Will TikTok and YouTube follow Meta’s new rules for teens?

0:00This episode of Equity is brought to you by Schneider Electric. Schneider Electric. Advancing energy tech. MET is paying$18 billion to make teen safety everyone's business and trying to pull TikTok and YouTube into the mix with an open letter. Also this week, a wave of robotic companies are raising billions in pursuit of their own chat GPT moment, and we're keeping an eye on a couple of deals this week, including Hugging Face reportedly exploring a$13 billion sale. So stick around.

0:33Hello and welcome back to Equity, TechCrunch's flagship podcast about the business of startups. Today is Friday, August 28th. I'm Kirsten Korosek, Transportation Editor here at TechCrunch. And I'm joined, as always, by Senior Reporter Special Project Sean O 'Kane. Howdy. And new this week, or back this week, is Senior Reporter Rebecca Balon. hey guys the gang's all back together the old transportation team yeah it's been a minute anthony hop will be back next week so sorry you're not here today anthony but we have a lot to talk about and you know top of the list well there was one story that certainly got my attention yeah and it's about p believe it or not i'll just say it uh no i mean listen i think as a rule we tend not to cover marketing campaigns, but I do think that our colleague Amanda did a really good job with this marketing campaign that Liquid Death put out earlier this month about, you know, sort of having Jason Kelsey joke about people's, I mean, maybe some people won't joke about it, sending their P data centers because, you know, to sort of essentially raise awareness of the environmental impact of and water usage of AI data centers.

1:52And Amanda actually did the work. Amanda went out and talked to some people who actually have thoughts on whether or not something like this could work, not should, but could work. And, you know, I love this story because it's just a really fun way into understanding this issue a little bit more. I mean, like really broadly speaking, the debate over water usage at data centers over the last like year or two has been very heated. And there's a lot of sort of misinformation flying around. this is a great way to get some basics in front of people in a really funny story. And there's also just, I won't bring them out, but like just some very choice quotes in here that I'm very jealous are not running in a story that I wrote.

2:31So I love it. Yeah, that was a good one. A classic, classic Amanda story. We were talking about it actually, you know, she noticed the campaign was riffing about it and, and, you know, she was like, can you actually use pee? I was like, actually, I think you can. Like, I think you, there are wastewater treatment, you know, plants that you can, I know people who use recycled gray water for gardening their plants and stuff. So it's definitely something that you can do. And so Amanda talked to a lot of people who know more about that than we do. So thank you, Amanda. Amanda also did some other good stories this week.

3:01Kirsten? Yeah. Well, this is a very much a departure from, you know, can you use P2, you know, within data centers. But it is a company that uses data centers. So there's my segue. And we're talking about Meta's$18 billion settlement, which was a massive story, but then it got bigger in my view, because it wasn't just the settlement, but it was this open letter that was sent to TikTok and YouTube, essentially encouraging other companies to follow what they have agreed to do. And to me, I don't know about the two of But to me, that was almost bigger than, quite frankly, the number, the$18 billion settlement.

3:45This is why I don't like settlements so much because it's not law, right? It's not written in the books that a company has to completely change their ways and that other companies won't be able to do. They're not setting a precedent for which other companies will have a standard. That is one thing that I take away from this. And the other thing is like Meta is trying to position itself as this like leader in responsible social media usage for kids being like, join us, TikTok and YouTube when it's like, are you kidding? Like you had it took so much work to get to a point where Meta is like kind of responding to the way that it has, you know, allegedly not allegedly addicted teens.

4:27And it's taken whistleblowers. It's taken a lot to get to this point. Meta has destroyed the Kids Online Safety Act. And yeah, it just felt kind of disingenuous to me. Yeah, I mean, we should say this is one of a number of lawsuits that Meta has faced and is facing. It's always interesting to me to see how far into the process a company will allow themselves to get before deciding to make this decision and sort of pull themselves out of that process. we should also say$18 billion is like what roughly a quarter's worth of profit for meta so you know we've seen fines levied on meta for myriad reasons over the years those fines tend to be far less than this this is something that is you know has a bit more bite but you know if I pull my feelings out of the situation there's part of me that acknowledges that you know it's a sort of They've come up with a, I don't know if clever is the right word, but an interesting way to make this work for them to a certain extent by trying to condition some of this settlement on these other companies following their, I don't want to say lead because it's not their lead.

