00:00:11
Speaker 1: Taylor, Rie, Kyle, I've been obsessed with heists in the AI era. And heists gone wrong are, of course, some of the great comic gold. Has anyone been following the stolen NVIDIA trucks?
00:00:21
Speaker 2: No.
00:00:22
Speaker 3: What's going on?
00:00:23
Speaker 4: It wasn't me, if that's what you're asking.
00:00:26
Speaker 1: So there are a couple of trucks with NVIDIA branding on them that some thieves thought were full of NVIDIA chips, which.
00:00:33
Speaker 5: Of course are very, very valuable.
00:00:36
Speaker 1: And so the trucks were stolen, but it happened they were just full of sand. And not the sand that's used to make chips. The trucks belong to Plus AI, who makes self-driving trucks. And in order to make self-driving trucks good, you have to simulate cargo loads. So they're doing that with tons of sand, 20,000 tons per truck. They also happen to have a partnership with NVIDIA. So the NVIDIA logos were just on the truck to kind of flex about the NVIDIA partnership, like the NASA logo on my computer.
00:01:09
Speaker 3: Good advertising.
00:01:10
Speaker 4: If I see someone selling sand on eBay, I will know where it came from.
00:01:14
Speaker 1: Well, actually, Plus AI managed to recover the trailers and all 40,000 tons of sand.
00:01:20
Speaker 5: So happy ending.
00:01:21
Speaker 2: Happy ending.
00:01:26
Speaker 5: Welcome to Tech Stuff.
00:01:27
Speaker 1: I'm Oz Veloshian, and this is The Week in Tech, where I'm joined by the world's most plugged-in reporters to.
00:01:32
Speaker 5: Break down what's really happening in tech right now.
00:01:35
Speaker 1: Today, we're joined by Reid Abregotti, tech editor for Semaphore and host of the new podcast, as of this week, Future Proof, Taylor Lorenz of UserMag, and back from leave, Kyle Chaker, who writes the Infinite Scroll column for The New Yorker.
00:01:48
Speaker 4: Welcome, all.
00:01:49
Speaker 2: Thanks for having us.
00:01:50
Speaker 5: Reid, welcome to the podcasting ranks.
00:01:53
Speaker 2: Can't hear you, Reid.
00:01:54
Speaker 5: Oh, we can't hear you?
00:01:57
Speaker 3: First you have to turn your mic on.
00:01:59
Speaker 4: I'm learning all the rules now.
00:02:01
Speaker 5: Don't mute your mic.
00:02:05
Speaker 1: Well, starting off with you, I mean, first, just give us a brief plug for the new podcast, but then I want to ask you what's going on in space.
00:02:10
Speaker 4: Yeah, well, it's called Future Proof. And, you know, we're going to interview people from all areas of the tech industry and go deep on on topics and try to teach people something. It'll be newsy, journalistic, but also really go deep on stuff and help listeners understand.
00:02:26
Speaker 5: I'm excited to check it out.
00:02:29
Speaker 1: This week's story for you feels a bit like the plot of a 90s Bond movie involving the US competing with Russia and China to build nuclear powered permanent bases.
00:02:39
Speaker 5: On the moon. What's going on? I mean, is this like a fantasy or?
00:02:44
Speaker 4: It depends on who you ask. I personally like, you know, I think everything in the space industry feels like fantasy until it actually happens. I remember when I first came out to the tech industry, SpaceX was sort of just becoming a thing. And people were talking about satellites, you know, in space providing internet. And really smart people told me that was completely impossible. It would never happen. Now I own, you know, two Starlinks. Um, So I just take the timelines for these things and I just extend them in my mind, but I don't dismiss anything outright. I mean, nothing here is totally impossible, right? But, you know, does it make economic sense? Will it make economic sense in the near future? That I don't, I just don't know.
00:03:28
Speaker 1: I think the US has tried it once before and Russia has been sending nuclear reactors into space since the 60s. But somehow this kind of new space race thing is particularly fascinating.
00:03:38
Speaker 5: Kyle, have you been following it?
00:03:40
Speaker 6: Just as far as the data centers in space. I mean, I think that whole concept is really appealing to me. And it kind of seems like the only answer. If we're kind of having all these fights over data centers on Earth, you know, maybe it would be better to put them in space and have them powered in some other way.
00:04:00
Speaker 3: So yeah, I don't know.
00:04:01
Speaker 6: I think it reads right about the timelines extending longer than you might think.
00:04:05
Speaker 4: But I mean, Kyle, do you know what a space... I've actually looked into this recently. A space data center is not like a floating data center. It is basically, you have a bunch of satellites that have a GPU or two or a rack, and they're spread out very far apart, and they're connected by lasers. So they communicate with each other by lasers instead of on Earth through fiber optic cable or copper, right? And because.
00:04:32
Speaker 6: It's not a Death Star. It's not like one floating massive data center.
00:04:36
Speaker 3: Right.
00:04:37
Speaker 4: But because, you know, there's no air in space, you can't cool these things in the same way you cool them on Earth. So the idea is that you'll somehow pull the heat away from these GPUs and dissipate it on a massive... They're going to essentially like a massive heat sink that's about the size of a tennis court. I mean, it is... like really kind of just the craziest concept you can possibly imagine. But it's not impossible. I mean, but I don't know if I agree that that's going to be the thing that happens before we just get our act together on Earth and build some data centers already.
00:05:11
Speaker 1: Yeah, I mean, the concept of space junk is kind of interesting. I mean, sending up this tennis court, or I think in some cases even football pitch or field size, kind of non-convective heat dispersing panels into space. I mean, you have to wonder... what the implications of all that stuff floating around in orbit are. And I think some of this, when I first saw the story, the headline about the nuclear reactors in space, I kind of assumed somehow it had to do with power generation for AI. But to my great surprise, it wasn't about that. It was about, go ahead, Taylor.
00:05:41
Speaker 2: I'm going to start a business that is a space junk removal cleanup business, because I think this is going to be huge business Going forward, I mean, we're about to have something like a million satellites or Elon wants to scale to a million Starlink satellites. And I think if we put data centers up there too, yeah, there's just going to be a lot of debris. And I think that poses risks.
00:06:05
Speaker 5: Especially when it's nuclear.
00:06:07
Speaker 3: I mean, right.
00:06:08
Speaker 4: It's kind of, well, nuclear doesn't have that much risk in space. I mean, you know, nuclear meltdown in space isn't... Unless it falls to Earth. Well, yeah, it'll burn up before it falls to earth. But I mean, what happens to these giant, you know, heat dissipating football field or tennis court size things if like a little, a little speck runs into it and puts a hole in it? Like, does it still work? And then if something breaks, they just drop it down and let it burn in the atmosphere, which is I'm sure horrifying to people who are here trying to get their hands on like a single GPU right now. But I thought the nuclear reactor thing is like, you got to build those on the moon, right? I mean, that was what you were talking about. Because if you want to send Mars missions... Staging post.
00:06:48
Speaker 3: Yeah.
00:06:49
Speaker 4: Is that what... That's what you were talking about, right?
00:06:50
Speaker 5: Yeah, exactly.
00:06:51
Speaker 1: And so the current like solar power cannot generate enough power on the moon to have actually warm habitable bases.
00:07:01
Speaker 4: Right.
00:07:01
Speaker 2: So.
00:07:01
Speaker 1: And China and Russia are collaborating. Russia have the best space nuclear technology. China, I guess, have good rocketry. So the new Cold War in space is on.
00:07:10
Speaker 4: Do you guys watch For All Mankind? Just out of curiosity, did you watch that show? No, I missed it. I mean, I think if you're interested in this topic, you've got to watch For All Mankind. Just go. It's on Apple TV. Go watch it. It's a sort of alternative history show about what would have happened if the if the Cold War sort of evolved a little bit differently and space just continued to be the thing that we put all of our, you know, human effort and ingenuity into. And like, I don't want to spoil the show for you guys, but you know, a lot of interesting stuff happens in space on a timeline that is not, that is a lot more accelerated than we have today. It's a really fun show. I think it's, you know, I've enjoyed it.
