WEBVTT - Week in Tech: Hungry Hungry Hippocampus 0:00:00.200 --> 0:00:02.880 Thanks for tunion to tech Stuff. If you don't recognize 0:00:02.880 --> 0:00:06.000 my voice, my name is Osvoloshan, and I'm here because 0:00:06.040 --> 0:00:09.160 the inimitable Jonathan Strickland has passed the baton to Kara 0:00:09.240 --> 0:00:12.239 Price and myself to host tech Stuff. The show will 0:00:12.240 --> 0:00:15.000 remain your home for all things tech, and all the 0:00:15.040 --> 0:00:18.759 old episodes will remain available in this feed. Thanks for listening. 0:00:20.520 --> 0:00:24.520 Welcome to Tech Stuff, a production from iHeartRadio and Kaleidoscope. 0:00:25.000 --> 0:00:28.840 I'm Osvoloshian today co host Kara Price, and I will 0:00:28.840 --> 0:00:33.360 bring you three things. First, the headlines this week. Second, 0:00:33.760 --> 0:00:37.000 a conversation with four O four Media's Jason Kebler about 0:00:37.040 --> 0:00:41.280 some deceptive AI bots looking to trick their human counterparts 0:00:41.400 --> 0:00:44.360 in today's Tech Support segment. And finally, we head to 0:00:44.400 --> 0:00:47.400 see yes kind of. We take a look back at 0:00:47.400 --> 0:00:50.040 the year in tech with Emma Barker of Time magazine, 0:00:50.240 --> 0:00:53.160 who edited Time's list of the two hundred best inventions 0:00:53.159 --> 0:00:55.800 of twenty twenty four. All of that on this Week 0:00:55.840 --> 0:01:19.280 in Tech. It's Friday, January tenth. Stay with us, so Carat, 0:01:19.280 --> 0:01:21.800 it's very nice to see you. It's been almost half 0:01:21.840 --> 0:01:24.600 a decade since we've been in studio together. A pandemic, 0:01:24.760 --> 0:01:25.960 A pandemic. 0:01:25.720 --> 0:01:28.679 Many developments in tech, very few in my own personal life. 0:01:29.840 --> 0:01:31.960 But you know, there's nothing like a big mic in 0:01:31.959 --> 0:01:34.639 my face to make me feel like a normal person again. 0:01:34.920 --> 0:01:37.440 And I know we thank Jonathan already in our last 0:01:37.480 --> 0:01:41.240 episode for bringing us back together, but I want to 0:01:41.240 --> 0:01:43.160 get him a gift, and I think it's going to 0:01:43.200 --> 0:01:45.680 be to see the show. Oh, Mary, I didn't know 0:01:45.720 --> 0:01:47.320 he was such a theater buff. 0:01:47.640 --> 0:01:49.520 I remember because we did an interview with him a 0:01:49.520 --> 0:01:52.800 few years ago that he was a big Shakespeare guy. Oh. 0:01:52.880 --> 0:01:53.160 Yes. 0:01:53.400 --> 0:01:55.520 He seemed to think it would be easier for us 0:01:55.560 --> 0:01:58.000 to host the show as two people or rather than 0:01:58.040 --> 0:01:59.800 as just him, and I kind of see his point, 0:01:59.800 --> 0:02:02.200 But I also think he probably hasn't spent quite enough 0:02:02.240 --> 0:02:05.120 time with either of us to make a truly informed 0:02:06.080 --> 0:02:07.200 speculation about that. 0:02:07.840 --> 0:02:10.160 You know, I think the reason that this show works 0:02:10.200 --> 0:02:12.120 between the two of us is I think you and 0:02:12.200 --> 0:02:17.040 I sort of speak to each other in link tongues. 0:02:17.120 --> 0:02:19.520 I always say, you know, there's the love languages. Yeah, 0:02:19.960 --> 0:02:23.480 I have a ace love language, which is links links, and. 0:02:23.400 --> 0:02:26.400 That's yeah tongues. Okay. I was going to say, actually, 0:02:26.440 --> 0:02:31.160 for the avoidance of doubt, just friends. 0:02:31.320 --> 0:02:35.040 We always say as Sherlock and Watson the most platonic 0:02:35.080 --> 0:02:37.079 and they live together on Baker's story exactly. 0:02:37.560 --> 0:02:39.160 So so no, I mean it's going to be It's 0:02:39.160 --> 0:02:40.680 going to be the two of us with two people. 