S.B.F Goes to Jail + Back to School with A.I. + Self-Driving Car Update

When Sam Bankman-Fried was arrested in December, he was confined to his parents’ house — but he was left free to roam the internet. Today, the New York Times reporter David Yaffe-Bellany talks about how access to the cyberworld allowed Mr. Bankman-Fried to violate his bail terms and land himself in jail. Then, how universities can manage a generative A. I. world. Plus: another look at autonomous vehicles.

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.

• 0:35Kevin Roose?

  It's funny, if you go to like the top podcasts on Spotify, most days it's like, you know, a mixture of like Joe Rogan and his friends, but then it's like sleep podcasts. Have you, have you seen these ones?

• 0:46Casey Newton?

  You're right, we are now below relaxing white noise.

• 0:50Casey Newton?

  It's unfortunate.

• 0:52Kevin Roose?

  They've had some really good guests recently.

• 0:53Casey Newton?

  They have some great guests.

• 0:55Kevin Roose?

  Should we do some white noise to open the show today? Just for anyone who's having trouble falling asleep and may use Hard Fork as a sleep aid. Okay.

• 1:02Casey Newton?

  Ready?

• 1:07Casey Newton?

  And then can you just loop that for 40 minutes?

• 1:10Casey Newton?

  Perfect. Podcast done.

• 1:19Kevin Roose?

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

• 1:21Casey Newton?

  I'm Casey Newsom from Platformer.

• 1:23Kevin Roose?

  And you're listening to Hard Fork.

• 1:25Casey Newton?

  This week on the show, Sam Bankman-Fried goes to jail. We'll tell you why and what's next. Then we go back to school with Wharton professor Ethan Malik, who tells us what teachers and students need to know about this year in AI. And finally, another wild week in the world of robotaxis. What happened? And what you told us about our autonomous vehicle episode last week.

• 2:12Kevin Roose?

  Casey, you know, on this show, we love to talk about Sam Bankman-Fried.

• 2:15Casey Newton?

  We do.

• 2:16Kevin Roose?

  And you know what he's having this week?

• 2:17Casey Newton?

  What?

• 2:18Kevin Roose?

  I mean, look, I've had a bad month.

• 2:19Casey Newton?

  Another one?

• 2:20Kevin Roose?

  Yeah.

• 2:21Casey Newton?

  Guys been through a bit of a rough patch.

• 2:23Kevin Roose?

  So Sam Bankman-Fried is having the, one of the worst weeks of a string of very bad weeks. Because this week, he went to jail. And here to talk with us about how that happened, what is going on in the ongoing saga of FTX, is my colleague and friend of the pod, David Yaffe Bellany. David, welcome to Hard Fork.

• 2:41Speaker 03

  Thanks for having me.

• 2:42Kevin Roose?

  Hi, David. So it has been a very active few weeks in the case of Sam Bankman-Fried. Sam Bankman-Fried, of course, is the former CEO of FTX, the crypto exchange that collapsed late last year after allegedly misusing a bunch of customer funds. You reported this week that Sam Bankman-Fried went to jail because a federal judge in New York had revoked his bail. So just catch us up on why Sam Bankman-Fried is sitting in a jail cell in New York now and not at his parents' home in California where he's been.

• 3:16Speaker 03

  So yeah, the important context here is that basically from the moment that SBF was released into home confinement, he's been pushing the boundaries of what you're allowed to do when you're out on bail. So first in January, he sends an email and message to a former FTX employee, basically saying, hey, let's talk, let's, you know, see if we can get on the same page about a few things. The prosecutors were really unhappy about that. They said this kind of reeked of potential witness tampering, that this was somebody who might testify in the case against Sam Bankman-Fried, and that therefore this was totally improper. Then like while the negotiations over whether to tighten his bail terms as a result

• 3:59Speaker 03

  of that were going on, he was caught using a VPN to access the internet, which the prosecutor said was a sign that he was trying to kind of evade detection in his online activities. So that was sort of the context of what led up to the last couple of weeks.

• 4:14Kevin Roose?

  And can I just do a sort of a clarifying aside here? When you are arrested and you get out on bail, whether you are a white collar, very wealthy defendant like Sam Bankman-Fried, or someone who is not wealthy or powerful, or the former CEO of a company, you have some conditions, right? The judge says, I will let you out on bail, but you have to, in some cases, pay some money or put up some collateral. And, you know, you can't commit any more crimes and there are certain conditions for you remaining free on bail. And in this case, my assumption is that because Sam Bankman-Fried was so online, for lack of a better word, like he was signaling people and playing video games, that like, one of

• 4:56Kevin Roose?

  the things that the judge had said was that in order to remain free on bail, you actually can't like do any of that stuff. Is that correct?

• 5:03Speaker 03

  That's essentially what happened. I mean, it came from this sort of interesting contradiction, which is that, you know, his original bail term said you have to stay in your parents' house and you have to wear an ankle monitor. So physically, he was confined, but in the world of cyberspace, he could kind of like roam free and do whatever he wanted. And, you know, most criminal defendants would just stay quiet, you know, try not to upset anybody while they're awaiting trial. But Sam's not a normal criminal defendant and he was signaling people, he was tweeting, he was writing Substack posts, and this was clearly really ticking off the prosecutors and ticking off the judge too. And so what ended up happening is that a whole new set of bail terms were imposed on him that limited which websites he could access. So there's actually, there was a list of maybe 25 or so websites that he was allowed

• 5:48Speaker 03

  to access, including NYtimes.com, of course. And that was what he was confined to doing. And he was also prevented from reaching out to people who were involved in the case, like, you know, former employees who might be involved in the trial. And so those were the limitations on him starting around last spring.

• 6:05Casey Newton?

  Now, when I hear that I would have to live in my parents' house and could only visit 25 websites, that sounds like a cruel and unusual punishment to me, but that's apparently not the case here?

• 6:16Speaker 03

  Well, it's a better alternative than than sitting in jail. I think most people would agree. Though maybe if you don't, Casey, I don't know your relationship.

• 6:23Casey Newton?

  Here's what I'll say. I want either want to have all of the internet or none of the internet. If you give me like 1% of the internet, that's going to give me an aneurysm.

• 6:30Speaker 03

  I'm also not sure if Platformer was on the list. You know, no, no offense.

• 6:34Casey Newton?

  And and by the way, my lawyers are looking into this.

• 6:38Kevin Roose?

  My favorite detail from your reporting on this bail violation is that he was apparently using a VPN to watch a football game. Is that correct?

• 6:49Speaker 03

  That is, that is true. At least that's what his defense has claimed. I mean, the prosecution sort of cited the fact that he'd used VPNs on certain dates as like evidence of potentially nefarious activity. And the defense came back and said, no, that was the date of the conference championships in the NFL and then the Super Bowl. And he was just trying to watch football. The reason he needed a VPN is because he'd gotten like an international NFL streaming account when he lived in the Bahamas and he needed to access it from his home in the US.

• 7:18Kevin Roose?

  If I was out on bail and one of my conditions of my bail was that I not use a VPN, like, maybe that's one where you just like skip the game and catch it on SportsCenter. Like, that's an option, too.

• 7:30Casey Newton?

  Maybe you just read the postgame report at NYtimes.com, one of your approved websites.

• 7:36Kevin Roose?

  Right. So, one of the most interesting and controversial wrinkles in this latest twist in the SBF case involves a story that you published, actually in late July, that was about basically the diary of Caroline Ellison, who ran the Alameda hedge fund and was also SBF's ex-girlfriend. And this was a story that reported on the existence of these sort of Google Docs that Caroline Ellison had written about her sort of stress and anxiety running this hedge fund, and some details of the relationship between her

• 8:18Kevin Roose?

  and SBF. This story came out and then was subsequently cited by prosecutors as evidence that Sam Bankman-Fried had violated the terms of his bail in talking with the media. So, David, what can you tell us about this wrinkle in the case and this diary of Caroline Ellison's?

• 8:41Speaker 03

  You know, we did a story a few months ago noting that people involved in the case had a handwritten diary that Caroline had kept and that she had also kind of reported her feelings and thoughts about various things on Google Documents. So we'd known that for a while and then this more recent story that we did actually had detailed excerpts from the Google Documents that she had written. And, you know, this included accounts of her kind of insecurity about her position running Alameda. Was she a good enough leader? Was she really suited to doing this type of job? You know, it sort of reflected a lot of the stress that she was feeling throughout the year. And it also got into some sort of intimate details of her romantic relationship with Sam Bankman-Fried.

