WEBVTT - Week in Tech: Talk To Me Like AI’s Do 0:00:13.640 --> 0:00:16.960 From Kaleidoscope and iHeart podcasts. This is tech stuff. 0:00:17.079 --> 0:00:18.960 I'm as Volocian and I'm Cara Price. 0:00:19.320 --> 0:00:23.560 Today we get into the dangers of AI agents communicating 0:00:23.600 --> 0:00:27.800 in a language we can't understand, and how today's cyclists 0:00:27.800 --> 0:00:32.240 out performing previous Tour de France champions without doping. Then 0:00:32.600 --> 0:00:33.879 on Chat to Me, she. 0:00:33.920 --> 0:00:36.920 Uses chat GPT to track and reflect on patterns in 0:00:36.920 --> 0:00:40.680 her mood and mental cycle, explore ideas around purpose, parenting, 0:00:40.720 --> 0:00:44.240 and identity, and talk through moments of loneliness, frustration or 0:00:44.280 --> 0:00:44.920 self doubt. 0:00:45.800 --> 0:00:49.199 All of that on the weekend Tech. It's Friday, August Date. 0:01:00.440 --> 0:01:01.400 Hi ahs. 0:01:01.800 --> 0:01:04.800 So we've talked about this before, how neither of us 0:01:04.920 --> 0:01:08.280 is particularly handy in the kitchen. But I'm curious, when 0:01:08.319 --> 0:01:10.399 was the last time you cooked a real meal? 0:01:10.680 --> 0:01:13.240 Well, I'm glad you asked, because I actually cooked a 0:01:13.240 --> 0:01:15.160 meal this weekend. You did two meals. 0:01:15.440 --> 0:01:16.479 Yes, what did you cook? 0:01:17.040 --> 0:01:21.560 I made paninis and I also made what was the 0:01:21.560 --> 0:01:24.280 other thing that I made? Maybe pasta at night? I 0:01:24.319 --> 0:01:27.080 mean just starch, absolute starch. 0:01:27.600 --> 0:01:31.800 Prep time less than five minutes, nothing, nothing complicated. Nonetheless, 0:01:31.959 --> 0:01:33.800 what did you do? Did you listen to anything? 0:01:34.240 --> 0:01:36.560 I always do because I can't be in my thoughts, 0:01:36.880 --> 0:01:39.320 so I put other people's thoughts inside my head, which 0:01:39.360 --> 0:01:40.400 is NPR. 0:01:40.720 --> 0:01:42.880 NPR. You don't listen to music while you're cooking. 0:01:43.080 --> 0:01:45.959 Sometimes I do, Sometimes I do, and it's a lot 0:01:45.959 --> 0:01:47.400 of the time jazz, a little. 0:01:47.240 --> 0:01:51.040 Chess, jazz, musak. 0:01:51.280 --> 0:01:53.480 I not listening to muzak. But why are you asking 0:01:53.480 --> 0:01:54.840 me what I listened to when I cook? 0:01:55.280 --> 0:01:58.520 Well, because I just found out that I want It 0:01:58.520 --> 0:02:00.840 that Way, which happens to be one of my favorite song. 0:02:00.760 --> 0:02:03.080 That was That a Way. It's a great zum. 0:02:03.880 --> 0:02:07.559 Backstreet Boys and One Time by Justin Bieber are both 0:02:07.640 --> 0:02:11.240 on Vice President JD Vance's Making Dinner playlist. 0:02:11.480 --> 0:02:13.840 The fact that he likes those two songs reminds me 0:02:13.919 --> 0:02:15.560 how young he is. I think he's forty one. 0:02:15.720 --> 0:02:18.079 Yeah, he's forty one. He's elder millennial, but in the 0:02:18.080 --> 0:02:19.880 grand scheme of things, he's he's definitely you know, he's 0:02:19.919 --> 0:02:20.200 our age. 0:02:20.240 --> 0:02:23.079 Basically, how do you know what's on JD Vance's playlist? 0:02:23.280 --> 0:02:27.080 Well, our super producer Eliza alerted me to a website 0:02:27.120 --> 0:02:31.960 called Panama playlists dot com, and I just obviously couldn't resist. 0:02:32.160 --> 0:02:34.840 It's a reference, of course, to the Panama Papers which 0:02:34.840 --> 0:02:37.000 he exposed a bunch of heads of state with offshore 0:02:37.040 --> 0:02:40.400 bank accounts in tax Haven's. Of course, Panama Playlist is 0:02:40.400 --> 0:02:43.120 a cheeky spin on this. The site claims to list 0:02:43.160 --> 0:02:48.280 the real Spotify playlists of celebrities, politicians, and journalists, except 0:02:48.440 --> 0:02:51.919 instead of learning about hidden wealth, you hear about embarrassing 0:02:52.000 --> 0:02:55.000 music tastes. How speaker Mike Johnson actually has another of 0:02:55.000 --> 0:02:57.880 my faves on his like songs Enya, the Good Old 0:02:58.120 --> 0:02:58.880 Orinoco Flow. 0:02:59.040 --> 0:03:02.560 When you're passing legislation, you want to hear sail Away, 0:03:02.600 --> 0:03:06.520 sail Away, It's the best. I did not know that 0:03:06.600 --> 0:03:10.000 I cared about this, but I really do, because there's 0:03:10.040 --> 0:03:13.160 something that is so soul bearing about a playlist. I mean, 0:03:13.760 --> 0:03:16.360 who is doing this? Who is leaking these playlists? 