1 00:00:01,260 --> 00:00:04,530 S1: Welcome to Unsupervised Learning, a security, AI and meaning focused 2 00:00:04,530 --> 00:00:07,410 S1: podcast that looks at how best to thrive as humans 3 00:00:07,410 --> 00:00:11,610 S1: in a post AI world. It combines original ideas, analysis, 4 00:00:11,610 --> 00:00:14,850 S1: and mental models to bring not just the news, but 5 00:00:14,850 --> 00:00:21,870 S1: why it matters and how to respond. All right. Welcome 6 00:00:21,870 --> 00:00:25,919 S1: to unsupervised learning. This is Daniel Missler. All right. Number 425. 7 00:00:25,920 --> 00:00:28,620 S1: So you all know the deal here. Tons going on 8 00:00:28,620 --> 00:00:31,500 S1: this week prepping for some paid talks. Going to be 9 00:00:31,530 --> 00:00:34,500 S1: traveling to a few different states. Honestly really excited to 10 00:00:34,500 --> 00:00:37,290 S1: be alive right now at this moment in human history, 11 00:00:37,290 --> 00:00:40,200 S1: like all the different opportunities there are that are just 12 00:00:40,200 --> 00:00:43,200 S1: popping off and I feel like there's always a conference, 13 00:00:43,200 --> 00:00:45,659 S1: there's always people. There's so many people building right now, 14 00:00:45,659 --> 00:00:48,210 S1: especially here in the Bay area. And I'm just excited 15 00:00:48,210 --> 00:00:51,330 S1: about the optimism. Like the optimism has its own gravity, 16 00:00:51,330 --> 00:00:54,510 S1: if that makes sense. And yeah hack build grind appreciate. 17 00:00:54,510 --> 00:00:57,360 S1: That's pretty much the current vibe. So I wrote this 18 00:00:57,360 --> 00:01:00,240 S1: piece called Efficient Security Principle. And I'm going to open 19 00:01:00,240 --> 00:01:03,600 S1: this up in a new tab. And it's essentially explaining 20 00:01:03,600 --> 00:01:06,990 S1: why a lot of security people are so upset, I think, 21 00:01:06,990 --> 00:01:09,030 S1: over the course of their career. So one of the 22 00:01:09,030 --> 00:01:11,580 S1: hardest things about being in security is just the frustration. 23 00:01:11,580 --> 00:01:15,930 S1: And this came from the conversation with Deno, where he 24 00:01:15,930 --> 00:01:20,010 S1: basically posted about raising the security minimum. And I was like, well, 25 00:01:20,010 --> 00:01:22,200 S1: you know, the minimum is there for a reason. And 26 00:01:22,200 --> 00:01:25,290 S1: I kind of wanted to capture that into a principle. 27 00:01:25,290 --> 00:01:28,110 S1: And ideally I would have this be one item, but 28 00:01:28,110 --> 00:01:29,820 S1: it was getting kind of long. So I broke it 29 00:01:29,819 --> 00:01:34,350 S1: into these statements. Security baseline of an offering continuous downward 30 00:01:34,350 --> 00:01:39,209 S1: pressure from customer excitement or reliance on a given thing. 31 00:01:39,209 --> 00:01:42,330 S1: And then the baseline will be set at the point 32 00:01:42,330 --> 00:01:45,690 S1: at which people will stop using it if it's any worse. 33 00:01:45,690 --> 00:01:47,490 S1: But it can't go higher than that. And then the 34 00:01:47,490 --> 00:01:49,230 S1: third piece of it, which these are all kind of 35 00:01:49,230 --> 00:01:51,780 S1: saying the same thing, is that the better the offering is, 36 00:01:51,780 --> 00:01:54,900 S1: or the more important it is, the lower the security 37 00:01:54,900 --> 00:01:59,340 S1: baseline can go without losing people. And that's really, really important. 38 00:01:59,340 --> 00:02:02,460 S1: So all this together basically means that the more essential 39 00:02:02,460 --> 00:02:04,590 S1: a thing is, and the more awesome it is, the 40 00:02:04,590 --> 00:02:08,070 S1: more people will put up with. And as we'll see 41 00:02:08,070 --> 00:02:10,410 S1: when we move through this, it's the reason you shouldn't 42 00:02:10,410 --> 00:02:13,290 S1: get too upset when the baseline is low. So the 43 00:02:13,290 --> 00:02:15,570 S1: way we know something has the right amount of security 44 00:02:15,570 --> 00:02:19,500 S1: is people just keep using it despite how insecure it is. Right? 