00:07:54
Speaker 3: Do we need more space propaganda?
00:07:56
Speaker 6: Like pro-space propaganda? pro-space content to encourage us all.
00:08:02
Speaker 4: It beats the propaganda we have right now. I mean, I'll take that over all the current propaganda.
00:08:09
Speaker 1: I've got a NASA sticker on my computer. What's the take on Jared Eisenman? I mean, the head of NASA, Trump had initially rejected him during the Trump-Elon Fodor, but then changed his mind and Eisenman was confirmed, obviously, a private sector billionaire. What's he doing at NASA? And what's the public-private relationship in space in the US today, Reid?
00:08:29
Speaker 4: Well, I mean, he's certainly helping Elon, right? There was a lot of tension with Elon and the Biden administration. And I think Jared's helping Elon. There's less tension between SpaceX and NASA. But, you know, I haven't heard any sort of controversies about him. He seems to be, you know, doing a fine job. We'll see what the effects of his term are. But I mean, the general trend, I think, in NASA is like more and more of this is becoming privatized. You know, there's good things and bad things about that, right? I think the public interest mission of space that I think we're all used to from our childhood of, you know, space shuttle going up and doing all these science experiments. I mean, some of that is for sure going to be going to be lost when the private sector sort of takes over. And I think that's kind of sad. But at the same time, it's incredible, the stuff that has been accomplished. I think it's amazing, not to mention just having a Starlink on my camper van in California. That's like the ultimate life.
00:09:25
Speaker 1: So I do want to just note that on the kind of safety issue with the nuclear reactors in space, there was a Russian nuclear reactor thing that came down to Earth. And US intelligence say that five people were killed. The Russians deny this. But just before we go to the break, Taylor, I mean, this idea of kind of cleaning space junk and including nuclear space junk, have you met any companies in this space?
00:09:51
Speaker 2: Yeah, I talked to Orbital, which is, you know, seeking to put the data centers in space recently and visited their offices in LA. So I do think that there's sort of like a space race with a lot of different startups right now trying to build out infrastructure in that realm.
00:10:07
Speaker 5: What did you learn at Orbital?
00:10:09
Speaker 2: What they're doing is cool. I mean, I think it's just like they're up against a lot of competition, as Reid said, too. It's like everything seems impossible until it somebody does it. And I personally think that the data center fight is not going away anytime soon, and we're going to need more compute. So yeah, if they can put them in space.
00:10:25
Speaker 3: Why not?
00:10:26
Speaker 4: Well, what's interesting about Orbital is the founder was the scooter guy, right? So he had a scooter startup in San Francisco. And then he kind of worked for Ford, I think.
00:10:36
Speaker 5: And then he.
00:10:37
Speaker 2: He sold it.
00:10:38
Speaker 4: And now he's like, I'm going to, I'm going to just make a space data center company because that's the obvious next move.
00:10:44
Speaker 2: Well, the reason why is because he has a lot of connection in manufacturing and in China. So I do think that like. building a scooter company ironically kind of prepared him to just do some sort of hardware startup generally.
00:10:58
Speaker 4: Yeah, I was being tongue-in-cheek, but actually I think that there is a point to be made, which is that the whole idea of SpaceX and Blue Origin and these companies to get infrastructure up into space is that you sort of enable a new crop of entrepreneur to work on space-related ideas and companies and So it's like, if you think of it as like all this infrastructure as a platform, it lets this guy who was a scooter entrepreneur, all of a sudden be a space entrepreneur, which never would have been possible if you didn't have the, you know, SpaceX rockets to get his equipment up into space, right? That he was never going to start a rocket company, but now he can start a space company. And I think that's, I think that's actually kind of a profound point. He, I talked to him too. He doesn't even want to go to space. I was like, do you want to go to, he's like, no, I have no interest in going to space. I'm like, that's, I think that's pretty profound.
00:11:52
Speaker 2: I mean, being in space kind of sucks. It's cold up there. It's like living in a submarine.
00:11:58
Speaker 3: Until you get the nuclear reactors on the moon.
00:12:00
Speaker 2: Right, right.
00:12:02
Speaker 1: To warm your hands by the warmth of a nuclear reactor on the moon. I have to say, there's always that classic question, what would you like the first line of your obituary to be? And I guess he was never going to build rockets is one of them. When we come back, who pays when chatbots lead you astray?
00:12:31
Speaker 5: Welcome back.
00:12:32
Speaker 1: Taylor, there was a piece in Futurism this week that there's a kind of golden oldie quality to it, a kind of tech made me do it story. I remember there was that episode of The Office where Michael Scott drives his car into the lake on purpose because GPS told him to. But nowadays, it's not GPS, but AI and cloud in particular that's leading not drivers, but hikers astray.
00:12:56
Speaker 2: Yeah, I actually read about this in the New York Times. They did a good story on it. But basically, this guy was hiking in British Columbia and he went off the trail. He used Claude as part of his planning process and ultimately had to be rescued off the side of a cliff because he was not a qualified enough hiker to, you know, to get himself out of the situation he had encountered. used. I don't think this is an AI problem necessarily. I actually ended up literally last week or two weekends ago when I was back in LA, I went to go on a hike and ended up in this like burned out section of Altadena where there was like no water, the trail was a mess. And I had read about it online and And even like looked it up, you know, ahead of time. And it's just that that route after the fire had changed a year ago and all this stuff. Of course, my friends were like, why wouldn't you use all trails? I think all trails is sort of the standard for this. And if you go anywhere else outside the internet, you're sort of risking it.
00:14:00
Speaker 1: Kyle, you nodded sagely about all trails. You seem like a trail guy.
00:14:04
Speaker 6: I'm not, which is the whole problem. I am not a hiking guy. I think I would end up in a bad situation. But I think if you have ever explored all trails, you know why relying on AI for your hiking knowledge is a terrible idea. All trails is the most human... intricate detailed set of commentary and data that you can possibly have i think in an app because it'll be like okay a hundred feet to your right like don't take this trail look for the rock and go the other way or if you aren't an experienced hiker maybe don't even bother uh so like all trails i don't know i want a person directly to tell me in their own words what I should and shouldn't do on one of these hiking trails.
00:14:54
Speaker 2: And a person recently.
00:14:55
Speaker 3: Yeah, yeah, it changes all the time.
00:14:58
Speaker 2: Like the hike that I went on a couple weeks ago, like it was rated really, people had blogged this, you know, like it was just not up-to-date information. And so, yeah, I just think that like, There's a lot of people interested in outdoors, but you need to rely on, this is like an all trails ad. Like, I think like, you know, he could have not used Claude and I don't know that he wouldn't have ended up in the same situation, especially when you're going off the trail, you know?
00:15:25
Speaker 1: He got rescued, sorry to interrupt, and the rescue manager said, Claude has never been in that specific area, has never been on that route, and has never been in the outdoors. Is this like a sticking rocks to the pizza with glue type story?
00:15:37
Speaker 5: Or is it like, what's the, not so much, Taylor?
00:15:39
Speaker 2: Yeah, and that was so silly, too. Like, I mean, I guess, like, especially Futurism is a website that, like, loves to write up this stuff as, like, it's just rage bait content to kind of be like, ha ha, you idiot, you know, look at you. And to me, it's like, look, it's a real problem with accuracy. I think this shows the value actually of these human curated, like very human updated apps. Like that's what it shows the value of. It's not that AI is inherently unreliable. I would say 99.999% of things that you ask AI and including that, like Google, you know, Gemini, I guess, review that had suggested you put glue on your pizza. And that was from a Reddit thread, you know, Early iterations of AI were certainly less reliable. They're getting smarter and smarter. Again, it just has to do with the fact that when you're going out into the physical world, hiking specifically, until we get drones, you know, going out and scouting these trails for us, I think, like, something like all trails is just the way to go.
00:16:34
Speaker 3: Or the humanoid robots.
00:16:37
Speaker 2: Right, we send the humanoid robots out to go chart the trails.