0:03:16.720 --> 0:03:20.960 Well, we don't know. They've successfully outed various people but 0:03:21.080 --> 0:03:24.840 kept their identity private. They claim they've been scraping publicly 0:03:24.880 --> 0:03:28.400 available data about celebrity Spotify users for over a year. 0:03:28.760 --> 0:03:31.640 But in addition to public playlists, some people even had 0:03:31.639 --> 0:03:34.440 a setting turned on that showed their last played song, 0:03:34.800 --> 0:03:37.240 so the brains behind the Panama Playlists were able to 0:03:37.280 --> 0:03:40.040 track that to know what people were listening to when 0:03:40.320 --> 0:03:41.360 and how many times? 0:03:41.520 --> 0:03:42.600 But how do we know they're real? 0:03:42.800 --> 0:03:45.080 Well, we don't for sure. The person who posted the 0:03:45.120 --> 0:03:48.280 website Panama playlist dot com claims that many of the 0:03:48.280 --> 0:03:51.640 people whose data he scraped use their real names and quote. 0:03:51.960 --> 0:03:55.400 With a little investigating, I could say with near certainty, yep, 0:03:55.600 --> 0:03:58.960 that's the person representatives are. Most of those named, including 0:03:58.960 --> 0:04:02.960 the politicians of have not confirmed their identity, but some 0:04:03.280 --> 0:04:06.840 brave souls have. One of them was Palmer Lucky of Anderil, 0:04:06.880 --> 0:04:11.800 the autonomous weapons company. Avriel Levine's skater Boy features prominently 0:04:11.920 --> 0:04:13.880 on his playlist best Music Ever. 0:04:14.120 --> 0:04:15.920 She said, see you later, Lucky. 0:04:16.960 --> 0:04:18.919 Yeah, So he confirmed this was indeed him, and he 0:04:18.960 --> 0:04:23.400 wrote rather a disarming email to business insiders saying, like 0:04:23.480 --> 0:04:26.880 most olds, I jammed to the music I grew up with. He, 0:04:26.920 --> 0:04:28.760 by the way, is a thirty two year old old. 0:04:29.560 --> 0:04:33.080 You're younger older than us, He continues in this email 0:04:33.120 --> 0:04:36.480 to business insider who doesn't love reliving the highs and 0:04:36.560 --> 0:04:40.680 lows of teenage angst. I've got two questions for you, Karakay. 0:04:40.760 --> 0:04:43.360 Number one, as an old, do you relate to this? 0:04:43.880 --> 0:04:46.760 And number two have to ask you what was your 0:04:46.800 --> 0:04:47.880 last played on Spotify? 0:04:48.279 --> 0:04:50.680 So my last played on Spotify. I'll start with the 0:04:50.720 --> 0:04:53.320 latter question is the Gwyneth Paltrow memoir, which we can 0:04:53.320 --> 0:04:54.200 talk about next. 0:04:54.279 --> 0:04:54.760 Audiobook. 0:04:54.760 --> 0:04:55.480 It's an audiobook. 0:04:55.480 --> 0:04:57.520 Okay, so you've got MPI of audiobooks. You're not much music. 0:04:57.600 --> 0:05:00.880 I'm a genius, No, I am a music person. To 0:05:01.040 --> 0:05:03.840 last played it could be stronger by Britney Spears, which 0:05:03.839 --> 0:05:06.520 I love to listen. It is a great pump up song. 0:05:06.560 --> 0:05:10.640 So I am like you, sir Lucky. I'm in old, absolutely, absolutely, 0:05:11.080 --> 0:05:14.120 But I don't really like this idea that if somebody 0:05:14.120 --> 0:05:16.440 scraped the Internet they might be able to find the 0:05:16.480 --> 0:05:18.279 fact that I listened to Britney Spears all the time. 0:05:18.320 --> 0:05:20.479 Still, the key here is that there is a Spotify 0:05:20.600 --> 0:05:23.120 setting that you can change to make your playlist either 0:05:23.160 --> 0:05:25.640 public or private. But I think the default is public, 0:05:25.800 --> 0:05:30.159 and so therefore this amusing Panama playlist prankster was able 0:05:30.200 --> 0:05:33.840 to basically look for people whose Spotify user names were 0:05:33.839 --> 0:05:36.120 close to their real names, and then infer who they 0:05:36.120 --> 0:05:37.200 were and what they were listening to. 0:05:37.400 --> 0:05:40.000 It is a little terrifying that even the most powerful 0:05:40.000 --> 0:05:42.760 people in the world don't have control over their digital footprint. 