Our 2025 Tech Predictions and Resolutions + We Answer Your Questions

Next year is going to be different: good and bad, but just different.

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  Support for this podcast comes from Dell Technologies. It's time to put AI to work. Is your business ready? The challenge isn't getting started with AI. It's scaling beyond pilots while maintaining control of your data, governance, infrastructure, and costs. Dell AI Factory provides the integrated foundation you need to scale AI securely and efficiently with one trusted partner, Dell, AI on your terms. From idea to impact. Learn more at dell.com/yourwaytoAI. That's dell.com/yourwaytoAI.

• 0:35Casey Newton?

  I'm not sure, have we ever had a guest who's under active federal investigation?

• 0:38Kevin Roose?

  We've had many who have since been on under federal investigation.

• 0:41Casey Newton?

  I want to get one while it's still going on. You know? In fact, that's one of my New Year's resolutions.

• 0:47Kevin Roose?

  Yeah.

• 0:47Casey Newton?

  I want to get somebody who's in severe legal jeopardy to come on Hard Fork and just air it out.

• 0:53Kevin Roose?

  Yeah.

• 0:53Casey Newton?

  Let's get their side of the story.

• 0:54Kevin Roose?

  Yeah, let's, let's get the FBI to send us their pipeline of upcoming investigations so we could do a little advanced planning.

• 1:00Casey Newton?

  Is Elizabeth Holmes still in trouble?

• 1:03Kevin Roose?

  I honestly, I would have her on.

• 1:05Casey Newton?

  Oh, I mean, completely.

• 1:07Kevin Roose?

  I need to ask her what happened with the dog.

• 1:08Casey Newton?

  What happened with the dog? Because she has a wolf dog.

• 1:11Kevin Roose?

  She had a wolf dog.

• 1:12Casey Newton?

  Yeah.

• 1:13Kevin Roose?

  It died under mysterious circumstances.

• 1:15Casey Newton?

  Who killed the wolf dog?

• 1:17Kevin Roose?

  Honestly, Serial, season six.

• 1:21Casey Newton?

  Who killed Elizabeth Holmes' dog? Honestly, that podcast will get a lot of downloads. A lot of downloads. Bad blood, too. Baltos, revenge.

• 1:35Kevin Roose?

  I'm Kevin Bruce, a tech columnist at The New York Times.

• 1:38Casey Newton?

  I'm Casey Noon from Platformer.

• 1:39Kevin Roose?

  And this is Hard Fork.

• 1:40Casey Newton?

  This week on the show, it's our predictions. We'll tell you what we got right and wrong about 2024 and tell you what we think is going to happen in 2025. Plus, we'll take some of your questions.

• 1:52Kevin Roose?

  Make them good.

• 2:02Kevin Roose?

  All right, well, Casey, Happy New Year.

• 2:05Casey Newton?

  Happy New Year to you, Kevin. I know last year was very dramatic and stressful. There were some terrible things that happened. There were some great things that happened. But what I can tell you is 2025 is going to be different.

• 2:18Kevin Roose?

  Different how, better or worse?

• 2:20Casey Newton?

  It's gonna be good and bad, but just different.

• 2:22Kevin Roose?

  Okay, wow. A bold prediction going into the new year.

• 2:27Casey Newton?

  I have a lot of hot predictions this year.

• 2:29Kevin Roose?

  And speaking of predictions, as has become our annual Hard Fork tradition, it is time to check in on our predictions from last year and lay out some new predictions for this year.

• 2:40Casey Newton?

  Yeah, and I'm really glad we're doing this because, you know, I think both of us identify as reporters, but I do think that we stray into punditry from time to time. And a criticism I have of punditry is they don't check in enough on the things that they said were gonna happen and say, hey, did that actually happen? So this week on Hard Fork, we are going to check in our predictions and see what we got wrong and if we got anything right.

• 3:02Kevin Roose?

  Yeah, it's time for some damn accountability on this show.

• 3:04Casey Newton?

  Yeah. It's about time.

• 3:06Kevin Roose?

  So last year, we broke down our predictions into confidence intervals. We had high confidence, medium confidence, and low confidence predictions. So what was your high confidence prediction for 2024? And how did that pan out?

• 3:19Casey Newton?

  All right, so my first prediction for 2024 was that Threads would overtake X, the former Twitter, in daily active users. And this one is a bit mixed, Kevin, but I feel pretty good about the prediction.

• 3:34Kevin Roose?

  So yeah, let's talk about this because you, you made the case to me that this had actually happened or may have happened. What are the numbers that you're looking at there?

• 3:43Casey Newton?

  Yeah, so this one is hard to determine with total accuracy because X is now a private company. They do not publish audited user numbers, so it's sort of very hard to compare. But what we know is that Threads recently reported that it has 275 million monthly users. Today, as we record this, it is actually the number two app in the App Store in the United States, and it has been at or near the top. Top of that chart for about the past month. And there was some reporting in Business Insider that if you just looked at the US daily active users, people who are using X versus Threads every day in the United States, that Threads actually had overtaken X. Now, again, these are estimates. We cannot say that these are completely true.

• 4:27Casey Newton?

  But I said with pretty high confidence, look, I think Threads is going to have a big 2024. And I do think you have to hand it to me on that one because Threads did have a big 2024.

• 4:36Kevin Roose?

  Yeah, I don't think I have to hand it to you on this one, actually. I am dubious of the numbers that both X and Threads are putting out there. And maybe we should do like a teeny little dive into these metrics that we're talking about. So daily active users, the metric that you predicted Threads would overtake X in last year, is basically the term in the industry for people who log in every day.

• 5:00Casey Newton?

  Yeah.

• 5:00Kevin Roose?

  Monthly active users, as you might expect, is people who log in at least once a month. Now, those could be sessions that are very long. People could be, you know, spending a lot of time on the apps. Or in the case of Threads, which is very tied to Instagram, and often Instagram is trying to sort of kick you over to Threads with these little like links in your feed. It could be someone who like accidentally clicked on a thread, went over to Threads. It counts that as a session and then they go back to Instagram where they meant to be.

• 5:26Casey Newton?

  Yeah, so what you're saying is these numbers seem like they have probably been juiced a little bit. And I'm willing to accept that, but I think there's juicing on both sides. And again, to me, the larger question is like, is X going down and another platform coming up? And I think the answer is basically yes. And like to the extent we have numbers, the numbers are on my side.

• 5:45Kevin Roose?

  Yeah, I would, I would sort of agree with that. I think that it has been a big year for Threads and Blue Sky, another, you know, X competitor. But I think that X is still very relevant. You know, I was thinking about this prediction as we were getting ready to tape the show. And I was trying to think of anything in culture this year that had sort of originated on Threads besides the viral post of you at gay E8 of the San Francisco airport. And Casey, I literally couldn't think of a single news story or trend in culture that originated on Threads. So while I do accept that it is probably growing in part because it's being thrust in front of Instagram users, I don't think it's nearly as relevant as X even today.

• 6:27Casey Newton?

  All right. So what I'm hearing is that you're going to be incredibly hard on me during this recording.

• 6:31Kevin Roose?

  So let's hear your. All I'm saying is, look, I think it's it's important to have clearly resolvable predictions. Because one thing that we did last year when we made these predictions is we also set up prediction markets on them on manifold. And these are play money prediction markets. No one's getting rich or or or losing money as a result of our predictions. But one thing that I have been yelled at by people over there on manifold for for doing is not having clear enough resolution criteria. So I think we should just say for this prediction, we just don't have enough good, reliable data about the popularity of these platforms.

• 7:04Casey Newton?

  All right, fine. What was your high confidence prediction?

• 7:07Kevin Roose?

  So my high confidence prediction from last year was that a lawless L L M, large language model, would get to 10 million daily active users. And I would say for this one, that it has the same problem as your high confidence prediction, which is this is it just that it is very hard to know, given the available data that is out there, which L L M's are getting how many daily active users?

• 7:31Casey Newton?

  Yeah, so I accept that. But to me, the spirit of this prediction was essentially, we may start to see a bit of a migration away from the big mainstream chatbots like a chat G P T and towards something that was a little edgier and less likely to refuse your request. And I'm just not sure we've actually seen that.

• 7:51Kevin Roose?

  No, I don't think we have. I think this one was was more wrong than right. You know, there are some large language models that are less restricted than, say, chat G P T that have become quite popular. Lama 3, the open source A I model from Meta, it's not lawless, but because it's open source, it's not hard to kind of make your own version of it, to fine tune it in a way that makes it much less restricted. And Meta has said that its A I assistant, which is based on Lama, now has nearly 600 million monthly active users.

• 8:26Casey Newton?

  All right. So we're trusting Meta's numbers for Lama, but not for threads.

• 8:29Kevin Roose?

  No, I don't trust them for either. But I would say like Grok, the language model from X is also pretty uncensored, but we also don't have good usage numbers for that either. So I would say in the absence of better data, this one was also a bust.

• 8:44Casey Newton?

