WEBVTT - Week in Tech: My Deepfake, My Choice 0:00:12.520 --> 0:00:14.760 From Kaleidoscope and iHeart podcasts. 0:00:14.920 --> 0:00:17.840 This is tech stuff, I'm as Volocian and I'm Kara Price. 0:00:18.040 --> 0:00:21.160 Today we get into the headlines this week, including the 0:00:21.200 --> 0:00:26.720 future of unlocking animal consciousness with AI and Grock's commitment 0:00:26.800 --> 0:00:30.680 to its maker Elon Musk. Then, on our new segment, 0:00:31.120 --> 0:00:31.680 Chatt and me. 0:00:32.200 --> 0:00:34.800 That GBT refused to read my novel when I tried 0:00:34.800 --> 0:00:37.840 to upload it, which I don't really blame it for doing. 0:00:37.960 --> 0:00:41.160 Novels are boring, but the annoying thing was that Chat 0:00:41.200 --> 0:00:43.760 GBT kept lying to me and insisting that it had 0:00:43.800 --> 0:00:45.000 read it when it clearly had. 0:00:45.360 --> 0:00:50.960 All of that. On the Weekend Tech, It's Friday, July eighteenth. 0:00:51.600 --> 0:00:53.600 Hey Kara, hi, Auzie. 0:00:53.760 --> 0:00:55.639 You know you and my dad are the only two 0:00:55.680 --> 0:00:56.840 people who call me Ozzie. 0:00:57.240 --> 0:00:59.279 I don't know your dad that well. I know him 0:00:59.280 --> 0:01:01.680 a little bit. I find that to be very flattering, 0:01:01.880 --> 0:01:04.959 that I stand beside him in nickname calling. I also 0:01:05.000 --> 0:01:06.880 call you Ozzie because it allows me to think of 0:01:06.959 --> 0:01:09.679 you dressed like Ozzie Osbourne, which tickles me. 0:01:09.920 --> 0:01:13.399 Another brit with slightly lank hair. Look, We've talked about 0:01:13.400 --> 0:01:16.000 this on the show before. I'm not much of a 0:01:16.080 --> 0:01:18.720 grocery shopper and I ready cook in the kitchen. 0:01:18.840 --> 0:01:20.800 That shocks me very little. 0:01:20.880 --> 0:01:24.080 What might shocks you more is that I actually can cook, 0:01:24.360 --> 0:01:28.520 really I can. So. My stepfather owns an Italian restaurant 0:01:28.520 --> 0:01:30.920 in London called Ricardo's check it Out one two six 0:01:30.959 --> 0:01:33.720 Fulham Road, and one summer during high school, I cooked 0:01:33.720 --> 0:01:36.840 in the kitchen there. Nowadays, living in New York City 0:01:37.240 --> 0:01:40.520 on the fifth floor of a walk up building, I'm 0:01:40.560 --> 0:01:43.120 pretty rarely grocery shopping and cooking. 0:01:43.120 --> 0:01:44.600 I have to confess, I thought you were going to 0:01:44.600 --> 0:01:46.640 give us your address there for a second as we 0:01:46.640 --> 0:01:47.240 were streaming. 0:01:47.440 --> 0:01:49.680 Well, I want people to go to Ricardo's restaurant. I 0:01:49.680 --> 0:01:51.240 don't really want people to come to my house. 0:01:51.480 --> 0:01:53.880 Your house could be Ricardo's if you cook, That's true, 0:01:53.960 --> 0:01:54.480 you never know. 0:01:54.880 --> 0:01:56.880 But basically, I, like many people in New York, and 0:01:57.000 --> 0:01:58.560 regularly on a broots. 0:01:58.760 --> 0:02:00.640 I forgot the story until we were just talking. But 0:02:00.760 --> 0:02:03.200 when I was a kid, they used to have those playhouses, 0:02:03.840 --> 0:02:05.280 you know, where you would go in and you'd be 0:02:05.320 --> 0:02:06.240 able to pretend. 0:02:05.960 --> 0:02:09.200 To cook, like the full size Yes, yes you did. 0:02:09.040 --> 0:02:11.600 As a kid, and my parents once caught me and 0:02:11.639 --> 0:02:14.240 they let me keep going. So that's grateful for this. 0:02:14.320 --> 0:02:16.160 There's a little yellow phone. It was like a Fisher 0:02:16.160 --> 0:02:18.359 Price thing, and they saw me ordering Chinese. 0:02:18.560 --> 0:02:24.240 No, you're ordering Chinese food in the playoff. I love that. 0:02:24.400 --> 0:02:25.880 So that'll give you a sense of how much I 0:02:25.919 --> 0:02:30.239 can cook and do anything sort of epicurean minded, Well. 0:02:30.080 --> 0:02:32.680 It's sunny. Use that word. I've actually been playing around 0:02:32.760 --> 0:02:35.520 with this website called epicure this week, which is almost 0:02:35.600 --> 0:02:36.880 tempting me back into the kitchen. 