• 9:25Speaker 03

  You know, they had a kind of on and off relationship. And so, you know, reflections on what it was like to be around him after they broke up and that sort of thing. And, you know, all of this is potentially relevant to the case because their relationship is at the heart of the relationship between these two companies. She's agreed to testify against him at trial. So any kind of insight into what she's thinking and feeling is helpful and interesting and sort of sheds sheds new light on the case. So that's the sort of background on on the on her writings and what was in them.

• 9:54Kevin Roose?

  And prosecutors have said that Sam Bankman-Fried leaked these documents to The New York Times. And that was apparently part of the judge's decision to revoke his bail and send him to jail. So what can you tell us about that?

• 10:08Speaker 03

  Sure. So within 24 hours of our story coming out, the prosecutors submitted a filing that basically said, we've confirmed with Sam's lawyers, they've admitted to us that Sam was a source for documents that were used in this story. And, you know, therefore, we want to put a gag order on him. We want to stop him from talking. He's already limited in what websites he can access, but we want to prevent him from talking to the media because this is a sort of improper intervention in the case that could sort of interfere with a fair trial, intimidate witnesses, that that sort of thing. So that was kind of their first salvo.

• 10:42Kevin Roose?

  So help me understand this because I'm a little confused here because on one hand you have prosecutors who are saying that Sam Bankman-Fried leaked these documents from Caroline Ellison to the media in order to engage in kind of character assassination or witness tampering or make her look bad ahead of her potential testimony at this trial this fall. And I read the story that you wrote. I read the excerpts from these writings. They were kind of embarrassing in the same way that many of our private writings would be embarrassing if they were leaked to the New York Times. But they didn't strike

• 11:24Kevin Roose?

  me as particularly damning. She wasn't saying, actually, Sam's innocent and I did all the crimes. So if prosecutors are correct in alleging that he did leak these diaries, why do you think he did that? Was there more to it than just trying to make her look bad?

• 11:41Speaker 03

  Yeah, so obviously I'm a little limited in what I can say because this involves confidential sourcing. But I can tell you what the prosecutors are claiming that he did and their rationale for why he did it. And what they're saying is that these were very sensitive writings, very personal, not the sort of anything that anybody would want to come out publicly, and that the prospect of similar material coming out regarding other witnesses could have this effect of intimidation or cause people to back off. It could also have that effect on Caroline, they're claiming, because she might think, oh, maybe Sam's got other stuff and I better be careful what I say at trial.

• 12:23Kevin Roose?

  That actually makes a lot of sense to me. It's like, nice diary you got there. Be a shame if something happened to it. So Sam Bankman-Fried had his bail revoked and was subsequently put in jail. What do we know about the hearing where this was announced? Were you there in the room with us? Can you tell us a little bit more about what that was? Like?

• 12:43Speaker 03

  So unfortunately, I was not at either of the hearings that have happened on this issue because I'm stuck out here on the West Coast. But I had colleagues at both hearings. So there's an initial hearing, which everyone thought was just about this gag order that the prosecutors were asking for. But the prosecution comes into the room and immediately changes tack. You know, seconds before the hearing starts, they tell the defense, we're going to be asking for his bail to be revoked and for him to be sent to jail immediately. So there's an initial hearing on that where Sam's lawyers basically say, we haven't had any time to prepare. You need to like, give us some more time. And so the judge calls a second hearing a couple of weeks later. That was this past Friday. And at that hearing, you know, there's about an hour of argument, you know, both sides kind of making their case. The judge says, you know, all things considered, I'm revoking bail.

• 13:28Speaker 03

  It's not just because of this recent story, it's because of the initial outreach to an FTX employee. It's partly this VPN thing, which the judge said kind of showed a mindset of trying to like evade limitations. And basically, it was the last straw for the judge.

• 13:43Kevin Roose?

  All right. So that's why SBF is in jail. You've also reported that there's been a new filing by prosecutors in this case detailing a bunch of new evidence against him. So tell us about what was in this filing and what that might mean for the trial.

• 13:57Speaker 03

  Yeah, so one of the things that was surprising about Sam's bail getting revoked is we're really in the home stretch before the trial. There's not a lot of time left. Trial starts October 2nd, and we're kind of in the final stage of sort of pre-trial wrangling in court, where both sides say, you know, this is the evidence we've got, and here's why we think it should be admissible in court. And so there's going to be arguments back and forth about that. And on Monday, the prosecution filed a long document, basically detailing with way more color than we've ever seen before, what they've got and how they plan to use it. So they said, you know, we've got these three close associates of SBF, including Caroline, who've all pleaded guilty already, and we expect them to testify.

• 14:41Speaker 03

  And we've got contemporaneous notes that they kept after conversations with Sam. That includes some of Caroline's writings, including a document titled Things Sam Is Freaking Out About. And so we can all look forward to learning more about that.

• 14:55Casey Newton?

  So you haven't seen it? We don't know what Sam was freaking out about?

• 15:00Speaker 03

  I have not seen the thing Sam is freaking out about.

• 15:02Casey Newton?

  Okay.

• 15:02Speaker 03

  However, it did say in the filing on Monday that the list included things like the bad press around the connections between FTX and Alameda. So we have sort of a vague sense, but.

• 15:13Casey Newton?

  By the way, if it were me, something I would be freaking out about, using customer funds to fund my crypto empire. But that would be things Casey was freaking out about. But go on.

• 15:25Speaker 03

  Well, maybe that's why you're not the one in MDC, right?

• 15:27Casey Newton?

  Well, not yet. Stay tuned.

• 15:32Speaker 03

  Anyway, so, so yeah, so there's also, there are a bunch of other interesting things, too. I mean, there's text messages that another high ranking FTX employee named Ryan Salem sent. He was the guy who donated tens of millions to Republican politicians and prosecutors have kind of encircling him for a while. He hasn't been, he hasn't been charged, but he's, he's clearly going to come up to some extent. There's also an audio recording of a meeting that Caroline held with Alameda employees right as the companies were collapsing last November, where she essentially confessed and said, I worked with, with Gary and Nishad, the two other executives that pleaded guilty, and with Sam, and Sam decided that we were going to take customer funds. She said that pretty explicitly. The fact that that meeting

• 16:16Speaker 03

  happened, we've known for a long time. I, I reported it back then, as, as did others. But the prosecution has a full recording of it with a transcript that was included in this filing.

• 16:26Casey Newton?

  Who, who recorded it? Who is, was someone wearing a wire? I would have recorded it. You, or we, any of us here would have recorded this. That is, you want to have that for the historical record. It just seems like everything at this firm was being recorded and put into Google Documents and voice memos and whatnot. Well, that's the funny thing, right? Is that they actually didn't do any record keeping for many, many of the important things that they should. But when it came down to writing down the crimes, like they were all over it.

• 16:50Speaker 03

  It was, yeah, it was an Alameda employee who, who recorded it apparently. I mean, this was a, you know, this was a staff meeting. The staff was pretty small, but there were enough people in the room that someone got the voice memos app out, presumably. And so, so yeah, I mean, that was a, it was a pretty damning bit of evidence when it kind of first emerged that she said these things at the meeting back in November when we were all writing about it. And now that they're planning to play that recording in, in court, it really, it, it doesn't look great for Sam.

• 17:15Kevin Roose?

  In one of your stories, there was a line that said that so far there have been millions of pages of evidence produced. I don't think you were being figurative. Are there literally millions of pages of documents? And if so, like, what is in those millions of pages? And how can one man's lawyers go through all of that?

• 17:35Casey Newton?

  And please describe all million.

• 17:36Speaker 03

  Yeah.

• 17:39Speaker 03

  There literally are millions of pages. But it's like, you know, at some point, the prosecution subpoenaed the whole contents of SBF's personal Google Drive from Google. So that's, you know, however many hundreds of thousands of pages. And, you know, it's like, all the documentation that any of the key people in this case ever had on their computers is in play. Their financial documents, spreadsheets, text message histories, Slack histories. They've got some Signal chats, I think, though the fact that SBF was on auto delete and advised employees to go on auto delete has been a factor in the case as well. So there could have been even more, potentially. And yet, it's just a ton of stuff.

• 18:16Casey Newton?

  Look, I know it is in the nature of federal prosecutors to just drown people in charges in hopes that they will reach a plea agreement and just end the whole thing without forcing a trial. And yet, that, in combination with all of the evidence that you've been describing for us, David, makes me wonder, why hasn't Sam Bankman-Fried pled guilty? Like, any thoughts on that?

• 18:37Speaker 03

  I mean, look, it's hard to, you know, read his mind on this. And of course, you know, I think probably, and this is all speculation, but one thing the prosecutors might be hoping is that by forcing him to spend a little time in jail, they might change his thinking about this. But, you know, all the evidence suggests that he's convinced that he has a chance of winning a trial and avoiding any prison time. And, you know, if you believe that, then our system allows you to fight the charges, as it should. And so, you know, we'll see, we'll see what comes of it. I mean, at various points throughout this process, people have pointed out that he seemed deluded about things, you know? He thought that he could save the company in the days before he was arrested, that he could just raise new money and make the whole disappear. And, you know, it could be that some of that optimism is playing into his thinking now.

