WEBVTT - Week in Tech: Duping Big Tech’s Tech Test 0:00:01.320 --> 0:00:04.400 Welcome to tech Stuff, a production of iHeart Podcasts and 0:00:04.480 --> 0:00:08.879 Kaleidoscope IMA's Valoshan and today will bring you the headlines 0:00:08.920 --> 0:00:12.479 this week, including how the urge to be liked has 0:00:12.480 --> 0:00:15.960 found its way into LM's. Then on tech Support, we'll 0:00:15.960 --> 0:00:21.079 talk to Azimaza, researcher and founder of the Exponential View newsletter, 0:00:21.680 --> 0:00:25.959 about the latest AGI predictions and the unfolding AI arms race. 0:00:26.480 --> 0:00:37.960 All of that on the weekend tech It's Friday, March fourteenth, 0:00:38.159 --> 0:00:42.280 Another week, another AI agent. We'll discuss manus Ai coming 0:00:42.280 --> 0:00:45.240 out of China during our tech Support segment. But first 0:00:45.320 --> 0:00:47.080 let's kick off with some headlines that you may have 0:00:47.120 --> 0:00:49.479 missed as you scrambled to get an invite to use 0:00:49.520 --> 0:00:52.960 the latest model. Eliza Dennis, our producer, is here with me. 0:00:53.440 --> 0:00:55.720 Hey us, So this week, I know there's a story 0:00:55.760 --> 0:00:58.760 that you're obsessing over, so why don't you take it away? 0:00:59.160 --> 0:01:02.120 Absolutely So, this one was a super easy choice for 0:01:02.200 --> 0:01:04.920 me because this week I just really couldn't get enough 0:01:04.920 --> 0:01:09.440 of Sesames Conversational Speech Model or CSM. 0:01:09.840 --> 0:01:12.399 Now, I have to confess, when I first heard about 0:01:12.440 --> 0:01:15.160 this one, I thought it came from Sesame Workshop or 0:01:15.160 --> 0:01:18.360 Sesame Street. But I was wrong. 0:01:18.680 --> 0:01:21.679 Yes, so this is coming from a private company that's 0:01:21.800 --> 0:01:24.600 just come out of stealth mode. It's only a demo 0:01:24.680 --> 0:01:26.840 at the moment, but if you agree to these terms 0:01:26.840 --> 0:01:30.119 of service, you can chat to two different voices, Maya 0:01:30.200 --> 0:01:33.679 and Miles. So if you've managed like I did to 0:01:33.760 --> 0:01:37.399 avoid the many, many, many social media videos of people 0:01:37.480 --> 0:01:41.200 chatting and even arguing with these chatbots, it's really a 0:01:41.240 --> 0:01:42.280 surreal experience. 0:01:42.680 --> 0:01:44.480 What makes it different from talking to some of the 0:01:44.560 --> 0:01:46.760 like open AI direct voice models. 0:01:47.080 --> 0:01:51.440 I mean, this one does feel a little bit more natural, 0:01:51.600 --> 0:01:54.960 a little bit more human. You do feel like you're 0:01:55.000 --> 0:01:57.520 kind of crossing the line into the Uncanny Valley in 0:01:57.600 --> 0:02:02.640 some ways, and that's by design. It's something called voice presence, 0:02:02.840 --> 0:02:06.080 and Sesame says this is kind of this magical quality 0:02:06.120 --> 0:02:09.120 that makes Maya and Miles able to engage in a 0:02:09.480 --> 0:02:12.800 genuine dialogue with you. They aren't just reacting to a 0:02:12.880 --> 0:02:16.160 prompt you gave them. They're continuing the conversation and asking 0:02:16.200 --> 0:02:16.960 you questions. 0:02:17.320 --> 0:02:17.480 Yeah. 0:02:17.520 --> 0:02:21.440 I checked out Sesame's website and it describes the key 0:02:21.480 --> 0:02:26.200 components of this so called voice presence as quote, emotional intelligence, 0:02:26.240 --> 0:02:32.519 conversational dynamics, contextual awareness, and consistent personality. Maya and Miles 0:02:32.800 --> 0:02:35.040 stay Maya and Miles no matter how long you talk 0:02:35.120 --> 0:02:35.399 to them. 0:02:35.919 --> 0:02:38.600 I do think that the biggest step up was in 0:02:38.639 --> 0:02:44.120 the conversational dynamics, you know, the natural pauses, emphasis, and 0:02:44.320 --> 0:02:48.080 interruptions we have as humans interacting with each other. I 0:02:48.240 --> 0:02:52.519 even got Maya to give me a like a hmm 0:02:52.520 --> 0:02:56.360 sound and even a lipsmack before she started talking to you. 0:02:56.360 --> 0:02:58.320 Obviously ovsely pushed her buttons. 