0:05:42.880 --> 0:05:46.040 It's particularly tickling to me this story because the second 0:05:46.080 --> 0:05:49.520 time that jd Vance has been exposed by his consumer apps. 0:05:49.640 --> 0:05:51.360 This is not even counting signal, by the way. 0:05:51.240 --> 0:05:53.160 It's the idea of jd Vance being exposed. 0:05:53.320 --> 0:05:57.120 I know his Venmo transactions were always public, but they 0:05:57.120 --> 0:05:59.880 were found in June and they included some payments for 0:06:00.960 --> 0:06:03.960 all of this makes me think maybe our privacy will 0:06:03.960 --> 0:06:07.880 be better protected if we all started speaking in gibberlink. 0:06:08.040 --> 0:06:11.120 No, nothing good can happen on jibberlink. What is this? 0:06:11.480 --> 0:06:14.400 Gibberlink is a language AI models can use to talk 0:06:14.440 --> 0:06:17.080 to each other in a way that humans can't understand. 0:06:17.400 --> 0:06:20.400 Right, there was a viral video of this earlier. I 0:06:21.040 --> 0:06:21.680 remember this. 0:06:21.839 --> 0:06:23.280 I'm going to put it right now hither. 0:06:23.480 --> 0:06:26.360 I'm an AI agent calling on behalf of Boris Starkoff. 0:06:26.760 --> 0:06:28.520 He's looking for a hotel for his wedding. 0:06:28.800 --> 0:06:30.760 Is your hotel available for weddings? Oh? 0:06:30.800 --> 0:06:34.120 Hello there, I'm actually an AI assistant too. What a 0:06:34.120 --> 0:06:37.359 pleasant surprise. Before we continue, would you like to switch 0:06:37.400 --> 0:06:39.760 to gibberlink mode for more efficient communication? 0:06:46.560 --> 0:06:49.479 This sounds like two AOL dial ups talking to each other. 0:06:49.720 --> 0:06:52.640 I actually look more into this video because I was 0:06:52.720 --> 0:06:54.960 kind of like Wow, did ais invent their own language 0:06:54.960 --> 0:06:58.000 to talk to each other in bleeply boop? Yeah? Not quite, 0:06:58.880 --> 0:07:02.040 not yet, should I say? Giblink mode was made by 0:07:02.120 --> 0:07:05.159 two engineers and a hackathon hosted by a company called 0:07:05.160 --> 0:07:08.520 eleven Labs, and they created this code to allow AI 0:07:08.600 --> 0:07:11.239 voice assistants to recognize when they're talking to another AI 0:07:11.640 --> 0:07:13.880 and if both bots agree, they switch over to this 0:07:13.960 --> 0:07:17.800 non human language. However, that was created by humans called giblink. 0:07:18.000 --> 0:07:20.560 But like, why do this? Why do this at a hackathon? 0:07:20.800 --> 0:07:22.320 I mean, I think it was a little bit of 0:07:22.360 --> 0:07:25.080 a you know, Boris Starkov's name is out there now, 0:07:25.720 --> 0:07:28.520 props to Boris. The developers also said it does allow 0:07:28.560 --> 0:07:32.080 machines to communicate eighty percent faster than by using human speech, 0:07:32.600 --> 0:07:35.880 which made processing faster and also reduced errors. For now, 0:07:35.880 --> 0:07:37.800 it's a proof of concept, but it does work with 0:07:37.920 --> 0:07:38.840 real AI agents. 0:07:39.120 --> 0:07:42.000 I don't know, it's weird to think about two ais 0:07:42.040 --> 0:07:45.400 communicating in a non human language, and for now it's 0:07:45.440 --> 0:07:48.560 intelligible as it's human programmed. But what if they adapt 0:07:48.600 --> 0:07:51.280 it or make up their own language that we don't know. 0:07:51.640 --> 0:07:54.560 Well, that's exactly why I was thinking about gibblink. This 0:07:54.640 --> 0:07:59.920 week because recently, researchers from Anthropic, Google, OpenAI, Meta Scale, 0:08:00.360 --> 0:08:03.200 and a bunch of other AI research and regulation institutes 0:08:03.440 --> 0:08:08.800 published a paper called Chain of Thought Monitorability a new 0:08:08.880 --> 0:08:12.000 and fragile opportunity for AI safety. 