  All right. Let's go to our medium confidence predictions, Kevin. My medium confidence prediction was that Google would mostly catch up to OpenAI in the quality of its large language model, neutralizing ChatGPT's lead. This is a bigger trend that quality differences will matter less and less, and distribution will matter more. So this one is a bit of a mixed bag. On the Google mostly catches up to OpenAI and LLM quality front, I think the answer is basically yes, they did. I see you have a note here that if you look at Chatbot Arena, which is where the chatbots compete in various benchmarking challenges, that as we record this today, Gemini is on top.

• 9:29Casey Newton?

  But did it neutralize ChatGPT's lead? On that front, I think it's much more mixed, and I sort of think it was wrong. ChatGPT said recently it has 300 million weekly users. If Google was doing any kind of numbers like that, we definitely would have heard about it by now. So this was actually quite surprising to me that in the end, ChatGPT's lack of a giant distribution channel like the Google search bar, right, didn't actually matter that much. And this was a year where ChatGPT's reputation as like the chatbot that everyone uses just grew and grew.

• 10:03Kevin Roose?

  Yeah, I think that's that part of it is totally right. Google may have caught up to OpenAI in some of the benchmarks that are used to measure these language models. But ChatGPT is still the industry leader when it comes to just how widely referenced it is in the culture. I think a lot of people don't even really know about Gemini or if they've encountered it, it's because, you know, it's been shoved into Google Docs or Gmail or some other Google product that they use. So it does not seem like Gemini has reached the level of mainstream awareness or usage that ChatGPT has.

• 10:34Casey Newton?

  And I wonder how much of that has just been that while it does seem to be performing well on these benchmarks, it also had some really bad launches, right? There was the famous case of search results telling people to put glue on their pizza or eat rocks. There was the case where it would not generate historically accurate-looking founding fathers. It kept spitting out racially diverse founding fathers. So they actually stopped it from generating all images of people for a while. So in that sense, I don't actually think that this is a year where Google caught up. And I'll be quite interested to see whether these new benchmarks that they've been hitting recently in the chatbot arena translate into the product because I have to say as a Gemini, I pay 20 bucks a month to use Gemini.

• 11:18Casey Newton?

  And I think it is the worst of the three that I pay for, for what it's worth.

• 11:21Kevin Roose?

  Yeah, I mean, I definitely use Gemini less than other models. I do like some stuff that's been built with Gemini, like Notebook LM. But in general, I've been pretty disappointed by just the the actual quality of the Gemini model. So I'm curious to see whether model quality ends up being a differentiating factor or whether the models have all kind of gotten good enough for most people to do most of what they want to do. And so it really will come down to like, how is the app designed? What is the distribution strategy, those kind of non-technical factors?

• 11:53Casey Newton?

  All right, what's your medium confidence prediction?

• 11:55Kevin Roose?

  Well, my medium conference prediction last year was that white collar workers would start unionizing to fight AI related job loss. And this one was a total dud. This was, this did not happen at all in any of the industries like finance or law or tech that I thought it might. We did not see substantial union activity related to fears of job loss. Now, we did have the port strike, which was partly about automation and workers fears of being replaced. But it wasn't really about AI and that's not a white collar industry anyway.

• 12:29Casey Newton?

  Well, Kevin, I don't think it was a bad prediction. I just think you might have been a little bit early on that one.

• 12:35Kevin Roose?

  Well, early is just as good as late in this business.

• 12:39Casey Newton?

  You're a really tough customer today.

• 12:42Kevin Roose?

  I hold myself to a high standard as I think we all should. So I really blew that one.

• 12:48Casey Newton?

  You're doing great, buddy. Okay, final prediction from last year. These were our low confidence predictions. So stuff that we thought there was a remote chance might happen. And my low confidence prediction was that the Apple Vision Pro would succeed enough to revive interest in mixed reality and the metaverse. So I'm going to say this. Was mostly wrong, but I have a couple of things I would say in my favor. One is that the information estimated that Apple would wind up selling about 420,000 units of the Vision Pro this year. That is a rounding error when you compare it to something like the iPhone, but it is enough for about 1.5 billion in revenue or just under that.

• 13:31Casey Newton?

  And if you were in any other company and you had a new product launch that generated 1.5 billion dollars of revenue for the first product, you would say, well, we should at least make another one, you know?

• 13:43Kevin Roose?

  Yeah, but I, this is Apple and everything that Apple does gets more promotion and marketing and also, you know, has higher expectations attached to it. So I was also open to the possibility that the Apple Vision Pro would be a huge success. And despite the steep price tag, people, you know, you'd be walking around and just you'd see, you know, tons and tons of people with their Vision Pros strapped to their heads. And like, I barely ever see anyone with one in public anymore.

• 14:10Casey Newton?

  Yeah, it's not a hit, but you know, I do think that interest in mixed reality was revived anyway, and it wasn't because of the Vision Pro, it was because of the Meta Ray-Ban glasses. It got the interest of some of the other big tech platforms, which are now working on glasses of their own that are quite similar. So the metaverse is definitely on hiatus right now, but mixed reality, I do think, is poised to continue kind of creeping into our lives. And I can see a world where it's going to be, you know, maybe a bit unusual to buy a pair of sunglasses that doesn't have some sort of computer inside.

• 14:48Kevin Roose?

  Yeah, I agree with that. All right, my low confidence prediction from last year was that Elon Musk would get his own Hunter Biden laptop scandal on X during the 2024 election cycle. And this one, I gotta say, I really nailed it.

• 15:02Casey Newton?

  Yeah, this was a banger, Kevin.

• 15:05Kevin Roose?

  So what I meant by Hunter Biden laptop scandal, if your memory doesn't extend as far back as 2020, is basically a politically motivated act of censorship taking place on X. Elon Musk and other conservatives have been worked up for years about Twitter's decision back in that election to suppress the reach and block links to a New York Post story about the Hunter Biden laptop because there was a belief that it might have been part of a Russian intelligence operation and a hack and leak. That kind of thing happened again in the 2024 election. There was a document, some called it a dossier, about J.D. Vance that was hacked from the Trump campaign.

• 15:48Kevin Roose?

  It's believed to have been linked to Iranian hackers. And the journalist Ken Klippenstein, who runs a Substack, was banned from X for posting links to this dossier. X said that he had violated the rules about posting unredacted personal information to the platform. But many, many people saw through that and said, this was just because Elon Musk didn't want people reading this thing.

• 16:14Casey Newton?

  Yes, it also goes against everything he said about how he was going to run this platform, which was with complete neutrality, is what he promised. He promised he was going to run it with complete neutrality. And then that was just never true. I mean, the thing that gets me the most about the story is that, yes, you were absolutely right. And I have not heard one peep about it since the day after it happened, right? I still hear people talking about the Hunter Biden laptop story, have not heard one person talking about the J.D. Vance dossier.

• 16:39Kevin Roose?

  Yeah, so that was our roundup of last year's predictions. But we also have some predictions for this year.

• 16:45Casey Newton?

  We do.

• 16:46Kevin Roose?

  For 2025. So Casey, what is your high confidence prediction about technology in the year 2025?

• 16:53Casey Newton?

  OK, now, this is a big one. Are you sitting down?

• 16:55Kevin Roose?

  I am.

• 16:55Casey Newton?

  OK, you know, Apple Computer?

• 16:57Kevin Roose?

  Yeah.

• 16:58Casey Newton?

  I'm predicting, maybe people are going to be mad. I'm predicting that they're going to release the iPhone 17.

• 17:05Kevin Roose?

  I don't buy it.

• 17:06Casey Newton?

  No, I've crunched the numbers. Let me walk you through this. This year, they released the iPhone 16.

• 17:10Kevin Roose?

  Yeah.

• 17:10Casey Newton?

  That really only leaves one option for them for next year.

• 17:12Kevin Roose?

  What did they do the year before?

• 17:14Casey Newton?

  I believe it was the iPhone 15. Yeah. No, OK, here's a second one. I'll give you a second one. You don't like that one? I'll give you a second one, Kevin. I think that this year, the AI culture war is going to begin. What do I mean by that? The last Trump administration, we got a real social media culture war. And the nature of that war had a lot to do with, are these systems biased against conservatives in particular? Are they privileging one set of politics over another? Are the employees woke? I think over the next year, as chatbots sort of enter more and more facets of Americans' lives, we're going to start to see the rumblings of a backlash here.

• 17:57Casey Newton?

  I can imagine there being congressional hearings about the way that ChatGPT responds to certain questions, for example. I can imagine frustrated conservatives using something like Llama to build a right-leaning chatbot that maybe actually starts to get some traction. I can imagine a big national conversation about the fact that so many people are now in these somewhat intimate relationships with chatbots, including both adults and children that we know are doing this. So there's a lot of sort of dry tinder there, and I don't know what the spark is going to be, but I'm telling you, everything is in place for this to have a moment in 2025.

• 18:37Kevin Roose?

