Skip to content

Podcast

Charlie Kirk and Online Rage + Inside Trump’s Chip Flip + This Week in A.I.

Hard Fork

Source ↗ ← All highlights
  • Borderline Content Drives Political Influence
    • Charlie Kirk built influence by producing borderline, engagement-optimized content that flirts with platform rules.
    • That tactic amplifies rage and polarizes political discourse across platforms. Transcript: Kevin Roose Well, Casey, we’re going to start the show on a more somber note than usual. Obviously, many of you, like us, have been reading about the assassination of Charlie Kirk, which happened while we were taping last week’s episode. So we couldn’t really get into it then. We also just didn’t know a lot. In the week or so since then, we have learned a lot more, and we’ve had a lot more time to think and reflect and try to figure out what, if anything, we could add to the conversation. So, Casey, we’re going to talk a little bit about what we’ve been talking about amongst ourselves, what we’re seeing, what we’re thinking about, and some of the connections between These Internet platforms that we talk about on the show and these violent events out in the physical world. And what, if anything, can be done about that? So Casey, how are you doing? How are you feeling? What has your sort of reaction been to all of these events over the past week? Casey Newton Well, I think like a lot of people, I was horrified by the event itself and have continued to be horrified as further events have unfolded. I was not a close student of Charlie Kirk’s work, but I know that he was a master of manipulating the platforms (Time 0:02:36)
  • Rage Optimization Corrodes Public Debate
    • Borderline content optimization harms civic discourse by normalizing outrage as a distribution strategy.
    • As more creators adopt it, the public square becomes dominated by material designed to upset viewers. Transcript: Casey Newton Now kind of the way that a wide swath of our politics works. Yeah. Kevin Roose Yeah. I think that’s a really good point. I remember this graph that Mark Zuckerberg once posted on his Facebook account. This was a much earlier version of Mark Zuckerberg who, you know, had not gone sort of, you know, full MAGA yet. But he was sort of sharing some of his observations about the kind of content that worked on Facebook. And maybe we can, you know, pull that graph up and put it in the show. But it basically showed that the closer you get to the line, no matter where the line is drawn, this was the interesting thing. It was like, no matter where a platform decides to draw the line and say, this content is unacceptable, the closer you get to that line, the more engagement you get. And he was sharing this not as sort of a celebration of this phenomenon. He was just kind of like, look, this is what we observe happening on our platforms. And it didn’t seem like he knew what to do about it then. And it doesn’t seem like anyone knows what to do about it now. Casey Newton Well, what platforms said in both, you know, Facebook at the time and YouTube said was that they were going to downrank this kind of thing. They said, hey, we see you edgelords out there. We’re not going to let you get away with this. So we’re going to try to restrict the spread of this borderline content. But as we look at the media landscape today, Kevin, it’s a lot of stuff that is just right on the border, right? It’s stuff that you kind of can’t believe that people are getting away with. So people are sort of always finding these new borders and they’re trying to exploit them. And they’re not always political. I’m thinking about, you remember the milk crate challenge on TikTok? Yes. People are stacking up milk crates and then trying to like walk across them, even though it’s an incredibly rickety and dangerous structure. You know, at the time, TikTok didn’t have a policy for that. But then TikTok watched a bunch of kids falling off milk crates, and they had to say, we’re not going to promote this sort of thing anymore. So this is just the nature of platforms that are looking for the most engaging thing for you to look at. (Time 0:06:36)
  • Engagement Incentives Push Edginess
    • Platforms reward content that approaches but doesn’t cross policy lines because it maximizes engagement.