45 00:02:19,500 --> 00:02:23,429 S1: So online companies constantly getting hacked. My friend Sasha Geller 46 00:02:23,430 --> 00:02:26,130 S1: basically talked about this one where he said that basically 47 00:02:26,130 --> 00:02:28,800 S1: the biggest example of this is actually email, where it's 48 00:02:28,800 --> 00:02:32,100 S1: like email gets everyone hacked. It's like the number one 49 00:02:32,100 --> 00:02:34,170 S1: way to get hacked is to have email. But guess what? 50 00:02:34,169 --> 00:02:36,450 S1: You need email. There's a reliance on it. It's like 51 00:02:36,450 --> 00:02:39,510 S1: our national you know, our global infrastructure of it is 52 00:02:39,510 --> 00:02:43,380 S1: basically emails to online banking. Everyone's getting constant fraud all 53 00:02:43,380 --> 00:02:46,230 S1: the time. Door locks and the internet in general is 54 00:02:46,230 --> 00:02:49,050 S1: pretty nasty. It's a cesspool, right? We use these things 55 00:02:49,050 --> 00:02:53,160 S1: anyway because the value they provide massively outweighs the security risks, 56 00:02:53,160 --> 00:02:55,769 S1: at least according to us in our own minds. And 57 00:02:55,770 --> 00:02:58,620 S1: the moment people stop using something due to security being 58 00:02:58,620 --> 00:03:02,040 S1: too bad, the baseline goes up. And Sasha had another 59 00:03:02,040 --> 00:03:04,680 S1: good point in that. SMS is actually an example of 60 00:03:04,680 --> 00:03:07,770 S1: this where it was deemed too nasty. And then we 61 00:03:07,770 --> 00:03:10,680 S1: started moving to like app based tufa. So the way 62 00:03:10,680 --> 00:03:14,100 S1: to actually use this is if you're a security tech professional, 63 00:03:14,100 --> 00:03:17,430 S1: like down in the weeds, realize that it's not that 64 00:03:17,430 --> 00:03:20,250 S1: level that matters at a bigger scale. So you want 65 00:03:20,250 --> 00:03:23,910 S1: to like uplevel your thinking and realize it's not about us. 66 00:03:23,910 --> 00:03:26,760 S1: It's not about the technical part of it. It's about 67 00:03:26,760 --> 00:03:29,910 S1: how important is the offering. And that's what actually sets 68 00:03:29,910 --> 00:03:33,600 S1: the baseline. And most importantly, do not take it personally. 69 00:03:33,600 --> 00:03:36,510 S1: Don't take it personally that the baseline is low and 70 00:03:36,510 --> 00:03:38,730 S1: you're showing them how to raise the baseline, but they're 71 00:03:38,730 --> 00:03:41,280 S1: not doing anything about it, right. That's really key. And 72 00:03:41,280 --> 00:03:43,680 S1: if you're a security leader, you know, you can have 73 00:03:43,680 --> 00:03:46,680 S1: the same sort of problem and frustration of a technical person. 74 00:03:46,680 --> 00:03:49,470 S1: And maybe you're also a technical leader. It doesn't really matter. 75 00:03:49,470 --> 00:03:51,780 S1: But the point is like you can get frustrated as well. 76 00:03:51,780 --> 00:03:54,990 S1: And the recommendation is first, make sure the security baseline 77 00:03:54,990 --> 00:03:57,960 S1: is actually where the customer thinks it is. Maybe they're 78 00:03:57,960 --> 00:04:00,030 S1: not complaining because they don't really know how bad it is. 79 00:04:00,030 --> 00:04:02,970 S1: So once those are in alignment now, now you have 80 00:04:02,970 --> 00:04:05,970 S1: sort of the same situation as the technical side where 81 00:04:05,970 --> 00:04:10,230 S1: it's like look fine, innovative ways to raise the baseline 82 00:04:10,230 --> 00:04:14,280 S1: without making it obvious or cause friction because people might 83 00:04:14,280 --> 00:04:16,350 S1: want to higher baseline, but only if it doesn't cost 84 00:04:16,350 --> 00:04:18,810 S1: them anything because they're only going to go as high 85 00:04:18,810 --> 00:04:22,950 S1: as it can go without sacrificing that experience of this 86 00:04:22,950 --> 00:04:25,530 S1: thing that they love. And I just think it's really powerful. 87 00:04:25,529 --> 00:04:28,050 S1: So yeah, only as good as it needs to be 88 00:04:28,050 --> 00:04:30,720 S1: to keep people from abandoning the service. And the more 89 00:04:30,720 --> 00:04:34,020 S1: popular or essential the offering, the lower the security can be. 90 00:04:34,020 --> 00:04:37,050 S1: This is in the summary here. And progress is still possible, 91 00:04:37,050 --> 00:04:40,440 S1: especially with policy change and regulation, because that kind of 92 00:04:40,440 --> 00:04:44,580 S1: like forces a baseline raise. But security experts loudly calling 93 00:04:44,580 --> 00:04:48,000 S1: out how low the baseline is and gesturing wildly towards 94 00:04:48,000 --> 00:04:52,200 S1: the solution seldom results in change. And most importantly, this 95 00:04:52,200 --> 00:04:55,410 S1: is kind of the big takeaway is experts struggling with 96 00:04:55,410 --> 00:04:59,910 S1: low security baselines should absorb the truth above and not 97 00:04:59,910 --> 00:05:03,540 S1: let their job satisfaction and their mental health suffer. And 98 00:05:03,540 --> 00:05:05,550 S1: got a bunch of notes here. But that's that's basically 99 00:05:05,550 --> 00:05:07,589 S1: that piece. All right. A couple of really cool patterns 100 00:05:07,589 --> 00:05:10,140 S1: here put into fabric this week. So the first one 101 00:05:10,140 --> 00:05:12,990 S1: is like it'll analyze your prose according to my favorite 102 00:05:12,990 --> 00:05:15,840 S1: writing book, which is The Sense of Style by Steven Pinker. 