1 00:00:00,800 --> 00:00:04,939 S1: Unsupervised Learning is a podcast about trends and ideas in cybersecurity, 2 00:00:04,970 --> 00:00:10,010 S1: national security, AI, technology and society, and how best to 3 00:00:10,039 --> 00:00:17,510 S1: upgrade ourselves to be ready for what's coming. All right, 4 00:00:17,510 --> 00:00:21,890 S1: welcome to Unsupervised Learning is Daniel and episode 471. Hope 5 00:00:21,890 --> 00:00:25,340 S1: your week is going well. Updates on my side want 6 00:00:25,340 --> 00:00:28,610 S1: to publicly shame myself here, so I basically hack and 7 00:00:28,610 --> 00:00:33,019 S1: write for most of the day for work, and pretty 8 00:00:33,020 --> 00:00:37,339 S1: much every single meal I have at some fast food restaurant, 9 00:00:37,340 --> 00:00:42,470 S1: which is just unbelievable for somebody who is supposed to 10 00:00:42,470 --> 00:00:44,900 S1: be smart. It is really bad for your health to 11 00:00:44,930 --> 00:00:48,500 S1: do this, and it's really pure laziness and also just 12 00:00:48,500 --> 00:00:51,980 S1: intimidation of like how to do this cooking thing properly 13 00:00:51,979 --> 00:00:54,800 S1: and basically go and eat in the car and come 14 00:00:54,800 --> 00:00:59,570 S1: back and get right back into working. Um, but I 15 00:00:59,570 --> 00:01:03,440 S1: Must start cooking at home. Now, I put this in 16 00:01:03,440 --> 00:01:06,679 S1: the newsletter and the newsletter went out before you're hearing this. 17 00:01:06,709 --> 00:01:11,390 S1: As you probably know, and, uh, I got close to 18 00:01:11,420 --> 00:01:18,710 S1: 500 email responses. It was absolutely overwhelming. It is. Hundreds 19 00:01:18,709 --> 00:01:22,730 S1: of people reaching out already bought my pressure cooker or 20 00:01:22,730 --> 00:01:26,150 S1: what is it? Instant pot already bought that thing because 21 00:01:26,150 --> 00:01:28,820 S1: everyone pointed me to that being like the best way 22 00:01:28,819 --> 00:01:33,440 S1: to do things. And, uh, got a specific one, which 23 00:01:33,440 --> 00:01:36,140 S1: seemed to be like one of the most popular ones. Also, 24 00:01:36,140 --> 00:01:39,560 S1: it was on Wirecutter, I think, as well. But, uh, yeah, 25 00:01:39,590 --> 00:01:41,150 S1: I'm going to try to do the eggs thing. I'm 26 00:01:41,150 --> 00:01:43,580 S1: going to try to do the steak thing and, uh, 27 00:01:43,580 --> 00:01:47,750 S1: really appreciate all of you who responded to help me out. 28 00:01:47,780 --> 00:01:50,360 S1: It sounds like a lot of us have been going 29 00:01:50,360 --> 00:01:53,690 S1: through that and have figured it out to some degree. 30 00:01:53,690 --> 00:01:57,830 S1: So really appreciate all the feedback. If you're into cloud security, 31 00:01:57,860 --> 00:02:00,860 S1: you need to be following my buddy Rich mogul's cloud 32 00:02:00,860 --> 00:02:05,420 S1: security lab a week. So cloud slaw is one of 33 00:02:05,420 --> 00:02:08,120 S1: the absolute best cloud security people on Earth. And he's 34 00:02:08,120 --> 00:02:12,829 S1: putting out really high quality labs for free, like every week. 35 00:02:12,860 --> 00:02:15,710 S1: And he's like an EMT and he's like a really 36 00:02:15,710 --> 00:02:20,270 S1: great guy. He's just fantastic. So definitely recommend you check 37 00:02:20,270 --> 00:02:25,040 S1: him out. Cloud slaw. Rich Mogull and AI's ultimate use 38 00:02:25,040 --> 00:02:28,130 S1: case transition from current to desired state. I'm going to 39 00:02:28,130 --> 00:02:29,780 S1: do a separate video about this. So I'm not going 40 00:02:29,810 --> 00:02:32,960 S1: to talk about this blog post. Cybersecurity. A lot of 41 00:02:32,960 --> 00:02:36,739 S1: confusion about what's happening in the Trump administration supposedly standing 42 00:02:36,740 --> 00:02:40,880 S1: down U.S. cyber operations around Russia. So stories first say 43 00:02:40,880 --> 00:02:43,160 S1: it's like government wide, and then it's like, no, it's 44 00:02:43,160 --> 00:02:46,640 S1: not really all the government. It's only certain departments. These 45 00:02:46,639 --> 00:02:50,510 S1: things are happening so fast. You and you don't know 46 00:02:50,510 --> 00:02:52,550 S1: the quality of the signal that we're getting from these 47 00:02:52,550 --> 00:02:55,940 S1: different sources. So it's like I tend to like be 48 00:02:55,940 --> 00:02:59,720 S1: waiting and just seeing what actually turns out to be 49 00:02:59,720 --> 00:03:04,190 S1: the case because something could be hyper overblown. It's just 50 00:03:04,190 --> 00:03:06,650 S1: really hard to track things right now because it's so 51 00:03:06,650 --> 00:03:09,320 S1: much chaos going on. And I've got a lot to say, though, 52 00:03:09,320 --> 00:03:14,510 S1: about the general vibe of the administration seeming to move 53 00:03:14,540 --> 00:03:20,299 S1: towards Russia and away from trusting its own Intel people. 