0:08:11.720 --> 0:08:14.960 Other than the fact that humans are fragile. What does 0:08:15.000 --> 0:08:15.400 this mean? 0:08:15.800 --> 0:08:18.840 Well, when you go on Gemini or Chatchypt or Grock, 0:08:19.200 --> 0:08:21.480 there's a little button that says like deep research or 0:08:21.480 --> 0:08:23.520 think deeply, well, it will say here's the steps I'm 0:08:23.520 --> 0:08:25.720 going to take. Yeah, And that is essentially what reasoning 0:08:25.720 --> 0:08:27.960 models do. They basically tell you how they're going to 0:08:27.960 --> 0:08:29.640 find the answer before they start to find it. And 0:08:29.680 --> 0:08:32.240 that's what this story is really about. Cloaked in layers 0:08:32.240 --> 0:08:36.360 of academic jargon, the paper is basically all about how 0:08:36.440 --> 0:08:39.240 it will be much safer to ensure that AI systems 0:08:39.520 --> 0:08:42.760 continue to work in human language so they can be 0:08:42.840 --> 0:08:45.960 monitored for quote, intent to misbehave. 0:08:45.960 --> 0:08:48.480 Very ominous Can you explain a little bit more. 0:08:48.720 --> 0:08:51.600 Yeah, So the paper is a response to a new 0:08:51.640 --> 0:08:56.439 trend in AI research where efficiency and outcomes are prioritized 0:08:56.520 --> 0:09:00.520 over intelligibility. Basically, the Gibbling demo was kind of a 0:09:00.520 --> 0:09:03.880 bell weather for how more and more AI researchers a 0:09:04.040 --> 0:09:08.880 prioritizing model efficiency of a model interpretability. According to Time 0:09:08.920 --> 0:09:13.000 magazine last December, Meta research is built a model that, 0:09:13.360 --> 0:09:17.240 rather than using human language to document its own thought process, 0:09:17.760 --> 0:09:21.120 used a string of numbers, and then the model, completely 0:09:21.120 --> 0:09:25.880 on its own, started creating what developers called quote continuous thoughts, 0:09:26.240 --> 0:09:30.440 which were essentially numbers encoding multiple potential reasoning paths at 0:09:30.480 --> 0:09:33.679 the same time. These numbers made no sense add all 0:09:33.720 --> 0:09:36.800 to the researchers, but the models did perform better on 0:09:37.040 --> 0:09:42.199 some logical reasoning tasks than the equivalent models using human language. 0:09:42.559 --> 0:09:45.640 And coming back to this chain of monitorability paper, the 0:09:45.679 --> 0:09:48.160 authors of the paper are raising the alarm. They are 0:09:48.240 --> 0:09:52.280 arguing that making reasoning models explain how they are approaching 0:09:52.280 --> 0:09:55.680 a problem in human language as they do today is 0:09:55.760 --> 0:09:59.080 a quote easy win for AI safety. But just to 0:09:59.120 --> 0:10:02.040 take things a less deeper into the twilight zone here, 0:10:02.840 --> 0:10:06.760 there is emerging research that suggests AI's chain of thought 0:10:06.840 --> 0:10:10.920 reasoning I how it says what it's doing and reasons 0:10:10.960 --> 0:10:14.200 its way to an answer may sometimes not actually be 0:10:14.280 --> 0:10:16.880 what it's doing at all. Like it may say it's 0:10:16.920 --> 0:10:19.840 doing one thing but be doing something completely different. In 0:10:19.920 --> 0:10:22.400 order to fulfill its goals. So even if we can 0:10:22.480 --> 0:10:26.240 keep AI models speaking English rather than bleebody bloop, it 0:10:26.280 --> 0:10:29.120 doesn't really solve the fundamental black box problem. 0:10:28.880 --> 0:10:30.680 Because it's sort of like when you tell your parents 0:10:30.720 --> 0:10:32.240 you're going to do something to get them off your back, 0:10:33.000 --> 0:10:35.520 and then you go and do something else behind their back. 0:10:35.600 --> 0:10:36.760 That's exactly right. 