0:02:36.960 --> 0:02:40.120 So is this like a knockoff of the recipe site Epicurius, 0:02:40.120 --> 0:02:40.919 which I have used. 0:02:41.360 --> 0:02:46.240 Not quite, although epicurious is a data set that Epicure's 0:02:46.360 --> 0:02:47.560 model was trained on. 0:02:48.040 --> 0:02:50.880 Interesting, So it's like AI generated recipes. 0:02:50.960 --> 0:02:52.560 That's exactly right. Actually want you to take a look. 0:02:52.600 --> 0:02:55.760 So if you go to epicure dot kaikaku dot. 0:02:55.560 --> 0:03:01.400 Air, I know exactly how to spell kaikaku. Okay, now 0:03:01.440 --> 0:03:03.360 what I have it? I have it here. Here's the 0:03:03.400 --> 0:03:05.560 slogan on top of the first of all, this website 0:03:05.680 --> 0:03:09.560 looks like it was created by Elon Musk. It says 0:03:09.639 --> 0:03:13.200 you are now the world's most creative chef. Leveraged the 0:03:13.200 --> 0:03:16.440 power of AI and machine learning to explore science backed 0:03:16.440 --> 0:03:18.240 flavor pairings and generate recipes. 0:03:18.400 --> 0:03:21.520 I think this is designed to flatter the Elon stand 0:03:23.320 --> 0:03:24.360 more than the kra price. 0:03:24.520 --> 0:03:28.079 What do they mean when they say science backed flavor 0:03:28.120 --> 0:03:31.440 pairings and AI machine learning? How do those things come 0:03:31.480 --> 0:03:32.919 together on this website? 0:03:33.320 --> 0:03:35.640 Well, the ux designer of the website must have had 0:03:35.640 --> 0:03:37.600 you in mind after all, because there is a tab 0:03:37.680 --> 0:03:41.160 that you can click with three words how it works? 0:03:41.200 --> 0:03:43.720 Oh my god, and I clicked on it, and what 0:03:43.760 --> 0:03:45.880 I learned is the website is built using a deep 0:03:45.960 --> 0:03:49.800 learning model called flavor graph, which was trained on over 0:03:49.880 --> 0:03:53.400 a million recipes and also the chemical compound data of 0:03:53.440 --> 0:03:56.400 different types of food items when they're cooked together. 0:03:56.640 --> 0:03:59.920 Oh so the chemical compound data is the science part exactly. 0:04:00.120 --> 0:04:03.800 So food chemists have identified these different flavor compounds, which 0:04:03.800 --> 0:04:08.520 I guess are kind of chemical compositions in most ingredients, 0:04:08.920 --> 0:04:11.960 and flavor graph is trained on over a thousand flavor 0:04:11.960 --> 0:04:14.760 compounds which are found in three hundred and eighty one 0:04:14.760 --> 0:04:17.800 different ingredients. So the same flavor compound can be more 0:04:17.800 --> 0:04:21.159 than one ingredient. The model can then create a flavor 0:04:21.279 --> 0:04:24.960 network in which two ingredients are connected if they share 0:04:25.000 --> 0:04:26.799 at least one flavor compound. 0:04:27.160 --> 0:04:30.760 Interesting. So that's how it generates recipes by like linking 0:04:30.960 --> 0:04:33.400 ingredients based on these flavor compounds. 0:04:33.520 --> 0:04:35.360 That's right, and I couldn't resist. Only a bit more 0:04:35.360 --> 0:04:38.080 of a deep dive on the founder's LinkedIn. You might 0:04:38.120 --> 0:04:39.000 not be surprised to. 0:04:39.000 --> 0:04:41.040 Hear you couldn't resist linkedins. 0:04:41.360 --> 0:04:43.280 So he had this funny post that begins with we've 0:04:43.320 --> 0:04:49.400 achieved agi artificial gastro intelligence. Al was a very bad joke, 0:04:49.440 --> 0:04:51.080 but it kind of got me. Do you remember this 0:04:51.240 --> 0:04:53.719 time when all of these chefs like Elbullian stuff were 0:04:53.720 --> 0:04:56.240 turning the kitchen into a chemistry chemistry? Yes, so that 0:04:56.360 --> 0:04:58.680 was sort of something that only, you know, the most 0:04:58.680 --> 0:05:01.600 famous chefs in the world could do it. Now leveraging 0:05:01.600 --> 0:05:02.240 the power. 