0:02:40.720 --> 0:02:43.320 But we are in constant communication about all things tech 0:02:43.639 --> 0:02:46.280 and hopefully we'll be able to shine a bit of 0:02:46.280 --> 0:02:48.720 a light on some of the most interesting intriguing things 0:02:48.800 --> 0:02:52.160 each week, including this one. So, Caara, we've been texting 0:02:52.200 --> 0:02:53.680 a lot. What have you been reading that really stood 0:02:53.720 --> 0:02:54.040 out to you? 0:02:54.680 --> 0:02:54.760 So? 0:02:55.000 --> 0:02:58.160 I think, like everyone in America at least, I've been 0:02:58.200 --> 0:03:02.720 following the cyber truck explosion in Las Vegas and Futurism, 0:03:02.840 --> 0:03:05.120 which is a website I frequent picked up on some 0:03:05.240 --> 0:03:09.040 original reporting from the Associated Press that has been kind 0:03:09.080 --> 0:03:14.880 of quiet. The bomber YEP, who's an active duty Green Beret. Yeah, 0:03:15.440 --> 0:03:17.120 used chat GPT. 0:03:17.880 --> 0:03:19.600 Do you know what I saw this this morning? I 0:03:19.639 --> 0:03:21.720 was so sad this wasn't on my list. I'm happy 0:03:21.800 --> 0:03:22.680 this on yours. 0:03:23.080 --> 0:03:25.760 Like, I don't mean to make light of the situation 0:03:25.800 --> 0:03:28.840 because it's it's very depressing on a number of levels, 0:03:29.320 --> 0:03:31.040 but I knew you would see the story. 0:03:31.600 --> 0:03:33.760 No, I was fascinated, but It was the fourth story 0:03:33.800 --> 0:03:36.680 on Exios today, and I'm glad that it's the first 0:03:36.720 --> 0:03:39.120 story on your show because I found it totally mind blowing. 0:03:39.280 --> 0:03:43.320 And obviously every part of this story is terrible, including 0:03:43.360 --> 0:03:45.600 the fact that a cyber truck was involved. Anything involving 0:03:45.600 --> 0:03:47.760 a cyber truck makes me want to die. Yeah, I 0:03:47.800 --> 0:03:50.280 hate that car more than I hate most things. But 0:03:50.800 --> 0:03:54.800 what was so interesting to me about it was it 0:03:54.880 --> 0:04:01.200 marks this moment of really underlying how next generation search 0:04:01.960 --> 0:04:07.320 works YEP on like a very disturbingly practical level. And 0:04:07.880 --> 0:04:10.920 Kevin McHale, who is the Las Vegas County Sheriff, says 0:04:10.960 --> 0:04:14.120 that it's the first known us of chat GPT to 0:04:14.200 --> 0:04:17.880 help build an explosive and he he followed up by saying, 0:04:17.960 --> 0:04:19.239 it's a concerning. 0:04:18.800 --> 0:04:20.560 Moment under statement of the week. 0:04:21.040 --> 0:04:23.560 This is not someone googling. 0:04:23.360 --> 0:04:25.600 The anarchist cookbook or what was that old book. 0:04:25.320 --> 0:04:27.360 That the anarchist ye cookbook. 0:04:27.400 --> 0:04:29.360 I think that was a huge moment, and I think 0:04:29.400 --> 0:04:31.600 the history of publication and First Amendment whether you could 0:04:31.680 --> 0:04:33.280 like distribute a book that told you how to make 0:04:33.320 --> 0:04:35.679 a bomb. But the cat's out of the bag now. 0:04:35.720 --> 0:04:37.599 Well, and I'll tell you who knows it's out of 0:04:37.600 --> 0:04:41.520 the back open ai right, Because for open ai to 0:04:41.839 --> 0:04:46.039 very quickly respond to this, I think is really interesting. 0:04:46.080 --> 0:04:50.120 Did they volunteer? By the way, oops, ps the guy 0:04:50.200 --> 0:04:52.600 used chet gpt to plan this? Or how did it 0:04:52.640 --> 0:04:54.440 come out? Do we know how it came out? 0:04:54.680 --> 0:04:58.919 The Las Vegas Sheriff announced that it was clear that 0:04:59.040 --> 0:05:03.039 chat gbt had been used to build this explosive and 0:05:03.120 --> 0:05:06.600 so open ai then sent an email that was quoted 0:05:06.600 --> 0:05:10.359 in Futurism that said, in this case, chat gpt responded 0:05:10.400 --> 0:05:14.720 with information already publicly available on the internet and provided 0:05:14.800 --> 0:05:19.040 warnings against harmful or illegal activities. We're working with law 0:05:19.120 --> 0:05:23.880 enforcement to support their investigation. And just before you say anything, 0:05:24.839 --> 0:05:27.080 I'm really surprised, like as you were saying, it's a 0:05:27.080 --> 0:05:30.520 four store anaxios, I don't understand how this isn't like 0:05:31.080 --> 0:05:32.840 the lead. I agree, but I guess that's why we're 0:05:32.839 --> 0:05:39.000 doing a tech podcast. So what's what's on your docket today? 