• 19:22Casey Newton?

  Yeah, sometimes when you have these very high-profile trials, the defendant becomes a kind of cause celebre, and perhaps they have a fan base that rallies to their side and sort of lobbies in the court of public opinion in the hopes that that might change the outcome. Is there an SBF fandom or constituency left? Are there folks out there saying, this guy's getting a raw deal?

• 19:45Speaker 03

  Not really. A lot of people have made up their mind about him. This is something that has sort of come up a little bit in the back and forth over the story we did on Caroline, where SBF's lawyers have argued that one of the reasons that he wanted to provide these documents was because he felt like he's been unfairly maligned and that he should have a right to defend his reputation. That's partly because, yeah, like, there really isn't like a pro-SBF contingent out there arguing for his innocence. I mean, crypto Twitter has certainly made up its mind that he's the worst villain in the history of the world and that he should be drawn and quartered, essentially. And I think most mainstream legal analysts are pretty convinced that he's guilty as well.

• 20:25Kevin Roose?

  So this is shaping up to be quite a trial in October. I imagine that you will be there, as will some of our other colleagues.

• 20:33Casey Newton?

  This is shaping up to be quite a trial in the way that the Harlem Globetrotters next match is shaping up to be quite a basketball game.

• 20:43Casey Newton?

  It's not looking good for the Washington Generals this time.

• 20:46Kevin Roose?

  I didn't say it was going to be a close trial. I just said it was going to be quite a trial. So what are you looking for or what are you most interested to see in the lead up to this trial?

• 20:57Speaker 03

  Well, one thing I would say, just to caution you on the Harlem Globetrotters comparison, you never know. I mean, this is why we have the system that we do. It only takes one juror to sort of swing the outcome. And the prosecution has made a lot of claims about Sam, but they've been sort of claims that haven't been fully fleshed out and tested. And so that's what we'll see in court in October, and it should be incredibly interesting. But what am I watching for? I mean, look, there's just the inherent drama of seeing three people as like top advisors at his company who were not just his top advisors. One was his girlfriend. The two others were two of his closest friends who lived with him, who'd been by his side for years. That group testifying against him, there's a certain, there's a drama in that that's

• 21:42Speaker 03

  pretty undeniable and that will be kind of fascinating to see play out. I mean, also for someone like me who's been obsessed with this case for almost a year and following every twist and turn, any new revelation of... Some new detail, you know, some document that Caroline had that we didn't know about is kind of thrilling as well. But it's also, you know, a test of like, can this crackdown that the US government is doing on the crypto industry actually yield results? So that's probably the more kind of important.

• 22:10Casey Newton?

  Yeah, I am looking forward to that moment when they all testify against him, though, because I think a very relatable feeling is wanting to see your boss go to prison. And those three are actually going to get to live that out.

• 22:20Speaker 03

  Well, Casey, would you flip on Kevin if it turned out that he was embezzling me from Hard Fork or something?

• 22:25Casey Newton?

  I would. I've told Kevin, buddy, you better be walking the straight and narrow, because when the cops come knocking at my door, let's just say I have a few Google Docs of my own, okay?

• 22:35Kevin Roose?

  Yeah, David, last question. If I were in possession of like a recording of a secret meeting at which like a podcast host and newsletter writer confessed to certain federal crimes, do you know any reporters I could send that to?

• 22:49Speaker 03

  You hit me up on Signal and we can discuss it.

• 22:51Kevin Roose?

  Okay, we'll just auto delete our conversation afterward. David Afegoni, thanks so much for coming back.

• 22:57Casey Newton?

  Thank you, David.

• 22:58Speaker 03

  Thanks for having me. Anytime.

• 23:03Casey Newton?

  When we come back, we're going back to school with AI.

• 23:29Speaker 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.

• 24:03Speaker 04

  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.

• 24:38Speaker 05

  Today's AI builders need infrastructure that scales with their ambitions. Irene is a vertically-integrated AI cloud platform, delivering data centers, compute, and software for AI training and inference, underpinned by its expansive portfolio of land and grid-connected power in renewable-rich regions globally. Irene's AI cloud gives direct access to Nvidia GPUs, scalable from hundreds to thousands. For builders, that means direct access to AI compute, rapid scaling, and predictable pricing. Irene, AI cloud for the next big thing.

• 25:13Kevin Roose?

  Casey, it's mid-August. You know what that means.

• 25:15Casey Newton?

  What does it mean, Kevin?

• 25:16Kevin Roose?

  It's back-to-school time, baby.

• 25:17Casey Newton?

  Oh, my God, Kevin, I haven't done the assigned reading. What happens at the end of Old Yeller?

• 25:23Kevin Roose?

  Nothing good.

• 25:23Casey Newton?

  Oh, no.

• 25:24Kevin Roose?

  So this year, one of the biggest questions facing schools as they reopen for the fall semester is what the heck do we do about generative AI? I think this is one of the biggest questions that schools have been wrestling with over the past year. You know, ChatGPT, it came out like the semester was already underway. It sort of landed as kind of an asteroid out of the sky, and schools really just scrambled to get through the year. And I really thought that this summer was going to be sort of when schools and universities kind of regrouped and put their heads together and figured out, how do we actually educate people in a world where this generative AI stuff exists? What does homework look like? What does admissions look like? What is the role of the faculty member anymore?

• 26:07Kevin Roose?

  And that just seems not to have happened in any big, organized way. A lot of schools Are still having meetings and organizing committees and task forces to try to figure out what to do about generative AI. And it just seems like there's an area where there are a lot more questions than answers.

• 26:22Casey Newton?

  Sure, and at the same time, I think that that might be OK. The technology is new. We don't totally know what we ought to do about it. And so I think a world where different teachers are taking different approaches and schools are being a little slow in how they craft policies might be to everyone's ultimate benefit.

• 26:39Kevin Roose?

  Right, but I do think this is an urgent issue for schools, especially going into this new school year. And so I wanted to kind of spend some time talking about that. And I wanted to talk to someone who actually does have a clear vision of how AI can and should transform education. And one person who stood out to me was Ethan Mollick. Ethan is an associate professor at the Wharton School at the University of Pennsylvania. He teaches innovation and entrepreneurship and also writes and thinks a lot about generative AI in the classroom. And he's been sort of documenting his experiments with AI. He has a Substack called One Useful Thing. And he has actually come up with a strategy that he thinks could help other schools adapt to the post chat GPT world.

• 27:21Kevin Roose?

  And so I wanted to bring him on today to talk about how schools should be thinking about generative AI and what they should be doing about it.

• 27:28Casey Newton?

  All right, let's hear what he has to say.

• 27:30Kevin Roose?

  Ethan Mollick, welcome to Hard Fork. Thanks for having me. So I first came across your work last year when you and I were both writing a lot about generative AI. And you have sort of become like, I would say, an AI guru inside the world of higher education. I know you've been talking with a lot of faculty members and administrators at schools across the country, who it's fair to say, I think, are confused and disoriented about what to do about all of this new technology. So I wanted to start with kind of a vibe check. Like, can you just paint a picture for us of what is going on with generative AI at schools right now as we head into the fall semester?

• 28:09Speaker 06

  I think the word I'd probably use would be chaos or apocalypse. I think that people are just starting to dawn on them what this means. And I think when we talk about what this means, I think level one of what this means is what's dawning on them right now, which is, oh, God, all my homework assignments don't work anymore. And people haven't started to think about the other implications fully yet. There's a lot of exceptions out there, but generally that's the vibe I'm getting.

• 28:33Casey Newton?

  Now, this technology has been available since last November. So what has happened between November and now where some of, you know, some other folks in higher ed or maybe secondary ed still aren't thinking about this?

• 28:44Speaker 06

  So I think for one thing, I think ChatGPT, the free version, GPT 3.5, still made enough kind of mistakes that, that, you know, it made it a little bit easier to ignore. I think the second set of things was that people are used to ignoring technology hype. So usually hype happens five years before technology comes out, right? So we're talking about Web three and you could just safely sit back and be a late adopter. I don't think people realized this was hype for a technology that already hit, which is a little bit unusual and people had to catch up to that. And then I think the third thing is enough alarm at the institutional level inside organizations that people who've ignored this have had to pay attention. So between all of those things, I think it's kind of created this bubble of anxiety and expectation.

• 29:28Casey Newton?

  Right now, one thing we should say is that if you're a teacher, something you can relax about is Web three. You can actually continue to ignore that one, but the AI stuff you should probably pay attention to.