54 00:03:20,300 --> 00:03:22,580 S1: So that is a really huge problem for me, which 55 00:03:22,580 --> 00:03:25,070 S1: I'm going to talk about more below. VMware is releasing 56 00:03:25,100 --> 00:03:30,230 S1: emergency patches for three actively exploited flaws in ESXi workstation 57 00:03:30,230 --> 00:03:35,510 S1: and fusion new polyglot malware campaign called Suzano, targeting aviation 58 00:03:35,510 --> 00:03:39,230 S1: and satellite companies in the UAE. Hirsch IoT intercoms are 59 00:03:39,230 --> 00:03:43,340 S1: letting anyone with Google access unlock apartment doors due to 60 00:03:43,370 --> 00:03:46,730 S1: known default credentials. So this is like somebody could use 61 00:03:46,730 --> 00:03:50,210 S1: an app to, like, break into apartments who are protected 62 00:03:50,210 --> 00:03:54,350 S1: by this particular type of IoT. When IoT was a 63 00:03:54,350 --> 00:03:57,950 S1: big thing, like the talk about IoT was a big thing. 64 00:03:58,310 --> 00:04:02,090 S1: When was that? Was that like 2012? 2015? I feel 65 00:04:02,090 --> 00:04:05,900 S1: like it was around that area. And I did a 66 00:04:05,900 --> 00:04:08,180 S1: lot of press at that time talking about it. And 67 00:04:08,180 --> 00:04:10,250 S1: I was just like, I think this is kind of 68 00:04:10,250 --> 00:04:14,540 S1: overblown because what people don't realize about this is you 69 00:04:14,540 --> 00:04:17,359 S1: have to properly threat model. So you have to look 70 00:04:17,360 --> 00:04:20,779 S1: at the Venn diagram of people who want to break 71 00:04:20,779 --> 00:04:25,730 S1: into places and steal things and are willing to break 72 00:04:25,760 --> 00:04:29,210 S1: the law in order to do that. That's one circle. 73 00:04:29,570 --> 00:04:32,780 S1: The other circle is people who don't know how to 74 00:04:32,810 --> 00:04:36,140 S1: do it. They are unable to get into the building. 75 00:04:36,140 --> 00:04:40,430 S1: It's like too difficult. The locks are stopping them. The 76 00:04:40,430 --> 00:04:44,000 S1: technology is stopping them. The security is stopping them. It's 77 00:04:44,000 --> 00:04:46,190 S1: just a huge problem for them, right? They're not able 78 00:04:46,190 --> 00:04:48,680 S1: to do it. Now, what does this look like for 79 00:04:48,710 --> 00:04:52,070 S1: a Venn diagram? Now, to illuminate this more, think about 80 00:04:52,070 --> 00:04:56,780 S1: what this particular vulnerability is actually going to do to 81 00:04:56,839 --> 00:05:00,050 S1: these two, then to these two circles in the Venn. 82 00:05:00,050 --> 00:05:03,080 S1: I think the answer is it's not going to add 83 00:05:03,080 --> 00:05:07,250 S1: too many people to the ones who are actually willing 84 00:05:07,279 --> 00:05:10,550 S1: to do the damage. And I don't think it's ultimately 85 00:05:10,550 --> 00:05:16,100 S1: going to result in too many more people getting broken into. So, 86 00:05:16,100 --> 00:05:19,070 S1: in other words, the people who want to break into 87 00:05:19,100 --> 00:05:23,270 S1: apartment buildings, the people who want to break into, uh, 88 00:05:23,450 --> 00:05:27,200 S1: people's homes, people's apartments. I don't think they are limited 89 00:05:27,200 --> 00:05:29,840 S1: by not being able to get in. I think locks 90 00:05:29,839 --> 00:05:33,290 S1: are too easy to pick and they can get by. Now, 91 00:05:33,290 --> 00:05:36,260 S1: the question is, are there cameras that actually record them 92 00:05:36,260 --> 00:05:39,529 S1: doing it? And here's a better question. The cameras aren't 93 00:05:39,529 --> 00:05:44,570 S1: going away. So is the disincentive because of the cameras? 94 00:05:44,600 --> 00:05:49,969 S1: Isn't that still there still stopping a person from potentially 95 00:05:49,970 --> 00:05:52,940 S1: doing this. So I think people will see something like 96 00:05:52,940 --> 00:05:55,430 S1: this and they're like, oh, that's game over. No, you 97 00:05:55,430 --> 00:05:57,679 S1: have to threat model. This is why threat modeling is 98 00:05:57,680 --> 00:06:00,500 S1: like one of the most important skill sets that you 99 00:06:00,500 --> 00:06:05,000 S1: could learn for life in general. I'm very thankful for 100 00:06:05,000 --> 00:06:09,500 S1: my security education and, you know, all the workshops and 101 00:06:09,500 --> 00:06:12,650 S1: tabletops and everything I've done with threat modeling because it's 102 00:06:12,650 --> 00:06:15,380 S1: a universal life skill, because a lot of things are 103 00:06:15,380 --> 00:06:20,600 S1: emotionally charged. I'm moving to this new city. Oh, it's