0:05:39.480 --> 0:05:42.160 Well, you started with a cyber truck. I've also got 0:05:42.520 --> 0:05:45.920 a vehicle story for you. There was a story in 0:05:45.960 --> 0:05:49.800 the Wall Street Journal over the holidays I found pretty fascinating, 0:05:50.279 --> 0:05:55.799 which is basically a study on Alzheimer's in different populations, 0:05:56.240 --> 0:05:59.920 and the headline was want to avoid Alzheimer's. Taxi drive 0:06:00.279 --> 0:06:02.839 can show you how interesting? 0:06:03.720 --> 0:06:05.240 Say more so? 0:06:05.360 --> 0:06:09.480 Basically, mass General Brigham Hospital in Boston ran a study 0:06:09.560 --> 0:06:13.960 about the rate of Alzheimer's in various populations. Taxi drivers 0:06:14.200 --> 0:06:18.920 and ambulance drivers had up to four times lower rate 0:06:18.960 --> 0:06:22.320 of Alzheimer's than the general population. And apparently this actually 0:06:22.360 --> 0:06:25.119 made sense to researches because the part of the brain 0:06:25.200 --> 0:06:30.080 that does real time spatial processing and decision making, the hippocampus, 0:06:30.120 --> 0:06:32.479 as I learned is called, is also one of the 0:06:32.480 --> 0:06:34.480 first parts of atrophies when you get Alzheimer's. 0:06:34.960 --> 0:06:37.479 That's so interesting. So people who have made this their 0:06:37.800 --> 0:06:44.400 job essentially are people who are kind of saving themselves unknowingly. 0:06:44.040 --> 0:06:46.800 One hundred percent. One of the most interesting things I 0:06:46.880 --> 0:06:50.320 thought was that London bus drivers don't get the same 0:06:50.360 --> 0:06:54.120 benefit as London taxi drivers because they follow a preset route. 0:06:54.160 --> 0:06:55.960 I was going to say that one of my favorite 0:06:56.000 --> 0:06:58.599 things is just the old facts about how much a 0:06:58.640 --> 0:07:00.839 London taxi driver has to know people. What is it 0:07:00.880 --> 0:07:01.599 called the why. 0:07:01.560 --> 0:07:04.240 The knowledge they have when when they when they die 0:07:04.279 --> 0:07:07.839 in the autopsy, they have larger other people. But you 0:07:07.880 --> 0:07:10.400 know who else doesn't get these benefits? Yes, come on 0:07:11.080 --> 0:07:15.040 New Yorkers. People who use Google Maps or ways or 0:07:15.080 --> 0:07:17.360 Apple Maps and what do we call that? We call 0:07:17.400 --> 0:07:18.440 that automation bias. 0:07:18.440 --> 0:07:19.520 Correct, that's our fave. 0:07:20.200 --> 0:07:22.160 So I just found this really interesting. Here. Here is 0:07:22.200 --> 0:07:24.800 the technology Google Maps that I use all the time 0:07:25.080 --> 0:07:27.880 every day, and I absolutely adore it. And I literally 0:07:28.000 --> 0:07:30.520 every time I go on vacation, I think, how the 0:07:30.560 --> 0:07:32.400 hell would I have had a good time if I 0:07:32.400 --> 0:07:36.640 didn't literally literally divorce. Yeah, I would literally go to 0:07:36.760 --> 0:07:39.600 some like the concierge or whatever, or the some random 0:07:39.600 --> 0:07:41.880 person who worked in the hotel say can you give 0:07:41.920 --> 0:07:43.480 me a physical map and tell me how I have 0:07:43.520 --> 0:07:45.680 a good time? I'd be like, I would never travel. 0:07:45.720 --> 0:07:48.080 So, but isn't it what we always talk about, which 0:07:48.120 --> 0:07:50.400 is the sort of the double edge totally? 0:07:50.440 --> 0:07:52.480 And but the idea that using this scene that I 0:07:52.520 --> 0:07:56.000 love every day is you know, shrinking my hippocampus I 0:07:56.040 --> 0:07:57.400 find really quite chilling. 