  Totally agree. I like this prediction a lot. I think that we are going to have many flare-ups in an AI culture war in 2025. What would you say is the one that you want to use as your resolution criteria here? Like, what would cause you to think, okay, we've had an AI culture war?

• 18:52Casey Newton?

  I would say if there's a congressional hearing about the response that a chatbot gives.

• 18:57Kevin Roose?

  I agree with that. Can't you just picture like a big poster printed out with like a ChatGPT's transcript?

• 19:04Casey Newton?

  Yes.

• 19:04Kevin Roose?

  In the halls of Congress?

• 19:06Casey Newton?

  Yes, absolutely.

• 19:07Kevin Roose?

  And like Jim Jordan yelling about it?

• 19:09Casey Newton?

  Yeah, yeah. Ted Cruz will say like, I asked ChatGPT to criticize me and it did. Explain that, Sam Altman.

• 19:17Kevin Roose?

  Oh, God, it's like, I can picture it now.

• 19:20Casey Newton?

  Right?

• 19:20Kevin Roose?

  Yeah. This is, this is a good high confidence prediction.

• 19:24Casey Newton?

  All right, Kevin, give us one of your own high confidence predictions.

• 19:27Kevin Roose?

  So my high confidence prediction for 2025 is that a newly released crypto meme coin will briefly reach $100 billion in market cap before crashing.

• 19:38Casey Newton?

  Now, is this inspired by the recent success of the Hawk Tua Girls meme coin?

• 19:42Kevin Roose?

  It sure is, Casey. So I was looking at the news recently and I saw that Haley Welch, who is known as the star of the viral Hawk Tua meme.

• 19:54Casey Newton?

  By the way, can you explain that to me? I've always wanted to know what it was about.

• 19:56Kevin Roose?

  Nope. People can go on the internet and look that one up, but I will not be doing the explaining there. But she had a digital crypto meme coin called Hawk that launched in December and briefly hit a market cap of almost $500 million. Again, this did not do anything. There was no like value attached to this thing. It was purely a kind of pump and dump operation. And it crashed within hours, losing more than 95% of its value. But I think that we are headed into a second golden age of speculation, of gambling. The Trump administration is going to be very crypto friendly.

• 20:37Kevin Roose?

  And I think that people are going to take that as a signal to try everything they can to cash in. What do you think?

• 20:43Casey Newton?

  I mean, when it comes to crypto and meme coin market caps, I basically believe anything is possible. I also think, and you sort of highlighted this, but one of the big themes that is unfolding in American life right now is the rise of gambling in more and more places. I think it's quite harmful and destructive. But when the Trump administration comes in, I do think it is going to be all bets are off on this gambling stuff. And so, yes, we're going to see many more speculators. And I do think that that is going to, at least briefly, juice a lot of market caps. So, yeah, good prediction.

• 21:18Kevin Roose?

  Okay, Casey, what is your medium confidence prediction for 2025?

• 21:22Casey Newton?

  Okay, so my medium confidence prediction is that 2025 is the year that Waymo goes mainstream. So this is something you and I have been talking about a fair bit recently. Like, I think I've said to you that, to me, when you step into a Waymo, that might be the first moment that you actually understand how AI is going to transform everything, right? There's something about a car driving itself that will sort of cause things to fall into place for you. And until Now Waymo has been extremely limited. You can use it in San Francisco, you can use it in Phoenix and now LA. But pretty soon you'll be able to use it in Atlanta and Austin. And they just announced that they're coming to Miami as well. And so you're going to see more of these cars in more big urban centers.

• 22:06Casey Newton?

  And I think as that continues, it's going to become a pop culture phenomenon. I think we're going to see memes. There are going to be so many viral clips everywhere. And if you're looking for a resolution criteria, maybe it's that there is a Waymo sketch on SNL. Right? Like to me, that will sort of be a moment where you think, okay, there's something happening here.

• 22:26Kevin Roose?

  Yeah, I'm surprised there hasn't already been a Waymo sketch on SNL. But that goes to my feeling about this, which is that the real mainstream spur for Waymo will be when it goes to New York City. Because most of the media still exists in New York City. And a lot of people, you know, I was just in New York and people there just genuinely do not understand how many Waymos there are on the streets of San Francisco or how how unremarkable it has become to walk around the streets of San Francisco and see just dozens of cars driving themselves.

• 22:55Casey Newton?

  Yeah, I do agree. And that probably is like the number one reason why SNL would not do a sketch about this. But I don't know, I'm just going to say, keep your eyes on this, right? I mean, like, I can imagine Waymo showing up in rap lyrics next year. Just like, it's gonna, it's gonna start to feel like it's a little bit more of the culture in 2025.

• 23:12Kevin Roose?

  I agree with that.

• 23:13Casey Newton?

  All right, give us a medium confidence prediction, Kevin.

• 23:16Kevin Roose?

  My medium confidence prediction for 2025 is that Apple will acquire Snap.

• 23:20Casey Newton?

  Okay.

• 23:21Kevin Roose?

  Now, this is something that people have been talking about for years. Snap is, of course, the company that makes Snapchat. And it has been, I would say, sort of, you know, chugging along for a couple of years now. It's not really growing much. The stock price is down about 23% year to date as of today. They did some layoffs earlier this year. And I would say this is a company that has always had really good product ideas and really creative, you know, use of technology, but that has never really managed to build it into an amazing business. And I think that pressure from From employees could force them to look for a buyer.

• 24:02Kevin Roose?

  I also think that it's going to be much easier to do tech deals and acquisitions during the Trump administration than it was during the Biden administration with Lena Khan at the FTC. And I think Apple and Snap are sort of culturally, they share some DNA, right? Evan Spiegel, the CEO of Snap is a big acolyte of Steve Jobs. I would say they have similar design philosophies. And Apple is also reportedly interested in developing smart glasses to compete with the smart glasses being made by Meta and Snap itself with its spectacles. So I think this would make a lot of sense for both Snap and for Apple, and I would not be surprised to see it happen in 2025.

• 24:43Casey Newton?

  Yeah, I mean, this is one that has made sense for a few years, at least in some ways. Snap has struggled a lot as a standalone company. It's been a while since they had a true big hit project. They do continue to be one of the default modes of communication for American teenagers. And that is an enduring source of strength for them, but it's been pretty hard to build a big business around it. If I'm Apple, my number one question is, do I want to be the default way that a bunch of teenagers communicate? Because it truly introduces so many annoying questions around privacy, security, safety, CSAM, right? All sorts of really tough stuff that all of a sudden Apple is going to have to operate and manage and answer for.

• 25:27Casey Newton?

  So I think there's a reason that Apple sort of likes keeping these social products at arm's length where it can continue patting itself on the back for being, you know, privacy warriors that does nothing but keep everyone safe all day while allowing all of these apps to, you know, roam free in its app store. But all that said, can I see it happening? Sure.

• 25:43Kevin Roose?

  All right, Casey, what is your low confidence prediction?

• 25:46Casey Newton?

  All right, so my low confidence prediction is that X, the former Twitter, will be merged into XAI. So XAI is Elon Musk's AI company. He currently plans to expand his giant supercomputer, which he uses to train Grok, to something like a million GPUs. And already, Kevin, XAI has been valued at $50 billion. You may remember that Twitter, when he acquired it, was only valued at $44 billion.

• 26:14Kevin Roose?

  That's wild to me because XAI does not really have a product yet.

• 26:19Casey Newton?

  No.

• 26:19Kevin Roose?

  So how is it valued at $50 billion?

• 26:21Casey Newton?

  It is a wish and a dream and people look at Tesla's valuation and they think, well, if he can do that for cars, surely he can do that for AI.

• 26:27Kevin Roose?

  And I'm gonna confess my ignorance here, but I did not actually know that XAI was a standalone company. I thought it was part of Tesla.

• 26:33Casey Newton?

  Well, it is a standalone company. Now, I understand your confusion though, because Elon Musk treats all of his companies as if they are all related already. So for example, this year, when he was building this big supercomputer, he had a bunch of GPUs that had been reserved from Nvidia that were supposed to go to Tesla to help Tesla work on self-driving. And Elon Musk said, no, actually Nvidia, just send all those over to XAI, right? And this is the sort of thing that like in normal times and normal circumstances would cause shareholders to revolt and say, Elon Musk, like Tesla and XAI are not the same company. You can't just buy a bunch of GPUs for one company and give them to another company, but it's Elon Musk, so there are no rules. So anyways.

• 27:13Kevin Roose?

  So why would they, why would he merge X into XAI?

• 27:17Casey Newton?

  Because I think the primary value for X going forward is just gonna be to generate training data for XAI. That this is just gonna be the sort of subsidiary that exists to help XAI grow bigger. I think the total value of something that like a truly powerful AI model could provide is just much greater than what diminished social network like X could provide. So might as well just bring them all in-house. Now, why wouldn't he do this? Well, as I said, he already treats his companies like they're related already and he probably just won't see the point in merging the two. But you know, if X continues to decline in some ways and it's feeling like a hassle in some ways, I can see him just saying, you know what? From now on, this is just a subsidiary of XAI.