00:16:41
Speaker 6: But it was striking to me, like, Hiking is a subjective thing. Difficulty is subjective. How much your body can take is subjective. AI, if you're not an experienced hiker, AI is not going to be able to tell you what you can handle in this situation. So maybe in this case, don't ask Claude, ask your friends.
00:17:03
Speaker 4: In my experience, I don't know if you guys go backpacking and stuff, but if something really bad doesn't happen where you think you might die, it's kind of a boring trip. Those are the things that you always remember, you know?
00:17:14
Speaker 3: I think.
00:17:15
Speaker 2: Okay, I am not trying to go on the hikes that Reid is going on. I love hiking a leisurely trail in the woods. I think also, you know, this is a teenager. I think he was 16 years old. So I think there's also just teenagers don't traditionally have a great track record of judging risk and... Potential for bodily harm.
00:17:35
Speaker 4: Yeah, people have always been doing dumb stuff.
00:17:37
Speaker 1: There is something in the automation bias, like AI psychosis adjacent thing here. I got my green fingers burnt earlier this year when I really went to town on pruning my roses with the encouragement of Gemini. And actually, I think I ended up killing them and I had to confess to the landlord. So I feel like you can lean in too hard to the reassuring advice.
00:17:57
Speaker 5: Of AI, certainly.
00:17:59
Speaker 1: The story that kind of sits behind this story, for me, though, is the kind of AI... responsibility and liability story. Taylor, I think this has been also in the news this week in terms of how insurance payments work out when AI does harm and who's held liable, like AI hugging face episode, for example, like in that case, hugging face didn't sue OpenAI, but if they had, who would have had the liability in that situation?
00:18:25
Speaker 2: Well, it's a company hacking another company. So the company has a liability. It intentionally hacked the other company.
00:18:33
Speaker 4: Yeah, exactly. I think the question though, there is an, I think there is a really interesting legal question that's going to have to be answered, which is like, let's say you're not open AI and there's got to be some cases where you could end up accidentally doing harm, right?
00:18:48
Speaker 2: And then I think- Here's the question, I think, or not to interrupt, but just to finish, or I think this was like the second part maybe that you wanted to get at, but like, here's the question that I think we're going to have to answer is when we move into this more agentic era of the internet, who is responsible for the agent's actions? And with OpenAI, it is literally their model. Of course, they're responsible. This is a test that they were doing. But if I use Grokbot, for instance, to go out into the world and do a lot of things for me, automate my lifestyle, and I end up causing harm to someone or... I lose, you know, banks are super, super concerned about this. I was talking to somebody in the banking industry yesterday about it, because if my grok bot goes out and spends lots of money, I am probably going to be inclined to declare that as fraud. And, you know, the bank will suddenly be held liable and banks are lobbying right now to say, hey, hey, hey, no, no, no, no. It's grok that's liable. So there's all these questions around agents. And I do think that there's a much more compelling case for liability for companies that are building agents to be liable for their agents. I think it just is a question too of how are you using that agent? Did you jailbreak the agent? Are you using it within terms of use? And I just think those are battles that we're going to see.
00:19:58
Speaker 4: Or what if we use open source, right? I think that's going to be another one where it's like, well, what if I'm using open source agents or open weight models?
00:20:04
Speaker 2: Then you're liable, right?
00:20:06
Speaker 4: But what if it's, let's say you get sued for something you did, And now there's the discovery process and you see all your prompts and your prompts show that you were acting in good faith and you reasonably thought that, you know, there was not.
00:20:18
Speaker 2: Well, you can unintentionally do harm to somebody else. And that's taken into account in terms of the legal fight. But if I build software that's my personal boutique software that I'm using and I use it to do something, maybe it was unintentional harm. And again, that can be taken into consideration. But I built that software and I'm using it for the purposes that I built it for.
00:20:42
Speaker 5: Right.
00:20:42
Speaker 4: But I think there, you know, you could see a world where, you know, maybe regulators want to hold whoever made the open weight models liable.
00:20:50
Speaker 2: That would be disastrous and deranged. I mean, hope that would be. And that is what I and you're right. People try to do that. I mean, this is the argument that they're trying to make against speech on content platforms as well. Right. Like, you know, or Chachi BT is responsible because somebody looked up how busy the student center was before doing a shooting. Therefore, Chachi BT is responsible for the shooting.
00:21:10
Speaker 4: Right.
00:21:11
Speaker 2: That sort of liability is incredibly dangerous. And we should not seek to build that system because then people won't have open access to information. They won't have open access to technology. They won't be able to build software. Then we're putting in speech restrictions. Because building software, of course, coding is speech. And if we start to restrict that speech, it goes down a dark road. But I think 99% of people are not going to build software. I think what will happen, I was talking to Schumer's chief of staff actually this week who has this AI agent act that I don't think is very good. But, you know, she was saying, like, there'll probably eventually be something like an app store where it's like, you know what, could you go off and do this yourself? Yes, you're taking a huge risk. But here's a bunch of trusted apps. And I think there's going to be market competition, right, for these sort of trusted agent technologies.
00:21:59
Speaker 1: What kind of policies, though, are like OpenAI and Grok and Anthropic taking out or trying to get new forms of coverage against, like, because for sure the lawsuits are coming, right, when people experience harm based on something that a bot has either done or recommended.
00:22:12
Speaker 4: I don't know. I actually think I've probably committed crimes with ChatGBT, like, inadvertently already. Like, I think it's probably happened. Because, like, these laws are so strict on what you can do with software. Like just like the Digital Millennium Copyright Act. Like you can break the law like so easily. I know. I mean.
00:22:31
Speaker 2: Which is, but this is why we need to obliterate copyright law.
00:22:36
Speaker 3: We're getting far.
00:22:37
Speaker 6: But it feels like as we use more and more personal agents, it's like there's going to be more and more mishaps.
00:22:46
Speaker 3: More and more like debates over the responsibilities.
00:22:49
Speaker 1: Actually, the copyright law is a perfect segue to Kyle after the break for his story.
00:22:53
Speaker 5: But before we go there, Kyle.
00:22:55
Speaker 1: I mean, you missed, while you were on paternity leave, a week-by-week, play-by-play coverage of the Hugging Face open AI hack and the AI extinction, all this for the last three months. So having not been talking about it with us every single week, what's... What's your kind of overall take on, on what's happened in this question about who's responsible when AI stuff goes wrong?
00:23:15
Speaker 5: Like where is the culture on that question?
00:23:16
Speaker 3: Yeah, I'm like Rip Van Winkle. I just woke up.
00:23:22
Speaker 6: The world has changed vastly in the time that I've been asleep. Um, yeah, I mean, I'm, I'm catching up a little bit on the nuts and bolts of it. Uh, but I think it matters who, who sets the agent on this task. Like who is, who is the user?
00:23:37
Speaker 3: Who is the actor?
00:23:39
Speaker 6: Behind a given operation and for open ai like they clearly set the the agents on this task that eventually went off the rails but that's like their responsibility i think it becomes harder when you have more and more different kinds of agents more and more systems more open models more things that kind of come from nowhere and aren't as trackable um and i kind of feel like those things must be happening right now already and we just haven't heard about it and don't quite know yet uh As more and more things go wrong, I feel like we're going to be taken by surprise.
00:24:16
Speaker 1: When we come back, Kyle, you have a hunch that maybe we won't need software subscriptions all that much longer.
00:24:21
Speaker 3: We can only hope.
00:24:22
Speaker 5: Stay with us. Welcome back, Kyle.
00:24:32
Speaker 1: I understand you've been following a trend in software where AI is changing all the rules. Tell us exactly what Photon Studios is and why you're interested in it.
00:24:41
Speaker 6: Yeah, I mean, I think this was coming up increasingly often on X. X is a great way to just follow what people in AI are talking about and sharing and posting as brags on a week-by-week basis. And I feel like, at least in my zone this week, was about decompilation, which is this word that was new to me, actually, of.
00:25:04
Speaker 6: Taking a given product, like whether it's a Photoshop or a video game, drawing it back to the source code and then recompiling that source code as something else with the help of AI. So you're essentially creating like a totally open sourced, remixable, hackable version of any kind of software or video game or a piece of code.