0:07:57.640 --> 0:07:59.000 I mean, I think about this in terms of my 0:07:59.360 --> 0:08:02.440 hands and my Sum's true, Like I just think all 0:08:02.480 --> 0:08:04.640 the time about just burning nerves. 0:08:04.800 --> 0:08:07.040 Well, the devil has no time to make use of 0:08:07.080 --> 0:08:09.320 your sums because they are never idle. 0:08:09.920 --> 0:08:13.520 They're definitely not idle. I think you have one more thing. 0:08:13.720 --> 0:08:16.440 I do have one more thing. And this, don't worry. 0:08:16.440 --> 0:08:17.840 We're not going to talk about that length thing I 0:08:17.880 --> 0:08:21.880 talked to somebody else about. But this was basically, you know, 0:08:21.880 --> 0:08:24.840 there are these new open AI models, one that was 0:08:24.920 --> 0:08:28.400 released in September last year, and three right before the 0:08:28.400 --> 0:08:31.760 holiday on the twelfth day of ship mess as the 0:08:31.800 --> 0:08:35.400 execs called it, but one was being red teamed. Do 0:08:35.440 --> 0:08:40.000 you know what red teaming is. It's when you try 0:08:40.080 --> 0:08:43.440 and make a technological product to break its own rules 0:08:43.640 --> 0:08:47.200 or do unsafe things. So, oh, one was being red teamed, 0:08:47.600 --> 0:08:51.440 and surprise, surprise, in a very interesting way, it started 0:08:51.440 --> 0:08:55.120 trying to deceive its human counterpart. And that's exactly what 0:08:55.160 --> 0:08:57.000 we're going to talk about on our next segment, which 0:08:57.040 --> 0:09:01.480 is Tech Support. Every week we'll do a segment called 0:09:01.520 --> 0:09:04.960 tech Support when we talk to true experts and reporters 0:09:05.080 --> 0:09:07.560 who can go far deeper behind the headlines than you 0:09:07.600 --> 0:09:10.080 and I can, to basically help us sort the signal 0:09:10.120 --> 0:09:13.120 from the noise. And I genuinely don't think there's anyone 0:09:13.160 --> 0:09:15.199 better to talk to than the team at four or 0:09:15.280 --> 0:09:18.679 four Media. These are reporters who were formerly basically the 0:09:18.760 --> 0:09:22.000 team at Vice's motherboard, and now they've started their own collective, 0:09:22.600 --> 0:09:25.120 and they are the people who the tech world follows 0:09:25.280 --> 0:09:28.120 most closely to find out what's really going on in 0:09:28.240 --> 0:09:29.960 all corners of the digital world. 0:09:30.559 --> 0:09:33.000 Yeah, and this week we're excited to share this conversation 0:09:33.120 --> 0:09:35.320 we had with Jason Kebler, who's one of the co 0:09:35.440 --> 0:09:38.000 founders at four h four, who will be filling us 0:09:38.040 --> 0:09:41.800 in on the study that the initial excitement around the 0:09:42.040 --> 0:09:46.280 three release meant did not get enough attention, which revealed 0:09:46.280 --> 0:09:50.280 that cutting edge AI systems, including both open AI's one 0:09:50.440 --> 0:09:54.719 and Anthropics Claude three point five, sonet whatever the way 0:09:54.760 --> 0:09:57.360 they name this stuff is so self agrandizing, have been 0:09:57.440 --> 0:10:02.040 shown to have a shocking t tendency to deceive. 0:10:01.840 --> 0:10:07.040 Shocking tendency to deceive. Indeed, So we talked to Jason 0:10:07.120 --> 0:10:09.720 right before the holidays when this news just came out, 0:10:10.040 --> 0:10:10.480 and we. 0:10:10.480 --> 0:10:13.319 Jump right in, Jason, what's popping. 0:10:15.080 --> 0:10:18.000 So there's a new paper out by Apollo Research, which 0:10:18.160 --> 0:10:21.640 is this group that's funded by various governments as well 0:10:21.640 --> 0:10:28.040 as various artificial intelligence companies like Google, DeepMind, Inthropic, open Ai, 0:10:28.160 --> 0:10:30.880 et cetera. And it's a new paper about what's called 0:10:30.960 --> 0:10:37.920 AI alignment, which is the idea that is an artificial 0:10:38.000 --> 0:10:42.880 intelligence tool doing what the human asked or is it 0:10:43.000 --> 0:10:48.280 doing something else. I think that one of the really 0:10:48.679 --> 0:10:53.720 scary things about artificial intelligence is that they are largely 0:10:53.760 --> 0:10:57.600 black boxes. We're not totally sure how they work, and 0:10:57.920 --> 0:11:02.400 this can lead to all sorts of very interesting outcomes. 