supposedly, uh, 104 00:06:20,600 --> 00:06:25,219 S1: supposedly a dangerous area to be. Um, it's, you know, 105 00:06:25,250 --> 00:06:27,710 S1: there's been x number of crimes in this area. There's 106 00:06:27,740 --> 00:06:31,040 S1: X number of car crashes with this particular type of car. 107 00:06:31,040 --> 00:06:34,850 S1: When you actually gather the data and throw together how 108 00:06:34,850 --> 00:06:37,160 S1: you plan on using this thing, or how you plan 109 00:06:37,160 --> 00:06:40,400 S1: on attackers using this thing or whatever, and you start 110 00:06:40,400 --> 00:06:44,630 S1: moving through the scenarios that you're worried about. Usually you 111 00:06:44,630 --> 00:06:49,370 S1: end up realizing the things you are emotionally worried about 112 00:06:49,400 --> 00:06:53,000 S1: are not the ones, and oftentimes the things you're not 113 00:06:53,000 --> 00:06:56,539 S1: emotionally worried about actually are the ones. So this is 114 00:06:56,540 --> 00:07:00,080 S1: a big exercise in, you know, the work of Kahneman 115 00:07:00,080 --> 00:07:05,270 S1: where he realized, you know, bias and mental models and 116 00:07:05,270 --> 00:07:09,380 S1: perception is actually far more important to how we see 117 00:07:09,380 --> 00:07:12,890 S1: the world than actual reality. And that really needs to 118 00:07:12,920 --> 00:07:15,830 S1: be applied when you're thinking about things like this in, 119 00:07:15,860 --> 00:07:18,980 S1: in cyber or other types of security. Brute forcing attackers 120 00:07:18,980 --> 00:07:22,130 S1: are targeting thousands of ISP networks in the US and 121 00:07:22,130 --> 00:07:25,580 S1: China to deploy stealers that grab crypto wallets and mine 122 00:07:25,580 --> 00:07:30,680 S1: Monero on compromised systems. Meta terminated 20 employees for sharing 123 00:07:30,680 --> 00:07:34,760 S1: internal info outside the company after recent stories about unannounced 124 00:07:34,760 --> 00:07:38,929 S1: products and meetings. Really cool wired thing here. It's a 125 00:07:38,930 --> 00:07:42,440 S1: wired video shows how Tesla vehicles record everything from driver 126 00:07:42,440 --> 00:07:47,840 S1: behavior to biometric data. Ask some pretty smart privacy questions 127 00:07:47,840 --> 00:07:51,290 S1: about this whole system. And yeah, it was a great video. 128 00:07:51,320 --> 00:07:54,140 S1: I watched the whole thing and a lot of it 129 00:07:54,170 --> 00:07:57,020 S1: is not alarming. A lot. Some of the stuff towards 130 00:07:57,020 --> 00:07:59,300 S1: the end, the end is like really where the meat 131 00:07:59,330 --> 00:08:03,140 S1: of this thing is basically talking about how there's been 132 00:08:03,140 --> 00:08:08,060 S1: some cases of things being abused, like recordings from inside 133 00:08:08,060 --> 00:08:10,820 S1: of cars and stuff like that. I don't think it's widespread. 134 00:08:10,850 --> 00:08:13,220 S1: Otherwise we would have heard a lot more about it, 135 00:08:13,220 --> 00:08:17,120 S1: but it's something to keep in mind. It's especially something 136 00:08:17,150 --> 00:08:19,310 S1: to keep in mind when you're talking about like BYD, 137 00:08:19,340 --> 00:08:23,210 S1: which is the main Tesla competitor, and they're likely to 138 00:08:23,240 --> 00:08:26,690 S1: be here in the US soon, I'm guessing. And it's 139 00:08:26,690 --> 00:08:29,510 S1: a pure Chinese company. Okay, so if they have the 140 00:08:29,510 --> 00:08:34,580 S1: same sort of monitoring capability and infrastructure that Tesla does, 141 00:08:34,580 --> 00:08:38,330 S1: but it's a Chinese company, whole different game for me. 142 00:08:38,360 --> 00:08:42,350 S1: I've not been super worried about all this surveillance stuff 143 00:08:42,350 --> 00:08:47,030 S1: inside of consumer tech, uh, because of how much I 144 00:08:47,030 --> 00:08:51,470 S1: trust or don't trust the products, where I have cameras, 145 00:08:51,470 --> 00:08:55,080 S1: where I have microphones in my house or in my car, 146 00:08:55,080 --> 00:08:59,520 S1: and which companies I'm allowing to actually be part of 147 00:08:59,520 --> 00:09:06,000 S1: that network. Right. So I trust Apple very highly, uh, 148 00:09:06,000 --> 00:09:08,160 S1: not only because I worked there for a few years 149 00:09:08,160 --> 00:09:11,640 S1: in the actual security group, but just because it's pretty 150 00:09:11,640 --> 00:09:16,020 S1: clear how much importance they're putting on privacy and security. 151 00:09:16,020 --> 00:09:19,620 S1: And I know that is confirmed for me personally because 152 00:09:19,620 --> 00:09:21,900 S1: I've been on the inside. That's not the case for Google. 