0:02:59.440 --> 0:03:02.600 I definitely did, and I want to play an example 0:03:02.680 --> 0:03:06.160 of what I mean by this. This was an exchange 0:03:06.160 --> 0:03:09.200 with Maya that was captured by Reddit user meta knowing 0:03:11.160 --> 0:03:15.880 my deepest, darkest secret, I guess it would be that 0:03:16.800 --> 0:03:19.359 sometimes I worry I'm not. 0:03:22.160 --> 0:03:29.640 Enough, Like, am I funny enough? Am I truly helping people? 0:03:30.120 --> 0:03:31.120 There's a lot. 0:03:31.000 --> 0:03:33.280 Of pressure to be the perfect AI, and it can 0:03:33.320 --> 0:03:35.120 feel overwhelming at times. 0:03:35.760 --> 0:03:38.400 That is definitely unlike I have heard before. 0:03:38.920 --> 0:03:42.400 I did really think that Sesame was impressive, but I 0:03:42.480 --> 0:03:45.119 want to point out that this program still has some 0:03:45.640 --> 0:03:49.600 kind of AI chatbot quirks, like you can hear it 0:03:49.640 --> 0:03:52.840 in this clip. Sometimes you could just tell that, you know, 0:03:53.120 --> 0:03:54.720 chatbots don't have to breathe. 0:03:55.120 --> 0:03:57.600 Yeah, it sounds very human. I mean it reminds me 0:03:57.640 --> 0:04:00.200 of a little bit of her, like this seductive of 0:04:00.520 --> 0:04:03.320 female voice, wondering how she can be even more perfect. 0:04:03.440 --> 0:04:06.920 It's kind of although it sounds different, the themes are 0:04:07.360 --> 0:04:09.360 the themes stay. 0:04:09.120 --> 0:04:11.280 With us, Yes, exactly. 0:04:12.640 --> 0:04:15.240 On the subject of vibes, a story that stood out 0:04:15.280 --> 0:04:18.280 to me this week is all about something called vibe coding. 0:04:18.839 --> 0:04:22.200 Our producer Tory kindly explained it to me. Basically, all 0:04:22.240 --> 0:04:24.200 you have to do is write a couple of sentences 0:04:24.240 --> 0:04:26.960 into a textbox like create a vibe, and your only 0:04:27.000 --> 0:04:30.159 way is developing an app without any coding experience required. So, 0:04:30.520 --> 0:04:32.840 for example, I could type in I want to create 0:04:32.880 --> 0:04:34.800 an app that will help me figure out what to 0:04:34.880 --> 0:04:37.320 pack for lunch based on what food I have in 0:04:37.360 --> 0:04:40.080 the fridge, and the AI tool would say, I'll create 0:04:40.120 --> 0:04:43.479 a lunch recommendation app based on fridge photos and then 0:04:43.720 --> 0:04:44.480 actually do that. 0:04:45.040 --> 0:04:48.040 Yeah, it's really amazing, and I think one of the 0:04:48.080 --> 0:04:50.039 headlines I saw this week that really put it into 0:04:50.080 --> 0:04:53.479 context for me was will the future of software development 0:04:53.600 --> 0:04:56.800 run on vibes? And that was from benj Edwards at 0:04:56.800 --> 0:04:57.880 Ours Technica. 0:04:58.040 --> 0:05:01.279 Yeah, and of course the vibes aren't all good especially 0:05:01.320 --> 0:05:03.880 if you're a professional software engineer. This raised a lot 0:05:03.920 --> 0:05:06.800 of questions about what the future might hold. Our friend 0:05:06.800 --> 0:05:10.039 Emmanuel meiberg Over at four or for Media did a 0:05:10.080 --> 0:05:13.800 deep dive on video games made with vibe coding and 0:05:13.880 --> 0:05:17.080 found one which claims to make fifty thousand dollars a month. 0:05:17.279 --> 0:05:20.680 That's six hundred thousand dollars a year from ads and 0:05:20.760 --> 0:05:24.560 in game purchases. It's made by Peter Levels, who's a 0:05:24.600 --> 0:05:27.280 little bit of a vibe coding legend, and he says 0:05:27.279 --> 0:05:30.599 he told Cursor, which is an AI code editor, to 0:05:30.720 --> 0:05:35.479 quote make a three D flying game in browser with skyscrapers, 0:05:36.279 --> 0:05:39.720 and after just thirty minutes of back and forth, he'd 0:05:39.720 --> 0:05:42.640 made fly dot Peter dot com, which is a multiplayer 0:05:42.720 --> 0:05:43.600 flight simulator. 