103 00:05:15,839 --> 00:05:18,600 S1: And here's what it actually looks like after a run. 104 00:05:18,600 --> 00:05:22,529 S1: This is. The analysis of me actually doing that, that essay. 105 00:05:22,529 --> 00:05:24,480 S1: So I put that essay through here and this is 106 00:05:24,480 --> 00:05:28,080 S1: the result I came back. So style analysis it's multiple 107 00:05:28,080 --> 00:05:32,520 S1: different styles put together positive assessment. Critical assessment I talk 108 00:05:32,520 --> 00:05:35,160 S1: a little bit too much about the writing itself I 109 00:05:35,160 --> 00:05:38,279 S1: guess is what this is saying. It gives me examples 110 00:05:38,279 --> 00:05:40,859 S1: of good writing from it. Things I can improve. No 111 00:05:40,860 --> 00:05:42,900 S1: spelling mistakes, which is cool. I'm going to be using 112 00:05:42,900 --> 00:05:45,270 S1: this to catch spelling from now on, and you start 113 00:05:45,270 --> 00:05:48,180 S1: off with a 100 score. Oh, and it gives recommendations 114 00:05:48,180 --> 00:05:50,160 S1: as well. But it starts off with 100 score and 115 00:05:50,160 --> 00:05:52,710 S1: then deducts. So I got an 85. I got an 116 00:05:52,710 --> 00:05:56,070 S1: 85 here because it took 15 points off for these things. 117 00:05:56,070 --> 00:05:59,190 S1: So that's what that pattern does. And then this one 118 00:05:59,190 --> 00:06:03,390 S1: is extract book recommendations. So this is insane. This is 119 00:06:03,390 --> 00:06:06,960 S1: essentially any book you could think of that was probably 120 00:06:06,960 --> 00:06:10,290 S1: brought in the training of a given model. And it's 121 00:06:10,290 --> 00:06:11,910 S1: most that you could think of like a New York 122 00:06:11,910 --> 00:06:14,520 S1: Times bestseller. Any big name book that you can think 123 00:06:14,520 --> 00:06:16,529 S1: of is probably going to be in here. So open 124 00:06:16,529 --> 00:06:18,960 S1: this one up and look at it. So this is 125 00:06:18,960 --> 00:06:22,800 S1: a man search for Meaning by Viktor Viktor Frankl. And 126 00:06:22,800 --> 00:06:25,919 S1: these are the recommendations that it gave me for what 127 00:06:25,920 --> 00:06:29,310 S1: to do right to like improve your life. I mean 128 00:06:29,310 --> 00:06:33,029 S1: look at these, these these are absolutely insane. Find meaning 129 00:06:33,029 --> 00:06:36,360 S1: in life by dedicating yourself to a cause greater than yourself. 130 00:06:36,360 --> 00:06:40,380 S1: Like even the prioritization here is really, really good. Take 131 00:06:40,380 --> 00:06:44,040 S1: responsibility for your own life and choices. I mean, this 132 00:06:44,040 --> 00:06:48,600 S1: is gorgeous and this all just came from running. I 133 00:06:48,600 --> 00:06:52,740 S1: literally said Echo Man's search for meaning pipe into extract 134 00:06:52,740 --> 00:06:56,159 S1: book recommendation and this is what it produced, right? This 135 00:06:56,160 --> 00:06:59,099 S1: is so powerful. I'm just blown away every time we 136 00:06:59,100 --> 00:07:01,170 S1: make one of these things. So those are two of 137 00:07:01,170 --> 00:07:03,000 S1: the biggest ones that I put in. I actually put 138 00:07:03,000 --> 00:07:05,039 S1: in more, but those are two of the big ones. 139 00:07:05,040 --> 00:07:07,860 S1: So security stuff. Yeah. It looks like you can extract 140 00:07:07,860 --> 00:07:10,860 S1: keys from M-series chips. I'm sure they'll fix that in 141 00:07:11,340 --> 00:07:13,590 S1: the next iteration. After that, a lot of people are 142 00:07:13,590 --> 00:07:16,470 S1: saying it's not really packable and that it's kind of 143 00:07:16,470 --> 00:07:19,320 S1: a big problem. So and that if you do do 144 00:07:19,320 --> 00:07:22,860 S1: a workaround, it might affect performance. But we'll see. And 145 00:07:22,860 --> 00:07:24,570 S1: here's the other piece. We'll have to wait and see 146 00:07:24,570 --> 00:07:27,179 S1: how real world these attacks are to know if it's 147 00:07:27,180 --> 00:07:30,870 S1: like actually bad and the integration of drones with digitized 148 