrast is what's that adion of the segment or CDP world because they specialize in selling to CMOs.
Speaker: That's a good point.
Speaker: I guess the one bull case that I have for the communication side of the business, they've had a couple of interesting product launches recently, which again, not to bring the conversation back to AI, but I
Speaker: I think they've had a couple of interesting products there where, you know, I think they started launched something of summarizing phone calls kind of product, which I think is an interesting opportunity for them to, you know, start differentiating more, maybe kind of double down more on their communications business.
Speaker: So I do think that on the communication side of the world, they're probably playing with a significant amount of data, which maybe opens up a roadmap of opportunities for us for them to double down on.
Speaker: I would say that I'd reframe the bull case to maybe frame them as Amazon a little bit.
Speaker: Imagine you are, you're already integrated with Amazon for all your web services and this new technology came out and you want to integrate one more web service.
Speaker: Well, are you going to use some random startup or are you going to hear that Amazon's going to release it next year?
Speaker: Apply that same logic to your texting and phone calls.
Speaker: Like maybe you have a really complex call center that defined in software that you built using Twilio.
Speaker: and doing what you described like ai for routing even or ai for understanding what people calling in want being able to reuse something that twilio can awful also offer to you out of the box rather than having to do some expensive integration i agree as as far as large language models or like this this ai boom being all about that human computer interface that i think it can it really makes warm and fuzzy for us is like
Speaker: You don't feel like you're chatting with a bot.
Speaker: Well, texting and a phone call is exactly the place where you, it would be nice if it didn't feel like you were talking to a bot.
Speaker: And so I think that contributes to your bold case.
Speaker: Absolutely.
Speaker: That was a good discussion, James.
Speaker: Two fun topics.
Speaker: That's it for today.
Speaker: Hope you enjoyed the show.
Speaker: If you have any comments for us, feel free to reply on Spotify.
Speaker: You can also email us at this is unpacked at substack.com.
Speaker: You could follow us on Spotify, Apple Podcasts, or YouTube, and we'll see you again next week.