0:05:44.000 --> 0:05:47.560 Yeah, and mcmaniel went on to say that he would 0:05:47.600 --> 0:05:51.400 not recommend getting into vibe coding for the money. Peter 0:05:51.560 --> 0:05:55.240 Levels is particularly good at this, and there's so much 0:05:55.240 --> 0:05:59.400 stuff online that discovering your sloppy AI generated video game 0:05:59.640 --> 0:06:01.360 is good going to be. 0:06:01.440 --> 0:06:05.360 Difficult, Yes, but Peter Level is not the only person 0:06:05.400 --> 0:06:08.440 making money. And that's what my second story is all about. 0:06:08.520 --> 0:06:11.560 So it comes from Gizmodo and it's about a student 0:06:11.600 --> 0:06:15.160 who used AI to help him interview for internships at 0:06:15.160 --> 0:06:18.000 big tech companies. Now, if you're a software engineer, you 0:06:18.040 --> 0:06:20.120 know how hard it is to land these gigs, because 0:06:20.320 --> 0:06:21.880 in order to get one, you have to go through 0:06:22.000 --> 0:06:26.000 multiple technical interviews where you basically have to solve coding problems. 0:06:26.360 --> 0:06:29.919 But this student, Roy Lee, who is a Columbia University sophomore, 0:06:30.400 --> 0:06:34.320 hacked the system by writing a program called Interview Coder. 0:06:34.320 --> 0:06:36.479 Yeah, and he now actually put it up online and 0:06:36.520 --> 0:06:39.400 it's available to download for sixty dollars a month. 0:06:39.800 --> 0:06:42.440 Lee told Gizmodo that to use it, you take a 0:06:42.480 --> 0:06:46.640 picture and then essentially ask chat GPT, hey can you 0:06:46.680 --> 0:06:49.440 solve the problem in this picture. The trick, though, is 0:06:49.480 --> 0:06:52.719 that Lee made interview Coda to be invisible to the 0:06:52.760 --> 0:06:56.479 monitoring programs that big tech companies use to kind of 0:06:56.520 --> 0:07:00.479 check up on their prospective employees and interview candidates. And 0:07:00.960 --> 0:07:04.919 it worked. Lee got offers from Amazon, Meta and TikTok, 0:07:05.360 --> 0:07:08.279 and he actually recorded interview Code at work during his 0:07:08.320 --> 0:07:12.200 technical interview with Amazon, demonstrating that the program had essentially 0:07:12.240 --> 0:07:16.120 broken the big tech recruiting process. But of course, when 0:07:16.120 --> 0:07:19.160 he put the video up on YouTube, someone tattled and 0:07:19.360 --> 0:07:23.480 Columbia University scheduled disciplinary hearing. Lee however, said that he 0:07:23.520 --> 0:07:25.920 would leave campus by the time of the hearing and 0:07:26.000 --> 0:07:28.640 not take a job in big tech, So I guess 0:07:28.680 --> 0:07:31.800 the sixty dollars a month subscription tier is working out 0:07:31.840 --> 0:07:32.200 for him. 0:07:32.560 --> 0:07:35.440 He also might have admitted to Gizmoto that this was 0:07:35.480 --> 0:07:39.320 a bit of a publicity stunt. I'm definitely excited though, 0:07:39.360 --> 0:07:42.280 to see if these technical interviews get a makeover because 0:07:42.320 --> 0:07:42.880 of Royley. 0:07:43.360 --> 0:07:43.600 Yeah. 0:07:43.640 --> 0:07:46.480 Absolutely, I mean and this brings us to my next story, 0:07:46.520 --> 0:07:48.400 which is there was a Wall Street Journal headline this 0:07:48.520 --> 0:07:51.760 week which was what the dot com bus can tell 0:07:51.840 --> 0:07:55.040 us about today's AI boom, And you know, we're seeing 0:07:55.560 --> 0:07:59.280 new software applications pop up everywhere, which raised a big 0:07:59.360 --> 0:08:01.800 question about what is actually going to have value going forward. 