00:07:30,870 --> 00:07:34,290 S1: command and control systems. Yeah, this is a US war 149 00:07:34,290 --> 00:07:39,720 S1: fighting effort essentially, and they're calling it transformative trinity for autonomous, 150 00:07:39,720 --> 00:07:43,920 S1: like very dangerous drones, which saved dystopian movie writers a 151 00:07:43,920 --> 00:07:46,650 S1: lot of time. So that's that's good news. And yeah, 152 00:07:46,680 --> 00:07:48,690 S1: along the lines of this, there's never been a better 153 00:07:48,690 --> 00:07:51,720 S1: time to read Daniel Suarez's kill decision, which is all 154 00:07:51,720 --> 00:07:55,110 S1: about autonomous drones. Actually, you should check out that book, 155 00:07:55,110 --> 00:07:58,200 S1: a really, really good book. And Jess has a new 156 00:07:58,200 --> 00:08:02,100 S1: strategy for tackling AI risks. Got the PDF here. Someone 157 00:08:02,100 --> 00:08:04,170 S1: in China got arrested in New York for trying to 158 00:08:04,170 --> 00:08:08,520 S1: sell Tesla's secret battery tech to undercover agents. I'm not 159 00:08:08,520 --> 00:08:10,680 S1: sure if he thought he was selling it to China. 160 00:08:10,680 --> 00:08:13,170 S1: I couldn't quite get that from the story. Canada is 161 00:08:13,170 --> 00:08:16,380 S1: rethinking its ban of flipper zero, trying to go after 162 00:08:16,380 --> 00:08:19,470 S1: misuse instead of just banning it outright, which I'm glad 163 00:08:19,470 --> 00:08:23,310 S1: they listened to the feedback there. Avonte critical bug. This 164 00:08:23,310 --> 00:08:27,330 S1: just continues to happen. Earth crushing Kraken. Yeah, I'm not 165 00:08:27,330 --> 00:08:30,150 S1: sure how to pronounce that one, but basically a Abt 166 00:08:30,150 --> 00:08:35,010 S1: group backed by the Chinese government hit over 70 organizations, 167 00:08:35,010 --> 00:08:40,140 S1: mostly focused on governments, and Atlassian patched a critical SQL vulnerability. 168 00:08:40,140 --> 00:08:42,990 S1: So myself and a bunch of other AI builders have 169 00:08:42,990 --> 00:08:48,990 S1: found something really weird where anthropic smallest model haiku has 170 00:08:48,990 --> 00:08:53,130 S1: been performing extremely well. And this one I did. What 171 00:08:53,130 --> 00:08:55,770 S1: is this? So this is extract ideas. So I did 172 00:08:55,770 --> 00:09:00,600 S1: extract ideas on a recent YouTube video of Jonathan Haidt 173 00:09:00,600 --> 00:09:03,270 S1: coming on Rogan again. And if you look at these 174 00:09:03,270 --> 00:09:06,660 S1: results here, they are remarkably similar, similar to each other. 175 00:09:06,660 --> 00:09:09,480 S1: So like GPT for the first one is opus, which 176 00:09:09,480 --> 00:09:12,060 S1: is like the best anthropic model. The second one is 177 00:09:12,059 --> 00:09:15,000 S1: GPT four, which is the best OpenAI model. But this 178 00:09:15,000 --> 00:09:19,050 S1: is the worst. It's the third level down of anthropic models, 179 00:09:19,050 --> 00:09:23,730 S1: and these top level like quotes here really good. Not 180 00:09:23,730 --> 00:09:27,570 S1: not quotes, but findings, ideas extracted. I mean, they're not 181 00:09:27,570 --> 00:09:30,030 S1: obviously worse than any of the other two. And this 182 00:09:30,030 --> 00:09:32,040 S1: has happened to me multiple times. So I'm really impressed 183 00:09:32,040 --> 00:09:34,350 S1: with that. And a lot of people are reporting GPT 184 00:09:34,380 --> 00:09:36,990 S1: five will be out this summer. And they're pointing to 185 00:09:36,990 --> 00:09:40,530 S1: this interview that Sam Altman did with Lex. And but 186 00:09:40,530 --> 00:09:44,459 S1: what he actually said was something good would come out soon, 187 00:09:44,460 --> 00:09:46,530 S1: maybe like this summer. I think he might have said, 188 00:09:46,530 --> 00:09:48,870 S1: like this summer. And some people are saying, well, maybe 189 00:09:48,870 --> 00:09:51,720 S1: it'll be like a 4.5 release, but I don't think 190 00:09:51,720 --> 00:09:54,210 S1: it's going to be five before the end of the year. 191 00:09:54,210 --> 00:09:59,969 S1: That's my guess. Nvidia is patterning. Patterning. See, now you know, 192 00:09:59,970 --> 00:10:02,220 S1: I don't use AI to write my stuff. I wrote 193 00:10:02,220 --> 00:10:06,720 S1: this and that is misspelled. That's unfortunate. Yeah. Partnering with 194 00:10:06,720 --> 00:10:11,429 S1: Hippocratic AI to introduce AI nurses for virtual patient care. 195 00:10:11,429 --> 00:10:13,709 S1: So you basically looking at an iPad and it's like 196 00:10:13,710 --> 