    • That incentive structure pushes creators to produce increasingly provocative material to gain reach. Transcript: Kevin Roose Yeah. I think that’s a really good point. I remember this graph that Mark Zuckerberg once posted on his Facebook account. This was a much earlier version of Mark Zuckerberg who, you know, had not gone sort of, you know, full MAGA yet. But he was sort of sharing some of his observations about the kind of content that worked on Facebook. And maybe we can, you know, pull that graph up and put it in the show. But it basically showed that the closer you get to the line, no matter where the line is drawn, this was the interesting thing. It was like, no matter where a platform decides to draw the line and say, this content is unacceptable, the closer you get to that line, the more engagement you get. And he was sharing this not as sort of a celebration of this phenomenon. He was just kind of like, look, this is what we observe happening on our platforms. And it didn’t seem like he knew what to do about it then. And it doesn’t seem like anyone knows what to do about it now. Casey Newton Well, what platforms said in both, you know, Facebook at the time and YouTube said was that they were going to downrank this kind of thing. They said, hey, we see you edgelords out there. (Time 0:06:42)
  • Leadership Shapes Platform Tone
    • X under Elon Musk amplified violent and revenge narratives by lifting moderation and broadcasting inflammatory posts.
    • That elevated perception of imminent civil conflict among platform elites. Transcript: Casey Newton Yes, and I think it is particularly true on X. And I think it might be worth drawing a distinction between how X handled this situation and how the Twitter of old may have handled a similar kind of murder. In the old days of Twitter, when there was a very violent video that was spreading, you would have teams that would either try to take it down or to ensure that it was hidden behind a screen. And they would probably downrank it, try to ensure that it was not being recommended to people. Twitter was never perfect at that sort of thing. But at the Twitter of old, there was an effort to try to prevent people who didn’t want to see this content from seeing it. Of course, Twitter also banned some of the most aggressive and violent right-wing accounts, right? People who had posted really racist and inflammatory stuff, they all got banned. Then Elon Musk buys it, turns it into X. And what does he do? Well, first of all, he brings back all those right-wing accounts, lets them post almost whatever they want. And then, crucially, he decides to use the platform to push his own very inflammatory views, right? In the immediate aftermath of this, before we know anything about the shooter, he tweets, the left is the party of murder. And so today, if you look at X, it is easy to get the sense that we are on the brink of civil war and the right wing is gearing (Time 0:12:21)
  • Use Bridging Algorithms To Reduce Polarization
    • Try deploying bridging-based algorithms that surface community notes accepted across political divides.
    • Use cross-spectrum consensus signals to reduce polarization in recommendation feeds. Transcript: Casey Newton It would not be a complete solution. It would just be something I’d like to see more platforms try. And it’s called bridging based algorithms. We talked about them on the show before. This is an idea from a guy named Aviv Ovaja, who’s come on hard for us, probably been a couple years since he was here. But he came up with the idea that is now at the heart of the community notes that you now see on X and on Meta’s platforms, right? You know how if you see something that’s wrong on X, you can add a note to it and say, well, no, actually, this is the truth and then link to whatever the truth is. And the way that bridging-based algorithms work is they show them to people across the political spectrum, right? They have various ways of figuring out sort of where you’re aligned politically. And they only show the note if people who are more on the left and more on the right agree, right? They sort of see a bridge between the two of you. And they think, well, if Republicans and Democrats both think this is true, this is likelier to be true. And so we’re going to show that note. And that is a way that using an algorithm, you can try to create some consensus, right? You can try to build a bridge. I’m not aware of any platform that has tried to do this in the sort of core feed. It may be that this is less engaging and a worse business than the ones that we have today. But we do know that there is a way to build technological systems that bring people together because they are already in use today on some of the worst platforms on earth. (Time 0:17:44)
  • Take Control Of Your Media Diet
    • Reassess your media diet and diversify sources to avoid rage-optimized feeds.
    • Choose calmer outlets and offline interactions to reduce emotional escalation. Transcript: Casey Newton Our emotions, and then deciding if we want to make a change there, right? Because perception can become reality in too many cases. And it just might be a moment to find something a little bit calmer to help you understand what’s going on in the world. For example, a podcast hosted by two friends who mostly like to tell jokes about technology companies. (Time 0:25:08)
  • Crypto Cash and Chip Deals Intersect
    • Two intertwined deals tied a $2B Emirati crypto investment to a separate push to export top AI chips to the UAE.