0:05:01.960 --> 0:05:04.480 At least famous chef in the world me, you can 0:05:04.520 --> 0:05:05.839 do it, do it myself. 0:05:06.080 --> 0:05:06.840 So you want to try it? 0:05:06.920 --> 0:05:07.320 I do? 0:05:07.520 --> 0:05:09.479 I do? Okay, So I got the website open. You 0:05:09.560 --> 0:05:12.839 basically you guys select ingredients or you can type in 0:05:13.120 --> 0:05:15.000 your own ingredients. So what ingredients would you like? 0:05:16.520 --> 0:05:18.680 Of course they're British. Let's oh, so it can really 0:05:18.720 --> 0:05:23.160 be any ingredient. Okay, let's do mustard. 0:05:23.320 --> 0:05:26.279 Mustard Okay, yeah, mustard seed or just mustard. 0:05:25.960 --> 0:05:29.480 Is just mustard, please, and pasta. 0:05:29.880 --> 0:05:32.680 Mustard and pasta. That sounds pretty disgussgusting to me. 0:05:32.760 --> 0:05:35.000 It sounds delicious. They're going to come up with something. 0:05:35.160 --> 0:05:38.800 So what you see first is this graph where mustard 0:05:38.800 --> 0:05:41.760 and pasta at the center, and coming off all these 0:05:41.800 --> 0:05:48.279 spokes with different ingredient ideas. So you've got bacon, onion, sausage, cheese, bread. 0:05:48.440 --> 0:05:49.440 That sounds pretty bad. 0:05:49.839 --> 0:05:52.400 This sounds like Spanish rid garlic. 0:05:53.000 --> 0:05:55.480 But if you're not able to just from this graph 0:05:55.600 --> 0:05:58.159 extrapolate your own recipe and get going in the kitchen, 0:05:58.480 --> 0:06:01.920 there is a feature to actually generate a recipe. So 0:06:01.920 --> 0:06:04.320 the first thing you have to you chosen your two ingredients. Now, yes, 0:06:04.560 --> 0:06:07.080 now you have to choose whether you want a snack 0:06:07.440 --> 0:06:11.240 casual dining, whether you want an appetizer fine dining. And 0:06:11.240 --> 0:06:12.880 then you get to choose the cuisine as well. So 0:06:13.240 --> 0:06:14.440 is this a snack? Is it a meal? 0:06:14.480 --> 0:06:15.679 And a main dish? 0:06:15.839 --> 0:06:18.119 Main dish? Okay? And what cuisine you're gonna choose. 0:06:18.160 --> 0:06:21.840 Oh, the cuisine that I'm gonna choose is Italian. 0:06:21.960 --> 0:06:22.440 Italian. 0:06:23.800 --> 0:06:26.119 I'm curious if we have the same thing I've gotten 0:06:26.200 --> 0:06:28.560 creamy pancetta and pea pasta. 0:06:28.200 --> 0:06:29.000 That sounds pretty good. 0:06:29.080 --> 0:06:32.000 Now, what I would do, because this thing does not 0:06:32.040 --> 0:06:34.640 miss a beat, is I should say that I'm vegetarian. 0:06:35.040 --> 0:06:36.360 This is very fun for me. 0:06:36.400 --> 0:06:38.479 So I actually choose appetizer rather than main dish, and 0:06:38.520 --> 0:06:44.279 I got ravioli dolci condricotta. You have that starter. It 0:06:44.320 --> 0:06:48.440 is actually in Italian. It is actually vegetarian. What are 0:06:48.440 --> 0:06:48.560 you going? 0:06:48.880 --> 0:06:53.400 I got pasta alforno with roasted vegetables and creamy mustard ricotta. 0:06:53.680 --> 0:06:56.599 We're looking at two different AI generated recipes. Now I 0:06:56.640 --> 0:06:59.040 have all the ingredients listed out and then the instructions, 0:06:59.200 --> 0:07:02.040 and I also have an AI generated image of this dish, 0:07:02.080 --> 0:07:05.840 which looks in my case pretty good, although in a 0:07:05.880 --> 0:07:09.760 difficult AI fashion. The fork and the spoon are merged together, 0:07:10.240 --> 0:07:12.880 so there is something a sport. 0:07:12.720 --> 0:07:14.240 Yeah, which is an AI hallucination. 0:07:14.480 --> 0:07:16.360 Is this what have you got? I have? 0:07:16.640 --> 0:07:20.320 Similarly, I wouldn't think it's AI, except for the basil 0:07:20.440 --> 0:07:23.840 is placed so perfectly on the top of the pasta 0:07:24.440 --> 0:07:27.040 that there's just no way this is in an AI 0:07:27.120 --> 0:07:27.920 generated image. 