00:25:25
Speaker 5: But what does that mean and how does it work?
00:25:28
Speaker 6: Well, it means that you're like, taking the code, translating it back to its basic form, and then retranslating it into a different form.
00:25:36
Speaker 5: Is this the same as distilling?
00:25:39
Speaker 6: Not in terms of model stuff, not in any way. This is a much more basic thing. We're just getting the source code of something that's been programmed, like a video game, and we're basically releasing it to the world. We're taking all those component parts, and we're putting them back out into the world in a form that anyone can kind of tinker with and reuse. So in the case of Photon Studios, I think, they did that to Photoshop. So they took the whole Adobe Photoshop architecture, the whole interface, the whole suite of tools, decompiled it using AI, and then put it back online just as a browser tool. So you can now go in your browser and open Photoshop exactly the same experience there's no subscription there's no signing up there's no anything it's just like here you go and that was kind of a mind-blowing experience to me as someone who had to like you know get the hacked software keys from kazaa or whatever when i was 12 years old we're all going to prison right this is this is kind of agents helping you break the law a little bit uh.
00:26:52
Speaker 4: I went down this rabbit hole too this week, Kyle. I'm writing something for the newsletter tomorrow actually about this, but the decompilation thing, which I hadn't really heard of, and then I heard about it because of the video game mashups. Have you seen these? Where people combine different games?
00:27:09
Speaker 6: Decompiling video games and then recompiling them, but in mashup versions. So I think not putting Mario in Elden Ring or Minecraft in Red Dead Redemption... And it looks perfect. You see these clips and it's like, literally, it's just Mario in the Old West or whatever.
00:27:28
Speaker 4: I'm not a gamer, but the comments were so eye-opening to me because gamers were just losing their minds about this capability and loving it. I saw these crazy comments like, oh my god, everybody hates AI, but this is finally the thing that makes me love AI. This is incredible.
00:27:47
Speaker 6: Finally, I can be Mario in Call of Duty. Cool, that's your dream come true.
00:27:53
Speaker 4: One guy was like, this is like AI curing cancer for gamers.
00:28:00
Speaker 3: It's kind of sad in a way. It's a small.
00:28:03
Speaker 6: thing to accomplish in a way the photoshop thing was more impressive to me because it was like it demonstrated how all software could become open source and how there was no defense against a decent amateur programmer decompiling and recompiling your corporate software products.
00:28:21
Speaker 1: I mean that's that's like the saspocalypse on steroids right like you don't have to vibe code an alternative you can just literally copy the underlying.
00:28:28
Speaker 4: Thing but i i saw people also commenting on that photoshop thing saying that like oh it doesn't work as well like it uses up too much memory you know that kind of stuff so i don't know have you actually.
00:28:40
Speaker 6: Tried it right yeah i was trying the web version and For me, who's just messing around in layers and doing type stuff, it's absolutely fine.
00:28:51
Speaker 3: And it's great. It's the kind of thing that I would.
00:28:53
Speaker 6: Use on a week-to-week basis and get rid of my Photoshop subscription. So that took me aback. But there are moats, right? Nintendo is already doing copyright violations to take down clips of Mario and Elden Ring, and Adobe will have defenses against this. Including that, like, the AI versions don't replicate all of their features and all of their, like, connectivity.
00:29:17
Speaker 4: They're going to have to stop Anthropic from allowing it, right? Because, like, otherwise everybody, you don't need to download it, right? You could just do it at home with your, you know, with Claude, right?
00:29:26
Speaker 5: But that's not, I mean, that's not a moat.
00:29:28
Speaker 1: That's more of a, like, legal recourse after the horse has bolted the stable, no?
00:29:32
Speaker 4: I mean.
00:29:32
Speaker 6: Yeah, and can you just tell them, oh, please code something that's, like, Photoshop for me?
00:29:39
Speaker 3: Like, probably.
00:29:39
Speaker 2: Yeah. yeah yeah and that's kind of the beauty i feel like of allowing more people to build software which is and i think that what's scary is like all of these people that want to restrict that.
00:29:51
Speaker 4: Process Did you ever play this game called Goldeneye, the James Bond game?
00:29:57
Speaker 1: Yeah, I was thinking about that game with the nuclear reactors on space. I feel like that would have been a perfect episode of Goldeneye, the game.
00:30:05
Speaker 4: Somebody decompiled Goldeneye before the AI age and they just finished. And I was reading about this in my rabbit hole this week and it took them like 10 years to decompile it. And I was like, I feel bad for those people because now they can just ask Claude to do it.
00:30:20
Speaker 6: Yeah, now it takes weeks. Like, I saw some charts, and it's like, oh, Super Smash Bros. took a decade, and now you can do it in three weeks. So it's like a really striking acceleration. Though, on Reed's note, there is this really interesting backlash from the handsy compilers, where they're like, no, we love doing this artisanal craft by hand and taking years to... replicate our favorite video games and now these shallow vibe coded decompiling versions are just cheap imitations it's like i don't know i think they're pretty.
00:30:55
Speaker 4: Good no the hand decompiler people that will be like a job in the ai future where it's like oh dude i got a hand decompiled copy of this game it's so good it's so much better.
00:31:06
Speaker 5: Who What's the business implication of this?
00:31:10
Speaker 1: Is it a huge, huge ongoing risk for companies like Adobe who have new proprietary software?
00:31:17
Speaker 2: No, I think it's about differentiating your product, but I do think that they can't lock you. And I do think in this world where everybody can build kind of basic level software, you really need to provide serious value to consumers. I think this is ultimately good for consumers. I think it just depends on how, again, we regulate it. And if we start doing these crazy restrictions on the ability to build software, that's going to be a problem. But I think like, look, you can build a cheap copy of Adobe Photoshop. It's not going to work as well. You're not going to get access to all these community, you know, features and events and things like that. And I just, I think that's kind of going to differentiate the big winners.
00:31:56
Speaker 4: I agree. I think this is like a really exciting time. And obviously, like, I don't want copyright to be, I want copyright law to still exist, but it would be really sad if the AI era just gets like smushed into the walled garden thing all over again. I think that would be a sad result.
00:32:16
Speaker 1: Of doing it illegal? And is it in the eye of the beholder? Could there be a debate? Was this decompiled or was this actually created from scratch using generative AI?
00:32:25
Speaker 6: That's a good question. I mean, I don't think decompiling itself is illegal. I think what they're already fighting back against is uploading it online. So the Nintendo clips, they fight back against that really fast. Photoshop, I feel like it's only a matter of time. I just Gemini-ed it.
00:32:43
Speaker 4: It says that it's not automatically illegal under the DMCA or US copyright law, as long as it's done for lawful purposes, such as achieving interoperability. So it sounds like it's illegal.
00:32:57
Speaker 5: Have any of you actually tried to do it?
00:32:58
Speaker 4: No. No.
00:33:00
Speaker 6: I'll have to think of something good. Like I was reading the blog post of someone who decompiled a really old Game Boy Advance game that they were obsessed with called Cronoa or something. And Claude discovered some Easter egg that no one had found before in the source code. That seems so fun and interesting to be able to go through that process so quickly and kind of break down old pieces of software. But in a video game, this is entertainment. It's culture. It's not as threatening to a business model as remaking Photoshop instantly or remaking Riverside.com into a better version.
00:33:42
Speaker 5: Well, that's all we have time for today. Thanks to our partners at Riverside.
00:33:45
Speaker 1: Thank you for joining us.
00:33:49
Speaker 3: Just kidding.
00:33:50
Speaker 4: All right.
00:33:50
Speaker 3: Great to be here.
00:33:51
Speaker 2: Thanks for having us.
00:33:55
Speaker 1: For Tech Stuff, I'm Oz Veloshin. This episode was produced by Liza Dennis. It was executive produced by me and Julia Nutter for Kaleidoscope and Katrina Norvell for iHeart Podcasts. Kyle Murdoch mixed this episode and wrote our theme song. A special thank you to Rita Brigotti, Kyle Chaker and Taylor Lorenz. Please check out all the work they put out into the world. We're lucky to call them friends of the pod.