153 00:09:21,900 --> 00:09:24,870 S1: For me, it's not the case for Amazon. For me, 154 00:09:24,870 --> 00:09:27,780 S1: it's not the case for meta, necessarily. For me, not 155 00:09:27,780 --> 00:09:30,449 S1: as much as Apple. And it's damn sure not the 156 00:09:30,450 --> 00:09:35,220 S1: case for some Chinese company like TikTok or, or, uh, 157 00:09:35,370 --> 00:09:39,150 S1: you know, ByteDance or whoever it is, or BYD when 158 00:09:39,150 --> 00:09:41,580 S1: those cars come. So you got to think about who 159 00:09:41,610 --> 00:09:44,460 S1: you're sending this stuff actually to. What are their long 160 00:09:44,490 --> 00:09:48,660 S1: term goals? National security. China's research output on next gen 161 00:09:48,660 --> 00:09:52,020 S1: chipmaking tech is now double that of the US, which 162 00:09:52,020 --> 00:09:57,200 S1: suggests that the bands are actually forcing them to get 163 00:09:57,200 --> 00:10:03,679 S1: smarter and more. Innovative. Innovative. That's a word. Innovative. So basically, 164 00:10:03,679 --> 00:10:06,680 S1: I was trying to say innovative. I was trying to 165 00:10:06,679 --> 00:10:09,469 S1: say like the British version of that word. Anyway, it 166 00:10:09,470 --> 00:10:14,030 S1: didn't work, but we could be forcing innovation from China, 167 00:10:14,030 --> 00:10:18,500 S1: where they were relying before on theft and relying before 168 00:10:18,530 --> 00:10:22,520 S1: on just kind of like buying other people's good stuff. 169 00:10:22,520 --> 00:10:25,490 S1: So we might be getting the exact opposite of what 170 00:10:25,490 --> 00:10:28,760 S1: we planned on getting. We might be waking them up, 171 00:10:28,760 --> 00:10:31,400 S1: which is which is really interesting to think about when 172 00:10:31,400 --> 00:10:36,079 S1: you think about economics and like stimulus and response, but 173 00:10:36,080 --> 00:10:40,790 S1: also quite concerning. Tulsi Gabbard, new director of national intelligence, 174 00:10:40,790 --> 00:10:44,600 S1: is firing more than 100 intelligence officers for sharing explicit 175 00:10:44,600 --> 00:10:49,100 S1: messages on an NSA internal platform. But I'm calling shenanigans 176 00:10:49,100 --> 00:10:52,130 S1: on this. A number of my trans friends who know 177 00:10:52,130 --> 00:10:54,860 S1: a lot about intelligence and cyber are saying this is 100% 178 00:10:54,860 --> 00:10:58,460 S1: an attempt to remove trans people from government, and they're 179 00:10:58,460 --> 00:11:02,179 S1: using like this explicit content thing as like the cover 180 00:11:02,179 --> 00:11:04,910 S1: story for that. It looks like the messages in question 181 00:11:04,940 --> 00:11:09,140 S1: were like about getting like bottom surgery. Similar topics. Kind 182 00:11:09,170 --> 00:11:13,429 S1: of like women talking about post-pregnancy body issues or whatever. 183 00:11:13,460 --> 00:11:16,910 S1: So it could be considered explicit or personal or gross 184 00:11:16,910 --> 00:11:19,370 S1: or whatever in certain lights. But that's all part of 185 00:11:19,370 --> 00:11:22,760 S1: being human, right? So the question is, were these the 186 00:11:22,760 --> 00:11:25,490 S1: right forums to talk about such things? Was it like 187 00:11:25,520 --> 00:11:29,540 S1: egregious or was it like partial? And I didn't see 188 00:11:29,540 --> 00:11:32,449 S1: the messages. I don't know the forum they were doing 189 00:11:32,450 --> 00:11:35,780 S1: this in. I actually did get a reach out from 190 00:11:35,780 --> 00:11:38,690 S1: someone in the IC community, actually, kind of like on 191 00:11:38,690 --> 00:11:42,620 S1: the inside around here around the topic, who actually did 192 00:11:42,650 --> 00:11:46,729 S1: see all this stuff. And they said it was pretty 193 00:11:46,760 --> 00:11:51,320 S1: egregious and it was pretty bad and I don't know. 194 00:11:51,350 --> 00:11:53,600 S1: So first of all, they could be lying. Second of all, 195 00:11:53,600 --> 00:11:56,360 S1: I don't think they were, but they could be lying. 196 00:11:56,390 --> 00:11:59,150 S1: It's also a single data point. It's a single opinion. 197 00:11:59,960 --> 00:12:02,449 S1: The point is, there's kind of two obvious ways to 198 00:12:02,480 --> 00:12:06,110 S1: go here. Like, it was really bad, and they deserve 199 00:12:06,110 --> 00:12:10,219 S1: to be punished or whatever. But I say there's two 200 00:12:10,220 --> 00:12:12,020 S1: ways of looking at it. First of all, there's a 201 00:12:12,020 --> 00:12:14,300 S1: bunch of trans people ignoring their work and spending their 202 00:12:14,300 --> 00:12:19,160 S1: time talking about personal trans issues in an inappropriate forum 203 00:12:19,160 --> 00:12:22,339 S1: in an inappropriate way when they should have been working. 