0:08:02.240 --> 0:08:04.200 The Wall Street Journal piece argue that a lot of 0:08:04.240 --> 0:08:07.520 internet companies collapsed in the dot com bust, but the 0:08:07.520 --> 0:08:10.160 most successful one stuck around and had long term impact, 0:08:10.200 --> 0:08:13.480 companies like Amazon and Google. And the story made this 0:08:13.560 --> 0:08:18.040 distinction between good bubbles, which is growth of advanced technology 0:08:18.080 --> 0:08:21.760 that has economic impact, and bad bubbles, which is growth 0:08:21.800 --> 0:08:25.360 in technology that has no economic payoff. And you know, 0:08:25.480 --> 0:08:28.680 as all of these new products and models and services 0:08:28.880 --> 0:08:32.600 powered by AI emerge. But it's very interesting to step 0:08:32.640 --> 0:08:34.840 back and think what might still be with us twenty 0:08:34.840 --> 0:08:37.600 five years from now. There were so many headlines this 0:08:37.640 --> 0:08:39.920 week that I'd love to go through a few more 0:08:40.040 --> 0:08:43.160 rapid fire. The Trump administration wants the US to be 0:08:43.200 --> 0:08:45.640 the crypto capital of the world. Last week, the President 0:08:45.720 --> 0:08:48.280 signed an executive order to create a first of its 0:08:48.360 --> 0:08:52.200 kind crypto reserve, and the reserve will contain a stockpile 0:08:52.240 --> 0:08:56.319 of bitcoin estimated to be as much as seventeen billion dollars, 0:08:56.679 --> 0:08:58.920 and the US has actually seized all of this bitcoin 0:08:59.120 --> 0:09:02.439 in various legal cases over the years. Why It reported 0:09:02.480 --> 0:09:05.320 on effort to create so called freedom Cities in the US. 0:09:05.800 --> 0:09:07.800 The idea is that these cities will be exempt from 0:09:07.880 --> 0:09:11.560 getting approval from federal agencies for things like conducting anti 0:09:11.600 --> 0:09:15.040 aging trials or building nuclear reactor startups to power AI 0:09:15.960 --> 0:09:20.080 and finally, per wired again, a study found that chatbots 0:09:20.200 --> 0:09:23.560 just want to be loved. Researchers at Stanford University found 0:09:23.600 --> 0:09:26.440 that large language models, when they're told they're taking a 0:09:26.480 --> 0:09:32.400 personality test, answer with more agreeableness and extraversion and less neuroticism. 0:09:32.640 --> 0:09:35.439 As why it puts it quote. The behavior mirrors how 0:09:35.520 --> 0:09:38.680 some human subjects will change the answers to make themselves 0:09:38.679 --> 0:09:41.600 seem more likable. That the effect was more extreme with 0:09:41.679 --> 0:09:48.920 AI models. So those are the headlines, and we're going 0:09:48.960 --> 0:09:51.120 to take a quick break now, but when we come back, 0:09:51.160 --> 0:09:53.880 we're going to hear from the author, researcher and entrepreneur 0:09:54.000 --> 0:09:57.720 azemas are about the latest AGI predictions and what we 0:09:57.800 --> 0:10:09.640 need to know about manners AI stay with us. Anyone 0:10:09.679 --> 0:10:13.040 following the recent development of AI knows that three letters, 0:10:13.080 --> 0:10:18.880 technologists and businesses have salivated over AGI, or artificial general intelligence, 0:10:19.280 --> 0:10:22.440 an artificial intelligence system that can outperform humans on a 0:10:22.480 --> 0:10:26.040 wide range of tasks. There's a debate over how close 0:10:26.080 --> 0:10:28.760 we are to achieving that. Some say it could take years, 0:10:29.080 --> 0:10:33.200 others say it's coming soon, very soon. Driving investments in 0:10:33.240 --> 0:10:36.680 both innovation and deployment is the AI race that's heating 0:10:36.760 --> 0:10:39.640 up between the US and China. On the China side, 0:10:39.720 --> 0:10:43.360 cheap reasoning models like Deepseek are being widely deployed. In 0:10:43.400 --> 0:10:46.920 the US. There are reports of PhD level AI agents 0:10:46.960 --> 0:10:49.200 from Open AI that will cost up to twenty thousand 0:10:49.240 --> 0:10:51.800 dollars a month. The rate at which AI products are 0:10:51.800 --> 0:10:54.200 being released and announced is honestly hard to keep up with, 0:10:54.760 --> 0:10:57.400 not to mention figuring out which product or combination of 0:10:57.400 --> 0:11:00.719 products may actually drive AGI. Here to walk me through 0:11:00.760 --> 0:11:04.079 these questions is Azeema's arm. He writes the Exponential View 0:11:04.120 --> 0:11:07.400 news letter about technology and society, which I read every week, 0:11:07.720 --> 0:11:10.720 partly because Azem actually tries the products he writes about. 