• 27:57Kevin Roose?

  And what would the actual ramifications of this merger be? Like if he already treats all his companies as if they're one big company, what would be meaningfully different if he did merge X, the social network, into XAI, the AI company?

• 28:10Casey Newton?

  I think that that is the right question and the practical answer might be not very much in the short term. In the long term, though, it would signal to me that he had finally decided to get more serious about the AI stuff and was going to stop wasting quite as much time posting on social networks.

• 28:24Kevin Roose?

  Yeah, or like create an AI agent to do that for him.

• 28:26Casey Newton?

  Yeah, exactly.

• 28:27Kevin Roose?

  All right. My low confidence prediction for 2025 is that at some point during the year, OpenAI will officially declare that they have achieved AGI, or artificial general intelligence. And there are a few reasons I think that this might happen in 2025. For starters, they want to get there first, right? This is a company that is very competitive, that is very motivated by wanting to reach these big milestones in AI ahead of their competitors. And Sam Altman, the CEO, has said basically that they think they are getting quite close to AGI. He said that superintelligence, which is sort of the step beyond AGI, might only be a few thousand days away. So I think that as soon as they have a model, whether

• 29:08Kevin Roose?

  they call it GPT-5, or, you know, O2, or sort of their next generation model, I think they might go ahead and just say, we've done it, we've built AGI. And the benefit of that for OpenAI would be that it would release them from their current deal with Microsoft. Because under the terms of that deal, once OpenAI reaches AGI as defined by its nonprofit board, Microsoft effectively loses access to any of its future models. It doesn't have to share them with Microsoft, which would effectively mean that OpenAI gets out of this deal altogether.

• 29:45Casey Newton?

  And why do you think they want that?

• 29:46Kevin Roose?

  Well, I think that they are eager to, you know, reduce their dependency on Microsoft. There's been some reporting that there's been some tension between Microsoft and OpenAI about things like compute allocation. But the real reason that I think this could happen in 2025 is that OpenAI is undergoing this restructuring process. And there's been some reporting recently in the Financial Times that OpenAI was weighing whether to basically get rid of this clause in this deal with Microsoft that would close off Microsoft's access to its models once it achieves AGI. So this could go a couple of ways. The most likely way that it might go is that Microsoft

• 30:26Kevin Roose?

  wants to strike this clause entirely so that they can keep using OpenAI stuff even after they say it's AGI. But I think the my low confidence prediction is that that will that will sort of blow up somewhere in the negotiation. And then instead, Sam Altman will just come out one day and say, we've done it. We no longer have to give you our models.

• 30:42Casey Newton?

  It's interesting. I mean, you know, so I also read the reporting that said that OpenAI was thinking about getting rid of this clause. And to me, the fact that that's under consideration suggests that OpenAI still needs Microsoft. They need access to Azure and the data centers and everything else. So I'm less inclined to believe that this is going to happen because I think that OpenAI and Microsoft still need each other.

• 31:03Kevin Roose?

  But Microsoft is also developing its own proprietary models. And I just don't know how that sort of works in the long term. If you've got Microsoft that's developing its own models, but also, you know, giving compute and and sort of, you know, data centers to OpenAI. And OpenAI is giving all of its models to Microsoft, which is then using it to improve it. It just feels very messy and like it may explode at some point.

• 31:28Casey Newton?

  Here's how it works. You know, Microsoft also makes its own proprietary web browser called Edge, and no one uses it. So that's how that's going to work.

• 31:37Kevin Roose?

  Okay, all right, fair point. And our standard disclosure, as always, The New York Times has sued OpenAI and Microsoft for copyright infringement. So those are our predictions for 2025. And if you want to kind of play along with these, you can log on to manifold markets. I will go in and create markets for each of these predictions and you can bet with play money on whether you think they will come true or not. When we come back, I resolve

• 32:08Casey Newton?

  to share my New York resolution with you, Kevin. Me too.

• 32:29Speaker 00

  Support for this podcast comes from Dell Technologies. It's time to put AI to work. Is your business ready? The challenge isn't getting started with AI. It's scaling beyond pilots while maintaining control of your data, governance, infrastructure, and costs. Dell AI Factory provides the integrated foundation you need to scale AI securely and efficiently with one trusted partner. Dell, AI on your terms. From idea to impact. Learn more at dell.com/yourwaytoAI. That's dell.com/yourwaytoAI.

• 33:03Speaker 03

  It's time to innovate at the speed of AI without compromising control. That's Outsystems. Take a look at what happened with one global retailer during a period of massive expansion. They used Outsystems to build AI agents to automate invoicing and translation processes, shifting 95% of their daily store operations to a new digital app, saving thousands of hours of manual work. Outsystems gives you the flexibility to deploy how you want and the governance to do it safely. Own your agentic future. Learn more at outsystems.com/NYT. That's outsystems.com/NYT.

• 33:38Speaker 04

  AI spend is growing fast, but most businesses still can't see exactly where that money is going. Ramp gives businesses visibility and control across AI providers with real-time guardrails by team and model, helping ensure spend goes to the highest ROI areas. Stop guessing where your AI spend is going. Get started with Ramp. And if you're new to Ramp, you can get $250 off at ramp.com/hardfork. Go to ramp.com/hardfork. Ramp.com/hardfork.

• 34:12Kevin Roose?

  So Casey, in addition to making predictions at the end of the year, we also do our resolutions for the new year.

• 34:18Casey Newton?

  Because we're always striving to better ourselves.

• 34:20Kevin Roose?

  That's true. So let's just quickly recap our resolutions from last year and see how those went, and then we can make some new ones for 2025.

• 34:29Casey Newton?

  All right, Kevin, remind me what you resolved to do this year.

• 34:32Kevin Roose?

  So my resolution for 2024 was more delight, less fright. Basically, I wanted to stop doom scrolling and warring against my phone and trying to like get it out of my hands as much as possible. And I wanted to make it into a more delightful experience.

• 34:48Casey Newton?

  Yeah, and for context, if you're a newer listener, one time Kevin just put his phone in a box in an effort to stop using it. So that's the kind of person Kevin is. So this has been a sort of podcast-long journey for him. How did it go for you this year?

• 34:59Kevin Roose?

  It went great, honestly. I have much less guilt about my phone use this time this year than I did this time last year. I've got, my phone is now presenting me on my home screen with my delights folder of photos of things that make me happy. And my screen time has stayed about the same. It has not gone way up or way down as a result, but I do just have a much better feeling about my phone. And I think that's good.

• 35:27Casey Newton?

  And did you have to do anything special to make this happen?

• 35:30Kevin Roose?

  No, basically, I mean, I did rearrange my phone. So I put some apps that make me happy in this photos widget on my home screen.

• 35:37Casey Newton?

  I remember you said you had put my face in your delights folder.

• 35:41Kevin Roose?

  It is, actually. I have a photo of you. Yeah. I mean, it's one of like 500, but but every couple of weeks it shows up and I, you know, I quickly scroll away. But yeah, you're a delight. So Casey, what was your resolution last year?

• 35:57Casey Newton?

  So my resolution was when you're watching YouTube, watch YouTube.

• 36:03Kevin Roose?

  And here's what I Don't just like leave it on in the background. That was what you had been doing before.

• 36:07Casey Newton?

  Yes, because, you know, I grew up in this house where like whenever there were like ads on TV, we would always mute the television. Like, you know, we would only turn on the TV when we were watching TV. And I always thought this was the right way to do it. And people that just sort of, you know, let the TV go on all Day, you know, we're, we're doing something wrong. And then I woke up like halfway through last year and I realized that I was doing this with YouTube. I would just be at my desk and I would open one video. I would immediately stop listening to whatever it was, even though it was a video I had chosen to watch. And I would play a video game, I would read a browser tab. And I thought, I am truly just destroying my own attention and this has to stop. So I did pretty good about this. I would give myself like an 85 out of 100. For the most part, I really did.

• 36:52Casey Newton?

  Now, I do think I started to slip a little bit toward the end of the year. I think I was stronger through, let's say, the first three quarters of the year than I was this last quarter. A big thing I did was I stopped playing video games on my laptop, basically. Like I used to have these like very simple games that I would do just to waste a little bit of time. Marvel Snap, when we've talked about a few times, I stopped doing that. So now it's like, if I'm gonna watch a video, I watch a video and I try not to change my attention too much. Now, do I look at my phone while I watch TV? That's a different story and maybe a resolution for another year.

• 37:24Kevin Roose?

  All right, well, I'm glad you got that under control, at least for most of the year. I'm curious if you think that being more intentional about YouTube has made you also more intentional about other things that you do on your phone or your laptop? Like, do you feel like you kind of command your own focus more?

• 37:40Casey Newton?

  I think that this year was pretty good for me in terms of doing more single tasking and less multitasking. Where I feel like I succeeded was in moving away from that place of, I am just going to let technology mindlessly steer me around, right? I think the big exception is anytime you're looking at a feed-based social network, you are letting an algorithm drag you around, but.