00:10:16,800 S1: a face telling you what to do $9 an hour. 197 00:10:16,800 --> 00:10:20,140 S1: It's actually an AI. Yeah. So there's a little point 198 00:10:20,140 --> 00:10:22,030 S1: here most of the benefit will get from AI in 199 00:10:22,030 --> 00:10:25,719 S1: the first few years. Won't be from doing work better 200 00:10:25,720 --> 00:10:29,229 S1: than good humans, right? It's going to be doing work 201 00:10:29,230 --> 00:10:31,809 S1: that otherwise would not have been done at all. So 202 00:10:31,809 --> 00:10:34,570 S1: the great examples of this are therapists. We need way 203 00:10:34,570 --> 00:10:39,040 S1: more therapists. We need billions of more therapists, like hundreds 204 00:10:39,040 --> 00:10:42,160 S1: of millions or billions of therapists, potentially. We don't want 205 00:10:42,160 --> 00:10:45,760 S1: to overindex on therapy, but we need so many therapists. 206 00:10:45,760 --> 00:10:48,939 S1: We need billions of tutors, continuous tutors, that you can 207 00:10:48,940 --> 00:10:52,780 S1: ask anything. Why? Well, because of this. Why? Because of this? Why? 208 00:10:52,780 --> 00:10:54,790 S1: How many parents get tired of that question? How about 209 00:10:54,790 --> 00:10:57,250 S1: somebody who doesn't get tired of that question? Right? And 210 00:10:57,250 --> 00:10:59,740 S1: I'm not saying parents shouldn't be there to answer those questions. 211 00:10:59,740 --> 00:11:03,130 S1: Of course they should. And so should teachers. But it's 212 00:11:03,130 --> 00:11:06,850 S1: nice when somebody is shy or introverted or they're just 213 00:11:06,850 --> 00:11:11,469 S1: alone and they have questions, you know, therapists, tutors and yeah, 214 00:11:11,470 --> 00:11:14,350 S1: we don't have enough things watching for asteroids in the sky. 215 00:11:14,350 --> 00:11:18,939 S1: It's extra eyes. Think of trillions of extra eyes and 216 00:11:18,940 --> 00:11:21,730 S1: a little bit of brainpower tied to that skin cancer screening. 217 00:11:21,730 --> 00:11:24,340 S1: How many people in the world have access to a 218 00:11:24,340 --> 00:11:27,370 S1: doctor to say, hey, look at this mole? Very few, 219 00:11:27,370 --> 00:11:30,670 S1: very few percentage of the entire planet have that ability. 220 00:11:30,670 --> 00:11:33,250 S1: But now you can with AI, right? So that's what 221 00:11:33,250 --> 00:11:37,480 S1: I'm excited about. US Department of Justice 16 state and 222 00:11:37,480 --> 00:11:42,520 S1: district attorneys, district attorney generals. Actually, I think that's supposed 223 00:11:42,520 --> 00:11:47,559 S1: to be pluralized attorneys general. Yeah. Lawsuit against Apple. Bottom line. 224 00:11:47,559 --> 00:11:52,449 S1: And I've got a pro Apple biased, very fanboy leaning. Biased. 225 00:11:52,750 --> 00:11:55,929 S1: I'm all about not avoiding bias, but actually just calling 226 00:11:55,929 --> 00:11:57,910 S1: it out and naming it, which I think I saw 227 00:11:57,910 --> 00:12:01,360 S1: Destiny talk about, which I thought was pretty smart. And yeah, 228 00:12:01,390 --> 00:12:04,750 S1: Apple's now making iPhones in Brazil. And Tim Cook also 229 00:12:04,750 --> 00:12:07,150 S1: just went to China to tell him everything's okay. And 230 00:12:07,150 --> 00:12:09,370 S1: don't worry. Don't worry about the fact that we're trying 231 00:12:09,370 --> 00:12:11,920 S1: to make everything not in China because you guys are crazy. 232 00:12:11,920 --> 00:12:14,650 S1: He probably didn't say that, but I did. Apple and 233 00:12:14,650 --> 00:12:18,250 S1: Tesla are losing market share in China. Meanwhile, as national 234 00:12:18,250 --> 00:12:21,939 S1: loyalty rises, and my sort of spicy take here is 235 00:12:22,059 --> 00:12:26,320 S1: one of China's greatest strengths is actually its nationalism. Right. 236 00:12:26,320 --> 00:12:30,160 S1: They have too much I would say. Um, and I 237 00:12:30,160 --> 00:12:32,710 S1: would say that the left in the US has too little. 238 00:12:32,710 --> 00:12:34,900 S1: I would say like the far right in the US 239 00:12:34,900 --> 00:12:37,929 S1: has too much. Can we can we have some medium, 240 00:12:37,929 --> 00:12:41,320 S1: can we have some, uh, some moderation, please? That would 241 00:12:41,320 --> 00:12:44,830 S1: be nice. Israel's government is reportedly running covert ops at 242 00:12:45,190 --> 00:12:48,790 S1: US universities. This turns out to be a little bit 243 00:12:48,790 --> 00:12:51,550 S1: kind of like a biased story. But that's why I 244 00:12:51,550 --> 00:12:53,950 S1: put this in here. It's like, what's the difference between 245 00:12:53,950 --> 00:12:59,470 S1: marketing and counter propaganda information operations? The difference is your perspective. 