• 29:37Kevin Roose?

  Right, well, and I think there's also this additional layer in education of like, well, these tech companies have been showing up for decades now and telling us that their tools are gonna transform the way we teach in the classroom and these Chromebooks and personalized learning software. And a lot of that has just been sort of empty hype. And so I think there was a reluctance on a lot of administrators' parts to just take this stuff seriously. But you are in conversations pretty regularly, I think, with people at universities who are trying to figure out what to do beyond sort of that first level, beyond the like, how do we stop students from cheating on their homework level? And I want to talk about that, but first I want to talk about the level one concern, which is can schools actually stop students from cheating with a generative AI software?

• 30:18Speaker 06

  No. I mean, that was a nuance, right? So the short answer is no. The long answer is AI uses undetectable and it detects people who don't speak English very well. It's terrible. So you can't use detectors. You can't ask AI to detect AI. It's just gonna lie to you. Like, every instinct we have about how to stop plagiarism doesn't work. So you can change how you teach, right? You could do Blue Book assignments. You could have people do oral exams. There are other ways of checking, but the old homework assignment is basically cracked by AI.

• 30:48Kevin Roose?

  I wonder if you could just take us inside one of these kind of faculty meetings where professors and school administrators are trying to figure out how to adapt to the post-chat GPT world. What are some of the common sort of things that you hear brought up in this world? And what are some of the sort of common objections to like? Why shouldn't we be changing our policies and procedures?

• 31:09Speaker 06

  So the first thing is the same problem everybody is suffering from with generative AI, which is no instruction manual. We're all figuring out as we go. There's literally papers coming out regularly about what kind of questions should you ask AI to get the best answers? We don't know whether it's really good at these tests or whether it's faking being good at these tests. So you come to a faculty meeting and the first 20 minutes are debunking rumors and then supporting others about how it learns and what it knows and is it stealing information and what's ethical? So there's a lot of that kind of discussion you'd expect among academics. Then there is sort of a discussion about, you know, how do you? Usually a punitive discussion about stopping plagiarism. And then there is this sort of more advanced discussion about what do we do now? Right? What do we, what do we tell our students? What does good instructional design look like?

• 31:53Speaker 06

  And I think that's the more profitable part of the conversation, but you have to get through the fact that nobody knows anything, you know, including, I'd hate to say all of us in the room. We're still kind of making it up and learning by experience, right? And trying to tell other people based on that experience, which is quite challenging.

• 32:08Kevin Roose?

  So, okay, let's talk about not just why schools are tying themselves up in knots about this, but what they should actually do. So you, me and Casey are starting a university tomorrow. Hard Fork University, it's a great school. Not accredited, but great school. And we get to craft the policy about generative AI and how it should be used. And we not only get to craft the policy, but we get to tell every instructor how to teach their class using this stuff in the best way possible. How would you run a school if you could make all the decisions about generative AI?

• 32:41Speaker 06

  So the cool thing about education is we're in for a couple of rough years, but I actually kind of have a sense of what the future looks like because we actually have a lot of research on how to teach and it happens to align really well with AI. And the secret is pretty simple. It's two concepts called the flipped classroom and active learning. So the idea of a flipped classroom is rather than learning the material inside of class and practice it outside of class, you learn the material outside of class and practice it inside. So the basic version that you might have seen is people watching videos of a math lecturer or Khan Academy outside of class. And then in class, you work on problem sets together. When you have trouble, the teacher comes around and helps. Some people present to the class about that. It's all about putting knowledge to use. It's all about the challenging yourself, pushing yourself into an active learning environment.

• 33:26Speaker 06

  So it flips the classroom experience where instead of focusing on lecturing inside of class and doing assignments outside, you do the reverse. We've known this is a really great way to teach for a long time, but the two problems have been what do we do in a flipped classroom setting? Do we just give people textbooks to read? Do they watch videos? That's never worked particularly well. Now we have AI tutors that will be able to fill in some of that basic instructional piece. And then in class, how do we design enough engaging experiences for people? Well, it turns out AI is really good at helping us create engaging experiences. So I actually think the classroom of the future looks remarkably like the classroom today, but you reverse what you're doing in it. And I think we could get a lot of the way there.

• 34:03Kevin Roose?

  Tell us what an engaging experience designed by an AI looks like, because I don't know that I would say I've had a lot of engaging experiences interacting with the bot so far.

• 34:12Speaker 06

  So I don't think it's about engaging experience with the bot. I think that it is about putting knowledge to use. The AI is remarkably good about taking theory and giving you practice from it. So if you say, I want to have an in-class activity for fifth graders to teach about entropy, and I've actually done this, it came up with a really cool idea of an in-class entropy activity involving balls and, you know, people standing still and running around. Actually tied really nicely into entropy and then it suggested a good classroom discussion that it actually built a game and coded the whole game for me that students could play with that activity. Now, I would want teachers to have input into that, but that's an active way of learning about a material that you learned outside of class and putting it to use in class and checking whether people understand the behavior.

• 34:53Speaker 06

  Having people discuss and having people act on it. So it's not about engaging with the AI. It's about having the AI help you figure out ways as an instructor to engage with the students.

• 35:02Kevin Roose?

  Yeah. Now, you have been using generative AI in your classrooms at Wharton since ChatGPT came out last year. What are some findings from that experience that might help other professors who are thinking about how to use this stuff in their classes?

• 35:18Speaker 06

  And I was actually using it a little bit before. So I was impressed enough by GPT three, which sort of wrote like a fifth grader. To have an assignment to sort of show them the future where they had to cheat with AI. And it was very funny because halfway through the cheating assignment, when half the students had turned in, GPT three point five came out and it definitely changed the game there. So I have all kinds of things that people do. I make AI mandatory. I teach an entrepreneurship class. So it's a little easier for me. If I was teaching English composition. I would probably be thinking a lot more about how do I do more in-class work where people are, you know, writing in blue books because you still need to learn to write and practice. And I can't flip the classroom that way. But in in terms of of my classes, it's been great. I actually literally require people to do at least one impossible assignment. That's in the syllabus now. So if you can't.

• 36:01Kevin Roose?

  What does that mean?

• 36:01Speaker 06

  So if you can't code, you have to write working software. If you've never done design work, I expect a fully built out graphic design product. It is literally, you have to do something that you think you cannot do. Every assignment has to be critiqued by at least three famous entrepreneurs through history. And that might sound like fun because it teaches them the how the AI works, but also is important because one of the defining characteristics of entrepreneurship is hubris. It is actually one of the number one predictors of entrepreneurial entry. And so things that break your hubris are actually quite useful. And the easiest way to do that is to have strong perspectives that do that. So it's let me increase the amount of work people do. It's let me push the kind of assignments they do. It's let me help, you know, adjust in a lot of ways. So AI mandatory works very well for the kind of classes I teach.

• 36:42Casey Newton?

  And we should say that when you you say that you're having your students talk with great entrepreneurs through through history, they are not actually speaking with the dead. They are going to the chat bot and say, sort of critique this in the the voice of Steve Jobs or something like that, right?

• 36:54Speaker 06

  Exactly, and it's one of several kinds of prompts that I give them, right? So there's I I give them tutor prompts, because otherwise they ask the AI to explain something like they're 10, and that's fine, but that's actually how good teaching works. The way good teaching works is you solicit explanations for people and you critique the explanations. You don't explain things to them and ask if they get it. So I give them prompts that we've created that actually do that kind of interaction. And I give them prompts that act as a mentor and help them with classroom problems and give them prompts that they can use for their teams to help do, you know, actual good team activities. So there is this kind of interesting role where I'm designing material that then interacts with them like a TA or a really good research assistant.

• 37:34Kevin Roose?

  In your experience, has has the introduction of generative AI into your classroom kind of changed the culture of your classroom at all? I mean, I I was recalling an episode that my colleagues at the at the Daily did earlier this year about basically ChatGPT in education. And they interviewed a professor who said that basically since ChatGPT came out, his whole attitude toward his students had changed, where it used to be that when he saw a surprising and skilled piece of student writing, he was delighted. It was like a cause for celebration. And now, you know, when he sees something that feels maybe a little too good for the student's ability level, he gets suspicious and and defensive and starts sort of saying, well, did they cheat on this using ChatGPT?

• 38:17Kevin Roose?

  It's just it just seems to have changed the the teacher-student relationship in that case. Are you finding that at all? Or maybe are students participating less in class because they know they can go, you know, ask a chat bot for clarification on something after they get back to their room? Or how is the culture of your classroom changed?