• 38:05Kevin Roose?

  TikTok, you mean?

• 38:06Casey Newton?

  For example, or Threads or Blue Sky, which I also spend probably even more time on. But when I'm not doing that, like, I'm relatively locked in, you know? I think my biggest attention-related challenges are, I find it pretty difficult to get through a lot of books. I find it difficult to read like academic research papers. I just feel like the attention leaching out of my brain when I try to do that. Other stuff, I feel okay about.

• 38:31Kevin Roose?

  Yeah, well, that is a good way to segue into our New Year's resolutions for 2025. You know, I enjoy the process of making New Year's resolutions. I don't hold myself to some impossible standard. I'm not one of these people who like, you know, you know, sort of needs to accomplish it, or I feel like a failure. But these are, I would say they're more intentions than resolutions. But did you make any resolutions about your tech use?

• 38:57Casey Newton?

  Yeah, so I have one. So I would like to get medium good at meditation using AI. And here's why. This year, more than others, I struggled with feelings of burnout, which was really surprising and challenging for me because I truly love what I do. I do not want to do less of what I do. But there were moments during the year where I was like, oh, gosh, I feel so tired. And so I thought, I'm going to do what people have been telling me to do for years, which I have just avoided, which was meditate. And when I started to do this, instead of reading a book, I went and I used a chatbot, in this case, Claude from Anthropic. And I said, hey, I want to get started with meditating. What should I do? And it gave me a bunch of instructions. And I went and I tried it.

• 39:42Casey Newton?

  And then I came back and I talked to it again. And I said, hey, I tried that. Here's what I noticed. And then it helped me refine and iterate and say, hey, why don't you try this different? Or you might want to try this different kind of meditation. I really enjoyed that feedback loop. So what I would like to do next year is to continue doing this because while the AI piece of it is interesting and makes it a little bit techie, meditation, of course, is the least techie thing in the entire world.

• 40:04Kevin Roose?

  Right. It's one of the oldest hobbies in existence.

• 40:07Casey Newton?

  Exactly. And like one of the most time-tested methods for just sort of like improving your, your mental health and your wellbeing. So to me, this feels like a good marriage of like a true goal that I have in my life, which is to like manage those feelings of feeling burnt out and give it just enough of a tech twist so that I, as a tech reporter, think, aha, I'm doing something very cool and futuristic.

• 40:25Kevin Roose?

  Now, can I ask you something about your AI meditation practice? So I have also struggled to meditate. I've never really successfully had a consistent meditation practice. And I was very optimistic when ChatGPT's

• 40:41Kevin Roose?

  Advanced voice mode came out that I would be able to have it basically be my meditation teacher. And so instead of just typing to it, I could actually say like, could you lead me on like a 15 minute guided meditation about this thing that I've been stressing out about? But it can't really do it because it doesn't sort of know how to insert all the right pauses. Like it wants to talk to sort of like fill the space. And so it's, it's not actually built in a way that has made it a good meditation partner for me. I know some startups are trying to do more AI meditation coaches, but do you ever use it for that for literally like leading you on a meditation? Or is it just sort of talking to you about an experience that you've had on your own?

• 41:23Casey Newton?

  It's the latter. I think about it as a journal that talks back to you, right? Which is like kind of what being coached in anything feels like, right? You think about your learning an athletic skill, something that I haven't done in years and might, might never do again. But you have a coach who's standing there with you and says, hey, go try this thing. And you do it and you come back and the coach says, next time do it this way. That is like essentially what the, the AI is doing. And because it is this general purpose technology, it can coach you pretty well in a lot of things. And one of the things I like about this is it gets around a common and true criticism of these chatbots, which is that they make a lot of mistakes or they, they hallucinate. All of that is true, but if you just want to like become a novice at meditating, it can handle that.

• 42:07Casey Newton?

  And actually it's like really good at it. And so I think it's important to find those, those chatbot use cases where it's not mission critical. No one's life or career is at stake and yet it can provide you this meaningful help. Because I think that actually is the truest story of AI that is unfolding right now is this expanding set of a positive and helpful and increasingly more powerful things. And if you're not sort of encountering that, I do think you're missing a big part of what's happening in Silicon Valley right now.

• 42:36Kevin Roose?

  Yeah, I agree. And I've been using AI to teach myself stuff this year a lot with pretty good success. I was trying to get really good at poker this year. That was like a hobby that I picked up. And I found that sort of like what you said, like, like is very. Good for for getting you from basically knowing nothing to to sort of having a sort of beginner's understanding of a of a topic if that thing is sort of widely represented on the internet in the training data. But I've found that there's a sort of limit to it, right? Where if I get good enough, if I want like, you know, more advanced strategy advice for poker, it can't actually help me with that. So are you worried that with meditation, you're going to kind of reach the limit of what Claude can do for you?

• 43:19Casey Newton?

  Yeah, well, first of all, it sounds like you should try no limit poker bot. That's a poker joke. But yeah, I absolutely will. And, you know, let me anticipate another criticism that I may get for this suggestion, which is Casey, why don't you read a dang book? That's a good point. I can and should read a book. In fact, you know, my boyfriend recommended me some good meditation books to read. And I and I probably will read them next year, honestly. But the thing about a book is that it can't talk back to you. You cannot ask questions of a book, right? You can do that with an AI. So I love books. I'll continue to read books. But like, this is something different. And it's really engaging. It brings you in because you are having a conversation. And that's just a powerful thing. So will I hit a limit? Yes. But like, that's OK. It's OK to hit those limits. That's just when you go deeper. And you know what to do when you want to go deeper.

• 44:04Kevin Roose?

  Well, I hope that resolution succeeds. I don't like you feeling burned out. We need you strong and kicking for all of 2025.

• 44:11Casey Newton?

  Thank you, Kevin.

• 44:12Kevin Roose?

  My 2025 resolution is to be the poster I wish to see in the world.

• 44:18Casey Newton?

  All right, I'm excited to hear. I'm excited to hear about this because I feel like you have had a somewhat distant relationship with posting this year.

• 44:26Kevin Roose?

  Yeah, so I was once a very active user of many social media platforms. I posted all the time. I was constantly on there, you know, arguing, posting jokes, you know, putting links to my stories and other people's stories up there.

• 44:43Casey Newton?

  And then vaccine misinformation?

• 44:46Kevin Roose?

  Yes, yes, snuff films, disinformation campaigns.

• 44:52Casey Newton?

  And then?

• 44:53Kevin Roose?

  I can't exactly tell when it happened, but maybe a year or two ago, I just kind of ran out of posts. And I felt like, you know what, I can promote our podcast, I can promote stories that I'm working on once in a while, I can go on and, you know, spend a few minutes doing a back and forth with someone. But I just kind of got tired.

• 45:13Casey Newton?

  Yeah.

• 45:14Kevin Roose?

  And I stopped really posting. And, you know, I think there are good reasons for that. I think I'm not the only person who's had this experience. But at some point recently, I began to feel like a hypocrite. Because I spend a lot of time complaining about social media and how all these platforms have their problems, and the people who are active on them are terrible, and, you know, they're spreading all this garbage. And at a certain point, I started to feel like, you know what, it is my job, if I want social media to be better, to roll up my sleeves and get in there and start posting what I want on social media.

• 45:48Casey Newton?

  And what do you want on social media?

• 45:50Kevin Roose?

  So, it's a mix of things. Like, I think part of what I want is just more casual engagement that is not self-promotion. Like, I want to go, I want to promote stuff that I like on the internet. I want to, you know, do the kind of thing that was much more common on Twitter a decade ago, which was just like, here's some interesting stuff that I'm reading. Or, here's a news story that just happened, and maybe a comment that I have about it. Like, that feels almost archaic in this day and age for people to do. But I think that is one of the best ways to use social media, is to tell the people in your network, like, what you are paying attention to.

• 46:22Casey Newton?

  Absolutely.

• 46:23Kevin Roose?

  And give them some sense of what you're thinking about it. So that is something that I have not done in a while, but I am going to get back into it. Not because I think I want my brain to be more hooked into social media, but I just, I feel like I can't complain about it unless I am prepared to do something to fix it.

• 46:38Casey Newton?

  You know, my case for doing this is, I just think it's the fastest way to get your your finger on the pulse of the conversation. Like, how are people understanding certain subjects? What are, like, the sort of third rails that no one ever touches? And what are the things that people can't stop talking about? The only way to really get a handle on that is to get in there and post. And, um, you know, I mean, it can be hard. I've been, obviously, canceled twice on Blue Sky this year, once in Portuguese, and it takes a toll. But- I think there's a way to do it that's really enjoyable.

• 47:09Kevin Roose?

  Yeah, and a way to do that that takes for granted the fact that if you're an active poster on social media, like, people are going to get mad at you.

• 47:16Casey Newton?

  Yeah.

• 47:16Kevin Roose?