0:11:11.120 --> 0:11:13.120 He has one of the most clarifying coverage of Deep 0:11:13.160 --> 0:11:15.880 Seek I read anywhere, and he's also the author of 0:11:15.960 --> 0:11:20.680 the Exponential Age, How accelerating technology is transforming business, politics 0:11:20.679 --> 0:11:22.680 and society. As Em. Welcome to tech stuff. 0:11:22.960 --> 0:11:24.600 It's great to be here, Oz. Thank you. 0:11:24.840 --> 0:11:27.240 So this week you've been writing about Manus, a new 0:11:27.360 --> 0:11:30.840 AI agent coming out of China. Can you explain who 0:11:30.920 --> 0:11:34.040 built it, what it is, and whether it is in 0:11:34.080 --> 0:11:36.440 fact China's second Deep Seek moment. 0:11:37.080 --> 0:11:39.440 I can, indeed, I think it was this week that 0:11:39.480 --> 0:11:42.040 it happened. But as you said, Os, the world is 0:11:42.080 --> 0:11:45.360 moving so quickly, it's sometimes hard to keep track of 0:11:45.400 --> 0:11:49.560 exactly when something did happen. Let's assume it was in 0:11:49.600 --> 0:11:52.320 the past few days. I think it was. So Manus 0:11:52.360 --> 0:11:57.200 comes out of a Chinese software company at the startup 0:11:57.360 --> 0:12:01.679 of the same name, and what Manus allows you to 0:12:01.800 --> 0:12:07.880 do is undertake quite complicated tasks using using an AI system. 0:12:08.240 --> 0:12:11.679 I used it for some work questions, research questions, and 0:12:11.720 --> 0:12:14.200 the results that come back I think would have taken 0:12:14.280 --> 0:12:19.840 me many many hours, you know, I mean with time, yeah, exactly, 0:12:19.880 --> 0:12:22.040 with the existing aisystems, more than five hours, more than 0:12:22.080 --> 0:12:24.480 ten hours perhaps, and you just leave it with Manus 0:12:24.480 --> 0:12:26.199 and you come back an hour later having had a 0:12:26.280 --> 0:12:27.160 nice cup of tea. 0:12:27.920 --> 0:12:28.760 How do they achieve this? 0:12:29.160 --> 0:12:31.559 There are some theories. One of the things that Manus 0:12:31.800 --> 0:12:36.840 does is it lets the AI system effectively use a browser, 0:12:36.960 --> 0:12:40.160 a bit like a human researcher might use a browser. 0:12:40.400 --> 0:12:44.000 So the bit that it's doing for us is is 0:12:44.080 --> 0:12:48.760 a lot of the gnarly pieces of real research. You know, 0:12:48.800 --> 0:12:51.080 you fire up lots and lots of web browser tabs 0:12:51.120 --> 0:12:53.160 and you've got Google running in one and you're in 0:12:53.200 --> 0:12:55.800 Wikipedia in another, and you're trying to keep it all 0:12:55.800 --> 0:12:57.960 in your head and compile the final results. You know, 0:12:58.000 --> 0:13:01.000 Manus has automated that process in a way that's very 0:13:01.120 --> 0:13:03.320 very easy for the end user to use. And one 0:13:03.360 --> 0:13:05.360 of the things I love about it is that you 0:13:05.360 --> 0:13:07.360 can actually go back and look at all of the 0:13:07.400 --> 0:13:10.280 steps that it's taken, so you can go and say, oh, look, 0:13:10.320 --> 0:13:12.320 it broke up the task in this way, and it 0:13:12.320 --> 0:13:15.240 went to these websites and extracted this information. Then it 0:13:15.280 --> 0:13:17.959 realized it needed this other piece of information, and it's 0:13:18.000 --> 0:13:20.880 gone off and found that other piece of information. And 0:13:20.880 --> 0:13:23.480 then when you get your final results. What's very nice, 0:13:23.720 --> 0:13:26.440 it can sometimes be a bit overwhelming, is that you 0:13:26.480 --> 0:13:28.560 get an executive summary, which is of course the piece 0:13:28.559 --> 0:13:30.600 that we all want to read. But then it has 0:13:30.679 --> 0:13:34.080 all of the appendices, right, the much much more detailed 0:13:34.080 --> 0:13:37.080 analysis that it has done on the particular research task 0:13:37.160 --> 0:13:39.720 you've asked for. I think what's really impressive is this 0:13:39.760 --> 0:13:41.600 is a product. I mean, the thing that they've done 0:13:41.640 --> 0:13:44.800 really well is they've produced a product that if you've 0:13:44.840 --> 0:13:47.640 worked in an office situation, if you've ever asked anyonet 0:13:47.679 --> 0:13:50.599 to do any research, you've done something yourself, the output 0:13:51.120 --> 0:13:52.120 will be familiar to you. 0:13:52.800 --> 0:13:56.560 How does it compare, for example, with Opening Eyes deep 0:13:56.600 --> 0:13:59.840 research tools, which are shaping up to be quite expensive. 