    • The transactions involved overlapping people and raised conflict-of-interest concerns. Transcript: Kevin Roose And this one did not get a ton of attention relative to the Charlie Kirk assassination story. But I thought it was a big, important investigation from my colleagues at The Times that deserved more attention than it got. Casey Newton I completely agree with you. This one is a really extraordinary piece of journalism, and so I’m glad we’re highlighting it today. Kevin Roose Yeah, so this story involves two seemingly intertwined deals between the Trump administration, World Liberty Financial, which is the Trump family part-owned cryptocurrency company, And the United Arab Emirates. And it raises just a ton of fascinating questions. Casey Newton Yeah, but it’s also a story about the future of AI and potentially AI safety. As you know, there are so many people all around the world who want to get their hands on state-of chips, but the United States has export controls in place that prevent some of our adversaries Or countries that we’re just kind of worried about from getting them. Yeah. Kevin Roose And I think it would be a really big deal in any other administration. I think it would have gotten, you know, months of airtime. It would have been all anybody would have talked about for weeks on end. There would have been hearings and, you know, calls for impeachment and things like that. (Time 0:28:33)
  • Reporting Sparked By A Dubai Conference Reveal
    • David Yaffe-Bellany began reporting after witnessing Zach Witkoff announce a $2B deal at a Dubai crypto conference.
    • That onstage moment prompted deeper scrutiny into how the transaction formed. Transcript: Kevin Roose Specifically, a super yacht off the coast of Sardinia, where you wrote about a meeting that took place this summer between the owner of the yacht, a man named Sheikh Tanun bin Zayed Al-Nayan, A member of the ruling family of the United Arab Emirates, and Steve Whitcoff, President Trump’s envoy to the Middle East. Now, before we get into the contents of this meeting, I have to ask DYB, what is a super yacht? What makes it different than a regular yacht? And how can I get invited on one? Casey Newton Doesn’t it just have guacamole and sour cream? David Yaffe-Bellany Well, did you guys read that New Yorker story from a few years ago, the haves and the have yachts? That’s very good. That’s my main reference point for all yacht-related questions, and I believe that there are actually very formal distinctions between yachts, super yachts, and mega yachts. And that it has to do with the length of the vessel. But clearly, you guys aren’t spending a lot of time on the coast of Sardinia. Casey Newton I heard it wasn’t about the length of the vessel, but how you use it. It’s about the motion of the ocean. That’s right. Kevin Roose Anyway, we’re getting off track here. We’re getting off track. Okay. So, DYB, let’s talk about this meeting. What were Sheikh Tannoun and Steve Whitcoff there on this yacht to discuss? David Yaffe-Bellany So the White House told us that this discussion centered on resolving international conflicts, essentially. But what our reporting showed is that the relationship between these two men, you know, spans both diplomatic matters like that and also, you know, essentially a business partnership Between Steve Witkoff’s family company and Sheikh Tachnun’s sort of family investment fund in the United Arab Emirates. (Time 0:30:50)
  • Chip Exports Carry National Security Risks
    • The AI chip deal would relax US export controls and risk chips reaching China via Emirati ties.