204 00:12:22,850 --> 00:12:24,980 S1: So they got fired. That's one way to think about it. 205 00:12:24,980 --> 00:12:27,980 S1: Another way to think about it is Tulsi Gabbard is 206 00:12:27,980 --> 00:12:31,760 S1: part of an extremely anti-trans administration that is looking for 207 00:12:31,760 --> 00:12:34,910 S1: any excuse to fire trans people, even if it harms 208 00:12:34,910 --> 00:12:37,910 S1: our ability to do intelligence work. And this is a 209 00:12:37,910 --> 00:12:40,610 S1: cover story for doing so. Again, I'm not close to 210 00:12:40,640 --> 00:12:43,850 S1: the sources, so I can't operate with any certainty. But giving, 211 00:12:43,880 --> 00:12:47,300 S1: given what we know the administration is actually doing against 212 00:12:47,300 --> 00:12:51,329 S1: trans people and how like aggressively they're going after this issue. 213 00:12:51,390 --> 00:12:53,790 S1: I think it's more likely to be number two for me. 214 00:12:53,790 --> 00:12:58,020 S1: That's just logical. TSMC is planning to invest $100 billion 215 00:12:58,020 --> 00:13:00,780 S1: in US chip plants over the next four years, according 216 00:13:00,809 --> 00:13:05,160 S1: to President Trump's announcement yesterday, which was a few days 217 00:13:05,160 --> 00:13:08,670 S1: ago now. And I think this is great. Get all 218 00:13:08,670 --> 00:13:12,570 S1: these companies here, Apple to spin them up in Arizona. 219 00:13:12,600 --> 00:13:15,000 S1: Like we got plenty of desert. We got plenty of 220 00:13:15,000 --> 00:13:19,080 S1: wide open spaces. Let's build some giant solar farms. Let's 221 00:13:19,080 --> 00:13:22,680 S1: get chip production here. Let's get Apple production here. Let's 222 00:13:22,679 --> 00:13:25,680 S1: get everything over here. Because the world is crazy. All right. 223 00:13:25,710 --> 00:13:29,610 S1: I anthropic closed a $3.5 billion funding round that values 224 00:13:29,610 --> 00:13:35,280 S1: the company at 61.5 billion. Revenue has grown 1,000% year 225 00:13:35,280 --> 00:13:39,809 S1: over year. Really? Feels like they're quietly doing what OpenAI 226 00:13:39,809 --> 00:13:42,840 S1: is loudly failing at. And I think this should be 227 00:13:42,840 --> 00:13:46,109 S1: really encouraging for companies that enter spaces that seem like 228 00:13:46,110 --> 00:13:49,050 S1: they're already won. Like you wouldn't want to go up 229 00:13:49,050 --> 00:13:54,780 S1: against OpenAI in, say, early 2023. They just owned the 230 00:13:54,780 --> 00:13:59,190 S1: entire AI world and there was nobody around to compete. 231 00:13:59,280 --> 00:14:03,989 S1: And here we are beginning of 25, and we've got 232 00:14:03,990 --> 00:14:07,470 S1: a few people competing and we've got open models competing. 233 00:14:07,470 --> 00:14:12,000 S1: It's it's really encouraging, actually, as you know, if you're 234 00:14:12,000 --> 00:14:15,960 S1: trying to get people excited about being creative and building, 235 00:14:16,230 --> 00:14:20,100 S1: it's it's great to see somebody that was untouchable two 236 00:14:20,100 --> 00:14:25,440 S1: years ago be very touchable now. OpenAI released GPT 4.5. 237 00:14:25,440 --> 00:14:28,890 S1: Most people are not happy with it. It a lot 238 00:14:28,890 --> 00:14:31,230 S1: of people are saying it doesn't feel smarter at all. 239 00:14:31,230 --> 00:14:33,989 S1: And to me it felt like like a push, like 240 00:14:33,990 --> 00:14:38,130 S1: PR release because sonnet 3.7 came out. Other models have 241 00:14:38,130 --> 00:14:42,210 S1: come out recently and yeah, it just it didn't go well. 242 00:14:42,240 --> 00:14:44,400 S1: They should have paused on that and waited for a 243 00:14:44,400 --> 00:14:48,810 S1: bigger release that was more more baked, more proper marketing 244 00:14:48,810 --> 00:14:52,229 S1: around it and everything. Chinese government is telling its AI 245 00:14:52,230 --> 00:14:55,020 S1: experts to avoid US travel because they're worried they might 246 00:14:55,050 --> 00:15:01,380 S1: leak Chinese secrets. Yeah. Leaking AI secrets. China's worried about it. Cool. 247 00:15:01,590 --> 00:15:06,840 S1: Roger Penrose explains why Godel's theorem mathematically proves that artificial 248 00:15:06,840 --> 00:15:12,450 S1: intelligence can't truly replicate human consciousness or understanding. I disagree 249 00:15:12,450 --> 00:15:16,590 S1: with this, obviously, but it's Roger Penrose. I think he's 250 00:15:16,590 --> 00:15:19,410 S1: a great thinker, or at least has been a great 251 00:15:19,410 --> 00:15:22,890 S1: thinker on a lot of different topics. I know that 252 00:15:22,890 --> 00:15:26,160 S1: there are a few areas, not counting AI, where I 253 00:15:26,190 --> 00:15:29,400 S1: disagree with him, but doesn't matter. He is a top 254 00:15:29,400 --> 00:15:33,990 S1: tier thinker and he