• 38:34Speaker 06

  I mean, it broke the old style culture, right? So I'm on the edge of a new technology, so I'm excited about that. I can understand why that's worrying. People don't raise their hands as much because one of the things we trust that people do in a classroom was show ignorance so we could explain things to them. Now, it's much better to not explain ignorance, show ignorance in front of 80 people. You'll ask the chatbot later about how to answer the question. It means that, you know, people always cheat, right? It's not a new thing. There's 20,000 people, at least prior to November, in Kenya, whose job was writing essays for people. This is not a new phenomenon, but now it's much more obvious.

• 39:06Casey Newton?

  Wait, what is this? Is this like a company? What is this company?

• 39:08Speaker 06

  Oh, no, it's not a single company. There's a paper that shows that estimates the number of people who are getting jobs writing essays, mostly for college students.

• 39:16Casey Newton?

  That's amazing.

• 39:17Speaker 06

  Yeah, it's pretty incredible. So cheating was pretty ubiquitous. Actually, it's been fascinating. Since the birth of the internet, the value of homework has dropped precipitously. I think cheating was already happening, we can ignore it, right? So this, again, another forcing issue. It forces us to grapple with actual changes that have already been happening in classroom environments and that we didn't have to worry about before. So it has changed the attitude. I would say that, you know, that that suspicion of writing is probably right, but I also no longer accept badly-written stuff in my classes. Why would I? And for every student that was a brilliant writer before, I had 18 students who were not brilliant writers. And some of them, English was their seventh language. Like, why should I expect them to write a beautiful essay and punish them or not based on their prose? So it does change things.

• 40:02Speaker 06

  We haven't figured it all out yet, but it is some positive as well as the negative.

• 40:06Casey Newton?

  Well, and I sort of want to pause on that because that's a very interesting point. What you're saying is that you used to get essays from kids that weren't particularly well-written, and you kind of give them a pass based on their individual circumstances. But now there is a tool that will instantly improve the quality of their prose, and so you just expect that they will use it, and to not do that is bad form.

• 40:27Speaker 06

  Well, not just that, I expect them to use it well. And so it turns out like a little bit of prompting knowledge goes a long way. And I'm, I require at least five prompts for everything they turn in. And they have to give me a paragraph reflection on the prompt. And if they want to use the AI for the paragraph reflection, they can. And then they have to tell me the prompts and reflect on that paragraph. But it's, it's hard to cheat on that piece. So the result is I want the writing to be different. It has to reflect their own writing. If I, if it has that chat G P D style where it says, and then in conclusion, I'm like, oh, come on, you didn't even cheat well.

• 40:56Kevin Roose?

  Now, you mentioned earlier that part of the reason you've been able to run all these experiments in your classes is because you teach entrepreneurship, which is sort of adjacent to AI in many ways. And, and you also have a very personal interest and affinity for this technology. What do you say to the, you know, 20 year English professor or the organic chemistry professor? Or the, the anthropologist who says, I actually don't want to make my class all about AI. I, I want to teach and I want my students to learn and I want them to show me their work, even if it's imperfect. And know that that reflects what's actually going on in their head and not some chatbots somewhere. What do you say to the professor who just says, you know, I, I don't really want to turn my classroom into an AI lab?

• 41:38Speaker 06

  And I, I think that's a huge number of people, right? And it makes complete sense. I think that stage one is recognizing that your homework broke. And that means that, you know, you may have to flip classrooms. You may have to hold people accountable with in-class exams, with having the wifi turned off. You know, your Chromebook in demo mode, there are ways of solving this problem in the short term. I think the bigger term, longer term problem is what does this all mean? What does this change about education? Now, I would actually argue in some ways, I think the only thing that carries us forward is expertise. And building expertise actually requires a lot of tedious fact learning and other material to get started. So I, I think that we're going to be able to justify some of these returns. But I think in the short term, it's acknowledging this thing is real.

• 42:22Speaker 06

  And then providing some guidance that, hey, I've been running up all of our assignments through AI. Here's the things that gets right and wrong. So just as you start to do it, you should recognize that it will give you wrong answers of these kind of chemistry problems. And then you flip the classroom a little bit and, and I think you could be okay in the short term.

• 42:39Kevin Roose?

  Right. We've talked a lot about how teachers and professors should be thinking and feeling about generative AI in the classroom. What about students? I mean, students, millions of them, are going back to school right now for the fall semester. I'm sure, you know, many of them have already been playing around with this stuff, but now they're confronting, you know, policies and restrictions. How should students be treating generative AI?

• 43:06Speaker 06

  I would demand clarity. I would demand clarity for what this means that AI is banned or acknowledged. Does this mean that I'm allowed to use AI to generate ideas? Could AI come with an outline that I work on? Can I ask for feedback from AI in my work? Because getting feedback is incredibly useful, and it's very good at providing feedback along the way. Am I allowed to use AI as a teammate? Can I ask the AI advice for something? You know, can I ask to explain why I got a question right or wrong? I think there is a request for clarity that's useful. But and I also think that, you know, the future AI that our students are going to graduate into is going to look very different than AI today. So I think the idea that we're teaching kids how to use AI is actually not that useful in and of itself. It's going to be much more self prompting. It's going to take away parts

• 43:48Speaker 06

  of work that we used to do before. So I think you are allowed as a student to ask for what does this mean, while being patient with your teachers that they haven't figured it out either, right? Nobody knows the answer.

• 43:59Kevin Roose?

  I'm curious, Ethan. One thing I hear a lot from educators when it comes to generative AI is this worry that it kind of flattens student creativity and output and effort that, you know, when everyone is a B plus writer, they're, and it's sort of producing kind of generic prose that is kind of like, sounds a little bit like Wikipedia almost. It's like you sort of sand down some of the edges of, you know, one student having a very different writing style, or another student communicating in a very different way. So I think there are a lot of worries, not just about what this is going to do to the classroom experience, but actually what it's going to do to the minds of students who are relying on this technology to, to help them think and write and work. Do you share any of that worry?

• 44:44Kevin Roose?

  Or sort of what are some of your worries about the long term effects of generative AI on students?

• 44:50Speaker 06

  I think we're going to lose a lot and I think we're going to need to figure out how to reconstruct that. I certainly think in the short term flattening is real. Another phenomenon that falling asleep at the wheel is real. We've had papers showing that people will tend to, when they use AI that's very good, they tend to not pay enough attention anymore and kind of disengage from the work itself. That's a real phenomenon. I mean, essays were useful, right? It's a shame to lose them as a homework assignment. Things are going to be lost. And I think some of this flattening effect is very true. So we need to teach people how to write with their own voice while still being able to use AI. And I think you've probably found as users, you can make AI do much more interesting things if you don't just do a generic prompt, right? The first prompt is always Wikipedia

• 45:32Speaker 06

  style, perfect English, but you can get to do some kind of neat things with some time. But I think it's part of that bigger issue. But by the time we have students at high school today, we have six or seven years before they're in the workforce, for those who are going to college, right? And so what does that mean for a moving target for the future is a bigger question. It does outsource some of our thinking, it's of our abilities in a way that people worry about with, you know, Google also and other phenomenon.

• 45:57Casey Newton?

  It's true. And yet I do want to say, as somebody who wrote essays in college at one point, a good number of those essays were written between the hours of, let's say, three and nine AM, the morning they were due, while I was, you know, hopped up on, you know, no-dose pills. And I'm not sure how much I learned during the experience. Now, on the whole, yes, essay writing was a huge part of my education, but, you know, not every student puts everything they have into every writing assignment.

• 46:23Speaker 06

  Absolutely. And I also want to apologize to the TA. In my English literature course in college, I took pass fail and never revealed to them it was pass fail. So I could get solid Cs. And every time they'd write these elaborate notes and apologize to me for giving me a bad grade, I'd be like, I wrote that in 25 minutes. So I totally get that, right? And I think that's another piece. It's like, we have to not be delusional about what has actually happened in education. And to go back to the issue, that's what the, I hope the gift of the AI piece is less the AI itself than this act of being deliberate in what we've reconstructed Hard Fork University of what parts of education matter. And we actually know a lot about that. And we have to figure out how to reconstruct those pieces.

• 47:00Kevin Roose?

  So back to Hard Fork University, our university that the three of us are starting.

• 47:04Casey Newton?

  Which, by the way, was recently voted to have the worst football team in the entire country.

• 47:08Kevin Roose?

  But the parties are amazing.

• 47:12Kevin Roose?

  Should we have an admissions essay? I mean, is that also one of the things that is on its way out?