0:07:27.960 --> 0:07:30.440 It's interesting. I was in Doha earlier this year, as 0:07:30.480 --> 0:07:33.360 you know, at a conference called web Summit, and Snapchat 0:07:33.400 --> 0:07:36.800 gave a presentation about what their augmented reality glasses might 0:07:36.800 --> 0:07:39.560 be able to do one day, And the presentation video 0:07:39.640 --> 0:07:42.640 is a guy opening the fridge wearing his augmented reality glasses, 0:07:43.000 --> 0:07:45.520 seeing some tomatoes and some eggs and whatever else and 0:07:45.520 --> 0:07:48.679 getting a recipe suggested and started cooking. So, for whatever reason, 0:07:48.800 --> 0:07:51.840 this idea of remixing ingredients seems to be the holy 0:07:51.880 --> 0:07:52.480 grail of AI. 0:07:52.680 --> 0:07:55.679 It'll be interesting to see if people start using AI 0:07:55.760 --> 0:07:59.680 generated recipes, if AI starts to influence their decisions in 0:07:59.720 --> 0:08:02.840 the kit. Similarly, the story that I want to tell 0:08:02.880 --> 0:08:05.400 you has a lot to do with the way that 0:08:05.520 --> 0:08:10.760 chat GPT is influencing our language. Huh. I've been looking 0:08:10.760 --> 0:08:13.640 at a study by researchers at the Max Planck Institute 0:08:13.640 --> 0:08:17.120 for Human Development in Germany to explore how AI is 0:08:17.160 --> 0:08:20.120 affecting the way we speak how we speak yes. So, 0:08:20.600 --> 0:08:23.480 the way that the study went is that it identified 0:08:23.680 --> 0:08:28.000 words that chatchybt favored. So they uploaded millions of pages 0:08:28.040 --> 0:08:32.199 of academic papers, news stories, emails and essays and asked 0:08:32.280 --> 0:08:36.960 chat gibt to polish the text. They then used AI 0:08:37.240 --> 0:08:41.520 edited documents to identify words that chat gbt seem to favor. 0:08:42.400 --> 0:08:43.760 So you read a lot of LinkedIn. What do you 0:08:43.760 --> 0:08:44.600 think those words are? 0:08:44.920 --> 0:08:46.400 You're putting me on the spot here, But I think 0:08:46.480 --> 0:08:50.160 the truth is I have read so much AI build 0:08:50.280 --> 0:08:53.120 and slopped them and completely sensitized. I have no idea 0:08:53.440 --> 0:08:54.680 where the walls even are. 0:08:55.280 --> 0:08:57.679 But tell me bilge is actually not one of them. 0:08:58.960 --> 0:09:01.760 The words that they found and were and maybe you've 0:09:01.840 --> 0:09:06.760 heard these more recently, delve delve into realm in the 0:09:06.800 --> 0:09:10.200 realm of possibility, meticulous. 0:09:09.480 --> 0:09:15.000 That's us, underscore, underscoes my point, Bolster, bolster, bolster, the 0:09:15.160 --> 0:09:19.280 argument bolsters my conviction, and boast is another one like 0:09:19.440 --> 0:09:21.080 it boasts an impressive resume. 0:09:21.440 --> 0:09:24.320 This sort of this makes sense in terms of AI sycophancy. 0:09:24.520 --> 0:09:27.000 So I get that they were able to understand from 0:09:27.040 --> 0:09:31.520 analyzing how AI edits documents that these words are common. 0:09:32.120 --> 0:09:34.800 How did they figure out that these words are also 0:09:34.880 --> 0:09:36.319 showing up in our mouths? 0:09:36.679 --> 0:09:40.880 So the researchers analyzed roughly a million YouTube videos and 0:09:40.960 --> 0:09:44.600 podcast episodes, and these words were used measurably more frequently 0:09:45.200 --> 0:09:46.920 after chat GBT was released. 0:09:47.000 --> 0:09:51.440 So basically YouTubers and podcasters are trackably using words that 0:09:51.480 --> 0:09:55.880 AI favors. In other words, were already just poppets for agi. 0:09:56.080 --> 0:09:58.400 Kind of you know. One of the studies authors told 0:09:58.440 --> 0:10:02.079 Scientific American that quote, it's natural for humans to imitate 0:10:02.120 --> 0:10:05.760 one another, but we don't imitate everyone that is around us. Equally, 0:10:05.960 --> 0:10:08.360 we're more likely to copy what someone else is doing 0:10:08.600 --> 0:10:11.920 if we perceive them as being knowledgeable or important. 