  Like, that is going to happen. You're gonna post something, it's gonna be a little out of pocket, or a little risque, or people are just gonna take it the wrong way. And so, part of what I'm trying to do, as part of this resolution, is just kind of prepare myself for the inevitability that something I do online is gonna piss people off, and that when that happens, I just have to sort of greet it with poise, and with understanding, and try to do better the next time.

• 47:40Casey Newton?

  Yeah, comes with the territory.

• 47:42Kevin Roose?

  If I ever told you what I think about, um, cancel culture, no, what's something that would get me canceled?

• 47:48Casey Newton?

  Like, the top, like, the top 10 dogs that your followers own that you think should be given away.

• 47:56Casey Newton?

  Because you don't think they're good pet owners.

• 47:59Kevin Roose?

  Yeah, I didn't say I wanted to be the shit poster that I wanted to see in the world, but I do think that there is a, a kind of defensiveness to the way that I and a lot of other people act on social media these days, because we've seen so many people just blow up their lives and careers by posting in an unhinged manner, that we've sort of retreated into this very comfortable, kind of boring, use of social media. And so, I'm gonna spice it up a little bit.

• 48:23Casey Newton?

  Yeah, the thing that I like about social media is that it just lets you show up and say, "Well, this is interesting," which I actually think is most of a journalist job.

• 48:30Kevin Roose?

  Yeah.

• 48:30Casey Newton?

  Like, "Well, this is interesting."

• 48:31Kevin Roose?

  But I do want to get your opinion on, on this, as I set out in this resolution, which is that what I don't want to have happen is to rot my brain by spending too much time on social media, and by over-indexing on what is happening on social media, because you and I have both seen plenty of examples, including some people in our own industry, of people who just spend way too much time on social media, who start to think in posts, whose every reference point becomes some meme, or some controversy on social media, and who kind of lose contact with reality. So, as I'm going into this "be the poster I wish to see in the world" year, how do I keep that from happening to me?

• 49:07Casey Newton?

  I think, you know, if you find yourself posting more than three or four times a day, check in with me, okay?

• 49:13Kevin Roose?

  Okay, something bad might happen, like, I might end up running SpaceX and Tesla.

• 49:20Casey Newton?

  You might wind up being the White House crypto czar.

• 49:24Casey Newton?

  Yeah, a lot of bad things, yeah. But no, I think it's you want to, and honestly, it's not really the post. Some people do fall into like posting too much territory, acting like that's quite rare. I think the more common thing is you just sort of can't stop looking at the feed. And that's something that only you can really decide for yourself. But yeah, try to get a sense of like, when you know that you've had enough. When do you have a sense of what the conversation that maybe that's like the heuristic is if you feel like you have a sense of like the contours of the conversation that day, whatever it might be, great, now you can move on.

• 49:51Kevin Roose?

  Great. And if I do slip into a social media brain rot, I want you to tell me.

• 49:56Casey Newton?

  Absolutely will. All right.

• 49:58Kevin Roose?

  Those are our resolutions. God help us all.

• 50:01Casey Newton?

  That's great. Based on this, we're going to be better people next year.

• 50:05Kevin Roose?

  I think so.

• 50:09Kevin Roose?

  When we come back, we'll answer some listener questions like, why is Casey so annoying?

• 50:13Casey Newton?

  Hey, I'll ask the questions around here.

• 50:30Speaker 00

  Support for this podcast comes from Dell Technologies. It's time to put AI to work. Is your business ready? The challenge isn't getting started with AI. It's scaling beyond pilots while maintaining control of your data, governance, infrastructure, and costs. Dell AI Factory provides the integrated foundation you need to scale AI securely and efficiently with one trusted partner. Dell, AI on your terms. From idea to impact. Learn more at dell.com/yourwaytoAI. That's dell.com/yourwaytoAI.

• 51:05Speaker 03

  It's time to innovate at the speed of AI without compromising control. That's OutSystems. Take a look at what happened with one global retailer during a period of massive expansion. They used OutSystems to build AI agents to automate invoicing and translation processes, shifting 95% of their daily store operations to a new digital app, saving thousands of hours of manual work. OutSystems gives you the flexibility to deploy how you want and the governance to do it safely. Own your agentic future. Learn more at outsystems.com/nyt. That's outsystems.com/nyt.

• 51:39Speaker 04

  AI spend is growing fast, but most businesses still can't see exactly where that money is going. Ramp gives businesses visibility and control across AI providers with real-time guardrails by team and model, helping ensure spend goes to the highest ROI areas. Stop guessing where your AI spend is going. Get started with Ramp. And if you're new to Ramp, you can get $250 off at ramp.com/hardfork. Go to ramp.com/hardfork. Ramp.com/hardfork.

• 52:13Kevin Roose?

  Well, Casey, we have one more thing to do on our very special beginning of the year episode.

• 52:18Casey Newton?

  This is where we commit a ritual sacrifice to the gods to protect us in the year to come.

• 52:22Kevin Roose?

  Yes, and we should also answer some questions from our listeners.

• 52:26Casey Newton?

  Yes, and you know, we always love hearing from our listeners. They send us so many good thoughts and questions every single week. And so what better way to kick off a fresh year of Hard Fork than by finding out what's on their minds?

• 52:38Kevin Roose?

  Yes, so we want to do something special, never before done today, which is to bring in one of our producers, Whitney Jones, to help us sort through all of our reader mail. So Whitney, welcome to Hard Fork.

• 52:51Speaker 05

  Hey, welcome to Hard Fork.

• 52:53Kevin Roose?

  We're breaking the fourth wall here. So tell us what our listeners have asked and what we should respond to.

• 53:00Speaker 05

  Yeah, I feel like I should have a giant mailbag, but actually I just copy and pasted these all into a document. They sort of fall into different categories, and so I want to take them sort of in categories. And the first one is just responses to segments that we've done on the show. There was one recently after the poi market elections betting segment that we did at the beginning of November. Listener Anne Lachey wanted to know more about VPNs that you guys mentioned. You mentioned that Americans were using these VPNs to get around restrictions on poi market to bet illegally on the election on the site. So Anne wanted to know, she writes, "How many people are using VPNs? Is it mostly for downloading movies, music, media without having to pay? Is it for gambling, as mentioned? Is it for political disturbances? Is it for hijacking Wi-Fi?

• 53:45Speaker 05

  I don't know," she writes. "Our company is concerned about it. What is the future for VPNs? Is anyone cracking down on it? What happens if you get caught using one?"

• 53:54Kevin Roose?

  That is so many questions about VPNs.

• 53:56Casey Newton?

  Yeah, I want to be clear. In the future, you're limited to one question.

• 54:00Kevin Roose?

  No, we will try our best to answer this one because I think it is a good one. Casey, what do you know about VPNs and how common they are and what people use them for?

• 54:08Casey Newton?

  So, virtual private networks have been around a long time and they are a pretty big market. I found one estimate that said that the market for them is well in excess of $40 billion. And I can just say anecdotally that when I go on YouTube and I'm watching videos, one of the most popular ads that gets inserted into creator content is ads for VPNs. So, to answer one of these questions, yes, I do think the primary reason that people use VPNs is to get around geographical restrictions on what kind of media they can consume.

• 54:43Kevin Roose?

  Yeah, and just if people have not used VPNs before, what they are is basically a kind of means of making it look like your traffic is coming from somewhere else, right? So, you're basically renting a server located someplace else and it sort of sends your traffic through that server to the website or the service that you're going to. So, if you're on a streaming site and you want to make it look like you are in London, but you're in California, you can use a VPN to accomplish that.

• 55:10Casey Newton?

  Yeah, now, it is the case that there is a political dimension to these. And often, we see in authoritarian countries, VPNs become quite popular. In fact, once the war in Ukraine started and Russia became a sort of a country non- grata on the global stage, various apps pulled out of Russia and the number of VPN downloads in Russia went through the roof because everybody wanted to use the internet that they were used to without all of the new geographic restrictions that had been placed on them. So, that's why I do think they can be a really important technology to sort of help people in those countries.

• 55:43Kevin Roose?

  Yeah, and I, like, we should say also that VPNs, there are many, many legitimate uses for them. I use one when I'm on a public Wi-Fi network. It makes it harder to sort of intercept your traffic. Corporations use them for employees to sort of keep their networks more secure if you're giving people remote access. So, lots of people use VPNs for lots of... Of totally normal and probably pretty privacy conscious reasons.

• 56:08Casey Newton?

  Yes, and if you want to know, can you use a VPN to hijack your neighbor's wifi? I don't really think you can do that. I mean, I get if you can hijack your neighbor's wifi, you don't need a VPN to do that, really.

• 56:18Kevin Roose?

  Right, but they are useful in some cases for things that are not listed. So if you want to gamble online in a jurisdiction where that is not allowed, you can sort of route your traffic through a place where it is allowed and do it that way. Although if you win a lot of money, you may not have an easy time cashing that out. And a bunch of gambling sites do actually block VPN traffic. There are ways to sort of figure out which traffic is coming from VPNs and block those.

• 56:47Casey Newton?