• 47:18Speaker 06

  Oh, absolutely. That's done for. I have been enjoying having it. If you have not had the experience of asking it to write admissions essays, justification, it's amazing at, right? So one of my recent favorites was, explain how stubbing my toe in the fourth grade taught me about adversity and why I want to be a neurosurgeon. And it was just amazing. I filled the details about how I realized that something as small as nerves could make such a difference in my life. And it was just like, wow, this is great. I mean, a lot of stuff's going to break, right? And so we have to decide what we're doing rather than fighting a fighting retreat against the AI that as it takes thing after thing. We have to really rethink what we want to do with admissions and other policies.

• 47:58Kevin Roose?

  Well, I mean, what I would say is that you have to have at least one TikTok go viral if you want to compete in this economy. So that's what I would be looking for.

• 48:05Speaker 06

  It's time to dance.

• 48:06Kevin Roose?

  OK, so we won't have take home essays and we won't have, you know, assignments that students can easily plagiarize using ChatGPT. Are all of our assignments going to be in-class Blue Book essays? Like, what other types of work will we assign at Hard Fork University?

• 48:23Speaker 06

  So I think we need to divide the kind of work that we want to do. Learning English composition for our English 101 course, that's going to be a lot of, you know, maybe reading stuff, getting critical feedback outside of class with a combination of AI help and human help. And then in class, you'll be doing a lot of writing. It's going to be writing workshops, right? That's our intro courses. I'm hoping that as our classes get more advanced, you're taking your 21 levels, you're taking more applied classes, that the in-class activities start to become very interesting. So the ability of students to get things done, to push past the frontiers that used to be a 101 frontier is fascinating, right? So we're going to push the power of AI to get people to do more than before. They're not just going to do basic stuff, they're going to do advanced stuff. They're going to be 10 times more productive by the time they get out of the program.

• 49:09Kevin Roose?

  Sounds compelling. Sign me up. Last question, Ethan. What would it take to get you to leave your job at Wharton and join the faculty of Hard Fork University?

• 49:17Speaker 06

  So do you offer tenure?

• 49:21Kevin Roose?

  We're working on it.

• 49:23Speaker 06

  I'd credit a tenured university. I feel like, you know, these are my colleagues who joined startup companies here. I've done that already. So I'm willing to be visiting faculty, though, if the location's nice.

• 49:34Kevin Roose?

  It's a dingy podcast studio for now, but we are going to be expanding to a beautiful campus soon.

• 49:39Speaker 06

  Not too late for the metaverse for this one.

• 49:43Kevin Roose?

  All right, Ethan Malik, thank you so much for coming on.

• 49:44Speaker 06

  Thank you, Ethan. Thanks for having me.

• 49:50Kevin Roose?

  After the break, another wild week in the world of self-driving cars, and some thoughts from you, our listeners, on our interview from last week.

• 50:13Speaker 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.

• 50:47Speaker 07

  Your sales team is at risk of missing quota. Don't panic. Just ask Rippling AI. Built on your real-time people and business data, Rippling AI can pull metrics from Rippling and your CRM into a meeting-ready dashboard showing quota attainment, headcount trajectory, and revenue to quota by region. In seconds, you'll see what's behind your quota risk and fix it before it's missed. Head to rippling.ai/hardfork to get the only AI built to give you full visibility and take complex actions across your whole business. That's R-I-P-P-L-I-N-G dot AI slash hard fork. Sign up today.

• 51:21Speaker 04

  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.

• 51:57Casey Newton?

  Well, Kevin, if I learned one thing from last week's episode, it is that people have strong opinions about autonomous vehicles.

• 52:04Kevin Roose?

  They really do.

• 52:06Casey Newton?

  I would say that not before in the history of the show have we gotten as strong a reaction to any interview that we have done.

• 52:13Kevin Roose?

  It's been really wild. I mean, we've gotten like so many emails and DMs and replies on social media about this episode.

• 52:22Casey Newton?

  Friends, skeets, Instagram stories. We got it all, baby.

• 52:25Kevin Roose?

  I did not realize that this is the most polarizing issue in America. It really feels like we accidentally stumbled into like, you know, a K-pop fan army, like.

• 52:33Casey Newton?

  Yeah, and, and look, I think, you know, we found our way into this story because there is this group, Safe Street Rebel, that is going around and they're putting traffic cones on the hoods of autonomous vehicles, which has the effect of disabling them. And they're trying to do that to draw attention to concerns they have around autonomous vehicles and cars in general. And we thought, well, that's interesting. Let's have a conversation with them about that. I think what we heard back from listeners, though, is that the issues here run deeper and our listeners really wanted to get into it into much more detail on a bunch of subjects related to AVs.

• 53:05Kevin Roose?

  Yeah, it was a very polarizing segment. We got a lot of people saying like they loved the segment. We got a lot of people saying they hated the segment. We got people saying your guests were correct and you guys are wrong and you're, you're too pro AV and anti-transit. We also got people saying your guests did not make their points clearly and you should have picked someone better to make the case. So it was just kind of all over the map, but very, very strong feelings on all sides.

• 53:28Casey Newton?

  So we want to talk about listeners' opinions about the segment because a lot of listeners raised some really great points. But first, a really extraordinary amount of news happened in the AV realm, including in San Francisco, just over the past seven days. And so we thought, let's take a quick beat and just talk about what's been going on.

• 53:46Kevin Roose?

  So the first thing that happened was we did manage to get it into last week's episode. It happened just before we published the episode, which is that the CPUC, the agency in San Francisco that was voting on the fate of these pilot projects for driverless cars, voted to allow Cruise and Waymo to expand in San Francisco. They can now run these. AVs 24/7 and charge money for them.

• 54:10Casey Newton?

  And that's a big deal, right?

• 54:11Kevin Roose?

  Huge deal.

• 54:12Casey Newton?

  Because in order to hail one of these taxis before, you sort of had to have special access. You could only access it at night primarily. But now these are about to become just a fact of daily life in San Francisco in a way they were not before.

• 54:23Kevin Roose?

  Right, and these companies are now going to start expanding into many more cities. Cruise just announced it is starting to offer driverless rides in Charlotte, North Carolina. They're also expanding to cities in Texas. So this is going to be coming to, if not a city near you, then at least, you know, a city within driving distance of you very soon.

• 54:41Casey Newton?

  That's right, but there's another story we should talk about, which is that over the weekend, a group of 10 Cruise cars essentially came to a halt blocking traffic in the North Beach neighborhood of San Francisco.

• 54:56Kevin Roose?

  Yeah, this was an amazing story. So right after this big vote by the CPUC, this pileup sort of happens in North Beach. And at first, it sounds like maybe it was related to this music festival, Outside Lands, that's sort of one of the big festivals in San Francisco. And at first Cruise said, well, this was due to wireless connectivity issues. But then, as the week went on, it looks more and more like this was a case of pedestrian interference. And in fact, that's the new explanation that Cruise has for why all these cars stopped in the middle of the street.

• 55:28Casey Newton?

  What I love about these explanations is that neither one of them makes any sense to me, okay? When it comes to wireless connectivity, these Cruise cars are miles away from the Outside Lands music festival. So I'm sure a lot of people were posting their skeets and their threads at Outside Lands, but I don't understand how they were doing it in large enough volume to stop a car miles away, okay? Cruise eventually comes to the same conclusion, right? It says, oh yeah, I guess it wasn't wireless connectivity, which doesn't exactly inspire a lot of confidence. But then they come up with explanation number two, which is pedestrian interference. And Kevin, I would just like to ask you, what do you think pedestrian interference is? And how does it stop 10 cars from moving?

• 56:07Kevin Roose?

  Well, we don't know because Cruz has not said a ton more and we don't have like the footage from the cars themselves. They have said, they told us that it was not a cone-related stoppage. This was not the fault of our organizer friends from last week that we know of.

• 56:22Casey Newton?

  As a pedestrian, I've interfered with traffic and here's how I've done it. I've stepped in front of a car because I wanted to cross the street. And at most, this has affected one car, okay? I mean, maybe the car behind it has to slow down too, but when I interfere as a pedestrian, I'm stopping one car for five seconds. So I would just like to know a lot more from Cruz about how this alleged pedestrian managed to stop 10 cars from moving for many minutes on end.

• 56:45Kevin Roose?

  Right. But whatever the reason for the traffic jam, I think it's fair to say this was not like one of the worst things that could happen in an autonomous vehicle. And it was actually cleared pretty quickly. In fact, it lasted only about 15 minutes.

• 56:54Casey Newton?

  Well, I do think this is one of the worst things that could happen to our argument that autonomous vehicles are good. You know, we've been trying to make the case and then 10 of these things come to a dead standstill and I found the whole thing very inconvenient.

• 57:06Kevin Roose?