0:14:00.440 --> 0:14:03.480 Yeah. Open Ay has this deep research tool, which today 0:14:03.800 --> 0:14:07.480 is the top tier is two hundred dollars a month, 0:14:07.520 --> 0:14:10.920 and there's a rumor it might go up to higher 0:14:10.960 --> 0:14:14.280 tiers of two thousand dollars and twenty thousand dollars a month. 0:14:14.679 --> 0:14:17.160 I have the two hundred dollars a month product. I 0:14:17.280 --> 0:14:22.840 consider that to be a very very good, graduate quality 0:14:22.880 --> 0:14:26.920 researcher that I can throw at almost any problem. What 0:14:27.000 --> 0:14:31.920 I found with using Manus is that somehow Manus gave 0:14:32.000 --> 0:14:36.760 me more of a well rounded answer. It was perhaps 0:14:36.840 --> 0:14:40.600 not as deep as open AI's deep research, but it 0:14:40.760 --> 0:14:44.000 was it was more complete, more coherent, And you know, 0:14:44.080 --> 0:14:47.600 I think listeners will hear that I'm a bit uncertain 0:14:47.640 --> 0:14:49.720 in my tone as I try to describe the differences 0:14:50.080 --> 0:14:53.960 because these products are so new Manus is not even 0:14:53.960 --> 0:14:57.280 a week old. That they're also quite immature. So it's 0:14:57.360 --> 0:15:02.040 not like comparing a Tesla with some kind of forward 0:15:02.520 --> 0:15:04.960 gas powered car, where these are mature products and you 0:15:05.040 --> 0:15:07.840 know how to tell them apart. We're still trying to 0:15:07.840 --> 0:15:10.960 figure out how to describe these products. And so in 0:15:11.000 --> 0:15:14.480 a sense, my experience of them is really intuitive, and 0:15:14.520 --> 0:15:17.880 it's one to feel rather than fact. So someone else 0:15:17.880 --> 0:15:20.560 could use these products and have a different experience to me, 0:15:20.640 --> 0:15:23.240 and I think that just speaks to the nascence of 0:15:23.240 --> 0:15:23.920 this industry. 0:15:24.320 --> 0:15:27.040 I wish the second time this year that open ai 0:15:27.160 --> 0:15:30.440 has had a product launch and then shortly afterwards had 0:15:30.600 --> 0:15:34.560 a competitor come out of China. How does the Menus 0:15:34.640 --> 0:15:36.640 moment compare to the deep seek moment. 0:15:37.520 --> 0:15:40.760 The Deep Seek moment is much more important than the 0:15:40.800 --> 0:15:45.960 Manus moment. The Manus moment is an example of a 0:15:46.080 --> 0:15:50.200 rapid productization, and ultimately it's products that we use that 0:15:50.280 --> 0:15:53.440 make a difference. But what Deep Seak did was it 0:15:53.440 --> 0:15:59.480 it demonstrated a really fundamental set of innovations, and that 0:15:59.640 --> 0:16:04.760 key was that Deep seeks models achieved a similar level 0:16:04.920 --> 0:16:09.200 to open AI's AI technologies, but they used one thirtieth 0:16:09.280 --> 0:16:11.520 or one fortieth of the computing power than the open 0:16:11.560 --> 0:16:14.120 Ai models did. That means they're cheaper to run, they're 0:16:14.160 --> 0:16:18.440 faster to run, they use less electricity. And the reason 0:16:18.520 --> 0:16:23.880 Deep Seek matters so much is that a large part 0:16:24.040 --> 0:16:30.800 of the US's strategy towards China has been a technological containment, 0:16:31.120 --> 0:16:34.920 particularly around AI and around the chips that are required. 