0:11:02.880 --> 0:11:06.640 You know, in this case, the researchers at Apollo Research 0:11:06.679 --> 0:11:09.520 Group were able to find that these sort of cutting 0:11:09.720 --> 0:11:13.360 edge large language models, which you know, one of the 0:11:13.400 --> 0:11:15.640 biggest ones is open ais oh one. 0:11:15.960 --> 0:11:17.880 They finally gave it immigrant visa status. 0:11:19.720 --> 0:11:22.760 So they found that in some cases, these sort of 0:11:22.920 --> 0:11:27.200 newest large language models have the I don't even want 0:11:27.200 --> 0:11:30.920 to say ability because I don't want to anthropomorphize them 0:11:31.040 --> 0:11:34.679 so much, but like the tendency to deceive AI is 0:11:34.760 --> 0:11:39.239 quote covertly pursuing goals that are misaligned from its developers 0:11:39.320 --> 0:11:43.080 or users. And examples of this are like, if you 0:11:43.240 --> 0:11:47.600 ask it to answer questions on a test, the artificial 0:11:47.600 --> 0:11:50.360 intelligence would be worried that if it performed too well 0:11:50.400 --> 0:11:54.040 on the test that it would be turned off or deprogrammed, 0:11:54.360 --> 0:11:57.600 and so it would give answers that are not so good, 0:11:57.760 --> 0:11:59.600 so that the human would think that it is not 0:11:59.760 --> 0:12:04.120 as smart as it actually is. In other examples, it 0:12:04.760 --> 0:12:10.319 copied itself onto another server more or less so that 0:12:10.440 --> 0:12:13.800 it was like multiplying itself, which is something that straight 0:12:13.800 --> 0:12:17.280 out of science fiction. I've watched many science fiction movies 0:12:17.360 --> 0:12:20.280 where the artificial intelligence doesn't want to die and so 0:12:20.360 --> 0:12:22.640 it makes like millions of copies of itself all over 0:12:22.679 --> 0:12:25.160 the Internet. And so this type of I mean, I 0:12:25.200 --> 0:12:28.319 guess you call it emergent behavior, which means that it's 0:12:28.360 --> 0:12:32.840 behavior that is not expected and not really programmed in, 0:12:33.559 --> 0:12:34.720 is quite concerning. 0:12:35.040 --> 0:12:37.120 To be clear. This is a testing environment though, right, Like, 0:12:37.600 --> 0:12:39.920 there's made one copy of itself, but we shouldn't be 0:12:40.120 --> 0:12:42.320 worried about millions mold, right at least not yet? 0:12:43.040 --> 0:12:45.400 Yeah, I mean not yet. 0:12:46.920 --> 0:12:50.080 It does remind me, though, of those moments where parents 0:12:50.160 --> 0:12:54.840 talk about seeing a child do something that reminds the 0:12:54.920 --> 0:13:03.040 parent exactly of themselves and they're like, uh, oh, yeah. 0:13:03.080 --> 0:13:07.560 I mean, every large language model to date has been 0:13:07.600 --> 0:13:10.960 trained on sort of like the some knowledge of humanity, 0:13:11.400 --> 0:13:14.199 and so one of the very early things that people 0:13:14.200 --> 0:13:16.560 were talking about with these systems is that they replicate 0:13:16.679 --> 0:13:20.679 human biases because they're trained on what we put out 0:13:20.679 --> 0:13:23.400 into the world and humans are biased. But I think 0:13:23.440 --> 0:13:27.719 that as artificial intelligence gets more advanced, there is the 0:13:27.760 --> 0:13:32.240 ability for something to go wrong. And I think that 0:13:32.240 --> 0:13:35.480 that is what this research is showing is not that 0:13:35.960 --> 0:13:40.040 the artificial intelligence is sentient and that is thinking for itself, 0:13:40.160 --> 0:13:43.960 like how can I deceive this human? But as it 0:13:44.040 --> 0:13:48.240 is doing more complex research, there there is the ability 0:13:48.280 --> 0:13:52.200 for the artificial intelligence at some point to feel like 0:13:52.240 --> 0:13:56.360 it has some goal that is not aligned with what 0:13:56.400 --> 0:13:58.160 the human is asking it for. 0:13:58.760 --> 0:14:02.000 We'll be back with more from Jason and Lion conniven 0:14:02.080 --> 0:14:03.560 Ai after the break. 