Speaker 1: Taylor, Rie, Kyle, I've been obsessed with heists in the AI era. And heists gone wrong are, of course, some of the great comic gold. Has anyone been following the stolen NVIDIA trucks?
00:00:21
Speaker 2: No.
00:00:22
Speaker 3: What's going on?
00:00:23
Speaker 4: It wasn't me, if that's what you're asking.
00:00:26
Speaker 1: So there are a couple of trucks with NVIDIA branding on them that some thieves thought were full of NVIDIA chips, which.
00:00:33
Speaker 5: Of course are very, very valuable.
00:00:36
Speaker 1: And so the trucks were stolen, but it happened they were just full of sand. And not the sand that's used to make chips. The trucks belong to Plus AI, who makes self-driving trucks. And in order to make self-driving trucks good, you have to simulate cargo loads. So they're doing that with tons of sand, 20,000 tons per truck. They also happen to have a partnership with NVIDIA. So the NVIDIA logos were just on the truck to kind of flex about the NVIDIA partnership, like the NASA logo on my computer.
00:01:09
Speaker 3: Good advertising.
00:01:10
Speaker 4: If I see someone selling sand on eBay, I will know where it came from.
00:01:14
Speaker 1: Well, actually, Plus AI managed to recover the trailers and all 40,000 tons of sand.
00:01:20
Speaker 5: So happy ending.
00:01:21
Speaker 2: Happy ending.
00:01:26
Speaker 5: Welcome to Tech Stuff.
00:01:27
Speaker 1: I'm Oz Veloshian, and this is The Week in Tech, where I'm joined by the world's most plugged-in reporters to.
00:01:32
Speaker 5: Break down what's really happening in tech right now.
00:01:35
Speaker 1: Today, we're joined by Reid Abregotti, tech editor for Semaphore and host of the new podcast, as of this week, Future Proof, Taylor Lorenz of UserMag, and back from leave, Kyle Chaker, who writes the Infinite Scroll column for The New Yorker.
00:01:48
Speaker 4: Welcome, all.
00:01:49
Speaker 2: Thanks for having us.
00:01:50
Speaker 5: Reid, welcome to the podcasting ranks.
00:01:53
Speaker 2: Can't hear you, Reid.
00:01:54
Speaker 5: Oh, we can't hear you?
00:01:57
Speaker 3: First you have to turn your mic on.
00:01:59
Speaker 4: I'm learning all the rules now.
00:02:01
Speaker 5: Don't mute your mic.
00:02:05
Speaker 1: Well, starting off with you, I mean, first, just give us a brief plug for the new podcast, but then I want to ask you what's going on in space.
00:02:10
Speaker 4: Yeah, well, it's called Future Proof. And, you know, we're going to interview people from all areas of the tech industry and go deep on on topics and try to teach people something. It'll be newsy, journalistic, but also really go deep on stuff and help listeners understand.
00:02:26
Speaker 5: I'm excited to check it out.
00:02:29
Speaker 1: This week's story for you feels a bit like the plot of a 90s Bond movie involving the US competing with Russia and China to build nuclear powered permanent bases.
00:02:39
Speaker 5: On the moon. What's going on? I mean, is this like a fantasy or?
00:02:44
Speaker 4: It depends on who you ask. I personally like, you know, I think everything in the space industry feels like fantasy until it actually happens. I remember when I first came out to the tech industry, SpaceX was sort of just becoming a thing. And people were talking about satellites, you know, in space providing internet. And really smart people told me that was completely impossible. It would never happen. Now I own, you know, two Starlinks. Um, So I just take the timelines for these things and I just extend them in my mind, but I don't dismiss anything outright. I mean, nothing here is totally impossible, right? But, you know, does it make economic sense? Will it make economic sense in the near future? That I don't, I just don't know.
00:03:28
Speaker 1: I think the US has tried it once before and Russia has been sending nuclear reactors into space since the 60s. But somehow this kind of new space race thing is particularly fascinating.
00:03:38
Speaker 5: Kyle, have you been following it?
00:03:40
Speaker 6: Just as far as the data centers in space. I mean, I think that whole concept is really appealing to me. And it kind of seems like the only answer. If we're kind of having all these fights over data centers on Earth, you know, maybe it would be better to put them in space and have them powered in some other way.
00:04:00
Speaker 3: So yeah, I don't know.
00:04:01
Speaker 6: I think it reads right about the timelines extending longer than you might think.
00:04:05
Speaker 4: But I mean, Kyle, do you know what a space... I've actually looked into this recently. A space data center is not like a floating data center. It is basically, you have a bunch of satellites that have a GPU or two or a rack, and they're spread out very far apart, and they're connected by lasers. So they communicate with each other by lasers instead of on Earth through fiber optic cable or copper, right? And because.
00:04:32
Speaker 6: It's not a Death Star. It's not like one floating massive data center.
00:04:36
Speaker 3: Right.
00:04:37
Speaker 4: But because, you know, there's no air in space, you can't cool these things in the same way you cool them on Earth. So the idea is that you'll somehow pull the heat away from these GPUs and dissipate it on a massive... They're going to essentially like a massive heat sink that's about the size of a tennis court. I mean, it is... like really kind of just the craziest concept you can possibly imagine. But it's not impossible. I mean, but I don't know if I agree that that's going to be the thing that happens before we just get our act together on Earth and build some data centers already.
00:05:11
Speaker 1: Yeah, I mean, the concept of space junk is kind of interesting. I mean, sending up this tennis court, or I think in some cases even football pitch or field size, kind of non-convective heat dispersing panels into space. I mean, you have to wonder... what the implications of all that stuff floating around in orbit are. And I think some of this, when I first saw the story, the headline about the nuclear reactors in space, I kind of assumed somehow it had to do with power generation for AI. But to my great surprise, it wasn't about that. It was about, go ahead, Taylor.
00:05:41
Speaker 2: I'm going to start a business that is a space junk removal cleanup business, because I think this is going to be huge business Going forward, I mean, we're about to have something like a million satellites or Elon wants to scale to a million Starlink satellites. And I think if we put data centers up there too, yeah, there's just going to be a lot of debris. And I think that poses risks.
00:06:05
Speaker 5: Especially when it's nuclear.
00:06:07
Speaker 3: I mean, right.
00:06:08
Speaker 4: It's kind of, well, nuclear doesn't have that much risk in space. I mean, you know, nuclear meltdown in space isn't... Unless it falls to Earth. Well, yeah, it'll burn up before it falls to earth. But I mean, what happens to these giant, you know, heat dissipating football field or tennis court size things if like a little, a little speck runs into it and puts a hole in it? Like, does it still work? And then if something breaks, they just drop it down and let it burn in the atmosphere, which is I'm sure horrifying to people who are here trying to get their hands on like a single GPU right now. But I thought the nuclear reactor thing is like, you got to build those on the moon, right? I mean, that was what you were talking about. Because if you want to send Mars missions... Staging post.
00:06:48
Speaker 3: Yeah.
00:06:49
Speaker 4: Is that what... That's what you were talking about, right?
00:06:50
Speaker 5: Yeah, exactly.
00:06:51
Speaker 1: And so the current like solar power cannot generate enough power on the moon to have actually warm habitable bases.
00:07:01
Speaker 4: Right.
00:07:01
Speaker 2: So.
00:07:01
Speaker 1: And China and Russia are collaborating. Russia have the best space nuclear technology. China, I guess, have good rocketry. So the new Cold War in space is on.
00:07:10
Speaker 4: Do you guys watch For All Mankind? Just out of curiosity, did you watch that show? No, I missed it. I mean, I think if you're interested in this topic, you've got to watch For All Mankind. Just go. It's on Apple TV. Go watch it. It's a sort of alternative history show about what would have happened if the if the Cold War sort of evolved a little bit differently and space just continued to be the thing that we put all of our, you know, human effort and ingenuity into. And like, I don't want to spoil the show for you guys, but you know, a lot of interesting stuff happens in space on a timeline that is not, that is a lot more accelerated than we have today. It's a really fun show. I think it's, you know, I've enjoyed it.