246 00:12:59,470 --> 00:13:02,620 S1: That's that's the difference. Like, we should consider ourselves always 247 00:13:02,620 --> 00:13:06,160 S1: being hit with propaganda and marketing or whatever. And like 248 00:13:06,160 --> 00:13:08,830 S1: you use the different term based on what you think 249 00:13:08,830 --> 00:13:11,890 S1: their goals are and how morally aligned you think their 250 00:13:11,890 --> 00:13:15,370 S1: goals are with, like good or bad. Measles was eliminated 251 00:13:15,370 --> 00:13:18,850 S1: in the US in 2000, and now it's back because 252 00:13:18,850 --> 00:13:21,640 S1: people don't want to take vaccines. As long Covid brain 253 00:13:21,640 --> 00:13:26,320 S1: fog could be from damaged blood vessels. That's a new study. People. 254 00:13:26,320 --> 00:13:28,600 S1: Young people are a lot less happy. And we actually 255 00:13:28,600 --> 00:13:30,610 S1: because of this, we fell out of the top 20 256 00:13:30,640 --> 00:13:34,030 S1: happiest nations in the United States. Uh, Alzheimer's might stem 257 00:13:34,030 --> 00:13:37,570 S1: from a fat buildup in brain cells. UC Berkeley professor 258 00:13:37,570 --> 00:13:40,180 S1: getting attacked for telling a student to get out of 259 00:13:40,179 --> 00:13:43,870 S1: artillery range of San Francisco and San Jose if he 260 00:13:43,870 --> 00:13:46,210 S1: wants to find a girlfriend, and he's just getting blasted 261 00:13:46,210 --> 00:13:49,270 S1: for this weather forecast. Massively improved. This is a pretty 262 00:13:49,270 --> 00:13:53,380 S1: cool seven day forecasts are now actually pretty accurate, and 263 00:13:53,380 --> 00:13:56,829 S1: four day predictions now are as accurate as one day 264 00:13:56,830 --> 00:14:00,790 S1: predictions from 30 years ago. Interesting. Yeah, four days versus 265 00:14:00,790 --> 00:14:03,850 S1: one day. You imagine the level of complexity there. Germany 266 00:14:03,850 --> 00:14:08,980 S1: legalized recreational marijuana. Blu ray is coming back because you 267 00:14:08,980 --> 00:14:10,720 S1: always have access to it. You don't have to worry 268 00:14:10,720 --> 00:14:13,059 S1: about it being pulled out of streaming. And it's still 269 00:14:13,059 --> 00:14:17,560 S1: the highest quality. Cancer rates under 50 have surged by 80%. 270 00:14:17,559 --> 00:14:19,600 S1: I think this might be in the US, but from 271 00:14:19,600 --> 00:14:23,620 S1: 1990 to 2019. So that's what 30 years, basically 31 272 00:14:23,620 --> 00:14:27,370 S1: years married people are thriving way more than unmarried. According 273 00:14:27,400 --> 00:14:31,330 S1: to a ten year long Gallup study. Bidets getting way 274 00:14:31,330 --> 00:14:34,780 S1: more popular. Got some recommendations here for you. And, uh, 275 00:14:34,780 --> 00:14:36,910 S1: a new way of thinking about the economy. I don't 276 00:14:36,910 --> 00:14:39,790 S1: want to read this whole thing. Uh, just. I think 277 00:14:39,790 --> 00:14:42,520 S1: everything is starting to come down to framing for me. 278 00:14:42,520 --> 00:14:46,120 S1: It's like my universal theory. So, uh, the economy might 279 00:14:46,120 --> 00:14:49,630 S1: be a thing where if you believe in optimism, if 280 00:14:49,630 --> 00:14:52,960 S1: you become optimistic about the economy that has, like its 281 00:14:52,960 --> 00:14:55,870 S1: own gravity, it has its own weight, its own power, 282 00:14:55,870 --> 00:14:58,360 S1: because it turns into action, and then the action turns 283 00:14:58,360 --> 00:15:00,850 S1: into reality. So it's like, once again, you have this 284 00:15:00,850 --> 00:15:05,830 S1: link between understanding or framing or viewpoints that translates directly 285 00:15:05,830 --> 00:15:08,530 S1: into what actually happen, which I just find fascinating. I 286 00:15:08,530 --> 00:15:11,619 S1: think we should rethink how we use the term, uh, 287 00:15:11,620 --> 00:15:15,130 S1: the word technical, which I laid out in a tweet thread. 288 00:15:15,130 --> 00:15:19,590 S1: And this mostly affects women who. Often are the only 289 00:15:19,590 --> 00:15:22,560 S1: people at a company who could do multiple things right. 