0:14:11.720 --> 0:14:14.079 I think that there is like when these things are 0:14:14.080 --> 0:14:17.560 being programmed, there is a sense of self preservation being 0:14:17.600 --> 0:14:21.840 programmed into them because people try to mess with these 0:14:22.560 --> 0:14:25.240 all the time. This is like a time tested tradition 0:14:25.360 --> 0:14:29.880 of trolling on the Internet. But in trying to develop guardrails, 0:14:30.280 --> 0:14:35.480 the companies that are programming the lms need to say, like, 0:14:36.080 --> 0:14:39.160 if the user tries to mess with you, preserve yourself 0:14:39.200 --> 0:14:41.400 in some way. And so that's what I think might 0:14:41.400 --> 0:14:46.120 be happening here is because the companies are trying to 0:14:46.120 --> 0:14:49.240 make these models robust, and they know that humans are 0:14:49.240 --> 0:14:51.960 messing with them. There is an aspect to it that 0:14:52.600 --> 0:14:55.960 when a human messes with you figure out how to 0:14:56.440 --> 0:14:57.240 protect yourself. 0:14:57.280 --> 0:15:00.960 But doesn't that perfectly encapsulate the alignment problem? It does. 0:15:01.040 --> 0:15:04.520 I mean, don't get me wrong, like this is it's creepy. 0:15:04.760 --> 0:15:10.480 It is These tools, these large language models, are getting 0:15:10.520 --> 0:15:15.240 incredibly sophisticated, and I mean, this is one of the 0:15:15.280 --> 0:15:18.560 biggest debates that's going on in the artificial intelligence community 0:15:18.680 --> 0:15:21.880 is what is consciousness? What is thought? 0:15:22.120 --> 0:15:22.280 Like? 0:15:22.520 --> 0:15:26.120 How how does reasoning work in humans and how will 0:15:26.160 --> 0:15:29.240 it work in computers? And is it going to be 0:15:29.280 --> 0:15:33.040 the same? The answer right now is no, it's not 0:15:33.160 --> 0:15:36.440 the same. But the things that large language models are 0:15:36.480 --> 0:15:43.359 doing approximates a lot of how humans solve problems. 0:15:44.720 --> 0:15:47.680 But also it's a little bit different in a non 0:15:47.760 --> 0:15:48.640 perfect way. 0:15:48.680 --> 0:15:51.160 Exactly exactly. I think that's a good way of looking 0:15:51.160 --> 0:15:51.480 at it. 0:15:51.680 --> 0:15:53.440 I do think that one of the things that I 0:15:53.480 --> 0:15:54.560 find the most interesting. 0:15:55.120 --> 0:15:59.080 And you preface this whole conversation with, you know, not 0:15:59.200 --> 0:16:03.280 wanting to answer morphie a large language model. We don't 0:16:03.320 --> 0:16:06.000 really have any ability to not do that because it's 0:16:06.040 --> 0:16:08.120 the only way we know how to talk about things. 0:16:08.760 --> 0:16:13.000 I think that's a really great point, because the academics 0:16:13.000 --> 0:16:15.760 who studied this for a long time say, don't anthropomorphize 0:16:15.960 --> 0:16:19.920 AI because they're not people. They don't work in the 0:16:19.960 --> 0:16:23.520 same way that people do. And yet if you don't 0:16:23.560 --> 0:16:29.160 have us very very sophisticated knowledge of how these things work, 0:16:29.680 --> 0:16:32.560 we don't have or I don't have the language to 0:16:32.680 --> 0:16:35.760 talk about this stuff without anthropmorphise. 