  Yeah, I think if we thought we had road rage problems before this, we have not seen like, if you're angry at someone for driving too slowly, you're going to be 10 times angrier if it's a robot driving too slowly or coming to a stop in front of you. But to my mind, the biggest story that has been published since we came out with our last episode was a story that ran in the San Francisco Standard with the title, "San Franciscans are having sex in robot taxis and nobody is talking about it." So this is a thing that I have been waiting to hear more about since these AVs launched, which was, you knew as soon as this happened, that people are going to start having sex inside the robot taxis. And the San Francisco Standard interviewed four people who claimed to have had sex inside

• 57:49Kevin Roose?

  Cruz AVs, including one source who they called Alex, which is not his real name, who they say claims to have performed at least six separate sex acts in robot taxis, ranging from impromptu makeout sessions to full-on sex no boundaries activities, a total of three times in a Cruz car.

• 58:08Casey Newton?

  First of all, are we considering a makeout session a sex act? That seems like a bit of a stretch to me.

• 58:14Kevin Roose?

  So I have questions about this.

• 58:16Casey Newton?

  An impromptu make out as opposed to one that's been scheduled to put on a Google Calendar?

• 58:20Kevin Roose?

  Oh, you don't put all your make out sessions on your Google Calendar?

• 58:23Casey Newton?

  Hey, I like to keep it loose, baby. Anyways.

• 58:25Kevin Roose?

  So I have questions about this.

• 58:26Casey Newton?

  All right.

• 58:26Kevin Roose?

  So we know that these cars have cameras inside them.

• 58:28Casey Newton?

  Yes.

• 58:29Kevin Roose?

  Right? Is some poor soul at Cruise headquarters just having to sift through hours of footage of horny people just climbing into these AVs and getting it on in the backseat?

• 58:38Casey Newton?

  You know, when we had the CEO of Cruise here, Kyle Vogt, we sort of asked him, like, isn't there a high potential for bad behavior inside these cars?

• 58:50Kevin Roose?

  Shenanigans.

• 58:50Casey Newton?

  Shenanigans, hijinks, antics, if you will, inside these cars because people, you know, because there's no Uber driver to sort of modulate people's behavior. And he said essentially just that, we have cameras in these cars. If anything gets out of control, we can look and see the cars. And it is amazing to me that San Franciscans are already saying, we don't care. Go ahead. Look all you want. There's a free show in the Cruise tonight.

• 59:13Kevin Roose?

  Yeah, I mean, there are lots of exhibitionists for whom the cameras are a feature, not a bug.

• 59:17Casey Newton?

  I want to say.

• 59:18Kevin Roose?

  Here's my other logistical question about this. So when we rode in a Cruise car, as you will remember.

• 59:23Casey Newton?

  Yeah.

• 59:23Kevin Roose?

  It would not move unless both of us had our seat belts on.

• 59:27Casey Newton?

  It was very firm on this point.

• 59:28Kevin Roose?

  So presumably, these people who are having sex inside the robo taxis are doing it with their seat belts on. And I just want to get your opinion on the logistics of that. How does that work?

• 59:40Casey Newton?

  This, we have a saying in this community, safety first, okay? If you want to have an impromptu make out session in a car, you need to take care of your partner. And that means that you do need to keep your seatbelt on. And look, those seatbelts have a lot of give, okay? I don't know about you, but I've been in a seatbelt. I've reached all the way over to the other side of the car to grab something from the driver's side. So yes, it's absolutely possible, and it's the way to do it.

• 1:00:01Kevin Roose?

  So that is the big news from the last week when it comes to autonomous vehicles.

• 1:00:06Casey Newton?

  Yeah.

• 1:00:06Kevin Roose?

  Now, let's talk about some of the things our listeners are feeling in response to last week's segment.

• 1:00:12Casey Newton?

  Yeah. So, you know, I want to make some room to get some of the criticisms in here that we got. And look, you know, I'll be the first to say, I'm not an expert on transit issues or autonomous. Vehicles issue and part of what we do on this show is we just lead with our curiosity. We bring in people, we ask them questions and yeah, we express our opinions, but we're like open to other ideas, right? And so, right now, we want to be open to our listeners' ideas.

• 1:00:36Kevin Roose?

  Right. This is something that people care a lot about and I want to respond to some of the notes we got about last week's segment and I think we should kind of sort it into a few flavors of criticism.

• 1:00:47Casey Newton?

  Yeah.

• 1:00:47Kevin Roose?

  The first one that we got a lot of responses on was basically listeners who were anti-self-driving car, who were on the side of the organizers we interviewed, but who didn't feel like they were the best ambassadors for that position or that we fully explored their point of view.

• 1:01:03Casey Newton?

  And I think, you know, there was an element that I saw in a lot of the messages, both in email and in social media, and it was basically that we didn't engage with the real thrust of the activist argument, which is that we need fewer cars on the road, period. And of course, there's a big tie-in between that idea and climate change, which I also think that readers think we were not taking that seriously enough. And if we were serious about climate change, we would want to be saying like, yes, absolutely, get all these cars off the road. So Kevin, what is your take on this idea that we need fewer cars on the road, period?

• 1:01:33Kevin Roose?

  Well, I think I share that view. I think car congestion and pollution are huge problems and as anyone who has ever tried to drive around San Francisco can tell you, it would be a lot more pleasant here if there were, let's say, 50% fewer cars on the roads. Now, I think there are some issues with that, namely, there are just some people for whom, you know, cars are a sort of a necessary fact of life. You know, we got a lot of notes from parents who said, you know, I support transit, but there is no way that I could lug my kids and all of their stuff around on an e-bike or on the bus, or what if you don't live on a transit route?

• 1:02:10Casey Newton?

  I got a note from somebody who said that cars are important if you have a disability and they make life much easier for folks who may struggle to use transit if they have a disability.

• 1:02:19Kevin Roose?

  Totally. So I think there are a lot of people who just need cars or are attached to cars as their daily mode of transportation, for whom switching those people over to mass transit is just not going to be very practical.

• 1:02:31Casey Newton?

  Yeah. Now, that said, could I spend more time? I'm imagining a car-free future. Yeah, like, I should. I grew up in America in the 80s and 90s. I'm extremely car-pilled and like, not by choice. It's just kind of the oxygen that I've been breathing. But I can tell you, the reason I live in San Francisco and the reason I love it so much is because I can walk almost anywhere. And I walk basically wherever I can. I love taking transit. I got rid of my car. So, in part, I really am trying to build this future, but we have a long way to go.

• 1:02:56Kevin Roose?

  Right. And the self-driving car companies would say that actually, self-driving cars help to get cars off the road because we know from a study that was done about 10 years ago that the typical car utilization is something like 5%. Which means that if you own a car, it sits there in your driveway or in a parking spot roughly 95% of the time, right? That is a horrible utilization rate. But with a driverless car system, that car can go pick up other people and take them where they need to go. You can get to, much closer to 100% utilization, which means that you theoretically would need fewer cars to do the same number of trips in a city.

• 1:03:35Casey Newton?

  Which would also mean that you need fewer parking garages and parking lots. And maybe we could use that to build more housing, right? So, at least, these are at least some ideas that are out there that I think are worth mentioning.

• 1:03:43Kevin Roose?

  Right. And the organizers we talked to last week, their point in response to that was, well, that's the same thing people said about Uber and Lyft, right? That if you have these ride-sharing services, people won't need to own cars. And anecdotally, I do know some people who have gotten rid of their cars because of Uber and Lyft.

• 1:03:59Casey Newton?

  I am one of those people.

• 1:04:01Kevin Roose?

  So, it does happen. But their point that Uber and Lyft have not actually decreased the amount of traffic congestion in cities is a good one.

• 1:04:09Casey Newton?

  And it's also true. And I agree, it's a good point. All right, what else did people tell us, Kevin?

• 1:04:13Kevin Roose?

  The other thing that people asked about was how good the data is on the safety of autonomous vehicles compared to human-driven cars.

• 1:04:21Casey Newton?

  Yeah, and this is a nerdy point, but I do think it's worth getting into it.

• 1:04:24Kevin Roose?

  Totally. Because I said on the show last week that the research that we have suggests that autonomous vehicles are substantially safer than human-driven cars. And that is true. We do have research that shows that that is the case. But there's a big asterisk there that I maybe should have spent more time talking about last week, which is that this data is not very good and it's not very complete. So, one of the issues with collecting data about autonomous vehicle safety is that there just aren't that many self-driving cars, right? In San Francisco, there have been several hundred of them between Cruise and Waymo compared to... Hundreds of thousands of human driven cars.

• 1:05:04Kevin Roose?

  And so if you're comparing the data between how often these various types of cars get into collisions or cause pedestrian injuries or something like that, you're going to just be comparing from two very different bases. And so we just don't have the kind of large scale data on autonomous vehicles that you would need to be able to make really good, really reliable comparisons.

• 1:05:27Casey Newton?