00:07:54
Speaker 3: Do we need more space propaganda?
00:07:56
Speaker 6: Like pro-space propaganda? pro-space content to encourage us all.
00:08:02
Speaker 4: It beats the propaganda we have right now. I mean, I'll take that over all the current propaganda.
00:08:09
Speaker 1: I've got a NASA sticker on my computer. What's the take on Jared Eisenman? I mean, the head of NASA, Trump had initially rejected him during the Trump-Elon Fodor, but then changed his mind and Eisenman was confirmed, obviously, a private sector billionaire. What's he doing at NASA? And what's the public-private relationship in space in the US today, Reid?
00:08:29
Speaker 4: Well, I mean, he's certainly helping Elon, right? There was a lot of tension with Elon and the Biden administration. And I think Jared's helping Elon. There's less tension between SpaceX and NASA. But, you know, I haven't heard any sort of controversies about him. He seems to be, you know, doing a fine job. We'll see what the effects of his term are. But I mean, the general trend, I think, in NASA is like more and more of this is becoming privatized. You know, there's good things and bad things about that, right? I think the public interest mission of space that I think we're all used to from our childhood of, you know, space shuttle going up and doing all these science experiments. I mean, some of that is for sure going to be going to be lost when the private sector sort of takes over. And I think that's kind of sad. But at the same time, it's incredible, the stuff that has been accomplished. I think it's amazing, not to mention just having a Starlink on my camper van in California. That's like the ultimate life.
00:09:25
Speaker 1: So I do want to just note that on the kind of safety issue with the nuclear reactors in space, there was a Russian nuclear reactor thing that came down to Earth. And US intelligence say that five people were killed. The Russians deny this. But just before we go to the break, Taylor, I mean, this idea of kind of cleaning space junk and including nuclear space junk, have you met any companies in this space?
00:09:51
Speaker 2: Yeah, I talked to Orbital, which is, you know, seeking to put the data centers in space recently and visited their offices in LA. So I do think that there's sort of like a space race with a lot of different startups right now trying to build out infrastructure in that realm.
00:10:07
Speaker 5: What did you learn at Orbital?
00:10:09
Speaker 2: What they're doing is cool. I mean, I think it's just like they're up against a lot of competition, as Reid said, too. It's like everything seems impossible until it somebody does it. And I personally think that the data center fight is not going away anytime soon, and we're going to need more compute. So yeah, if they can put them in space.
00:10:25
Speaker 3: Why not?
00:10:26
Speaker 4: Well, what's interesting about Orbital is the founder was the scooter guy, right? So he had a scooter startup in San Francisco. And then he kind of worked for Ford, I think.
00:10:36
Speaker 5: And then he.
00:10:37
Speaker 2: He sold it.
00:10:38
Speaker 4: And now he's like, I'm going to, I'm going to just make a space data center company because that's the obvious next move.
00:10:44
Speaker 2: Well, the reason why is because he has a lot of connection in manufacturing and in China. So I do think that like. building a scooter company ironically kind of prepared him to just do some sort of hardware startup generally.
00:10:58
Speaker 4: Yeah, I was being tongue-in-cheek, but actually I think that there is a point to be made, which is that the whole idea of SpaceX and Blue Origin and these companies to get infrastructure up into space is that you sort of enable a new crop of entrepreneur to work on space-related ideas and companies and So it's like, if you think of it as like all this infrastructure as a platform, it lets this guy who was a scooter entrepreneur, all of a sudden be a space entrepreneur, which never would have been possible if you didn't have the, you know, SpaceX rockets to get his equipment up into space, right? That he was never going to start a rocket company, but now he can start a space company. And I think that's, I think that's actually kind of a profound point. He, I talked to him too. He doesn't even want to go to space. I was like, do you want to go to, he's like, no, I have no interest in going to space. I'm like, that's, I think that's pretty profound.
00:11:52
Speaker 2: I mean, being in space kind of sucks. It's cold up there. It's like living in a submarine.
00:11:58
Speaker 3: Until you get the nuclear reactors on the moon.
00:12:00
Speaker 2: Right, right.
00:12:02
Speaker 1: To warm your hands by the warmth of a nuclear reactor on the moon. I have to say, there's always that classic question, what would you like the first line of your obituary to be? And I guess he was never going to build rockets is one of them. When we come back, who pays when chatbots lead you astray?
00:12:31
Speaker 5: Welcome back.
00:12:32
Speaker 1: Taylor, there was a piece in Futurism this week that there's a kind of golden oldie quality to it, a kind of tech made me do it story. I remember there was that episode of The Office where Michael Scott drives his car into the lake on purpose because GPS told him to. But nowadays, it's not GPS, but AI and cloud in particular that's leading not drivers, but hikers astray.
00:12:56
Speaker 2: Yeah, I actually read about this in the New York Times. They did a good story on it. But basically, this guy was hiking in British Columbia and he went off the trail. He used Claude as part of his planning process and ultimately had to be rescued off the side of a cliff because he was not a qualified enough hiker to, you know, to get himself out of the situation he had encountered. used. I don't think this is an AI problem necessarily. I actually ended up literally last week or two weekends ago when I was back in LA, I went to go on a hike and ended up in this like burned out section of Altadena where there was like no water, the trail was a mess. And I had read about it online and And even like looked it up, you know, ahead of time. And it's just that that route after the fire had changed a year ago and all this stuff. Of course, my friends were like, why wouldn't you use all trails? I think all trails is sort of the standard for this. And if you go anywhere else outside the internet, you're sort of risking it.
00:14:00
Speaker 1: Kyle, you nodded sagely about all trails. You seem like a trail guy.
00:14:04
Speaker 6: I'm not, which is the whole problem. I am not a hiking guy. I think I would end up in a bad situation. But I think if you have ever explored all trails, you know why relying on AI for your hiking knowledge is a terrible idea. All trails is the most human... intricate detailed set of commentary and data that you can possibly have i think in an app because it'll be like okay a hundred feet to your right like don't take this trail look for the rock and go the other way or if you aren't an experienced hiker maybe don't even bother uh so like all trails i don't know i want a person directly to tell me in their own words what I should and shouldn't do on one of these hiking trails.
00:14:54
Speaker 2: And a person recently.
00:14:55
Speaker 3: Yeah, yeah, it changes all the time.
00:14:58
Speaker 2: Like the hike that I went on a couple weeks ago, like it was rated really, people had blogged this, you know, like it was just not up-to-date information. And so, yeah, I just think that like, There's a lot of people interested in outdoors, but you need to rely on, this is like an all trails ad. Like, I think like, you know, he could have not used Claude and I don't know that he wouldn't have ended up in the same situation, especially when you're going off the trail, you know?
00:15:25
Speaker 1: He got rescued, sorry to interrupt, and the rescue manager said, Claude has never been in that specific area, has never been on that route, and has never been in the outdoors. Is this like a sticking rocks to the pizza with glue type story?
00:15:37
Speaker 5: Or is it like, what's the, not so much, Taylor?
00:15:39
Speaker 2: Yeah, and that was so silly, too. Like, I mean, I guess, like, especially Futurism is a website that, like, loves to write up this stuff as, like, it's just rage bait content to kind of be like, ha ha, you idiot, you know, look at you. And to me, it's like, look, it's a real problem with accuracy. I think this shows the value actually of these human curated, like very human updated apps. Like that's what it shows the value of. It's not that AI is inherently unreliable. I would say 99.999% of things that you ask AI and including that, like Google, you know, Gemini, I guess, review that had suggested you put glue on your pizza. And that was from a Reddit thread, you know, Early iterations of AI were certainly less reliable. They're getting smarter and smarter. Again, it just has to do with the fact that when you're going out into the physical world, hiking specifically, until we get drones, you know, going out and scouting these trails for us, I think, like, something like all trails is just the way to go.
00:16:34
Speaker 3: Or the humanoid robots.
00:16:37
Speaker 2: Right, we send the humanoid robots out to go chart the trails.