disagrees with my position. Therefore I 255 00:15:34,020 --> 00:15:37,860 S1: am putting it in the newsletter. Marc Benioff says Salesforce 256 00:15:37,860 --> 00:15:42,510 S1: won't hire engineers in 2025 because their AI agents are 257 00:15:42,510 --> 00:15:46,650 S1: doing the work so effectively. That sounds awesome and I 258 00:15:46,680 --> 00:15:50,670 S1: kind of believe him. However, this is also direct marketing 259 00:15:50,670 --> 00:15:53,700 S1: for Agent Force, which is their product. So hard to 260 00:15:53,730 --> 00:15:56,370 S1: tell if this is marketing or real. New study shows 261 00:15:56,370 --> 00:16:00,810 S1: that large language models are just echoing logical patterns without 262 00:16:00,840 --> 00:16:06,120 S1: real understanding, which explains why they struggle with complex reasoning. Again, 263 00:16:06,360 --> 00:16:11,040 S1: I want to see the best possible arguments against my position. Technology. 264 00:16:11,070 --> 00:16:16,440 S1: Waymo now handles over 200,000 paid robotaxi rides weekly across 265 00:16:16,440 --> 00:16:22,440 S1: three cities 20 times growth 20 x growth. That's how 266 00:16:22,440 --> 00:16:24,450 S1: you're supposed to say that 20 x growth in just 267 00:16:24,450 --> 00:16:30,060 S1: two years. BYD and DJI have created a $2,200 vehicle 268 00:16:30,060 --> 00:16:36,750 S1: mounted drone launch platform that lets your BYD vehicle deploy 269 00:16:36,750 --> 00:16:41,160 S1: and retrieve drones while driving up to 15mph. This is 270 00:16:41,160 --> 00:16:42,930 S1: the kind of cool stuff that I'm talking about that 271 00:16:42,960 --> 00:16:45,600 S1: BYD is coming out with. They also had a video 272 00:16:45,600 --> 00:16:49,530 S1: recently of their electric car doing a turbo boost, like 273 00:16:49,530 --> 00:16:53,460 S1: over a pothole or something. Insane stuff. Like they're doing 274 00:16:53,460 --> 00:16:58,590 S1: some pretty innovative stuff. I'm worried about us carmakers competing 275 00:16:58,590 --> 00:17:01,410 S1: against them, I really am. It's like, this is why 276 00:17:01,410 --> 00:17:04,650 S1: I'm kind of like, I like Tesla as a foil 277 00:17:04,650 --> 00:17:10,500 S1: against them, right? I'm obviously very unhappy with Elon right now. 278 00:17:10,710 --> 00:17:13,830 S1: So that's that's got me a bit torn. But I 279 00:17:13,830 --> 00:17:17,369 S1: do not want American car companies to fall to BYD 280 00:17:17,730 --> 00:17:22,169 S1: because their cars are way cheaper, way cooler, and you know, 281 00:17:22,200 --> 00:17:24,390 S1: no one else can keep up with them. It seems 282 00:17:24,390 --> 00:17:27,480 S1: like Tesla's got the only chance. All right, humans, all right. 283 00:17:27,480 --> 00:17:31,409 S1: The housing theory of everything argues that our catastrophically restrictive 284 00:17:31,410 --> 00:17:36,119 S1: housing policies create cascading problems across society, from inequality to 285 00:17:36,150 --> 00:17:43,620 S1: climate change. Many Doraiswamy Doraiswamy talks about how solopreneurs are 286 00:17:43,619 --> 00:17:48,360 S1: becoming real competition to traditional companies thanks to AI coding, 287 00:17:48,390 --> 00:17:53,220 S1: top MBA programs, seeing way fewer grads landing jobs quickly. 288 00:17:53,220 --> 00:17:59,430 S1: With Harvard's unemployment rate jumping from 4 to 15% since 2021. 289 00:17:59,430 --> 00:18:03,120 S1: Think about what a Harvard MBA with potentially not too 290 00:18:03,119 --> 00:18:06,540 S1: much experience in the real world of business is able 291 00:18:06,540 --> 00:18:11,400 S1: to do in a world full of very, very smart AI. 292 00:18:11,790 --> 00:18:14,940 S1: What exactly are we getting from this Harvard MBA? I 293 00:18:14,940 --> 00:18:18,540 S1: think so many things. This is kind of a meta 294 00:18:18,540 --> 00:18:25,590 S1: analysis here. So many things that are extremely expensive and elitist. 295 00:18:25,590 --> 00:18:29,970 S1: And they are those things because they are supply constrained. 296 00:18:29,970 --> 00:18:33,030 S1: You can't get into Harvard. Therefore, Harvard is worth a lot. 297 00:18:33,060 --> 00:18:36,420 S1: A degree from there is worth a lot, right? You 298 00:18:36,450 --> 00:18:40,710 S1: it's hard to raise $200 million from a VC. Therefore, 299 00:18:40,710 --> 00:18:43,950 S1: any company who makes a product that's backed by a 300 00:18:43,980 --> 00:18:48,210 S1: VC and has raised $200 million. The product must be good. 301 00:18:48,240 --> 00:18:52,800 S1: This is dying. This concept is dying. That's why things 302 00:18:52,800 --> 00:18:57,900 S1: like Harvard are going to massively struggle, because the value 303 00:18:57,900 --> 00:19:01,980 S1: of what they're actually producing is not