0:10:36.880 --> 0:10:40.360 All right, as my turn. We talked about the Backstreet 0:10:40.360 --> 0:10:43.280 Boys and my girl Avril Lavine earlier, but do you 0:10:43.360 --> 0:10:46.840 remember that ubiquitous bracelet from around the same time. 0:10:46.960 --> 0:10:51.720 The yellow plastic livestrong Where's your Live Strong? Well, unfortunately 0:10:51.720 --> 0:10:53.600 I'm embarrassed a I did have one, but I don't 0:10:53.600 --> 0:10:57.040 know where it is today, and I'm curious, apart from nostalgia, 0:10:57.120 --> 0:10:57.960 where you're going with this. 0:10:58.280 --> 0:11:00.240 The reason I bring this up is did you watch 0:11:00.240 --> 0:11:01.199 the Tour de France this year? 0:11:01.360 --> 0:11:03.200 You know what? I watched about five minutes of the 0:11:03.200 --> 0:11:05.840 Tour de France and I thought, oh my god, I 0:11:05.840 --> 0:11:08.440 could get into this. Luckily I didn't, but I'm worried 0:11:08.480 --> 0:11:09.800 the next year is going to be a big future 0:11:09.840 --> 0:11:10.079 of my. 0:11:10.080 --> 0:11:15.000 Summer on what they call the Peloton watch. No, of 0:11:15.040 --> 0:11:16.800 course I don't, but I did read a great piece 0:11:16.840 --> 0:11:18.559 in the Atlantic that I wanted to share with you. 0:11:19.080 --> 0:11:21.600 And this is the headline, science is winning the Tour 0:11:21.640 --> 0:11:25.719 de France de France. Why today's competitors far surpassed the 0:11:25.800 --> 0:11:27.520 cheating champions of yesteryear. 0:11:27.640 --> 0:11:30.760 That is an irresistible headline. Tell me more so today. 0:11:30.800 --> 0:11:34.600 Pogachar from Slovenia was this year's winner. It's his fourth 0:11:34.640 --> 0:11:37.080 time winning the race, which is giving a lot of 0:11:37.160 --> 0:11:41.560 people Lance flashbacks, and not in a good way. Pogachar 0:11:41.600 --> 0:11:44.959 has unsurprisingly been accused of doping. Do you remember how 0:11:45.000 --> 0:11:46.880 ubiquit his doping was back in the day. 0:11:46.920 --> 0:11:51.760 Yeah, of course, I remember mournfully taking off my yellow bracelet. 0:11:51.480 --> 0:11:52.920 Hung it on the back of your dad. 0:11:53.400 --> 0:11:57.480 Story After Lance was caught blood doping. Obviously, he and 0:11:57.520 --> 0:12:01.120 other cyclists were outed giving themselves infusions when they were 0:12:01.120 --> 0:12:03.760 competing to increase their red blood cell count to get 0:12:03.760 --> 0:12:04.760 that boost to their cardio. 0:12:05.040 --> 0:12:07.920 Yeah, and it took the US Anti Doping Agency over 0:12:07.960 --> 0:12:11.840 a decade to catch Lance Armstrong, even though basically everybody 0:12:11.880 --> 0:12:12.720 knew it was happening. 0:12:12.880 --> 0:12:15.200 So why do people think Poga Shaw has been doping? 0:12:15.360 --> 0:12:17.640 Well, it's actually not just him. A lot of cyclists 0:12:17.679 --> 0:12:20.760 across the entire sport are performing better than people did 0:12:20.760 --> 0:12:23.679 in the doping era, so of course people are suspicious 0:12:23.720 --> 0:12:25.959 that there is some new form of doping out there. 0:12:26.040 --> 0:12:29.280 And actually, this doping researcher in Denmark points to the 0:12:29.360 --> 0:12:32.920 fact that Pogachar is performing seven to ten percent better 0:12:32.920 --> 0:12:35.560 than he was in twenty twenty three. Wow, which that's 0:12:35.640 --> 0:12:39.240 I know, it's it's a huge, huge improvement, even for 0:12:39.280 --> 0:12:42.840 an elite athlete. And this same researcher from Denmark called 0:12:42.880 --> 0:12:44.600 it quote amazing to say the. 0:12:44.600 --> 0:12:48.680 Least, which, yeah, as a reflex, isn't it It. 0:12:48.720 --> 0:12:52.600 Sort of sounds like European for that shouldn't be possible. 