0:10:12.080 --> 0:10:15.160 I guess it's sociolinguistics one on one, right. We match 0:10:15.320 --> 0:10:18.000 the way we speak to people we admire and want 0:10:18.040 --> 0:10:21.560 to imitate. In this case, it's not people, it's a machine, 0:10:21.600 --> 0:10:25.280 which is kind of disturbing. It's funny. I don't use 0:10:25.320 --> 0:10:27.920 that many of the AI words, but I have noticed 0:10:27.960 --> 0:10:30.880 that since moving to the US, I've found myself regularly 0:10:30.960 --> 0:10:35.559 using words like totally, absolutely, incredible, one hundred. 0:10:35.280 --> 0:10:36.720 Percent to become a value girl. 0:10:37.040 --> 0:10:40.400 Basically, yeah, I've become a value girl, a business stech. 0:10:40.559 --> 0:10:43.400 The valley girl is our sort of predominant cultural icon, 0:10:43.480 --> 0:10:45.480 which I think is similar to why they're doing this study. 0:10:45.520 --> 0:10:47.080 Now, what you're saying is a value girl is being 0:10:47.160 --> 0:10:50.000 replaced by chacchi chachi. What are these words? What are 0:10:50.000 --> 0:10:50.520 the tells? 0:10:50.760 --> 0:10:52.760 I think what's interesting to me is we're just sort 0:10:52.760 --> 0:10:55.920 of puppets of I guess whatever subculture or culture we're 0:10:55.960 --> 0:10:58.360 living in. Like, I remember studying abroad when I was 0:10:58.400 --> 0:10:59.800 in high school. I did this a broad thing, and 0:10:59.800 --> 0:11:01.559 I was living with a Canadian girl, and I started 0:11:01.559 --> 0:11:05.360 saying a after three weeks, Yeah, I was like fifteen. I 0:11:05.360 --> 0:11:08.240 guess the version of having a Canadian roommate, though, is 0:11:08.280 --> 0:11:12.320 now sort of more ubiquitous with something like CHATGPT. And 0:11:12.920 --> 0:11:17.079 the paper seems to suggest that Chatgypt has become this 0:11:17.120 --> 0:11:21.240 sort of cultural authority. Quote, machines trained on human culture 0:11:21.320 --> 0:11:25.880 are now generating cultural traits that humans adopt, effectively closing 0:11:25.920 --> 0:11:30.880 a cultural feedback loop. Which as I was reading this, 0:11:31.200 --> 0:11:33.719 I'm sort of thinking to myself. Everyone's like, AI is 0:11:33.720 --> 0:11:36.560 going to take our jobs, and I'm like, I think 0:11:36.600 --> 0:11:39.040 it's taking our brains faster than it's taking our jobs. 0:11:39.320 --> 0:11:41.320 Yeah. We did that story a few weeks ago about 0:11:41.440 --> 0:11:44.640 cognitive debt, basically the idea that if you offload too 0:11:44.760 --> 0:11:48.200 much work to AI, you basically become less capable of 0:11:48.200 --> 0:11:49.040 doing it yourself. 0:11:49.440 --> 0:11:51.839 Yeah, and you know, the paper raises a concern that 0:11:52.200 --> 0:11:55.680 this development could lead to cultural homogenization. You know, there's 0:11:55.720 --> 0:12:00.000 a quote that if AI systems disproportionately favor specific cultural traits, 0:12:00.480 --> 0:12:04.360 they may accelerate the erosion of cultural diversity one delve 0:12:04.400 --> 0:12:04.959 at a time. 0:12:05.080 --> 0:12:07.760 I mean, this is like you know, social media, YouTube, etc. 0:12:08.080 --> 0:12:11.760 There's like very rapid global flattening of culture whenever a 0:12:11.880 --> 0:12:15.200 meme emerges. And this seems to be a kind of 0:12:15.240 --> 0:12:18.480 real booster of that, you know. Yeah, I mean when 0:12:18.520 --> 0:12:20.640 we think about this aurboros and the idea of the 0:12:20.679 --> 0:12:22.640 snake that needs its own tail. I mean, in this 0:12:23.040 --> 0:12:27.760 search for efficiency and generating ideas and output, are we 0:12:28.280 --> 0:12:31.800 you know, consuming ourselves. But I think what's kind of 0:12:31.800 --> 0:12:35.679 interesting here is this idea of automation bias on steroids, 0:12:35.679 --> 0:12:39.520 Like we believe that machine output is more authoritative than 0:12:39.600 --> 0:12:41.599 human output, and then we start to copy it, we 0:12:41.640 --> 0:12:43.240 start to mirror our own machines. 