00:16:41
Speaker 6: But it was striking to me, like, Hiking is a subjective thing. Difficulty is subjective. How much your body can take is subjective. AI, if you're not an experienced hiker, AI is not going to be able to tell you what you can handle in this situation. So maybe in this case, don't ask Claude, ask your friends.
00:17:03
Speaker 4: In my experience, I don't know if you guys go backpacking and stuff, but if something really bad doesn't happen where you think you might die, it's kind of a boring trip. Those are the things that you always remember, you know?
00:17:14
Speaker 3: I think.
00:17:15
Speaker 2: Okay, I am not trying to go on the hikes that Reid is going on. I love hiking a leisurely trail in the woods. I think also, you know, this is a teenager. I think he was 16 years old. So I think there's also just teenagers don't traditionally have a great track record of judging risk and... Potential for bodily harm.
00:17:35
Speaker 4: Yeah, people have always been doing dumb stuff.
00:17:37
Speaker 1: There is something in the automation bias, like AI psychosis adjacent thing here. I got my green fingers burnt earlier this year when I really went to town on pruning my roses with the encouragement of Gemini. And actually, I think I ended up killing them and I had to confess to the landlord. So I feel like you can lean in too hard to the reassuring advice.
00:17:57
Speaker 5: Of AI, certainly.
00:17:59
Speaker 1: The story that kind of sits behind this story, for me, though, is the kind of AI... responsibility and liability story. Taylor, I think this has been also in the news this week in terms of how insurance payments work out when AI does harm and who's held liable, like AI hugging face episode, for example, like in that case, hugging face didn't sue OpenAI, but if they had, who would have had the liability in that situation?
00:18:25
Speaker 2: Well, it's a company hacking another company. So the company has a liability. It intentionally hacked the other company.
00:18:33
Speaker 4: Yeah, exactly. I think the question though, there is an, I think there is a really interesting legal question that's going to have to be answered, which is like, let's say you're not open AI and there's got to be some cases where you could end up accidentally doing harm, right?
00:18:48
Speaker 2: And then I think- Here's the question, I think, or not to interrupt, but just to finish, or I think this was like the second part maybe that you wanted to get at, but like, here's the question that I think we're going to have to answer is when we move into this more agentic era of the internet, who is responsible for the agent's actions? And with OpenAI, it is literally their model. Of course, they're responsible. This is a test that they were doing. But if I use Grokbot, for instance, to go out into the world and do a lot of things for me, automate my lifestyle, and I end up causing harm to someone or... I lose, you know, banks are super, super concerned about this. I was talking to somebody in the banking industry yesterday about it, because if my grok bot goes out and spends lots of money, I am probably going to be inclined to declare that as fraud. And, you know, the bank will suddenly be held liable and banks are lobbying right now to say, hey, hey, hey, no, no, no, no. It's grok that's liable. So there's all these questions around agents. And I do think that there's a much more compelling case for liability for companies that are building agents to be liable for their agents. I think it just is a question too of how are you using that agent? Did you jailbreak the agent? Are you using it within terms of use? And I just think those are battles that we're going to see.
00:19:58
Speaker 4: Or what if we use open source, right? I think that's going to be another one where it's like, well, what if I'm using open source agents or open weight models?
00:20:04
Speaker 2: Then you're liable, right?
00:20:06
Speaker 4: But what if it's, let's say you get sued for something you did, And now there's the discovery process and you see all your prompts and your prompts show that you were acting in good faith and you reasonably thought that, you know, there was not.
00:20:18
Speaker 2: Well, you can unintentionally do harm to somebody else. And that's taken into account in terms of the legal fight. But if I build software that's my personal boutique software that I'm using and I use it to do something, maybe it was unintentional harm. And again, that can be taken into consideration. But I built that software and I'm using it for the purposes that I built it for.
00:20:42
Speaker 5: Right.
00:20:42
Speaker 4: But I think there, you know, you could see a world where, you know, maybe regulators want to hold whoever made the open weight models liable.
00:20:50
Speaker 2: That would be disastrous and deranged. I mean, hope that would be. And that is what I and you're right. People try to do that. I mean, this is the argument that they're trying to make against speech on content platforms as well. Right. Like, you know, or Chachi BT is responsible because somebody looked up how busy the student center was before doing a shooting. Therefore, Chachi BT is responsible for the shooting.
00:21:10
Speaker 4: Right.
00:21:11
Speaker 2: That sort of liability is incredibly dangerous. And we should not seek to build that system because then people won't have open access to information. They won't have open access to technology. They won't be able to build software. Then we're putting in speech restrictions. Because building software, of course, coding is speech. And if we start to restrict that speech, it goes down a dark road. But I think 99% of people are not going to build software. I think what will happen, I was talking to Schumer's chief of staff actually this week who has this AI agent act that I don't think is very good. But, you know, she was saying, like, there'll probably eventually be something like an app store where it's like, you know what, could you go off and do this yourself? Yes, you're taking a huge risk. But here's a bunch of trusted apps. And I think there's going to be market competition, right, for these sort of trusted agent technologies.
00:21:59
Speaker 1: What kind of policies, though, are like OpenAI and Grok and Anthropic taking out or trying to get new forms of coverage against, like, because for sure the lawsuits are coming, right, when people experience harm based on something that a bot has either done or recommended.
00:22:12
Speaker 4: I don't know. I actually think I've probably committed crimes with ChatGBT, like, inadvertently already. Like, I think it's probably happened. Because, like, these laws are so strict on what you can do with software. Like just like the Digital Millennium Copyright Act. Like you can break the law like so easily. I know. I mean.
00:22:31
Speaker 2: Which is, but this is why we need to obliterate copyright law.
00:22:36
Speaker 3: We're getting far.
00:22:37
Speaker 6: But it feels like as we use more and more personal agents, it's like there's going to be more and more mishaps.
00:22:46
Speaker 3: More and more like debates over the responsibilities.
00:22:49
Speaker 1: Actually, the copyright law is a perfect segue to Kyle after the break for his story.
00:22:53
Speaker 5: But before we go there, Kyle.
00:22:55
Speaker 1: I mean, you missed, while you were on paternity leave, a week-by-week, play-by-play coverage of the Hugging Face open AI hack and the AI extinction, all this for the last three months. So having not been talking about it with us every single week, what's... What's your kind of overall take on, on what's happened in this question about who's responsible when AI stuff goes wrong?
00:23:15
Speaker 5: Like where is the culture on that question?
00:23:16
Speaker 3: Yeah, I'm like Rip Van Winkle. I just woke up.
00:23:22
Speaker 6: The world has changed vastly in the time that I've been asleep. Um, yeah, I mean, I'm, I'm catching up a little bit on the nuts and bolts of it. Uh, but I think it matters who, who sets the agent on this task. Like who is, who is the user?
00:23:37
Speaker 3: Who is the actor?
00:23:39
Speaker 6: Behind a given operation and for open ai like they clearly set the the agents on this task that eventually went off the rails but that's like their responsibility i think it becomes harder when you have more and more different kinds of agents more and more systems more open models more things that kind of come from nowhere and aren't as trackable um and i kind of feel like those things must be happening right now already and we just haven't heard about it and don't quite know yet uh As more and more things go wrong, I feel like we're going to be taken by surprise.
00:24:16
Speaker 1: When we come back, Kyle, you have a hunch that maybe we won't need software subscriptions all that much longer.
00:24:21
Speaker 3: We can only hope.
00:24:22
Speaker 5: Stay with us. Welcome back, Kyle.
00:24:32
Speaker 1: I understand you've been following a trend in software where AI is changing all the rules. Tell us exactly what Photon Studios is and why you're interested in it.
00:24:41
Speaker 6: Yeah, I mean, I think this was coming up increasingly often on X. X is a great way to just follow what people in AI are talking about and sharing and posting as brags on a week-by-week basis. And I feel like, at least in my zone this week, was about decompilation, which is this word that was new to me, actually, of.
00:25:04
Speaker 6: Taking a given product, like whether it's a Photoshop or a video game, drawing it back to the source code and then recompiling that source code as something else with the help of AI. So you're essentially creating like a totally open sourced, remixable, hackable version of any kind of software or video game or a piece of code.