    • Granting NVIDIA-level chips to the UAE is strategically fraught given security and transfer concerns. Transcript: David Yaffe-Bellany And this is a tentative agreement between the United States and the Emirates for the U.S. To allow the export of American-made chips to the UAE. These are the most valuable, powerful chips that fuel AI. They’re developed by NVIDIA. And without them, you can’t really become sort of a tech powerhouse in the way that the UAE wants to be. And, you know, the Emirates has been trying to get access to these chips for years and years, and the issue has always been that U.S. National security officials worried that if we give these chips to the UAE, then they’ll end up in the hands of China, because there are really tight ties between those countries. Yeah, tell us about some of those ties. Joint military exercises, technology sharing agreements. At the center of this is G42, which is another kind of Emirati tech firm, also run by Sheikh Tachnun. And so, yeah, the fact that the Emirates is getting these chips was a big deal and not necessarily an inevitability, given the posture of the Trump administration toward China and given The resistance in the Biden administration toward giving free access to these chips. Casey Newton And there was resistance in the Trump administration as well to giving the UAE those chips. David Yaffe-Bellany Is that right? Absolutely. This is something that we found in our reporting, that there were disputes within the White House and people who were very concerned about the national security implications of allowing Access to these chips. You know, I should be clear, you know, this is like a preliminary agreement and there might be new security guarantees that are built in when those chips actually leave the U.S. And go to the UAE. (Time 0:34:17)
  • Staff Changes Altered Chip Policy Path
    • Firing export-control advocates cleared the way for pro-export officials to advance the UAE chip agreement.
    • Personnel changes inside the NSC materially shifted policy trajectories. Transcript: Kevin Roose Let’s talk about some of the other figures involved in this story. In particular, I want to ask you about the role played by David Sachs, who’s the AI and crypto czar in the Trump administration. He was also, as you report, involved in some of these deals. So what is David Sachs’ role in all this? David Yaffe-Bellany So David Sachs, he’s the AI and crypto czar. And, you know, he’s interesting because, you know, similar to Wyckoff, he has a foot in both worlds. He’s simultaneously working in government and he has this tech job at Kraft Ventures, his VC firm. And what we found in our reporting is that he was one of the driving forces behind the chip steal with the UAE. He was pushing forward aggressively in public and also behind the scenes within the White House. And that raised concerns from some of his White House colleagues about potential conflicts of interest. The reason being, you know, Sachs is a tech investor. He’s still working as a tech investor. His company has invested in AI ventures in the past. And one of the original backers of Kraft was another kind of Emirati investment firm that is currently chaired by Shake Talk Noon. And so those concerns sort of bubbled up within the administration while this deal was being negotiated. I should also say that, you know, Sachs received ethics waivers from the White House that basically said, you know, you’re divesting from a lot of these stakes, you know, what’s left Is sort of de minimis and isn’t going to affect your work and sort of it’s okay for you to kind of play in these areas that kind of intersect with your business. So he did get that sort of permission slip from the Trump administration, but there were still concerns from colleagues. Casey Newton I have to say, very little is funnier to me in this moment than the idea of getting an ethics waiver from the Trump administration. Just what is that process like? Kevin Roose Okay, I have one more character that I need to ask you about, David, who is another David, David Feith. He was Trump’s senior director for technology on the National Security Council. So a member of the Trump administration who was known inside the government as being sort of the export controls guy for chips, the guy who thought it was a really bad idea for the U.S. And its companies to be shipping the most powerful AI chips overseas. This was a view that he shared with some folks in the Biden administration and its Security Council. And he was fired earlier this year as part of sort of a clearing of the House at the NSC by President Trump. So what did you report about that firing and what may have motivated it? David Yaffe-Bellany I mean, that was a crucial moment, you know, during these chip negotiations, where a kind of, you know, potential roadblock to a deal going through was abruptly removed. And it kind of cleared the field in a way that, you know, allowed David Sachs to eventually kind of take control of these negotiations. (Time 0:40:18)
  • AI Drafts Threaten Trust And Jobs
    • Newsrooms experimenting with AI-generated first drafts risk reputational harm and job automation.