  Yeah, so that's just something that we're going to have to keep an eye on.

• 1:05:30Kevin Roose?

  Totally. But there is a report that compares human drivers to AVs. Cruise put out a report this year that said that in their first 1 million miles of self-driving car data, they found that their autonomous vehicles had 54% fewer collisions than human rideshare drivers, 92% fewer collisions where the autonomous vehicle was the primary contributor to the collision, and 73% fewer collisions with a meaningful risk of injury. It also said that of the collisions that it did encounter in its first million miles of autonomous driving, 94% of them were caused by the other party.

• 1:06:08Casey Newton?

  All right. And do we trust this data? Is this just being reported by the companies themselves? Or is this sort of using something that we can check against, like, I don't know, the Department of Motor Vehicles records or something?

• 1:06:17Kevin Roose?

  So it's a good question because this is obviously self-reported data from these companies, which have an interest in making their vehicles look very safe. But in Cruise's case, they partnered with the University of Michigan Transportation Research Institute and the Virginia Tech Transportation Institute, which are independent academic settings that are studying this data too. So I give that a little more credibility. So I think we should say upfront that these studies, they are conducted with the blessing and the data from these companies. And it's also very hard to compare, say, data about self-driving cars in San Francisco to national data because it's so different, right? It's driving in San Francisco in a crowded city with lots of other cars versus, you know-

• 1:07:01Casey Newton?

  Like driving on the highway.

• 1:07:02Kevin Roose?

  Driving on the highway or driving in a small town in rural America somewhere.

• 1:07:07Casey Newton?

  You could write a great song about that.

• 1:07:10Kevin Roose?

  Yeah, so it's just, it's not quite apples to apples and we just don't have enough good data about autonomous vehicles. But I would say the early indications in San Francisco are that these cars are getting into collisions pretty infrequently. And when they do, that it is often the result of the human in the other car, not the autonomous vehicle itself.

• 1:07:30Casey Newton?

  What I'm taking away from this is that the early data looks good, but let's just be careful with what sorts of safety claims we're making around AVs as we continue to collect this kind of data.

• 1:07:40Kevin Roose?

  Yes, absolutely.

• 1:07:41Casey Newton?

  All right. Let me read another piece of feedback that we got. This is from a listener named Dirk in Rhode Island, and he writes to us, quote, "I don't think you realize that Casey and Kevin came off as Luddites, not the guests. Car-free cities really are the future, as seen by the younger generation. The next generation doesn't want the SFF today, they want Zurich, Paris, Madrid to shape the future of the bay. The AV itself is an afterthought, and you may be inside of an industry bubble. It's not how will AVs change our lives, it's how can I live in a way that I would rarely slash never need to use an AV?" So what do you think, Kevin? Are we just sort of not dreaming big enough when we talk about AVs as playing a major role in the city of the future?

• 1:08:19Kevin Roose?

  I think there's some truth to that. I mean, I have been to cities outside the US where there is much more transit.

• 1:08:25Casey Newton?

  Well, la di da, Kevin. Tell us more about your international travels.

• 1:08:29Kevin Roose?

  Can I tell you, the Swiss trains and the tram system there is unbelievable. So good, very efficient.

• 1:08:35Casey Newton?

  I will also say, I've never seen a more expensive country in my entire life than Switzerland. These people are charging 40 euros for a cocktail? Come on.

• 1:08:42Kevin Roose?

  So I'm torn here because on one hand, like, I do believe that getting rid of many cars on our streets would make them safer, would make city life more pleasant, and would ultimately be a better future. At the same time, people are very wedded to their cars, right? We see this all the time. People, even in cities with good mass transit, want cars, for whatever reason. Maybe they don't want to wait around for the bus, maybe they need to take their kids, maybe they work a job where they have to transport large things and they can't really do that on mass transit. People are just really, really connected to cars and car Our culture, it is sort of a fundamentally American trope, but I think it is probably

• 1:09:24Kevin Roose?

  true. And so I'm of two minds here because I think there's an argument to be made that, yes, we should have many fewer cars. And in order to sort of green light autonomous vehicles, we should make sure first that they actually are going to reduce the number of cars on the road. But I don't think people are going to easily give up cars. So in the meantime, shouldn't we try to make them safer and greener by making them, you know, electric and autonomous? To me, it feels a little bit like the debate over eating meat where you have people who say, you know, we need to eat a lot less meat in America. And the way to accomplish that is by convincing millions of Americans to switch to a plant-based diet. And then you have the technology-driven approach, which is, well, let's assume that people are

• 1:10:07Kevin Roose?

  not going to give up meat easily. And let's try to make meat that has less of an impact on animals and on the environment. Let's try to grow it in a lab so that people can still have their hamburgers, but it just doesn't involve factory farms.

• 1:10:20Casey Newton?

  You know, what I think all of these sort of similar issues come down to is that if you are an activist who's rallying around this issue, or you're just a citizen who believes in one of these ideas, I think a skill that you want to develop is the ability to show people a path from how to get from here to there, right? And sometimes just talking about this radical vision that seems so different from what we live in today can be very effective, right? It sort of bumps that overton window a little bit closer to your side. And that's a good thing. But the risk of it is that you do sometimes make people, and I think this happened to me on our last episode, kind of throw up our hands and say, this feels like kind of unrealistic to me. Like, if you're going to tell me that we're going to live in this sort of Jetsons future, I needed to sketch out a few things first that are

• 1:11:05Casey Newton?

  going to get us from today to there. So at the same time, what I'll say is, I am just going to take the note to be more open-minded about this.

• 1:11:14Kevin Roose?

  Yeah, I'll take that note, too. But I will say that I think there's an issue here of making the perfect the enemy of the good, right? I think if we could get to a place where, you know, say 10% of the vehicles in San Francisco on a daily basis were Self-driving, they were safe, they were electric, and they were not getting stuck in intersections in North Beach. I think that actually would be a substantial safety improvement over the status quo. And so I think the people who are the hardliners on this issue should maybe allow for the possibility that transitioning everyone off cars is not going to be a realistic goal, at least in the short term, and should look for some sort of incremental wins along the way.

• 1:11:54Casey Newton?

  Because if nothing else, these AVs can be a really fun place to have sex.

• 1:12:00Kevin Roose?

  So that is, I would say, a fitting follow-up to last week's episode. Thank you to everyone who wrote in, even the guy who told us to go suck a tailpipe or lick an EV battery if you love cars so much.

• 1:12:14Casey Newton?

  And now, is that considered a sex act inside one of these AVs? Because, no. But seriously, look, I love hearing from our listeners, even when we make them mad. And one of the things I like about our show is that it actually can be a dialogue. If we say something and you hate it, tell us, and maybe we'll talk about it more, and maybe you'll shift our views.

• 1:12:31Kevin Roose?

  Yeah, please don't put cones on our heads.

• 1:12:32Casey Newton?

  Don't put cones on our heads.

• 1:13:04Speaker 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.

• 1:13:38Speaker 07

  Your sales team is at risk of missing quota. Don't panic, just ask Rippling AI. Built on your real time people and business data, Rippling AI can pull metrics from Rippling and your CRM into a meeting-ready dashboard showing quota attainment, headcount trajectory, and revenue to quota by region. In seconds, you'll see what's behind your quota risk and fix it before it's missed. Head to rippling.ai/hardfork to get the only AI built to give you full visibility and take complex actions across your whole business. That's R-I-P-P-L-I-N-G dot AI slash hard fork. Sign up today.

• 1:14:11Speaker 04

  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.

• 1:14:48Casey Newton?

  Hey, before we go, Hard Fork is hiring. We're looking for a video producer to help us bring this show to YouTube. So if you or someone you know does that kind of thing, let us know. We're [email protected]. Hard Fork is produced by Rachel Cohn and Davis Land. We're edited by Jen Poyant. This episode was fact-checked by Caitlin Love. Today's show was engineered by Sophia Landman. Original music by Dan Powell, Marian Lozano, and Rowan Nemisto. Special thanks to Paula Schuman, Puywink Tam, David McCraw, Noel Galogly, Kate LaPreste, and Jeffrey Miranda. You can email us at [email protected] to let us know what Kevin said this week

• 1:15:33Casey Newton?

  that you think was really stupid and needs a whole segment about the show to talk about.

• 1:15:39Casey Newton?

  Please, no more.

• 1:16:21Speaker 08

  You can't undo everything in life, but you can undo prediabetes. More than 2 in 5 adults in the US have prediabetes. Most don't know they have it. But with early diagnosis and action, you can delay or even prevent type 2 diabetes. Visit doihaveprediabetes.org to take the 1-minute risk test and start undoing the things you can. Brought to you by the Ad Council and the Centers for Disease Control and Prevention.