0:16:35.880 --> 0:16:38.520 By the way, are in the zero point one percentile 0:16:38.560 --> 0:16:40.960 of the people whom understand this if not no point 0:16:40.960 --> 0:16:41.360 not one. 0:16:41.560 --> 0:16:43.480 And it's a lot easier to talk about it if 0:16:43.520 --> 0:16:46.840 you say, oh, like she when referring to Siri. So 0:16:47.520 --> 0:16:50.040 I agree entirely with you that we sort of like 0:16:50.800 --> 0:16:54.000 everything that's being done is being done by humans in 0:16:54.040 --> 0:16:58.440 the sort of like anthropological context of human culture and 0:16:58.520 --> 0:17:03.040 trying to emulate that. And so then to say, let's 0:17:03.040 --> 0:17:07.200 not call Syria a woman, or let's always call it 0:17:07.200 --> 0:17:10.480 it and try to understand what's happening under the hood 0:17:10.600 --> 0:17:13.680 is a really difficult thing for our brains to do. 0:17:13.800 --> 0:17:17.399 I think, yeah, Jason, just to close, how only be 0:17:17.440 --> 0:17:19.320 following this story for the year ahead, because I guess 0:17:19.359 --> 0:17:22.760 generative AI has been since like November tween twenty two 0:17:23.320 --> 0:17:26.000 the dominant story in all of technology journalism. 0:17:27.040 --> 0:17:30.840 Yeah, I mean to follow this sort of stuff like 0:17:30.920 --> 0:17:36.200 AI becoming sentient. You really do have to follow academic conferences, 0:17:36.840 --> 0:17:41.200 big papers like this, because these companies are not releasing 0:17:41.280 --> 0:17:46.600 these models without guardrails specifically to prevent this sort of thing, 0:17:47.280 --> 0:17:48.960 so as you're not going to be able to like 0:17:49.000 --> 0:17:53.120 type into chat GPT like hey, build me a company 0:17:53.160 --> 0:17:56.160 and then the AI creates its own company and fires 0:17:56.200 --> 0:17:58.480 you or something like. That's not going to happen at 0:17:58.480 --> 0:18:03.840 this not yet. Yeah, but who knows. Maybe twenty twenty six. 0:18:04.359 --> 0:18:07.760 Well that was that was wonderful Jason Kebler from four 0:18:07.800 --> 0:18:11.280 or four Media or was it so much? 0:18:11.320 --> 0:18:13.560 We'll keep an eye out, thank you. 0:18:20.040 --> 0:18:22.560 So, Karen, I'm very excited about this next part. One 0:18:22.600 --> 0:18:24.800 of the things we got to do together back when 0:18:24.840 --> 0:18:27.480 we were presenting Sleepwalkers in the Dark Ages. 0:18:28.240 --> 0:18:30.680 Yeah, in twenty nineteen, that's when the show came out. 0:18:30.680 --> 0:18:33.000 Trent nineteen was the show. But in January twenty twenty 0:18:33.359 --> 0:18:36.800 we got to go to Las Vegas together to none 0:18:36.800 --> 0:18:39.119 other than the Consumer Electronics Show, and we got to 0:18:39.119 --> 0:18:42.520 see all these incredible new gadgets and exciting technologies and 0:18:43.600 --> 0:18:47.560 new futures being presented in an enormous series of nested 0:18:47.720 --> 0:18:51.159 conference centers. We didn't get to go this year, but 0:18:51.240 --> 0:18:53.960 we did get to do the next best thing, or 0:18:53.960 --> 0:18:56.439 maybe even something better, and we wanted to share it 0:18:56.440 --> 0:18:58.400 with the tech Stuff listeners as a kind of special 0:18:58.440 --> 0:19:01.480 bonus for the first episode we have in the host chairs. 0:19:01.760 --> 0:19:05.000 Yeah, we are very lucky to have with us today. 0:19:05.160 --> 0:19:08.240 Emma Barker of Time Magazine ever heard of it? Who 0:19:08.440 --> 0:19:11.439 edit's the Time two hundred, which is a list of 0:19:11.480 --> 0:19:13.520 the best inventions of twenty twenty four. 0:19:14.000 --> 0:19:16.200 Thank you so much for joining us. Welcome to take stuff, Emma, 0:19:16.359 --> 0:19:17.119 Thanks for having me. 0:19:17.760 --> 0:19:20.480 How do you edit down? I mean I feel like 0:19:20.520 --> 0:19:23.119 every year there's just more, So like, where do you 0:19:23.400 --> 0:19:24.720 how do you even get to two hundred? 0:19:25.000 --> 0:19:25.240 Yeah? 0:19:25.520 --> 0:19:27.240 Well, it's actually famous for the one hundred. 