    • Early adoption may save time but invites credibility and employment trade-offs. Transcript: Kevin Roose Stories of the week. Let’s go through them. Number one, Business Insider tells reporters they can use AI to generate first drafts of articles. This was first reported by Oliver Darcy in his great status newsletter. This is about an internal memo that went out to the staff of Business Insider last week from editor-in Jamie Heller, telling them that they could use AI not just for research and various Other things, but for writing the first drafts of their stories as long as reporters, quote, make sure your final work is yours. Casey, what did you make of this? Casey Newton Very curious what she means by make sure your final work is yours. If the AI did all of the research and wrote the first draft, you, what, change the placement of a couple of commas? Look, I’m very curious to see what happens with this one. The cynic in me suspects that this sort of thing is going to become a lot more common. And as AI writing tools improve, I think more reporters are going to feel like maybe I should let it take a stab at a first draft. I should say for myself, I do use AI systems to edit my copy. So not to do the writing, but to catch the typos and some of the factual errors. So there is kind of a spectrum of what I think you and I think is like, okay, right now, this goes further than I would, but this just seems like one of those frontiers that keeps advancing. And before long, a lot of other folks are going to do it. (Time 0:56:20)
  • Italian Brain Rot Became A Roblox Phenomenon
    • An Italian Brain Rot–inspired Roblox game hit a record of about 23.4 million concurrent players.
    • The trend shows Gen Z enthusiasm for AI-generated characters and rapid cultural recycling. Transcript: Kevin Roose That is, of course, the hit game creation platform where, to several accounts that track Roblox activity on X, a new game recently broke the all-time record for most concurrent players On the platform. Something like 23.4 million people were simultaneously playing a game based off the viral AI sensation of Italian brain rot. Casey Newton Italian brain rot, I think one of our favorite finds of the year. And you’re telling me it’s a video game now. Kevin Roose Yes. So this game, which I guess you can build games inside Roblox. I’m not myself a big Roblox user. Is called Steal a Brain Rot. And I did spend some time on YouTube trying to understand this phenomenon. Yesterday, I watched a video about a guy who spent 24 hours playing Steal Brain Rot inside Roblox. That sounds good for you. Yes. The video, thankfully, was not 24 hours long. But basically, this is a video game that users of Roblox, or one user in particular, has created where you kind of just go around stealing people’s characters from the Italian Brain Rot franchise, which, as you will remember, as you introduced me to earlier this year, was created on TikTok as kind of a self-referential form of Zoomer humor. Casey Newton So is this kind of like play on Pokemon, except instead of stealing Bulbasaur, you’re stealing Tung Tung Tung Sahur and Ballerina Cappuccino? Kevin Roose Yes, you can go around, you can buy various characters with your Robux. You can buy a Nubini Pizzanini and use it to do various things. You can then go around stealing other people’s characters that they’ve bought. This has become a huge sensation. 23.4 million people is massive. (Time 1:02:57)
  • Protect Teens With Age-Aware Chat Experiences
    • Build age-aware experiences by detecting under-18 users and offering restricted chat versions.
    • Integrate content filters and escalation paths to protect vulnerable teens using chatbots. Transcript: Casey Newton On Tuesday, OpenAI said it’s building a system that will try to identify users who are under 18 and will give them a more limited version of ChatGPT that will block graphic sexual content And in rare cases potentially involve law enforcement. So what did you make of this story when you saw it? Kevin Roose Positive development. Like, I am very worried about the increasing frequency with which young people especially are using these chatbots as sort of emotional support companions, are forming these intimate Relationships with them. And as we’ve discussed, like for some set of them, this does really appear to be taking a toll on their mental health. So I am all in favor of OpenAI doing what it can to determine which of its users are under 18 and to give those users a different experience. They also have been under some pressure. Last Thursday, the FTC launched an inquiry into these AI chatbots, including OpenAI’s. And the company is also facing a lawsuit from the family of Adam Rain, a 16-year boy who died by suicide after talking with ChatGPT. So this did not come out of nowhere. They have been under a lot of pressure to do something for their younger users. But I think this is probably a good and overdue step. Casey Newton Yeah, I agree with that. At the same time, I worry that this truly is not going to be enough. You know, right now, I think one reason why younger people and vulnerable people are talking and are sharing their most intimate thoughts with ChatGPT is because they’re confident That those communications are private. Now that they know that those could potentially be shared with your parents, with law enforcement, my guess is those teens and vulnerable people are simply going to go elsewhere. They’re going to go to other platforms that don’t have those same protections. So while I agree with you, I think this is the right thing for OpenAI to do. I think we need to look around the industry at platforms that do not have these protections and ask them, well, what are you going to do? And even after all of them do it, there’s the risk that teenagers and vulnerable people are just going to start talking to large language models that live on their laptops that don’t Communicate with anyone, right? (Time 1:06:34)
  • Chatbots Replace Search And Companionship
    • OpenAI finds most ChatGPT use is non-work and leans toward companionship and practical guidance.