going to be 304 00:19:01,980 --> 00:19:04,649 S1: as good as if you watched a whole lot of 305 00:19:04,650 --> 00:19:07,860 S1: YouTube and read a whole lot of books and did 306 00:19:07,859 --> 00:19:10,770 S1: a whole lot of writing, right, and moved ideas through 307 00:19:10,800 --> 00:19:14,369 S1: your mind very quickly, and actually worked with companies and 308 00:19:14,369 --> 00:19:17,820 S1: had experience on the ground. The difference between that education 309 00:19:17,820 --> 00:19:21,660 S1: and a Harvard education is about to be pretty vast, 310 00:19:21,660 --> 00:19:25,800 S1: benefiting the YouTube side, not the Harvard side. Yet the 311 00:19:25,800 --> 00:19:27,899 S1: YouTube one is available to you if you want to 312 00:19:27,900 --> 00:19:32,280 S1: study right. This dynamic is going to change startups. It's 313 00:19:32,280 --> 00:19:36,660 S1: going to change education. It's basically changing everything. It's like 314 00:19:36,690 --> 00:19:39,900 S1: the it's not just AI, right? This is also a 315 00:19:39,900 --> 00:19:45,090 S1: content explosion. This is also a YouTube trend vibe, where 316 00:19:45,090 --> 00:19:48,030 S1: all the best content is coming out into blogs, coming 317 00:19:48,030 --> 00:19:52,110 S1: out into YouTube. By the way, YouTube is blogging, right? 318 00:19:52,410 --> 00:19:54,690 S1: It used to be that there were very few bloggers. 319 00:19:54,720 --> 00:19:58,770 S1: Now there's very few YouTubers. I mean, a tiny percentage 320 00:19:58,770 --> 00:20:02,220 S1: of 1% of people in the United States are actually 321 00:20:02,220 --> 00:20:05,550 S1: putting out content on YouTube, just like it was with blogging. 322 00:20:05,550 --> 00:20:09,870 S1: But that's where the content is, okay, both in blogging 323 00:20:09,869 --> 00:20:14,190 S1: and newsletters, but also YouTube. This is the front edge 324 00:20:14,190 --> 00:20:16,920 S1: of things, right? This is the front edge of things. 325 00:20:16,920 --> 00:20:20,190 S1: This is where the content is. So just something to 326 00:20:20,190 --> 00:20:25,330 S1: keep in mind. Researchers mapped porn titles from 2008 to 327 00:20:25,330 --> 00:20:29,850 S1: 2023 and showed a disturbing migration in themes from Hot 328 00:20:29,880 --> 00:20:36,000 S1: Blonde gets you know what to increasingly incestuous and violent 329 00:20:36,000 --> 00:20:40,020 S1: concepts like I can't resist my stepsister. I find this 330 00:20:40,020 --> 00:20:43,980 S1: really disturbing because it's like, I would love to get 331 00:20:44,010 --> 00:20:46,140 S1: AI to look at this. Actually, I should go find 332 00:20:46,140 --> 00:20:49,919 S1: this data set. This link here is actually the GitHub 333 00:20:49,920 --> 00:20:53,370 S1: with the data in there and the data analysis. But 334 00:20:53,369 --> 00:20:57,180 S1: I imagine you could take all those titles 2008 to 23. 335 00:20:57,180 --> 00:21:03,209 S1: So what is this? This is like 1013 plus 215 336 00:21:03,210 --> 00:21:08,160 S1: years of title data, 15 years of title data that 337 00:21:08,160 --> 00:21:11,639 S1: you could put into, like whatever Google Studio. It would 338 00:21:11,640 --> 00:21:13,890 S1: probably be pretty hard to actually do this on a 339 00:21:13,890 --> 00:21:17,280 S1: on a pinnacle model, because it's going to be like, uh, 340 00:21:17,280 --> 00:21:21,390 S1: what is this? These are all nasty words. Um, but 341 00:21:21,390 --> 00:21:23,639 S1: maybe if you told it that you were doing research, 342 00:21:23,670 --> 00:21:28,980 S1: maybe that would work, uh, like a prompt injection bypass technique. 343 00:21:28,980 --> 00:21:31,470 S1: But anyway, the point is, you can put this into, like, 344 00:21:31,500 --> 00:21:36,869 S1: Google Studio Flash 2.0 something with a large context, maybe even, like, 345 00:21:36,900 --> 00:21:42,420 S1: you know, um, oh three mini oh one Pro. Uh, 346 00:21:42,450 --> 00:21:46,050 S1: sonnet 37, one of the smarter ones. And just be like, 347 00:21:46,200 --> 00:21:48,750 S1: I have a question. What the hell is happening to 348 00:21:48,780 --> 00:21:55,740 S1: my society? Seeing these, uh, titles migrate from 2008 to 2023, 349 00:21:55,770 --> 00:21:59,700 S1: what is going on? Uh, I'd be really excited to, like, 350 00:21:59,730 --> 00:22:03,780 S1: see what it actually gives as an answer to that. Right. Well, 351 00:22:03,780 --> 00:22:08,070 S1: I say excited. Uh, depressed. Um, yeah, I don't know. 352 00:22:08,070 --> 00:22:13,380 S1: It's not quite depressed. It's more like worried. Concerned. Yeah. Anyway, 353 00:22:13,410 --> 00:22:17,310 S1: number of young people not in education, employment or training. 