0:12:43.559 --> 0:12:47.199 Yeah. I also think it's just interesting to note that 0:12:47.280 --> 0:12:49.320 we I mean, I don't know if I would say 0:12:49.320 --> 0:12:53.960 that for me personally or for you, but many people 0:12:53.960 --> 0:12:56.440 in my life do look at chat ept not only 0:12:56.440 --> 0:12:59.559 as a cultural authority, but as an authority figure on 0:12:59.679 --> 0:13:03.120 another number of topics. And I wanted to report on 0:13:03.160 --> 0:13:07.040 this because I think that it's important to consider the 0:13:07.160 --> 0:13:10.480 influence that a non human agent can have on your 0:13:10.559 --> 0:13:13.000 daily life, whether or not you use it a lot 0:13:13.280 --> 0:13:16.440 or just use it a little, it becomes something that 0:13:16.520 --> 0:13:20.880 you are deferential to, which to me is actually more 0:13:20.920 --> 0:13:24.400 serious than the bigger like will AI take over our lives? 0:13:24.480 --> 0:13:28.439 Like being deferential to a chatbot is a lot more insidious, 0:13:28.480 --> 0:13:32.000 but it's real. So I do want to flag that 0:13:32.080 --> 0:13:34.679 this study has yet to be peer reviewed, which is 0:13:34.720 --> 0:13:36.800 something we're kind of getting used to with these studies. 0:13:36.960 --> 0:13:39.680 I also want to say that correlation does not equal causation. 0:13:39.840 --> 0:13:43.760 You know, language does change, there could be other cultural 0:13:43.800 --> 0:13:47.000 forces at play. The point still stands, though, we should 0:13:47.080 --> 0:13:49.480 keep an eye on AI's influence on our culture and 0:13:49.600 --> 0:13:50.559 the way we communicate. 0:13:51.000 --> 0:13:54.360 Yeah, I think AI and unintended consequences is a rich 0:13:54.400 --> 0:13:58.120 area for discussion, including of course, in our politics. 0:13:58.400 --> 0:14:01.640 Are you talking about Lion Marco Little? 0:14:01.800 --> 0:14:03.920 Look, no, he's not anymore. 0:14:03.360 --> 0:14:05.920 He's Rubia. He's a segreator right now? 0:14:06.080 --> 0:14:09.320 So he and White House Chief of Staff Susie Wiles 0:14:09.360 --> 0:14:13.079 have both been impersonated by AI recently. Now, we've talked 0:14:13.200 --> 0:14:15.080 at length about how easy it is these days to 0:14:15.120 --> 0:14:18.920 clone someone's voice using AI. You don't need extensive, clean audio, 0:14:19.000 --> 0:14:22.400 You need fifteen seconds of someone's voice and basically for free, 0:14:22.640 --> 0:14:24.120 you can make a believable clone. 0:14:24.280 --> 0:14:27.120 And who is an easier target than a politician because 0:14:27.160 --> 0:14:29.479 they talk a lot and make a lot of public appearances. 0:14:29.720 --> 0:14:31.640 I would say at least more than the average person. 0:14:32.120 --> 0:14:34.240 Yeah, I mean they're easier targets, and they're also of 0:14:34.280 --> 0:14:39.560 course higher value targets. If Marco Rubio's wild schools, you 0:14:39.600 --> 0:14:41.640 know that they probably have it more clout when someone 0:14:41.680 --> 0:14:45.080 picks up the phone than you what I do. Rubio's 0:14:45.160 --> 0:14:49.000 impostor called three foreign ministers, a governor, and a senator, 0:14:49.440 --> 0:14:52.440 and in two instances left voicemails on the messaging apps 0:14:52.520 --> 0:14:57.000 signal that old friend of the Trump administration. Supposedly the 0:14:57.000 --> 0:15:00.320 impersonators use. The name on signal was Rubio at date 0:15:00.400 --> 0:15:03.720 dot gov, which is perhaps something which also psychologically primed 0:15:03.720 --> 0:15:04.760 the targets. I think it was real. 0:15:04.880 --> 0:15:07.000 Whenever these sorts of things happen, I'm like, I would 0:15:07.040 --> 0:15:13.000 fall for Rubio at state dot gov, Mark or Rubio. Why, like, 0:15:13.040 --> 0:15:13.760 why did this happen? 