  And that's all we know about VPNs. Thank you, Aaron. We hope you're happy now.

• 56:53Kevin Roose?

  Have fun gambling from Lithuania or wherever you're gonna do it from.

• 56:58Speaker 05

  All right. Another second we got a lot of email about was the interview you did with Stephen Johnson about Notebook LM. And Bob Flint wrote recently asking, how do I know that you guys aren't bots? Thanks to you, I'm now familiar with Notebook LM. How can I know you're not using similar technology to produce your podcast, albeit with a certain degree of human intervention?

• 57:19Casey Newton?

  Well, let me ask you a question, Bob. How do I know that you're not a bot? How do I know that you didn't use ChatGPT to write that whole thing? To complete this game, bud.

• 57:28Kevin Roose?

  But Casey, this is a I like this question because there are actually, you know, companies that are building AI podcasting tools that will allow you to like clone your own voice. And we actually do have sort of a laborious process here where if we mess something up when we're originally taping the show, we will go back and record a little insert. And some eagle eared listeners have sometimes picked up on these. We try to make them sound as smooth as possible. But, you know, that kind of thing could be easier to do if we just had AI clones of our voices and our producers could just kind of change what we say. But we don't do that now, do we?

• 58:06Casey Newton?

  No, we don't. Now, you know, is it true that due to the existence of these technologies, it now does just get harder? To tell which of the media that you're consuming is synthetic in some way. Yeah, it does mean that. That's one of the big concerns I have about the rise of AI. And we have to keep our eyes on it. In the meantime, all I can do is tell you something that GPT-2 would never tell you. And that's that two plus two equals four. So hopefully that gives you some confidence.

• 58:34Kevin Roose?

  Yeah.

• 58:35Casey Newton?

  Wait, I'm being told that we're now at GPT-4, so I'm actually not sure.

• 58:39Kevin Roose?

  Yeah, we should do like a captcha at the beginning of every episode to just prove that we are humans. No, this is interesting, actually, and I'm glad you wrote in with this, Bob, even though I think this question was mostly a joke. Because a thing that I am starting to hear is that people are sort of getting suspicious of podcasts that sound too much like the Notebook LM podcasts. Have you heard about this?

• 59:00Casey Newton?

  I haven't heard about this.

• 59:01Kevin Roose?

  So I was recently talking with a friend and I was asking them that they were talking about a podcast they recently started listening to. And they were like, I think this might be a Notebook LM thing. Like, it sounds very similar. And it wasn't like I, I, I know the people who make this podcast, but because of the way that these podcasts sound with the kind of back and forths and the, the disfluencies in them, um, there is starting to be kind of this question about like, is this a real person?

• 59:25Casey Newton?

  Yeah, I mean, maybe the only answer is that like all podcasts will have to move to three or more people so that it no longer sounds like the two people podcast.

• 59:33Kevin Roose?

  Oh, you don't think Notebook LM is working on that?

• 59:35Casey Newton?

  Oh, I hope not. Well, the whole team just quit, so hopefully that'll help.

• 59:39Kevin Roose?

  That's true. Okay, what's next, Whitney?

• 59:42Speaker 05

  Um, another category of questions that we get a lot are just general questions about different, um, technologies. Like, why is my chatbot behaving in this particular way? I've got one from a listener, Raphael Holmes. Uh, had a question about AI image generators that I wanted to run by you guys. He says, uh, hi Kevin and Casey. I was trying to show my octogenarian dad the wonders of generative AI. And his request was to try to draw a loon in a bathtub.

• 1:00:04Casey Newton?

  And it says a loon in the bathtub and it drew Kevin.

• 1:00:06Kevin Roose?

  Hey!

• 1:00:08Casey Newton?

  No, I'm sorry. Go ahead, Whitney.

• 1:00:09Speaker 05

  It says, as it turns out, Dolly has no problem putting a terrifying five foot loon in a bathtub. See attached images, one of many, but it can't get the correct proportion to one another. Even with tons of wrangling, still only monster loons. My sister-in-law tried Gemini and got very similar failures. Would you try it with a few image generation tools? The world needs to know. And so I have a bunch of images here for you guys, if you want to have a look at these. So these are the ones that the listener Raphael sent in to us. This is a giant loon in a bathtub. There's this one, which you can see the prompt says, "A tiny loon that appears almost invisible in a huge bathtub." And this is the image that comes back with.

• 1:00:51Kevin Roose?

  And like, for reference, like, how big is an actual loon?

• 1:00:55Speaker 05

  Well-

• 1:00:55Kevin Roose?

  I'm not a big loon guy.

• 1:00:58Speaker 05

  Loons appear to be a bit bigger than a duck.

• 1:01:00Kevin Roose?

  Okay, but in these images, they're like, filling up the whole bathtub. Like, these are giant loons.

• 1:01:05Speaker 05

  It's either a very big loon, or a very tiny tub.

• 1:01:09Kevin Roose?

  Yeah, so the images you're showing us, Whitney, are these AI-generated images, like the ones that our listener described, of just these, like, monstrous loons that are taking up most of the bathtub.

• 1:01:21Speaker 05

  Correct.

• 1:01:21Casey Newton?

  Okay.

• 1:01:22Speaker 05

  I went to Meta AI, Meta AI did the same thing.

• 1:01:25Kevin Roose?

  So, so do we know why this happens?

• 1:01:27Casey Newton?

  Yes, we do know why. Here's why. When you use a text-to-image generator, it's trying to find the statistical average that satisfies your prompts, right? It's kind of like a sculptor. It's trying to take away everything from the image that is not a loon in a bathtub, and get to the, like, median image that it can, you know, sort of conceive of.

• 1:01:49Casey Newton?

  And the thing is, loon in a bathtub is probably not a very high-volume request. I don't think, a lot of painters out there have painted a lot of normal-sized loons in normal-sized bathtubs. And so, this is just a classic case of asking a model to do something that it is not well-suited to do. You know, when you are using a text-to-image generator, you are throwing a wish into a fountain. And sometimes the wish is granted, many times the wish is not. It is not your problem, you're doing everything right, but the model cannot do this yet. Maybe someday it will, but it can't yet.

• 1:02:19Kevin Roose?

  I think it will be able to do it soon. And I put this question to Claude about, like, why these image generators are having trouble doing a loon in a bathtub. And it drew me a, like, diagram of a loon in a bathtub. It actually coded, like, a diagram of the common loon at 26 to 28 inches, and the standard bathtub at about 60 inches. So, this kind of thing is possible, but, you know, I think that the reason that this is having trouble with current models is because of these- This thing's called contextual scaling issues. Basically, if two or more objects do not frequently appear together in the training data of whatever this system was trained on, it may not sort of understand the proportions of one and the other.

• 1:03:03Kevin Roose?

  And so often, people will notice that it just, these image generators, they have a hard time with proportions, with sizes, and that will probably improve somewhat in future models. But right now, I would not use it for anything mission critical when it comes to loons and bathtubs or any other sort of like juxtapositions.

• 1:03:20Casey Newton?

  Also, this is a great time to just learn how to draw a loon in a bathtub. You have the power, listener. All right, what's next?

• 1:03:27Speaker 05

  The next one says, a couple of different listeners asked about how, what's hot and what's not when it comes to data sources for training AI right now. One listener, Asa Strong, says, do they want satellite data? Ben Stone says, are they using data from home assistants and security cameras? So any new news on training data and what's in demand right now?

• 1:03:47Casey Newton?

  So, I mean, in short, they want all the data that they can handle. Do we know what data is hot? No, not really. And the reason is because they don't tell us what data they put into the models anymore. They used to, but then they stopped. Let's just say some certain newspapers got a little bit irritated about what they were reading about the training data that was going into those models. And so now, there's a lot less transparency. I wish there were. It would really help us understand these models. If I told you, if I could tell you one kind of data that is becoming increasingly popular in this world, though, it is video data. There is a lot of thinking among AI researchers that the final frontier of developing models that have something approaching human or even superhuman understanding is the kind of knowledge in the world that you only get by moving through the world.

• 1:04:32Casey Newton?

  And so they're starting to adjust a lot more video to understand motion and depth and everything else, reasoning, everything else that you can learn by just sort of fixing cameras on the world, running them through models, and trying to understand what's happening there.

• 1:04:46Kevin Roose?

  Yeah, I do know a little bit about the training data that's in demand right now because unlike Casey, I've done reporting on this.

• 1:04:52Casey Newton?

  Oh, let me guess, there was a huge demand for Kevin Roos columns. They couldn't get enough of them over there.

• 1:04:59Kevin Roose?

  No, but I did talk to someone who is working on a project where they're basically going into university libraries and archives and digitizing a bunch of stuff there that has not been previously digitized. Because the sort of low hanging fruit, the stuff that's online, the stuff that's, you know, in these sort of repositories that are widely used, that that stuff is, you know, good, but has already been used. And so now they're looking for new sources, these AI companies. And a lot of what they're finding is that there's just a lot of stuff that hasn't been digitized. And if you can go in and go into a library and just put everything on there online, you will maybe improve the resulting models.