5:44They had to go through a trial to get to this point and all that. So that's what's been running through my mind over the last day since we saw this settlement come out. And I don't know that there are other cases that will get to this point and force Meta to make decisions like this. So it seems like a pretty important moment then as far as social media in the United States goes. Yeah, I think it's worth going over some of the rules that I'm reading off of this. but there's a two-hour daily default limit on Facebook and Instagram, overnight and school hour restrictions, hiding likes and reactions from teens, restrictions on certain beauty filters, stronger parental controls and protections from adults.

6:27And here's the one that's probably the most controversial, which is more aggressive age verification. Because while there's absolutely advocates for age verification as the stopgap to make sure that these rules are followed, There is also the privacy sort of data collection and surveillance piece of it and folks who, you know, want to keep kids safe but are opposed to those sort of actions. So these are the quote unquote new rules that they agreed to in the settlement and that they're encouraging others to take. And to your point, Sean, it's like, you know, of course they're going to go for other companies to follow this because they want to have, they want to try to level the playing field, if you will, competitively speaking, and not have to be the only company out there.

7:10I will say that it is very likely that other cases will come up against other companies besides Meta that would try to push for these same rules. And so it could happen. It could be litigated and the same result could happen. And whether YouTube and TikTok and others are going to follow based on Meta's encouragement, I'm not so sure. Yeah. And it also just feels like a way for Meta to try to like these, you know, agreeing to these rules, trying to maintain some semblance of a hold in that market share. Right. Because there's a lot of countries that are banning and I think maybe some states, but definitely a lot of countries that are banning social media use for kids.

7:50under 16. New Zealand just became, you know, one of the more recent countries to do this. Australia has done it. You know, Turkey, Malaysia, Indonesia, the UK is thinking about it. So is France and Denmark. So this is, you know, it's becoming a problem for them, I would say. And then I also just want to add that Meta has spent a lot of money advocating like millions and millions of dollars lobbying pretty much against a lot of like online safety protections for teens. They have spent a lot of money advocating for this other bill called the App Store Accountability Act, ASAA. And that is their attempt to act like, look, we want to protect the kids.

8:32But it really just pushes the blame or the responsibility to App Store owners like Google and Apple, you know, shifting the legal responsibility for age verification and parental consent away from them and placing it on the App Store. So, you know, some people have set this as a distraction from COSA and their previous pushback against the bill, which now they're kind of supporting. But there's a lot of policy wonky stuff I won't I won't get into there. Well, I will say that our producer is reminding us that eight states have enacted laws banning minors from social media. So thank you. Or requiring parental consent.

9:10And the question for both of you, just very briefly before we move on, is does this specific settlement change anything in terms of how kids interact with social media? Just this specific case. Who knows? It's too soon to tell. I don't know. I mean, no kids are on Facebook because that's in the words of Ted Lasso for racists and old people. But Instagram, I guess maybe they're all now have Finstas now. I don't know what the kids are doing. I think they're all on TikTok anyway. Yeah, I think we're going to have to wait to really find out whether it'll have an impact. I mean, if any of these changes that were listed out as part of the settlement do go into effect, I mean, yes, it will definitely have an impact on the feeling of using the app, the experience of connecting with people on it.

10:01It's also putting Meta in sort of weird positions. I mean, some of these restrictions really do bump up against First Amendment arguments. And, you know, I could see them having sort of branching trouble that comes from this if they do go through some of these things. My question, you know, the age verification piece of this is, you know, another great story written by Amanda this week is about just that problem in general. And we've talked about that on the show before. You know, it's a nice idea if you're not really thinking too hard about it. But when you think about how it's achieved, the current solutions to that are not great in, I think, most of our eyes.

10:38and so you know we'll have to find out yeah and last thing is like it's tbd how good they're you know whatever they're using for age verification whether it's ids or or you know trying to suss out you know based on the user's behavior if they're a kid or not like they could just lie about their age you know a lot of reports that i've seen from last year from um you know advocacy groups that are that are looking into how you know meta protects or doesn't protect kids, we're able to like really easily make fake profiles like as adults pretending to be teens. And, you know, so it's unclear how well Meta's enforcement of this will be, but I guess we'll see.