    • This shift makes chatbots competitive with general web search and interpersonal support roles. Transcript: Kevin Roose Yeah, I thought this was really interesting. This is the kind of data that’s been relatively sparse. We just don’t know a lot about how people are actually using these chatbots. And the companies that make the chatbots are the ones, of course, with the access to the best data about that. But we got reports this week from both OpenAI and Anthropic who talked about the various ways that people are using their models. We should also say that these are not exactly apples to apples. OpenAI’s report just looked at the sort of consumer version of ChatGPT, so not counting all of the, like, business and enterprise users who are using it through the API. Anthropics report did include both the sort of consumer cloud version as well as the sort of more enterprise-focused coding tools. But let’s talk about a few things from this. One thing that stuck out to me was that with ChatGPT, there is a closing gender gap in usage. So a year or two ago, it was the case that many more men than women were using ChatGPT. But now, according to OpenAI, more than half of ChatGPT’s users appear to be women, which is surprising to me. Casey Newton That’s so interesting. Now, did they share any speculation on what has managed to close that gap? No. Kevin Roose The researchers who conducted this study did not give any hypotheses for why this gender gap has narrowed, but they did say that by July of 2025, the share of ChatGPT users with sort of Typically feminine names had risen to more than half of all ChatGPT users, 52%. They also said that this has become a fast-growing tool in low- and middle-income countries. Maybe there’s some sort of relationship there. They said that by May 2025, ChatGPT was growing much faster in the lowest-income countries than in higher-income countries. So there seems to be kind of an expansion of use of ChatGPT, both among women and people outside the sort of richest Western countries. Casey Newton It’s interesting to see. I mean, and I would say that if there remained this huge gender gap that we were seeing a year or two ago, to me, that would be like a really negative signal about AI, right? Because if this technology is as useful and general purpose as executives are always telling us that it is, there should actually be relative gender parity in terms of who’s using it. Yeah. What else was in this report? Kevin Roose Another interesting thing from the OpenAI version of this report was that something like 73% of ChatGPT messages are not work-related. This was surprising to me. I thought that maybe the usage of ChatGPT for sort of personal and work-related tasks would be split pretty evenly. I certainly, you know, would guess that my own usage of ChatGPT is about half work and half non-work. But apparently, the non-work use cases of ChatGPT are growing really quickly, and they comprise now the majority, about three-quarters of all ChatGPT usage. (Time 1:10:03)
  • Different Paths: Consumer Chat vs Enterprise API
    • Anthropic’s Claude sees enterprise-driven adoption focused on coding and automation via API use.
    • OpenAI and Anthropic are diverging: mass consumer vs. developer-enterprise focus. Transcript: Kevin Roose They found that adoption of Claude was being driven primarily by coding and what they called automation usage patterns in businesses using Claude through the API. Anthropic also found that their highest Claude use per capita was in Singapore and Israel. And within the US, the leading per capita users of Claude were not in California or New York, as you might expect, but in Washington, DC and Utah. Casey Newton The Mormons love Claude. How about that? To me, what this suggests to me is that OpenAI and Anthropic are really diverging in who their customer bases are. OpenAI wants everyone. It’s like a ChatGPT in every house, on every phone. Anthropic is increasingly going after the businesses and the enterprises. (Time 1:14:06)