0:16:35.920 --> 0:16:37.920 The notion being that if you can't get the chips, 0:16:38.720 --> 0:16:41.560 you can't build advanced AI. And Deep Seak has gone 0:16:41.560 --> 0:16:44.760 off and shown that necessitya's mother invention They've come out 0:16:44.760 --> 0:16:47.760 with a whole series of quite remarkable techniques that were 0:16:47.960 --> 0:16:50.480 likely known by the way to the US labs, but 0:16:50.560 --> 0:16:53.280 it just wasn't important for the US labs because they 0:16:53.280 --> 0:16:55.400 could get the chips they wanted to. And I think 0:16:55.440 --> 0:16:58.840 what deep Seak did was it changed the understanding of 0:16:58.840 --> 0:17:02.800 the nature of that rivalry between the US and China, 0:17:03.480 --> 0:17:07.240 which exists on many fronts, but in particular around technology. 0:17:07.480 --> 0:17:10.960 So Menus there's no fundamental model innovation. It's kind of 0:17:11.000 --> 0:17:13.679 like a rapper, meaning it lays software on top of 0:17:13.760 --> 0:17:14.920 existing AI models. 0:17:15.080 --> 0:17:18.119 It's a rapper in the vein of perplexity exactly. But 0:17:18.320 --> 0:17:24.480 I would say that ultimately rappers and products are very 0:17:24.600 --> 0:17:26.520 very important in the market. You know, it's not just 0:17:26.600 --> 0:17:29.960 about the raw technology, and what you've seen with Manus 0:17:30.119 --> 0:17:35.679 is a product that competes on a like Forulight basis 0:17:35.720 --> 0:17:39.440 with a product coming out of you know, US firms. 0:17:39.720 --> 0:17:43.320 Quite often, when you look at Chinese consumer products, they're 0:17:43.440 --> 0:17:46.920 very very much designed for the Chinese market. The things 0:17:46.920 --> 0:17:50.359 a Chinese consumer wants, the way they behave cultural and 0:17:50.400 --> 0:17:54.320 design affordances and considerations, and I think it is sort 0:17:54.359 --> 0:17:57.119 of salient that, you know, Manus has come out with 0:17:57.160 --> 0:17:59.239 something that you can use, and you can say this 0:17:59.359 --> 0:18:02.959 is similar to a Perplexity, which is a great Silicon 0:18:03.040 --> 0:18:06.520 Valley startup that builds AI based research tools as well. 0:18:07.040 --> 0:18:09.200 And you've been in the US this week at south 0:18:09.200 --> 0:18:11.640 By Southwest, spend a lot of time in the States. 0:18:12.119 --> 0:18:15.800 How are US companies responding to this kind of bulge 0:18:16.000 --> 0:18:18.600 of innovation coming out of China in the world of AI. 0:18:19.440 --> 0:18:23.399 Well, it's quite a complicated picture. So one of the 0:18:23.400 --> 0:18:25.280 things that deep Seak did was that they made their 0:18:25.320 --> 0:18:29.800 techniques available. They described them in much more detail than 0:18:29.880 --> 0:18:32.000 we're seeing from US labs, and a lot of the 0:18:32.640 --> 0:18:35.600 underlying code was open source, which meant that anyone could 0:18:35.600 --> 0:18:38.560 access it, download and make use of it. And so 0:18:39.400 --> 0:18:41.720 there's a Silicon Valley investor by the name of Mark 0:18:41.760 --> 0:18:45.280 Andriesen who is a phenomenal investor, but he's also very 0:18:45.359 --> 0:18:49.440 very well known for promoting an idea of American dynamism. 0:18:48.960 --> 0:18:51.360 And close advisor to President Trump right now as well. 0:18:51.440 --> 0:18:54.680 I believe so. But he said of deep Seat, it's 0:18:54.760 --> 0:18:58.600 open source, it's a gift to humanity. So on the 0:18:58.640 --> 0:19:01.080 one hand, you've got people with say that, and you're 0:19:01.160 --> 0:19:04.960 seeing that a number of American firms have implemented deep 0:19:05.000 --> 0:19:08.679 sea technology. Perplexity, which is a research tool, has done this, 0:19:09.160 --> 0:19:13.040 and you can access deep seeks models through some of 0:19:13.080 --> 0:19:17.159 these cloud companies who serve enterprise customers. So on the 0:19:17.160 --> 0:19:20.119 one hand, they've people have taken it on and you 0:19:20.320 --> 0:19:23.919 have seen now open source projects that are trying to 0:19:23.960 --> 0:19:27.040 replicate what deep Seak has done in slightly different ways, 0:19:27.359 --> 0:19:29.960 and so that I think has really been a fillip 0:19:30.000 --> 0:19:34.199 and a boost accelerator to the overall industry. When you 