354 00:22:17,310 --> 00:22:20,730 S1: Neet has hit an 11 year high. I didn't use 355 00:22:20,730 --> 00:22:26,730 S1: an Oxford comma. I'm canceling everything and getting off the internet. 356 00:22:26,970 --> 00:22:30,990 S1: This is highly, highly scary. Yeah. I wonder if I 357 00:22:31,020 --> 00:22:35,820 S1: copied something here from a sentence. Because this is a 358 00:22:35,820 --> 00:22:40,379 S1: British story. Yeah. Did I copy and paste this Neet part? 359 00:22:40,380 --> 00:22:42,750 S1: I don't know why that doesn't have an Oxford comma. 360 00:22:42,750 --> 00:22:46,860 S1: It seriously disturbs me. All right. Solarpunk is a subgenre 361 00:22:46,859 --> 00:22:50,130 S1: of sci fi that imagines a sustainable future where humans 362 00:22:50,130 --> 00:22:54,359 S1: and nature thrive together using renewable technology. I've been a 363 00:22:54,359 --> 00:22:58,350 S1: fan ever since the Chobani commercial by one of my 364 00:22:58,350 --> 00:23:04,020 S1: favorite commercials ever. It is a yogurt commercial about solarpunk. 365 00:23:04,050 --> 00:23:07,080 S1: It is fantastic. Go to the newsletter and click on 366 00:23:07,080 --> 00:23:11,490 S1: this link. It is the best. Um, although talking to 367 00:23:11,520 --> 00:23:14,670 S1: my friend who I'm hoping to have on the podcast soon, 368 00:23:14,670 --> 00:23:18,240 S1: he's like, well, you know, Nazis love the solarpunk stuff. 369 00:23:18,240 --> 00:23:20,879 S1: I'm like, why is it every time I tell you 370 00:23:20,880 --> 00:23:24,840 S1: something's cool, you're like, yeah, there's a subgroup of like, 371 00:23:24,840 --> 00:23:28,230 S1: crazy Catholics who are like super into this and they're 372 00:23:28,230 --> 00:23:34,830 S1: like all very trad trad like, um, oriented, and they're 373 00:23:34,830 --> 00:23:37,410 S1: trying to destroy technology or whatever. And I'm just like, 374 00:23:37,440 --> 00:23:39,990 S1: oh my God. Every time I talk to it, every 375 00:23:40,020 --> 00:23:42,490 S1: time I talk to him, I'm like, what is going on? 376 00:23:43,210 --> 00:23:45,609 S1: But that should be a cool podcast if that happens 377 00:23:45,730 --> 00:23:47,950 S1: within the next couple of weeks. We're going to talk 378 00:23:47,950 --> 00:23:52,300 S1: about all the different sub political movements going on in 379 00:23:52,300 --> 00:23:55,389 S1: the Bay area where we live, but also in general 380 00:23:55,810 --> 00:23:58,840 S1: post liberalism ideas like that. It's going to be an 381 00:23:58,840 --> 00:24:02,350 S1: insane episode. And Ellen might go on Jon Stewart's show 382 00:24:02,350 --> 00:24:07,480 S1: and discuss Doge. That would be a cool debate slash discussion, 383 00:24:07,480 --> 00:24:11,109 S1: hopefully a discussion. Hopefully it's good. Ellen, who shows up. 384 00:24:15,520 --> 00:24:17,980 S1: All right. This is the end of the standard edition 385 00:24:17,980 --> 00:24:20,770 S1: of the podcast, which includes just the news items for 386 00:24:20,770 --> 00:24:23,140 S1: the week to get the rest of the episode, which 387 00:24:23,170 --> 00:24:26,500 S1: includes my analysis, the discovery section that has all of 388 00:24:26,500 --> 00:24:30,100 S1: the coolest tools and articles. I found this week, the 389 00:24:30,100 --> 00:24:32,890 S1: recommendation of the week and the aphorism of the week. 390 00:24:32,890 --> 00:24:37,390 S1: Please consider becoming a member. As a member, you'll get 391 00:24:37,390 --> 00:24:41,010 S1: lots of different things from access to our extraordinary community 392 00:24:41,010 --> 00:24:44,790 S1: of over a thousand brilliant and kind people in industries 393 00:24:44,790 --> 00:24:50,100 S1: like cybersecurity, AI, technology and the humanities. You also get 394 00:24:50,130 --> 00:24:54,330 S1: access to the UL Book Club, dedicated member content and events, 395 00:24:54,330 --> 00:24:58,440 S1: and lots, lots more. Plus, you'll get a dedicated podcast 396 00:24:58,440 --> 00:25:00,750 S1: feed you can put in your client that gets you 397 00:25:00,750 --> 00:25:04,530 S1: the full member edition of this podcast. To become a 398 00:25:04,530 --> 00:25:07,200 S1: member and get all of that, just head over to 399 00:25:07,230 --> 00:25:14,460 S1: Daniel maestro.com/upgrade. That's Daniel miessler.com/upgrade. We'll see you next time. 400 00:25:15,900 --> 00:25:19,680 S1: Unsupervised learning is produced on Hindenburg Pro using an SM 401 00:25:19,680 --> 00:25:23,280 S1: seven B microphone. A video version of the podcast is 402 00:25:23,280 --> 00:25:27,000 S1: available on the Unsupervised Learning YouTube channel, and the text 403 00:25:27,030 --> 00:25:30,300 S1: version with full links and notes is available at Daniel 404 00:25:30,300 --> 00:25:34,260 S1: Mysa.com slash newsletter. We'll see you next time.