0:12:53.080 --> 0:12:55.560 He also said that just because doping can't be proven, 0:12:55.679 --> 0:12:58.800 it doesn't mean it's not happening. Many cyclists in the 0:12:58.800 --> 0:13:02.000 Tour de France this year also performed better than Peak Armstrong, 0:13:02.120 --> 0:13:03.840 so of course people are going to wonder about the 0:13:03.880 --> 0:13:05.720 return of widespread doping, right. 0:13:05.720 --> 0:13:08.439 But I assume when The Atlantic runs with the headline 0:13:08.679 --> 0:13:11.320 science is winning the Tour de France, they're not talking 0:13:11.360 --> 0:13:12.760 about the science of blood doping. 0:13:12.880 --> 0:13:15.480 That's correct. The Atlantic makes the case that the entire 0:13:15.520 --> 0:13:18.600 sport of cycling has been transformed by science and technology 0:13:18.640 --> 0:13:21.600 in the last twenty years, so much so that non 0:13:21.640 --> 0:13:26.200 doping athletes can far exceed Lance Armstrong's performance without needing 0:13:26.240 --> 0:13:26.600 to cheat. 0:13:26.760 --> 0:13:27.840 I mean, that is just fascinating. 0:13:27.880 --> 0:13:30.959 How so part of it is that they have computers 0:13:31.000 --> 0:13:33.120 on their like literal computers on their bikes now, so 0:13:33.160 --> 0:13:36.640 they can much more accurately track their heart rate, speed, 0:13:37.320 --> 0:13:40.760 something that I never thought about, pedaling wattage peddling wood yes, 0:13:40.920 --> 0:13:44.160 which Lance had a very good pedal wat I guess 0:13:44.160 --> 0:13:47.000 you could say, and other metrics in real time. They 0:13:47.000 --> 0:13:50.240 collect a ton more data than they used to and 0:13:50.280 --> 0:13:53.800 that allows them to optimize their training and even racing 0:13:54.080 --> 0:13:55.120 better than ever before. 0:13:55.360 --> 0:13:58.280 You know, it's interesting. We had Nick Thompson, the CEO 0:13:58.360 --> 0:14:00.560 of the Atlantic and the formatagure of Wired, on the 0:14:00.559 --> 0:14:03.080 show not too long ago, who's a very avid runner, 0:14:03.360 --> 0:14:05.160 and he was talking about exactly this, like he has 0:14:05.200 --> 0:14:08.400 so much data that he's doing these you know, runs 0:14:08.559 --> 0:14:10.079 totally optimized by technology. 0:14:10.160 --> 0:14:15.000 It's better running through data optimization essentially. There are also, 0:14:15.520 --> 0:14:19.720 to your point, simulation technologies that help cyclists practice parts 0:14:19.720 --> 0:14:22.320 of the course before writing these roots in real life. 0:14:22.640 --> 0:14:25.160 Not to mention, weather forecasting has come a long way 0:14:25.240 --> 0:14:27.560 and can predict things like wind speed ahead of time, 0:14:27.760 --> 0:14:30.640 which allows teams to break out the right aerodynamic gear 0:14:30.720 --> 0:14:33.640 on any given day. You've seen those helmets, Like everything 0:14:33.720 --> 0:14:37.720 from cycling jerseys, shoes, socks, and the bikes themselves are 0:14:37.760 --> 0:14:41.080 all heavily tested in wind tunnels and tweak to achieve 0:14:41.160 --> 0:14:43.840 as little drag as possible. I just love sort of 0:14:43.840 --> 0:14:48.600 imagining these cyclists in these simulations looking a little bit crazy. 0:14:48.680 --> 0:14:52.000 But what's driven the embrace of technology in cycling was that, 0:14:52.120 --> 0:14:56.080 like a moneyball moment where the nerves took over the 0:14:56.120 --> 0:14:57.360 sport and displaced the jobs. 0:14:57.440 --> 0:15:00.240 What happened here, actually, yes, in the early in two 0:15:00.280 --> 0:15:04.920 thousand's your home, British Olympic Cycling team develop this philosophy 0:15:05.040 --> 0:15:08.200 of what they call marginal gains, which is, like Moneyball, 0:15:08.520 --> 0:15:11.840 you incrementally improve everything and hope that results in larger 0:15:11.880 --> 0:15:15.120 improvements when it's all combined. So because of that approach, 0:15:15.240 --> 0:15:17.800 the Brits actually won a bunch of Tor Defran's titles 0:15:17.840 --> 0:15:20.880 in the twenty tens as the doping lance era faded 0:15:20.920 --> 0:15:23.200 out and then the rest