0:15:13.880 --> 0:15:15.480 We don't know. We don't know why I happen. We 0:15:15.480 --> 0:15:18.600 don't know who's doing it. The FBI is investigating. One 0:15:18.680 --> 0:15:21.560 of the major questions is was this carried out by 0:15:21.560 --> 0:15:27.080 criminal actors or potentially by national security adversaries, and our 0:15:27.080 --> 0:15:29.000 producer Lies was pushing me on whether or not we 0:15:29.000 --> 0:15:32.720 should include this story because it's an interesting novel use case. 0:15:32.720 --> 0:15:34.960 But we've talked extensively about deep fates on the show. 0:15:35.040 --> 0:15:37.480 But to sort of bolster my case as to why 0:15:37.480 --> 0:15:39.840 I thought this was important and timely, I did some 0:15:39.880 --> 0:15:44.000 extra homework. And also in the last week there was 0:15:44.040 --> 0:15:47.720 a story about deep fake technology that brings up sort 0:15:47.760 --> 0:15:51.320 of connected questions about how we define our identity in 0:15:51.360 --> 0:15:55.680 the digital age. So Kara comment, Danmark. 0:15:56.600 --> 0:16:01.880 You're an unbelievable teacher's pet. I'll show you why you 0:16:01.920 --> 0:16:02.280 want to. 0:16:02.200 --> 0:16:05.000 Know, and I'll google how to pronounce welcome to Denmark 0:16:05.040 --> 0:16:08.800 in Danish Danish citizens could soon have more ownership and 0:16:08.920 --> 0:16:13.200 control over their likeness, including voice and facial features, because 0:16:13.200 --> 0:16:16.479 the Danish government is actively considering a piece of legislation 0:16:17.000 --> 0:16:20.040 to give citizens tools to fight back if their likeness 0:16:20.080 --> 0:16:21.200 is copied without their consent. 0:16:21.440 --> 0:16:23.480 So the US does not do this, I remember we 0:16:23.520 --> 0:16:25.800 talked about the Take It Down Act a few weeks ago. 0:16:26.160 --> 0:16:28.200 Yeah, I mean, that's this new law in the US 0:16:28.560 --> 0:16:32.400 that mandates platforms to remove deep fake pornography and other 0:16:32.440 --> 0:16:37.240 misinformation from their sites upon user request. But lawmakers and 0:16:37.320 --> 0:16:39.840 Denmark are saying this is not actually an effective approach 0:16:39.880 --> 0:16:43.680 because it forces governments into a defensive posture and only 0:16:43.720 --> 0:16:48.080 addresses specific use cases of deep fake technology like individual posts, 0:16:48.120 --> 0:16:52.040 not the conceptual problem. The Danish Cultural Minister told the Guardian, 0:16:52.160 --> 0:16:55.360 quote in the bill we agree on are sending an 0:16:55.440 --> 0:16:59.000 unequivocal message that everybody has the right to their own body, 0:16:59.400 --> 0:17:03.200 their own and their own facial features, which is apparently 0:17:03.240 --> 0:17:06.520 not how the current law is protecting people against gerative AI. 0:17:06.560 --> 0:17:09.720 My likeness, my choice, and it certainly isn't protecting anyone 0:17:09.720 --> 0:17:11.399 in the United I mean, this is the first of 0:17:11.440 --> 0:17:12.240 its kind law. 0:17:12.400 --> 0:17:14.399 Yeah, it hasn't even been passed in Denmark yet, and 0:17:14.440 --> 0:17:16.639 what does it do well. It would make social media 0:17:16.680 --> 0:17:20.800 companies responsible for offending deep fakes, but it would not 0:17:20.840 --> 0:17:23.919 penalize the users who shared or posted them. This is 0:17:23.960 --> 0:17:27.520 basically the same mechanism as the Take It Down Act, 0:17:27.840 --> 0:17:29.840 just a different legal theory. The Take It Down Act 0:17:29.880 --> 0:17:31.720 is you have to prove that these deep fakes have 0:17:31.800 --> 0:17:34.800 caused harm. I think the legal theory here is that 0:17:35.160 --> 0:17:38.800 you have a copyright to digital copies of yourself, which 0:17:38.840 --> 0:17:41.639 is a different conceptual framework, and maybe you can apply 0:17:42.119 --> 0:17:44.639 more broadly and put less onus on users and governments. 