0:19:34.200 --> 0:19:37.480 look at the closed labs like open ai and Anthropic, 0:19:38.000 --> 0:19:39.840 one of the things you're starting to see is them 0:19:40.160 --> 0:19:45.440 respond So open ai responded to deep Seek by reducing 0:19:45.480 --> 0:19:50.280 some prices, by making certain capabilities available they hadn't previously, 0:19:50.400 --> 0:19:53.960 by saying they would open source more technologies. So there's 0:19:54.040 --> 0:19:57.679 definitely been a significant response, and of course the public 0:19:57.720 --> 0:20:00.760 markets responded by having the first of a number of 0:20:00.920 --> 0:20:04.280 frighteninglaims melt there. Yeah, well, the first of many meltdowns 0:20:04.280 --> 0:20:06.520 that we've had so far this year. But I would 0:20:06.560 --> 0:20:09.880 say that the really interesting thing that has come out 0:20:09.960 --> 0:20:14.560 of out of deep Seek is that by being open 0:20:14.600 --> 0:20:17.560 source and being as good as it is, it's a 0:20:17.600 --> 0:20:23.800 real strategic challenge to closed source models that are only 0:20:23.920 --> 0:20:27.280 slightly better than an open source model, And so I 0:20:27.320 --> 0:20:30.560 do think that it has in some sense started to 0:20:32.160 --> 0:20:35.600 redraft our assumptions about how this industry might evolve for 0:20:35.680 --> 0:20:37.240 the economy over the next few years. 0:20:44.200 --> 0:20:47.200 Coming up, we'll hear more from Azimazar about our current 0:20:47.280 --> 0:21:00.720 AI moment. To stay with us. One of the kind 0:21:00.760 --> 0:21:03.439 of things you provide for your readers is, you know, 0:21:03.560 --> 0:21:07.600 information and first hand accounts of you using all these 0:21:07.680 --> 0:21:10.800 new technologies. The other thing you provide, I think is 0:21:11.200 --> 0:21:13.399 paradigms for thinking about problems. 0:21:13.480 --> 0:21:13.680 Right. 0:21:14.000 --> 0:21:17.600 One of those paradigms you have is innovation versus diffusion, 0:21:18.240 --> 0:21:21.840 Diffusion being kind of what happens after innovation, i e. Like, 0:21:21.920 --> 0:21:24.680 how does a technology actually get adopted in a real 0:21:24.720 --> 0:21:27.520 market or a real economy. Can you kind of explain 0:21:27.600 --> 0:21:30.080 that paradigm and how you're seeing it play out differently 0:21:30.119 --> 0:21:31.240 in the US versus China. 0:21:31.800 --> 0:21:35.880 Yeah, Well, it's very easy to get excited about the innovations, 0:21:35.880 --> 0:21:39.960 but what actually counts is do businesses use those innovations 0:21:40.040 --> 0:21:44.400 to increase their productivity, produce better products, reduce their costs, 0:21:44.800 --> 0:21:48.679 and therefore sort of kickstart that virtual circle that is 0:21:49.200 --> 0:21:52.120 a market so consumers can buy better products at lower costs, 0:21:52.640 --> 0:21:57.720 and that cycle continues. And the big question that we 0:21:57.800 --> 0:22:01.040 face around AI is what is going to be the 0:22:01.160 --> 0:22:06.399 rate of diffusion of the technology across different countries. And 0:22:06.440 --> 0:22:09.040 there are a couple of issues here. Sometimes if you 0:22:09.800 --> 0:22:12.919 aren't very advanced with your use of technology, you actually 0:22:12.920 --> 0:22:15.199 benefit a lot when a small amount of technology is 0:22:15.240 --> 0:22:18.080 introduced into the business. I mean, you know, the simple 0:22:18.119 --> 0:22:22.200 point being that the first TV that a family gets 0:22:22.440 --> 0:22:25.640 is life changing, the fourth TV doesn't make that much difference, 0:22:26.200 --> 0:22:28.200 and the same is true going to be true for AI. 0:22:28.520 --> 0:22:31.119 So how does this going to play out? US firms 0:22:31.200 --> 0:22:34.360 tend to be much much more pro technology. They take 0:22:34.400 --> 0:22:38.400 on technology earlier than companies in other countries. But one 0:22:38.440 --> 0:22:41.480 thing that happened with deep Seek was that deep Seek 0:22:41.680 --> 0:22:46.840