• 1:05:37Casey Newton?

  That's great. Maybe you should write a story about that because I think it's time you saw the inside of a library.

• 1:05:43Casey Newton?

  What else do we got?

• 1:05:45Speaker 05

  Mitzi had a question about security and voice cloning technology. She writes, "Both of my investment institutions use voice verification to ID customers on the phone. The password is literally the voice saying, 'My voice is my password.' In this era of AI, it seems foolhardy. Am I being paranoid?"

• 1:06:02Casey Newton?

  No, you're not being paranoid. Tell your bank to knock that off. Go to a new method of authentication immediately.

• 1:06:08Kevin Roose?

  Yeah, this is a known issue. There was a story in 2023 in Vice by Joseph Cox about how he broke into a bank account with an AI-generated voice. These voice verification systems, they are not secure and it is very easy to clone someone's voice using just a small snippet of audio from that person. So yeah, I would move as quickly as I could away from using your voice for verification.

• 1:06:36Speaker 05

  Cool. Do you guys want to close out with one ethical hard question?

• 1:06:39Casey Newton?

  Sure. Yes.

• 1:06:41Speaker 05

  Let's do it. Dylan writes, "Here's my ethical dilemma. Someone I was recently dating, but now no longer, left their HBO Max account logged into my computer after a movie date. I only realized it was still logged in after we had stopped seeing each other. I was going to log out of their account, but then ended up binge watching House of Dragon, which I had been dying to see. Was that wrong? And if not, could I watch The Last of Us next?"

• 1:07:03Casey Newton?

  Well, I mean, it was wrong to watch the second season of House of Dragon if you'd watched the first season because it was really bad. And I thought, I really thought it should have been canceled. But as far as, you know, the ethics of using a logged in HBO Max account, I say go nuts.

• 1:07:16Kevin Roose?

  So I have feelings about this because I've been on both sides of this this year. Because I, well, not not in the dating.

• 1:07:23Casey Newton?

  Did you break up with Dylan?

• 1:07:25Kevin Roose?

  No, but I, I, I left my Amazon Prime video logged in at an Airbnb, like several years ago and only discovered it this year because I was getting, I like was checking my credit card statements and I found like a BritBox subscription and like other stuff that I had just not, like movie rentals that I just not subscribed to. So someone had been purchasing things on my Amazon Prime video account at this Airbnb that I had stayed in several years ago. So I would say the ethics do not extend to purchasing, but I would say if you're just watching the stuff that is like included for free, I would say that's kosher with one exception.

• 1:08:08Casey Newton?

  What's that?

• 1:08:09Kevin Roose?

  Which is that if you are watching a show on a pilfered streaming account, that the owner of that account is also watching, you may not skip episodes, you may not, you may not watch that show until the owner of the account has also watched that show because you know what happens. And this has happened to me and I've, I've actually accidentally done this to people. You're borrowing their account, you're using their account and you start watching, you know, a popular show that just came out, like the show about chimps on Netflix or something on,

• 1:08:38Casey Newton?

  Yeah, we love the chimp show.

• 1:08:40Kevin Roose?

  And they are watching it at the same time. And by watching these things on the same account at the same time, you were actually screwing up all of their timestamps. When they go into resume watching, it's going to take them to a totally different episode of the show. So don't do that. Just keep your, your hygiene consistent when you're sharing these accounts.

• 1:08:57Casey Newton?

  I think that's good advice. And you know, I think it's really brave of you after having recently made fun of me for having wine on tap at my house to have just admitted that you ignored thousands of dollars of purchases of videos over the years.

• 1:09:12Kevin Roose?

  It was not thousands of dollars.

• 1:09:13Casey Newton?

  You literally don't even notice when people are renting thousands of dollars worth of movies from your Amazon Prime account. Wow. Must be nice, Ruze. Must be nice.

• 1:09:23Kevin Roose?

  And if you are the person who bought BritBox and rented movies on my Amazon Prime video account at this Airbnb, I will track you down. This is not over.

• 1:09:34Speaker 05

  Can I ask a follow-up question?

• 1:09:36Kevin Roose?

  Yes.

• 1:09:37Speaker 05

  Do you think the ethics of borrowing logins changes if you're no longer in a relationship with somebody?

• 1:09:44Casey Newton?

  No, I don't. I don't. Who is the victim here? Who is being harmed?

• 1:09:48Kevin Roose?

  Yeah, it's the victimless crime. And in fact, I've heard of people actually sort of continuing to voluntarily split accounts with exes after they break up. So, you know, you might even not need to hide it.

• 1:10:00Casey Newton?

  And here's what else I would say. As long as there is a single logged in HBO Max account between the two of you, there's a chance you could get back together. There's a fiber of something there that could turn into something actually really special.

• 1:10:12Kevin Roose?

  Ooh, that's, I hadn't considered that, but it might give you some hints. You know, if you know they're watching, you know, House of the Dragon, you know, you might just spark up a conversation.

• 1:10:22Casey Newton?

  Right. Like, imagine you've broken up with someone and then you like go back into your HBO Max and they're halfway through a movie and the name of the movie is "I Really Miss My Ex." All of a sudden, the wheels start turning. Hmm, maybe I should text that person. Maybe there was something there.

• 1:10:36Kevin Roose?

  Yeah.

• 1:10:37Casey Newton?

  So, uh, yeah, keep watching it. Just, but last, the Last of Us, that's a, that's a hard watch. I'll say it. That's a hard watch.

• 1:10:44Kevin Roose?

  I couldn't do it. It was too dark.

• 1:10:46Casey Newton?

  Yeah. Did I tell you about my idea for a sequel to The Last of Us?

• 1:10:48Kevin Roose?

  No.

• 1:10:48Casey Newton?

  It's called The Second to Last of Us.

• 1:10:52Casey Newton?

  Anyways.

• 1:10:56Kevin Roose?

  Well, on that note, Whitney, uh, thank you so much. And we should also just say it's delightful to have a producer on the show.

• 1:11:04Casey Newton?

  It is.

• 1:11:04Kevin Roose?

  Our, our team, Whitney, Rachel, Jen, Caitlin, you know, Chris, Ryan. Everyone works so freaking hard all year to make this show and we are just so, so appreciative. So, thank you, Whitney. And, um, yeah, don't let this be the last time.

• 1:11:23Casey Newton?

  Yeah, come back, come back anytime.

• 1:11:24Speaker 05

  It was fun to be on.

• 1:11:36Speaker 00

  Support for this podcast comes from Dell Technologies. It's time to put AI to work. Is your business ready? The challenge isn't getting started with AI. It's scaling beyond pilots while maintaining control of your data, governance, infrastructure, and costs. Dell AI Factory provides the integrated foundation you need to scale AI securely and efficiently with one trusted partner. Dell, AI on your terms. From idea to impact. Learn more at dell.com/yourwaytoAI. That's dell.com/yourwaytoAI.

• 1:12:10Speaker 03

  Companies are pouring money into AI, but few leaders can explain how those AI investments work with the systems you already have. You know, the ones that actually run your business? Without Systems, you don't have to start over with AI. You can choose the projects where AI will have the most impact. And OutSystems gives you the platform to make it agentic on your terms with complete control every step of the way. So you can trust AI with the mission-critical systems that run your business. Own your agentic future. Learn more at outsystems.com/NYT. That's outsystems.com/NYT.

• 1:12:45Speaker 04

  AI spend is growing fast, but most businesses still can't see exactly where that money is going. Ramp gives businesses visibility and control across AI providers with real-time guardrails by team and model, helping ensure spend goes to the highest ROI areas. Stop guessing where your AI spend is going. Get started with Ramp. And if you're new to Ramp, you can get $250 off at ramp.com/hardfork. Go to ramp.com/hardfork. Ramp.com/hardfork.

• 1:13:20Kevin Roose?

  Hard Fork is produced by Whitney Jones and Rachel Cohn. We're edited by Jen Poyant. We're fact-checked by Caitlin Love. Today's show was engineered by Alyssa Moxley. Original music by Alicia Baty-Tup, Marian Lozano, Rowan Nemisto, and Dan Powell. Our audience editor is Nell Gallokey. Video production by Ryan Manning and Chris Schott. You can watch this full Episode on YouTube at youtube.com/hardfork. Special thanks to Paula Schuman, Quiwing Tam, Dahlia Hadad, and Jeffrey Miranda. You can email us, as always, at [email protected].

• 1:14:16Speaker 04

  AI isn't just opening new business opportunities. It's changing the way businesses get work done, the questions they ask, the decisions they make, and the routine tasks they can hand off. And nobody understands that better than NetSuite. That's why they put together a free expert guide you need to grab today. Aligning for the agentic era: How AI is Changing Everyday Work. Download it and you'll see exactly how AI is reshaping work across every part of your organization and how to make sure you come out ahead. Get your free guide at netsuite.com/NYT. netsuite.com/NYT.