11:15And there's a lot of eyes on Meta at the moment. Yeah. I mean, I think this goes back to my original point when I was listing all the rules, which is the age verification piece and how that is achieved is going to end up being the big question, not just for 2026, but in the years to come because social media is not going away as much as many of us wish it would. And so I think that that's going to be like the method of how it is done. And in order to circumvent or to prevent someone circumventing those rules and just lying about it, does that age verification become so aggressive that then it starts becoming like an extreme privacy concern?

11:51And so, you know, to me, that's what this is all going to come down to. And we'll have to watch it play out. I'm sure we'll talk about it again. What moves our world forward? Energy. History has been shaped by it, revolutions won with it, and progress dependent on it. For 190 years, Schneider Electric has been a leading force in this story, creating the technologies that harness energy and transform the competitive landscape. And today, in the dawning era of energy and industrial intelligence, Schneider is helping customers build more efficient, resilient, and sustainable operations across every sector, from data centers to homes, from buildings to industry.

12:28Schneider Electric, your energy technology partner. Discover more at se.com. I do want to get into robots, one of my favorite subjects, and really how they're chasing, specifically robot makers, like humanoid robot makers in particular, are chasing their own chat GPT moment. Yes, Generalist is a robotic startup. They raised some fresh capital,$200 million, and that's valuing them at$3 billion. They are one of the companies that are coming up trying to be like this, you know, chasing robot brains, right? They're trying to not necessarily build the hardware behind it, although maybe they are. But there's a lot of talk around robots almost getting to this chat GPT moment, right?

13:14That robots will soon be able to perform general tasks without being explicitly trained for each one. so normally in the world of robotics the way that you kind of train a robot to do a specific physical task is you train it on that one task forever and ever and ever you get a ton of data on it it learns it it learns how to not mess up that task and then you know you deploy it onto a thing and then you watch it then maybe you fine-tune it and you have to update the software and then redeploy it so the this is trying to get it to a point where it's like okay we have a base model, a foundational model, if you will.

13:49And that model is good enough that when the robot comes to a new task, it could just pick it up a lot more easily and doesn't have to be like fine-tuned and trained again. So we've seen sort of the same idea kind of percolate with autonomous vehicles, which is, you know, training these models, foundational models or operating systems that can be used on various different machines, in this case, robots. And it's interesting if you look at the background of Generalist. This is a company that was only founded two years ago, by the way, by former Google DeepMind researchers, Pete Florence and Andy Zhang.

14:25You know, the interesting thing is that DeepMind, Google DeepMind is actually working with Boston Dynamics right now, essentially trying to create a more robust and let's say flexible sort of model or brain, if you will, for Boston Dynamics, aka Hyundai's brain. So there's a lot of activity. Generalist is certainly not the only company that is working on this. And it is, to me, the more interesting piece. I think, you know, Sean, I don't know if you have an opinion on this. I'm sure you do. But there's been a lot of focus in the years leading up to this with, you know, robots running faster than a human or, you know, doing backflips and dancing.

15:09And to me, where the real breakthroughs are going to happen are, you know, what's happening in terms of its quote unquote brain. And it seems like there's a lot of opportunity here. I mean, I think the thing underlying all of these stories and all of these fundraisers is just nobody knows the right way to approach this, right? Like, I mean, you mentioned there are some similarities to autonomous vehicles, but I know that in Tim Fernholz wrote a good story for this week after going to a conference last week and talking to a bunch of these people about how there's a big difference between those two worlds in that autonomous vehicles, their kind of number one job is other than moving people to avoid things and robots, whether they're humanoid or more industrial, the main job is to actually interact with things.

15:54So you're sort of fundamentally accomplishing very different tasks, regardless of how specific that task is. And so, you know, I think it's important to keep that in mind. No one has the right solution yet. And so I think that's one of the reasons we're seeing so much money rush into this part of the robotics world is to try and figure out the answer. Because it's, you know, when you think about a lot of the more software focused artificial intelligence that we talk about a lot here, so much of that is really down to feed it every book, feed it every YouTube video, feed it, you know, there's, there's language, and that's what it needs to understand.