00:25:25
Speaker 5: But what does that mean and how does it work?
00:25:28
Speaker 6: Well, it means that you're like, taking the code, translating it back to its basic form, and then retranslating it into a different form.
00:25:36
Speaker 5: Is this the same as distilling?
00:25:39
Speaker 6: Not in terms of model stuff, not in any way. This is a much more basic thing. We're just getting the source code of something that's been programmed, like a video game, and we're basically releasing it to the world. We're taking all those component parts, and we're putting them back out into the world in a form that anyone can kind of tinker with and reuse. So in the case of Photon Studios, I think, they did that to Photoshop. So they took the whole Adobe Photoshop architecture, the whole interface, the whole suite of tools, decompiled it using AI, and then put it back online just as a browser tool. So you can now go in your browser and open Photoshop exactly the same experience there's no subscription there's no signing up there's no anything it's just like here you go and that was kind of a mind-blowing experience to me as someone who had to like you know get the hacked software keys from kazaa or whatever when i was 12 years old we're all going to prison right this is this is kind of agents helping you break the law a little bit uh.
00:26:52
Speaker 4: I went down this rabbit hole too this week, Kyle. I'm writing something for the newsletter tomorrow actually about this, but the decompilation thing, which I hadn't really heard of, and then I heard about it because of the video game mashups. Have you seen these? Where people combine different games?
00:27:09
Speaker 6: Decompiling video games and then recompiling them, but in mashup versions. So I think not putting Mario in Elden Ring or Minecraft in Red Dead Redemption... And it looks perfect. You see these clips and it's like, literally, it's just Mario in the Old West or whatever.
00:27:28
Speaker 4: I'm not a gamer, but the comments were so eye-opening to me because gamers were just losing their minds about this capability and loving it. I saw these crazy comments like, oh my god, everybody hates AI, but this is finally the thing that makes me love AI. This is incredible.
00:27:47
Speaker 6: Finally, I can be Mario in Call of Duty. Cool, that's your dream come true.
00:27:53
Speaker 4: One guy was like, this is like AI curing cancer for gamers.
00:28:00
Speaker 3: It's kind of sad in a way. It's a small.
00:28:03
Speaker 6: thing to accomplish in a way the photoshop thing was more impressive to me because it was like it demonstrated how all software could become open source and how there was no defense against a decent amateur programmer decompiling and recompiling your corporate software products.
00:28:21
Speaker 1: I mean that's that's like the saspocalypse on steroids right like you don't have to vibe code an alternative you can just literally copy the underlying.
00:28:28
Speaker 4: Thing but i i saw people also commenting on that photoshop thing saying that like oh it doesn't work as well like it uses up too much memory you know that kind of stuff so i don't know have you actually.
00:28:40
Speaker 6: Tried it right yeah i was trying the web version and For me, who's just messing around in layers and doing type stuff, it's absolutely fine.
00:28:51
Speaker 3: And it's great. It's the kind of thing that I would.
00:28:53
Speaker 6: Use on a week-to-week basis and get rid of my Photoshop subscription. So that took me aback. But there are moats, right? Nintendo is already doing copyright violations to take down clips of Mario and Elden Ring, and Adobe will have defenses against this. Including that, like, the AI versions don't replicate all of their features and all of their, like, connectivity.
00:29:17
Speaker 4: They're going to have to stop Anthropic from allowing it, right? Because, like, otherwise everybody, you don't need to download it, right? You could just do it at home with your, you know, with Claude, right?
00:29:26
Speaker 5: But that's not, I mean, that's not a moat.
00:29:28
Speaker 1: That's more of a, like, legal recourse after the horse has bolted the stable, no?
00:29:32
Speaker 4: I mean.
00:29:32
Speaker 6: Yeah, and can you just tell them, oh, please code something that's, like, Photoshop for me?
00:29:39
Speaker 3: Like, probably.
00:29:39
Speaker 2: Yeah. yeah yeah and that's kind of the beauty i feel like of allowing more people to build software which is and i think that what's scary is like all of these people that want to restrict that.
00:29:51
Speaker 4: Process Did you ever play this game called Goldeneye, the James Bond game?
00:29:57
Speaker 1: Yeah, I was thinking about that game with the nuclear reactors on space. I feel like that would have been a perfect episode of Goldeneye, the game.
00:30:05
Speaker 4: Somebody decompiled Goldeneye before the AI age and they just finished. And I was reading about this in my rabbit hole this week and it took them like 10 years to decompile it. And I was like, I feel bad for those people because now they can just ask Claude to do it.
00:30:20
Speaker 6: Yeah, now it takes weeks. Like, I saw some charts, and it's like, oh, Super Smash Bros. took a decade, and now you can do it in three weeks. So it's like a really striking acceleration. Though, on Reed's note, there is this really interesting backlash from the handsy compilers, where they're like, no, we love doing this artisanal craft by hand and taking years to... replicate our favorite video games and now these shallow vibe coded decompiling versions are just cheap imitations it's like i don't know i think they're pretty.
00:30:55
Speaker 4: Good no the hand decompiler people that will be like a job in the ai future where it's like oh dude i got a hand decompiled copy of this game it's so good it's so much better.
00:31:06
Speaker 5: Who What's the business implication of this?
00:31:10
Speaker 1: Is it a huge, huge ongoing risk for companies like Adobe who have new proprietary software?
00:31:17
Speaker 2: No, I think it's about differentiating your product, but I do think that they can't lock you. And I do think in this world where everybody can build kind of basic level software, you really need to provide serious value to consumers. I think this is ultimately good for consumers. I think it just depends on how, again, we regulate it. And if we start doing these crazy restrictions on the ability to build software, that's going to be a problem. But I think like, look, you can build a cheap copy of Adobe Photoshop. It's not going to work as well. You're not going to get access to all these community, you know, features and events and things like that. And I just, I think that's kind of going to differentiate the big winners.
00:31:56
Speaker 4: I agree. I think this is like a really exciting time. And obviously, like, I don't want copyright to be, I want copyright law to still exist, but it would be really sad if the AI era just gets like smushed into the walled garden thing all over again. I think that would be a sad result.
00:32:16
Speaker 1: Of doing it illegal? And is it in the eye of the beholder? Could there be a debate? Was this decompiled or was this actually created from scratch using generative AI?
00:32:25
Speaker 6: That's a good question. I mean, I don't think decompiling itself is illegal. I think what they're already fighting back against is uploading it online. So the Nintendo clips, they fight back against that really fast. Photoshop, I feel like it's only a matter of time. I just Gemini-ed it.
00:32:43
Speaker 4: It says that it's not automatically illegal under the DMCA or US copyright law, as long as it's done for lawful purposes, such as achieving interoperability. So it sounds like it's illegal.
00:32:57
Speaker 5: Have any of you actually tried to do it?
00:32:58
Speaker 4: No. No.
00:33:00
Speaker 6: I'll have to think of something good. Like I was reading the blog post of someone who decompiled a really old Game Boy Advance game that they were obsessed with called Cronoa or something. And Claude discovered some Easter egg that no one had found before in the source code. That seems so fun and interesting to be able to go through that process so quickly and kind of break down old pieces of software. But in a video game, this is entertainment. It's culture. It's not as threatening to a business model as remaking Photoshop instantly or remaking Riverside.com into a better version.
00:33:42
Speaker 5: Well, that's all we have time for today. Thanks to our partners at Riverside.
00:33:45
Speaker 1: Thank you for joining us.
00:33:49
Speaker 3: Just kidding.
00:33:50
Speaker 4: All right.
00:33:50
Speaker 3: Great to be here.
00:33:51
Speaker 2: Thanks for having us.
00:33:55
Speaker 1: For Tech Stuff, I'm Oz Veloshin. This episode was produced by Liza Dennis. It was executive produced by me and Julia Nutter for Kaleidoscope and Katrina Norvell for iHeart Podcasts. Kyle Murdoch mixed this episode and wrote our theme song. A special thank you to Rita Brigotti, Kyle Chaker and Taylor Lorenz. Please check out all the work they put out into the world. We're lucky to call them friends of the pod.