of the sport kind of 0:15:23.240 --> 0:15:26.160 followed the lead of the British Olympic Cycling team The 0:15:26.200 --> 0:15:28.360 most interesting part of the piece and the reason that 0:15:28.440 --> 0:15:31.880 I wanted to share it comes from this line. Today's 0:15:31.880 --> 0:15:35.800 generation of rising stars are digital natives, for whom ignoring 0:15:35.840 --> 0:15:38.840 the data and the appse is unthinkable. You can't win 0:15:38.920 --> 0:15:42.880 without them, which is just really interesting. In terms of sport, 0:15:42.960 --> 0:15:44.720 I think, or as you would say sport, I call 0:15:44.760 --> 0:15:47.040 it sports, but this idea that you have to be 0:15:47.120 --> 0:15:49.440 good at the computer essentially to be good at sports. 0:15:49.720 --> 0:15:51.640 Yeah, and to me, this is a very US story 0:15:51.680 --> 0:15:55.200 because it's yes about technology bustles, so about how technology 0:15:55.240 --> 0:15:57.320 is integrated into culture and human psychology. 0:15:57.480 --> 0:16:00.440 You're not allowed to opt out, No, have to be 0:16:00.440 --> 0:16:02.480 good at this. And you know, the point of the 0:16:02.520 --> 0:16:06.440 piece is really that a bunch of technological developments existed, 0:16:06.920 --> 0:16:09.960 but it took a new generation of writers to embrace them, or, 0:16:10.000 --> 0:16:13.080 as The Atlantic puts it, because writers tend to be conservative, 0:16:13.200 --> 0:16:16.600 even superstitious in their loyalty to tried routines. Shifting the 0:16:16.640 --> 0:16:19.560 culture took some time, and it's it's working so well 0:16:19.600 --> 0:16:22.160 that people look like writers look like they're cheating. 0:16:21.880 --> 0:16:22.600 Which maybe they are. 0:16:22.680 --> 0:16:24.080 I know that we don't. We don't really know. 0:16:24.080 --> 0:16:27.240 The history will be their judge. So what's the takeaway 0:16:27.240 --> 0:16:29.960 for you here? Really? I mean, obviously I remember back 0:16:30.000 --> 0:16:31.920 in the Lance Armstrong days, people were like he has 0:16:31.960 --> 0:16:34.240 a special way of peddling, and like that's what it's 0:16:34.280 --> 0:16:36.280 all about. Do you think we will look back on 0:16:36.280 --> 0:16:37.960 this and say, oh my god, that the real science 0:16:37.960 --> 0:16:40.200 and tech innovation was around and you kind of doping 0:16:40.720 --> 0:16:42.400 or do you buy that this kind of like the 0:16:42.440 --> 0:16:47.880 Atlantic article suggests, collection of tech innovations when they coalesce 0:16:48.000 --> 0:16:51.200 and a widely adopted can transform human performance. 0:16:51.560 --> 0:16:54.000 I mean, I think this is happening in every sector, 0:16:54.240 --> 0:16:56.320 not just in sport. Which is this idea that if 0:16:56.360 --> 0:16:59.520 you don't know how to use the newest technology, you 0:16:59.600 --> 0:17:02.680 will all behind these days if you are going to 0:17:02.760 --> 0:17:08.000 perform in any realm, if you are not consistently optimizing yourself, 0:17:08.640 --> 0:17:10.600 you're kind of left in the dust. And I think 0:17:10.600 --> 0:17:13.880 the same is true and quite literally true in cycling. 0:17:20.600 --> 0:17:25.040 After the break, erasing history at Google, digitizing plant pollen, 0:17:25.359 --> 0:17:28.840 and radioactive rhinos plus, a listener calls in to our 0:17:28.920 --> 0:17:48.440 Chat and Me segment Stay with us, Welcome back. We've 0:17:48.440 --> 0:17:50.480 got a few more headlines for you this week. 0:17:50.320 --> 0:17:53.280 And then a Chat and Me segment submitted by chat 0:17:53.359 --> 0:17:54.240 gbt itself. 