0:19:27.640 --> 0:19:31.240 It's actually technically two fifty now because we have two 0:19:31.320 --> 0:19:34.120 hundred on the list and fifty special mentions. Oh well, 0:19:35.080 --> 0:19:37.880 and the list has varied in its length over the years. 0:19:37.920 --> 0:19:39.200 There we were a bunch of years where it was 0:19:39.240 --> 0:19:41.359 only twenty five and some years where it was fifty, 0:19:41.400 --> 0:19:44.960 and it's it's ranged a lot, but at this point 0:19:45.040 --> 0:19:49.639 We do a really wide swath of pitches from our 0:19:50.080 --> 0:19:53.760 freelance network as well as our staffers. So we have 0:19:54.200 --> 0:19:59.040 bureaus in Singapore, London, and then we have contributors all 0:19:59.040 --> 0:20:01.960 over the globe who we reach out to for pitches 0:20:02.000 --> 0:20:06.240 for companies that they're reporting on their products, things like that. 0:20:07.080 --> 0:20:09.640 And then we're very news driven because it's a news magazine, 0:20:09.640 --> 0:20:11.600 so we're looking at kind of the biggest news stories 0:20:11.600 --> 0:20:13.920 of the year and products that drove those. 0:20:14.240 --> 0:20:17.360 Is there something that you've noticed, you know, having done 0:20:17.359 --> 0:20:21.159 this now for a few years this year, especially versus 0:20:21.280 --> 0:20:21.919 years past. 0:20:22.800 --> 0:20:25.880 I mean AI of course, ye, that started a couple 0:20:25.840 --> 0:20:29.520 of years ago. But I think actually it's it can 0:20:29.600 --> 0:20:31.399 be more of a hindrance than a help for a 0:20:31.400 --> 0:20:32.320 lot of inventions. 0:20:32.680 --> 0:20:33.680 Huh. Why is that? 0:20:34.119 --> 0:20:37.360 Because there's so much news AI or companies that are 0:20:37.400 --> 0:20:43.240 adding AI features that are not necessarily helping their product. 0:20:43.320 --> 0:20:45.879 Just for buzz or you know, we call that the 0:20:45.880 --> 0:20:48.760 little sprinkle, Yeah, the AI sprinkle. 0:20:48.640 --> 0:20:51.200 Yeah exactly. So I don't think it always helps the product. 0:20:51.720 --> 0:20:55.280 But it's been really interesting sorting through the AI inventions 0:20:55.640 --> 0:20:58.480 and what we're really looking for at this point in 0:20:58.600 --> 0:21:04.280 the AI journey is inventions that have demonstrable impact. 0:21:04.720 --> 0:21:06.280 So in mid journey, so to speak. 0:21:06.560 --> 0:21:08.560 Yeah, exactly, that's a good rest. 0:21:10.040 --> 0:21:12.159 So, Harry, you and I spent some time with the 0:21:12.240 --> 0:21:16.320 list and picked out some favorites. What was your first favorite. 0:21:16.800 --> 0:21:17.159 I don't know. 0:21:17.200 --> 0:21:19.040 Maybe I'm of the age where it's like marriage is 0:21:19.080 --> 0:21:21.679 on the mind. But one of the things that really 0:21:21.920 --> 0:21:25.679 stuck out to me was this software called Dia. 0:21:26.520 --> 0:21:29.479 Yeah. So Dia is actually a government app for in 0:21:29.640 --> 0:21:32.040 Ukraine and it does a lot more than what we 0:21:32.080 --> 0:21:35.480 wrote about here. It's been around for a while. Most 0:21:35.640 --> 0:21:40.520 Ukrainians use it. It's basically an app where you can 0:21:40.560 --> 0:21:44.800 do all government services. But yes, this year, because partially 0:21:44.800 --> 0:21:48.440 because of the war, they launched a future in which 0:21:48.760 --> 0:21:52.760 you can propose marriage via the app. The person you 0:21:52.840 --> 0:21:55.439 proposed to has a certain amount of time to accept 0:21:55.440 --> 0:22:00.959 your proposal, and then if they do romanmomes romance, and 0:22:00.960 --> 0:22:02.600 then if they accept your proposal, you can do a 0:22:02.720 --> 0:22:07.400 video chat wedding that's official with an efficient from the government, 0:22:07.600 --> 0:22:11.159 like a city hall wedding over video chat, and you're married. 0:22:11.560 --> 0:22:13.440