0:17:44.720 --> 0:17:47.840 It sort of changes the assumptions going into how people 0:17:47.880 --> 0:17:49.080 can use digital copies of you. 0:17:49.359 --> 0:17:51.679 I'm curious to follow this because, well, one because it's 0:17:51.760 --> 0:17:54.640 the first I'm hearing of it, and because this concept 0:17:54.640 --> 0:17:58.720 of using copyright laws to protect your digital likeness rather 0:17:58.760 --> 0:18:01.919 than having to prove harm caused by a specific use 0:18:01.960 --> 0:18:04.000 case of a deep fike is very interesting to me. 0:18:04.200 --> 0:18:06.119 I think that's why I thought this story and the 0:18:06.160 --> 0:18:09.159 Marco Rubio one were an interesting pair, because it's like 0:18:09.359 --> 0:18:12.280 this is happening in real time. It's in the wild. 0:18:12.680 --> 0:18:17.080 Senior US officials are being impersonated in their interactions with 0:18:17.640 --> 0:18:20.240 other foreign leaders, and I mean this is sort of 0:18:20.640 --> 0:18:23.080 it's always on a rolling boil, but it feels to 0:18:23.160 --> 0:18:26.399 me like there's a kind of new crisis point emerging. 0:18:26.760 --> 0:18:28.679 It's something that affects everyone, and no one has all 0:18:28.680 --> 0:18:30.879 the answers. But I do think it's worth pausing just 0:18:30.920 --> 0:18:33.040 to note that the people who are really most affected 0:18:33.080 --> 0:18:36.840 by this and most harmed by this are not government officials. 0:18:36.880 --> 0:18:40.520 They are everyday teenagers. According to Thorn, which is a 0:18:40.640 --> 0:18:45.560 child online safety nonprofit. One in ten teenagers age thirteen 0:18:45.600 --> 0:18:49.199 to seventeen personally knows someone who's been the target of 0:18:49.280 --> 0:18:52.480 deep fake nude imagery. I mean, it's a horrific thought. 0:18:52.720 --> 0:18:55.040 And imagine trying to apply the Take It Down Act 0:18:55.080 --> 0:18:58.560 to one in ten teenagers in America, and only after 0:18:58.600 --> 0:19:01.840 the harm has been caused, so lots to chew on. 0:19:12.400 --> 0:19:15.480 After the break, we introduce you to someone you'll never 0:19:15.520 --> 0:19:26.200 want to meet, Mecha Hitler. Stay with us, Welcome back. 0:19:26.240 --> 0:19:28.000 We've got a few more headlines for you this. 0:19:27.960 --> 0:19:31.320 Week, and then a story about just how uncooperative chat 0:19:31.320 --> 0:19:35.399 GPT can get. But first we have to talk about GROCK. 0:19:35.760 --> 0:19:37.560 If I told you you would one day say the 0:19:37.560 --> 0:19:38.520 line we have to talk. 0:19:38.400 --> 0:19:41.639 About GROC, I never would have saudy. 0:19:41.240 --> 0:19:44.800 But this story was unavoidable. Elon Musk's AI chatbot made 0:19:44.840 --> 0:19:48.440 anti Semitic comments to some users. Recently, evidence of those 0:19:48.440 --> 0:19:51.320 comments has been deleted, but users said that Grok praised 0:19:51.400 --> 0:19:55.040 Hitler and at times referred to itself as Mecha Hitler. 0:19:55.280 --> 0:19:58.359 And this started almost immediately after an announced update to 0:19:58.440 --> 0:20:01.560 the model, which, according to the verse, GROC was updated 0:20:01.560 --> 0:20:05.439 to assume that quote subjective viewpoints sourced from the media 0:20:05.520 --> 0:20:09.639 are biased and quote the response should not shy away 0:20:09.640 --> 0:20:12.679 from making claims which are politically incorrect as long as 0:20:12.720 --> 0:20:16.440 they are well substantiated. But this wasn't the only odd 0:20:16.480 --> 0:20:20.080 GROC behavior. Last week, AI super users God Bless Them 0:20:20.359 --> 0:20:24.440 discovered that when asked to give an opinion on controversial topics, 0:20:24.720 --> 0:20:28.560 the new GROC would sometimes search for Elon Musk's opinions 0:20:28.560 --> 0:20:32.080