16:29And then it can move forward from there. That's obviously extremely reductive, but it's just a little bit, in some ways, a little bit more straightforward than this, where there's just so much about physical manipulation. And so every time I think about one of these startups, whether these ones that are raising a ton of money or even smaller ones, like some of the ones I've written about, I'm always curious how they're thinking differently about this problem of collecting this data and making sure that it's valuable. I mean, speaking of that is general intuition, They have a$6 billion valuation.

17:01This is an AI startup that is also pushing into robotics. And to your point of like how they're getting there, they're trying to turn video game behavior into training data for AI agents that can understand and act in the physical world. So you're correct in that there is a lot of different approaches ongoing. I haven't come to a conclusion of one that really rises to the top, but I will say video language models are certainly a big part of that. Yeah. And I think the interesting thing about general intuition, and we've had their founder and CEO Pim Do It on the podcast as well recently. So everyone should go check that out.

17:37But one thing that they think is kind of a secret sauce for them, and it's not the same with every company, but their secret sauce is the action labels. So they have, you know, this is a spin out for metal TV, which is essentially like Twitch for gamers. so they have not only a ton of video game data like of actual videos of the video games being played but also how you were like what you were clicking and when on the remote right and so much of robotics is actually controlled by like video game remotes so that action label data creates they think creates this kind of general intuition of how you're supposed to move through space and time so that's you know one thing that i'm realizing from talking to a few of these companies I spoke to skilled AI as well, and they recently came out with their own foundational model called S1 that dropped this week.

18:25And they are saying that they can teach the robot to do a specific task just by showing it one video of it. So a task that they weren't trained on at all, like making coffee or flipping a pancake just based on watching one video of it. And the thing that I've heard from skilled AI and general intuition is the most important thing is pre-training. So getting the pre-training data right, doing a ton of pre-training, and then you have to do a lot less, if none at all, of post-training, which is really how you get to that moment where you can kind of prompt a foundation model and it will give you an output without having to go back and fine-tune it.

19:05Rebecca, since you've been talking to some of these companies more than I think either of us, like, this is maybe a weird question, but is the focus right now still just entirely on the ability to do the task? Because like, I see that as a massive challenge. And then I also think about how like, even just doing the task isn't enough, especially if you're talking about humanoid robots, that are going to be existing alongside people in whatever capacity that is, whether that's in a factory or in a home or in an office building. there's like a social element to this that often gets overlooked in silicon valley and how things are integrated into the fabric of society and how you know it's not about being able to flip a pancake it's about flipping it at the right time so that it gets to somebody's table like you know hot and fresh you know that's a really silly example but is there any thought being given to that or is the focus just entirely on we need to make this work and then we can go from there So like the question is, are they thinking about how these bots will interact with humans?

20:04Yeah, I think so. It's not what they're talking about, right? Like right now they're mostly talking about, they're excited about being able to accomplish those tasks. And no one's really at like a deployment level yet, right? Like no one's deploying at scale. Anyone that's doing it is like kind of something like above a pilot, right? I mean, I think it will be really interesting once we start seeing these bots in homes and in factories alongside humans. Like, we've already seen how agents, you know, with cybersecurity, hugging face, open AI, yada, yada, we've seen how agents can kind of decide to do something that we didn't ask them to do.

20:37So what does robot alignment look like? Like, could they potentially, you know, hurt a human in the process of, you know, trying to complete a task? There's going to be some serious, I mean, we'll be keeping an eye out on the OSHA complaints. Am I right? Well, yeah, we will. I will make a quick point, which is that like to your comment earlier about pilot stage, I mean, we have Hyundai, probably the best example of humanoid and also probably agility robotics, too. But humanoid robots that are now in pilot phase, they're in warehouses, they're being used. And Hyundai does plan on having Atlas. It's sort of next gen.

21:15And this is the one that DeepMind is kind of partnered up with them on in factories, but completely removed from people. So the big step will be once they are interacting with people, they are separating these robots from people. And once they do start interacting with people, to your point, yes, OSHA, I think databases will be something that we'll be keeping an eye on. And we already have someone who's doing that, but not on the robots side. Well, let's say robot car side. Sean, you recently looked at a lot of data from OSHA and discovered that test drivers, Zooks and Waymo specifically, were getting injured at pretty high rates if you compare it to where they're testing and how.