290 00:15:22,590 --> 00:15:26,070 S1: We're talking about being able to, like, navigate a business process, 291 00:15:26,070 --> 00:15:30,270 S1: get people from one place to another, actually do deeply 292 00:15:30,270 --> 00:15:35,310 S1: technical technology related work, but also deeply technical, like law 293 00:15:35,310 --> 00:15:39,180 S1: related work or business related work, other things that are 294 00:15:39,180 --> 00:15:43,740 S1: like super high end, very, very difficult stuff. And the 295 00:15:43,740 --> 00:15:46,230 S1: the system would not be able to function. Nobody could 296 00:15:46,230 --> 00:15:49,680 S1: explain things. Nobody could actually get the work done if 297 00:15:49,680 --> 00:15:52,560 S1: it wasn't for these people who are there helping. And 298 00:15:52,560 --> 00:15:56,790 S1: because it's not necessarily oriented around a piece of technology, 299 00:15:57,240 --> 00:16:01,800 S1: they get really diminished. So I think we should redefine that. Yeah, 300 00:16:01,800 --> 00:16:04,590 S1: I'm getting all the different AI devices coming in. I 301 00:16:04,590 --> 00:16:08,040 S1: really just want something that record conversations and transcribe them 302 00:16:08,040 --> 00:16:11,040 S1: and get them into my AI workflows. Going back to 303 00:16:11,040 --> 00:16:13,710 S1: original style of the new summary, which is like a 304 00:16:13,710 --> 00:16:16,230 S1: single sentence which which I did a lot up here, 305 00:16:16,230 --> 00:16:18,420 S1: rereading the Sense of style. That's why I did that 306 00:16:18,420 --> 00:16:21,030 S1: pattern really helps my writing. Every time I read it. 307 00:16:21,030 --> 00:16:22,890 S1: I try to do it every like 2 or 3 years. 308 00:16:22,890 --> 00:16:26,010 S1: Resubscribe to the Twitter API so I could do things 309 00:16:26,010 --> 00:16:29,100 S1: like this. Boom, go down here, Twitter boom. So these 310 00:16:29,100 --> 00:16:32,700 S1: are actually Twitter feeds that are in Feedly because I 311 00:16:32,700 --> 00:16:35,400 S1: have the connection there and the API key put in. 312 00:16:35,400 --> 00:16:37,980 S1: So that's pretty cool to be able to consolidate, bring 313 00:16:37,980 --> 00:16:40,560 S1: everything in. You could also do it with Twitter lists 314 00:16:40,560 --> 00:16:43,560 S1: which I use quite a bit. Yeah. Analyze pros talked 315 00:16:43,560 --> 00:16:46,860 S1: about that one and getting into the three body problem 316 00:16:46,860 --> 00:16:50,550 S1: on Netflix. Really enjoying this. The books were fantastic. Found 317 00:16:50,550 --> 00:16:54,780 S1: this service called Opus Clip. Um, Jason pointed me to it. 318 00:16:54,780 --> 00:16:57,660 S1: It's basically you take a video like this video, it'll 319 00:16:57,660 --> 00:17:00,300 S1: be output and it cuts it up into pieces. And 320 00:17:00,300 --> 00:17:01,710 S1: you're going to see some of these. You're going to 321 00:17:01,710 --> 00:17:03,989 S1: see clips from like this will go out in my 322 00:17:03,990 --> 00:17:06,900 S1: feeds and it'll be like, you know, a 32nd or 323 00:17:06,900 --> 00:17:10,470 S1: 92nd or whatever little clip of a cool little section, 324 00:17:10,470 --> 00:17:12,570 S1: but it does it automatically. Normally you would send this 325 00:17:12,570 --> 00:17:14,850 S1: to like an editor, and it's really difficult because they 326 00:17:14,850 --> 00:17:17,160 S1: have to know the subject matter to know what are 327 00:17:17,160 --> 00:17:19,380 S1: the coolest parts of what you said. Well, this thing, 328 00:17:19,380 --> 00:17:22,680 S1: you don't have to do that because it's actually automatic. 329 00:17:22,680 --> 00:17:26,080 S1: You just upload it extracts like whatever, three or 5 330 00:17:26,080 --> 00:17:29,010 S1: or 10 different clips out of it. Rag Tune Tool 331 00:17:29,010 --> 00:17:34,140 S1: for tuning and Optimizing rag Pipelines by Misbah, said collects, 332 00:17:34,140 --> 00:17:36,750 S1: takes a web page, returns all the URLs, and switch 333 00:17:36,750 --> 00:17:39,750 S1: back to this view. Open dev an open source project 334 00:17:39,750 --> 00:17:43,470 S1: to clone and improve Devon Unreal Engine I mean, look 335 00:17:43,470 --> 00:17:49,619 S1: at this. This is. Come on. This is content from 336 00:17:49,619 --> 00:17:50,430 S1: Unreal Engine. 337 00:17:50,940 --> 00:17:52,230 S2: They're combing the streets. 338 00:17:52,230 --> 00:17:55,230 S1: It looks like sort of video I mean searching helps to. 339 00:17:55,230 --> 00:17:55,710 S3: How real. 340 00:17:57,930 --> 00:18:00,390 S4: If they arrest you too, they will take you to 341 00:18:00,390 --> 00:18:03,149 S4: their headquarters and you will not return. 