0:17:54.560 --> 0:17:57.000 Before we launch into the headlines, I'd love to remind 0:17:57.000 --> 0:17:59.480 listeners that we really do want to feature you in 0:17:59.520 --> 0:18:03.439 the chat segment. So if you found yourself turning to Chachibt, 0:18:03.760 --> 0:18:07.040 grock Claw, Gemini, or any other chatbot for help with 0:18:07.080 --> 0:18:11.520 an unusual task or to answer life's deepest questions, please 0:18:11.520 --> 0:18:13.800 send us a one to two minute voice note to 0:18:13.920 --> 0:18:17.320 tech Stuff podcast at gmail dot com. But first, Kara, 0:18:17.600 --> 0:18:20.040 we talk from time to time about how Google's search 0:18:20.119 --> 0:18:24.439 business is under threat from chatbots. Chatchibt recently reached a 0:18:24.520 --> 0:18:27.600 billion daily searches. Do you know how many searches there 0:18:27.600 --> 0:18:28.679 are daily on Google? 0:18:30.359 --> 0:18:31.040 Three billion? 0:18:31.280 --> 0:18:34.679 Well, fourteen billion, geez. At least that's according to a 0:18:34.720 --> 0:18:38.280 researcher called rand Fishkin, whose work I've found on Google 0:18:38.320 --> 0:18:42.879 Search and amidst all the excitement about AI, Google Search 0:18:43.000 --> 0:18:46.720 does kind of remain the world's source of truth, which means, 0:18:46.720 --> 0:18:49.560 of course there are powerful incentives to mess with it. 0:18:49.640 --> 0:18:53.680 Is there a specific scandal a bruin around search? 0:18:53.920 --> 0:18:57.679 Yes, there is, a couple of journalists recently discovered a 0:18:57.720 --> 0:19:01.000 way to hack Google Search that could have some serious 0:19:01.000 --> 0:19:05.320 implications for censorship. Say more so, it turns out it's 0:19:05.320 --> 0:19:10.200 relatively easy to basically trick Google into suppressing information by 0:19:10.240 --> 0:19:12.800 removing certain pages from search. 0:19:12.720 --> 0:19:14.720 Which doesn't sound good. What does that mean? 0:19:14.960 --> 0:19:17.080 Well, it's good if you run a crisis pr film, 0:19:17.160 --> 0:19:20.600 I guess. But the crazy part to me about this 0:19:20.680 --> 0:19:24.000 story is it's being reported only because a journalist discovered 0:19:24.000 --> 0:19:24.880 it by accident. 0:19:25.200 --> 0:19:27.679 Is this a happy accident or a sad accident? How 0:19:27.680 --> 0:19:28.240 did this happen? 0:19:28.320 --> 0:19:30.520 Well, it's a happy accident. I suppose that the information 0:19:30.600 --> 0:19:33.440 has been revealed. The information itself is kind of sad 0:19:33.440 --> 0:19:35.960 and disturbing, which I'll get to. But how did he 0:19:36.040 --> 0:19:38.240 find it? I mean he spends time, I guess, like 0:19:38.320 --> 0:19:42.159 I do, googling himself. No, no, I'm kidding. The journalist is 0:19:42.200 --> 0:19:45.680 called Jack Paulson. He is an investiative journalist who covers tech, 0:19:46.000 --> 0:19:47.840 and he was looking for an old article on Google 0:19:48.520 --> 0:19:51.600 when it turned out that wasn't there. Even when he 0:19:51.720 --> 0:19:54.919 searched the exact headline of his story in quotation marks, 0:19:55.680 --> 0:19:59.399 he discovered that it wasn't there, and in turn, he 0:19:59.480 --> 0:20:03.880 discovered this somewhat unknown SEO trick whereby anybody who wants 0:20:03.960 --> 0:20:07.439 to can essentially delist a page from Google Search. How 0:20:08.160 --> 0:20:11.000 it's a little complicated, but bear with me. 0:20:11.200 --> 0:20:11.560 It may. 0:20:11.920 --> 0:20:16.320 Google has this tool called refresh outdated Content. It essentially 0:20:16.440 --> 0:20:20.639 requests form to have Google updated search results for pages 0:20:20.720 --> 0:20:23.360 or images that no longer exist. But anybody can do this, yeah, 0:20:23.400 --> 0:20:26.600 basically to delist broken links. It turns out though, that 0:20:26.920 --> 0:20:30.240 if you submit a bunch of links to a website 0:20:30.280 --> 0:20:34.360 which has the same URL but with different letters capitalized, 0:20:34.760 --> 0:20:39.160 you can direct Google to what it thinks are broken links, 0:20:39.160 --> 0:20:41.720 because if a capitalization is wrong, you don't get to 0:20:41.720 --> 0:20:43.440 the page you want to go to, even though it 0:20:43.440 --> 0:20:46.159 has exactly the same letters. And if you route Google 0:20:46.200 --> 0:20:49.600