22:05And that's an example of interesting OSHA data that we might be able to find involving robotics of a different kind. Yeah. Yeah. It's a story I've been working on for a little while. Folks should just go read it. I won't dive too deep into it here. But I think for a long time, we've known about the general process of training autonomous vehicles involving these sort of test drivers or supervisors, whatever you want to call them. And it was just a bit surprising to me when I started looking into that earlier this year, finding that there were a bunch of these folks getting hurt, not just on the job, like that tends to happen at different rates in different industries, but getting injured as a result of the sort of behavior of these vehicles, harsh braking, sudden movements that were unexpected, and often, you know, in response to things that maybe weren't, you know, it wasn't necessarily they were trying to avoid a crash and they got hurt, you know, phantom braking, things like that.

23:00So I'd encourage people to go read the story. And if you have any thoughts about it, certainly reach out to me because it's something I want to learn more about. I have one question for you, though, that I think that we should address. And it came up in your story. So please read the whole story. But what did you discover as to why Zooks and Waymo, you know, were at the top of the list, but other names weren't included in there? Because there's a lot of AV developers out there on the road. Yeah, I mean, it's hard to say what you don't know. But what I do know is that, you know, these reports are only public because of a few things.

23:33One, injuries happened at facilities that have 100 employees or more. And then also, it really depends on sort of how the companies self-identify what industry they work in. And so OSHA forces better disclosure for these, you know, quote unquote, higher hazard injuries. If you're considering yourself to be like a limo, a taxi operator versus just like an R &D lab where, you know, you're still maybe getting data sent to OSHA, but it's not going to be as public. So that's one reason why we only focused on Waymo and Zoox in this story. But, you know, I hope people are reading. If they know more, let me know.

24:10Yes. Sean's signal is everywhere. So, you know, we've talked a lot about actually a bunch of different deals sort of scattered throughout here. But I do want to highlight a few more that got my attention and one in which I wrote about. And that's Gaddix. Speaking of autonomous vehicles, this is a self-driving trucks company and specifically their box trucks. So what they're doing is driving actually commercially and driverlessly, so no human safety operator behind the wheel, on fixed and dynamic routes now in a number of cities, specifically for PepsiCo and a few other companies like Loblaw in Canada and Walmart.

24:53And these box trucks are going from distribution centers to retail stores. And they raised$200 million. dollars. Rebecca, back when you were on the transportation beat, you wrote about them a fair amount. Was that surprising to you that they, you know, raised so much money or their progress? They've been at it for a few years now. It just, it feels like we could have written this exact headline three years ago or four years ago. You know what I mean? It is interesting to me. It seems like they've just been, I'm surprised. I don't want to say I'm surprised they're still around, but I think that I'm surprised they're just moving slow and steady.

25:28You know, they're just moving slow and steady. You can throw shade if you want, but I will provide the counterpoint. It's been a while since I caught up with them. Yes. I will provide a counterpoint, which is, I think that sometimes we can get a little cynical here because we see all these deal flows come through and like, what are the results? And so the interesting thing about Gatic, and I think it is a fair question, like they have been a little bit under the radar. We don't really hear about them too much. They've carved out a niche for themselves in which they don't have a ton of competition.

25:55A lot of companies are going towards that like sort of long haul trucking piece, but they have something like 600 million in contracted. So this is not proof of concept. This is not pilot programs. 600 million in contracted revenue with companies like PepsiCo. So that's real. To me, what's interesting is that they're still raising money in spite of that projected revenue. And the answer I was given was, well, we want to scale up and we want to hire a lot more engineers. We want to, you know, expand within the areas we are already, you know, working in and in new cities and that costs money. Yeah, I guess there's like two things that kind of stand out to me about this.

26:33One is that, you know, cool if they're under the radar and it's working and they're bringing in money and they're able to like fund themselves without having had to raise a ton. That's cool. But also it's like the fixed commercial routes, repeatable trips. Like I'm just curious, like how much their technology has advanced in the last few years that they're still kind of doing like fixed route stuff. Like that's like technologically an easier problem to solve. They are going dynamic though now. So that's another thing. I feel like I'm defending this company that I've been writing about, but like they are doing dynamic routing in Phoenix.