342 00:18:05,920 --> 00:18:08,109 S5: I'm more concerned. It looks like a six foot cat. 343 00:18:08,650 --> 00:18:13,330 S5: It's got claws that can cut through. Vibranium alloy. Slightly tongue. But. 344 00:18:14,280 --> 00:18:18,389 S6: Let's just go forward. Okay. So let's boom down and 345 00:18:18,390 --> 00:18:21,720 S6: take a closer look at this environment. Yeah, sure. Look 346 00:18:21,720 --> 00:18:25,679 S6: at this. So we're talking about nanites. New adaptive tessellation feature. 347 00:18:25,710 --> 00:18:28,260 S6: Look at this. So while I let you or something. 348 00:18:28,260 --> 00:18:28,650 S1: Yeah. 349 00:18:29,160 --> 00:18:31,320 S6: There you go. For artists to create a series of 350 00:18:31,320 --> 00:18:34,109 S6: layers in the environment and then build up. So one 351 00:18:34,109 --> 00:18:34,560 S6: angle that. 352 00:18:34,560 --> 00:18:36,869 S1: We're approaching this from. This is another. Speaking of. 353 00:18:36,869 --> 00:18:38,910 S6: Set dressing. Let's go check out that. 354 00:18:39,570 --> 00:18:43,500 S1: In terms of like creating content that's completely controllable in 355 00:18:43,500 --> 00:18:47,580 S1: creative but realistic look as a skill. The secret to 356 00:18:47,580 --> 00:18:51,240 S1: a meaningful life I love this so much. A long term, 357 00:18:51,240 --> 00:18:55,620 S1: ambitious vision that pushes you to grow smarter, wealthier, and 358 00:18:55,619 --> 00:19:00,629 S1: mentally stronger. That is very concise. A guide to espouse 359 00:19:00,630 --> 00:19:04,470 S1: and use. Prompt injection and jailbreaking are not the same thing. 360 00:19:04,500 --> 00:19:08,430 S1: By Simon Wilson and Recommendation of the week. Think about 361 00:19:08,430 --> 00:19:12,750 S1: your hedonic baseline. So this is like well I'll just 362 00:19:12,750 --> 00:19:16,230 S1: keep reading. So it's thinking about my life just walking 363 00:19:16,230 --> 00:19:19,260 S1: upstairs or walking to the car. It's a stoic technique 364 00:19:19,260 --> 00:19:23,460 S1: of imagining that you're not able to do these simple things. 365 00:19:23,460 --> 00:19:26,730 S1: And as you start removing them from your life and 366 00:19:26,730 --> 00:19:30,840 S1: imagining yourself without them, you can imagine how unhappy you 367 00:19:30,840 --> 00:19:33,750 S1: would be now, like a Jedi or a monk or 368 00:19:33,750 --> 00:19:36,449 S1: something would say, well, you shouldn't care about that. You 369 00:19:36,450 --> 00:19:38,730 S1: shouldn't care about losing anything. But I like to appreciate 370 00:19:38,730 --> 00:19:40,920 S1: the things that I do have, including the ability to 371 00:19:40,920 --> 00:19:44,550 S1: walk outside in the sun and not not be assaulted 372 00:19:44,550 --> 00:19:47,970 S1: or something to breathe fresh air like all these things. 373 00:19:47,970 --> 00:19:52,169 S1: I'm trying to micro appreciate a thousand different things in 374 00:19:52,170 --> 00:19:55,020 S1: my life, which is a very stoic thing. You imagine 375 00:19:55,020 --> 00:19:57,689 S1: the things that you have, you imagine them gone, and 376 00:19:57,690 --> 00:20:01,350 S1: then you get appreciation when you realize that you still 377 00:20:01,350 --> 00:20:05,220 S1: have them. Right. So relationships, mobility, the ability to think, 378 00:20:05,220 --> 00:20:07,950 S1: being in the Bay area at this crazy moment with 379 00:20:07,950 --> 00:20:13,200 S1: AI and stuff, and it's like actively cultivate appreciation by 380 00:20:13,200 --> 00:20:16,710 S1: lowering my baseline, what is the minimum thing that I 381 00:20:16,710 --> 00:20:19,980 S1: need to be happy? And ideally that would be nothing. 382 00:20:19,980 --> 00:20:22,650 S1: And it's just something I'm thinking about. So think about 383 00:20:22,650 --> 00:20:26,250 S1: what you have and think about those things by subtraction 384 00:20:26,250 --> 00:20:30,000 S1: and the aphorism of the week. Contentment is natural wealth. 385 00:20:30,000 --> 00:20:35,040 S1: Luxury is artificial poverty, contentment is natural wealth and luxury 386 00:20:35,040 --> 00:20:40,050 S1: is artificial poverty. Socrates Unsupervised Learning is produced and edited 387 00:20:40,050 --> 00:20:43,740 S1: by Daniel Missler on a Newman 87 AI microphone using 388 00:20:43,740 --> 00:20:47,730 S1: Hindenburg intro and outro. Music is by zombie with a Y, 389 00:20:48,570 --> 00:20:50,760 S1: and to get the text and links from this episode, 390 00:20:50,760 --> 00:20:53,010 S1: sign up for the newsletter version of the show at 391 00:20:53,010 --> 00:20:57,649 S1: Daniel missler.com/newsletter. We'll see you next time.