27:05So that is new and, you know, obviously progress is slow, but it is happening. 200 million is nothing. You guys want a big number? I'll give you a big number. How about Hugging Face reportedly almost on the verge potentially of being acquired by NVIDIA for$13 billion. How do you like that? I love it. I love it so much. I'm sure our hugging face loves it. Gotta be, what a roller coaster couple months for them, huh? Yeah. Well, here's my question for you. Like, I think you haven't reported on it, but you've obviously been following it, which is, you know, how real is this? Is this for sure happening?

27:41Is this just, you know, rumor? What is actually happening here? Well, we've had a couple of reports in the last couple of days about, you know, this being on the verge of coming to fruition. And the information is one of the outlets that reported at Business Insider also said that it was, you know, basically about to happen. But, you know, that's kind of where we're at until, you know, unless it happens while we're recording, you never know. NVIDIA certainly has plenty of money to play with and is no stranger to working closely with companies that could possibly need its own sort of products to keep moving forward.

28:18This, I think, would be a new twist in the naughty circular nature of the modern AI world, though, if they were to go and acquire this company. So I don't know, Rebecca, you may have a better read on what that actually would mean for Hugging Face to go sort of inside NVIDIA and how that changes the market dynamics. But that's what I'm looking at. I don't know. I think that they will probably still operate as Hugging Face and do their own think they'll just have um they'll just have a lot more money to to do what they want to do i i don't know i was surprised about this and then now i'm not so surprised because i was surprised because hugging face turned down a 500 million dollar investment offer from nvidia earlier this year they said they didn't want just one investor controlling everything right but now i'm thinking they're pulling a whiz they're pulling a cursor you know what i mean they were like oh no we couldn't possibly but like once they see a bigger number they're like yes like this is a negotiation tactic, if it is in fact, you know, if the information is reporting is down here.

29:16And NVIDIA hasn't confirmed as of the time of this recording. So it's interesting. I mean, when I think more about it, I'm like, oh, I guess this kind of makes sense. Like NVIDIA and Hugging Face have had a long history together. They're both like really publicly aligned on this open source push, right? Clem DeLung, the CEO and founder or co-founder of Hugging Face said that, you know, they used an NVIDIA modified version of a Chinese open source model to defend against the open AI attack. So there's a few reasons why I could see this aligning more. Yeah. Any thoughts from you, Kirsten? Well, we've seen a lot of acquisition type of taking various shades here, some on the acqui-hire where the leadership is pulled into a much larger company, and then the other existing company is left to sort of fend for itself.

30:10I don't think that that's what's going to happen here, to be clear. But we are seeing a lot of M &A activity. It sort of reminds me a little bit of the Stripe open router acquisition for$7 billion. Rebecca, I think in the past you said that you think that you're going to see more of these sort of big enterprise SaaS companies swiping up sort of open model infrastructure providers. So I don't know if you see if this is a through line to that. To your point earlier, it doesn't totally surprise me, but I always kind of pictured Hugging Face as a company that wanted to stay truly independent. But perhaps in this case, as Sean opened up the segment on this, it was just too big of a number to turn away.

30:56And maybe they reassessed their philosophy and strategy moving forward. Yeah, definitely think that I was also surprised because of that, because I had Clem again on the podcast a few weeks ago. If you haven't listened, take a listen. And, you know, he really talked about how important it was for his, you know, the community of Hugging Face and doing right by, you know, the people in his community and the developers in the community. And so it felt interesting that the company would sell itself to such a big player like NVIDIA. But, again, there are reasons to do it. And to your point, Kirsten, I did think that we're going to see more of these types of acquisitions coming for like the infrastructure.

31:36We're seeing really big valuations for these. They're not even making the models. They're just platforms, right? Open Router is just a platform. Hugging Face is a platform. And so I think that this platform play, because they've got that distribution element, they've got the tech already worked in to help other businesses and companies and developers adopt open source AI. I think we're going to start seeing that space ramp up a lot. I'm hearing that there are a few other acquisition targets in the works, whether that'll come from like a Salesforce or maybe, you know, maybe like a meta, right? Like maybe, you know, or a neocloud company that wants to offer a full service.

32:10I'd be curious to see what happens next. So if anyone has any tips, I'm on Signal too. Well, if this teaches us anything, it's that if you're comfortable in AI right now, you're probably going to be surprised because everything is moving very fast and everyone is being bought by everyone. And with that in mind, we're just about out of time, but Equity will be back next week. And so make sure to follow us on X and Threads at the handle EquityPod.