1 00:00:22,363 --> 00:00:25,633 S1: Welcome to Unsupervised Learning, a security, AI and meaning focused 2 00:00:25,633 --> 00:00:28,513 S1: podcast that looks at how best to thrive as humans 3 00:00:28,513 --> 00:00:32,713 S1: in a post AI world. It combines original ideas, analysis, 4 00:00:32,713 --> 00:00:35,953 S1: and mental models to bring not just the news, but 5 00:00:35,953 --> 00:00:43,093 S1: why it matters and how to respond. All right. So 6 00:00:43,093 --> 00:00:46,123 S1: the first thing for this week is I finally turned 7 00:00:46,123 --> 00:00:50,713 S1: my 9000 word I predictions essay into an actual video. 8 00:00:50,713 --> 00:00:53,263 S1: So if you click on this thing all right I'm 9 00:00:53,263 --> 00:00:55,333 S1: going to try to do something crazy right now I'm 10 00:00:55,333 --> 00:00:59,653 S1: going to try. This is the whole video I go through. 11 00:00:59,653 --> 00:01:01,123 S1: I thought it was going to be like 30 minutes. 12 00:01:01,123 --> 00:01:03,283 S1: I'm like, knows what we want because it has all 13 00:01:03,283 --> 00:01:06,013 S1: the prefaces in, it knows our journal, it knows all 14 00:01:06,013 --> 00:01:08,473 S1: the stuff that we plus like two hours of editing, 15 00:01:08,473 --> 00:01:11,323 S1: which I never should have done. But anyway, it is 16 00:01:11,323 --> 00:01:15,793 S1: a full walkthrough of the illustrated essay, and I even 17 00:01:15,793 --> 00:01:20,803 S1: give a lot more sort of expansion narration on top 18 00:01:20,803 --> 00:01:23,173 S1: of the text that's in the essay, so it's quite good, 19 00:01:23,173 --> 00:01:25,093 S1: if I do say so myself. I've had a lot 20 00:01:25,093 --> 00:01:27,763 S1: of people say they enjoyed it, so I would say, 21 00:01:27,763 --> 00:01:30,763 S1: go check that out. It gets me excited to even 22 00:01:30,763 --> 00:01:33,523 S1: think about it. Talk about it. It's like the fun stuff. 23 00:01:33,523 --> 00:01:36,373 S1: Plus I remember like doing the art. It's just super 24 00:01:36,373 --> 00:01:39,253 S1: good time. Okay. Security. Yeah. But so go check out 25 00:01:39,253 --> 00:01:43,633 S1: the video. All right. Security. So Israel's top spy, chief 26 00:01:43,633 --> 00:01:48,763 S1: Jose Ariel, accidentally revealed his identity through an Amazon book sale. 27 00:01:48,763 --> 00:01:52,153 S1: So I guess we're just doxing him now. That's rude. Well, 28 00:01:52,153 --> 00:01:54,133 S1: I guess he's not going to be that anymore, because 29 00:01:54,133 --> 00:01:55,903 S1: they're going to be like, you obviously can't do this. 30 00:01:55,903 --> 00:01:57,733 S1: So they probably got rid of him and replaced him 31 00:01:57,733 --> 00:02:01,663 S1: with someone who is still anonymous. So now he's no 32 00:02:01,663 --> 00:02:04,213 S1: longer the spy chief. He's just a guy doing an 33 00:02:04,213 --> 00:02:08,863 S1: Amazon book sale. All right, cvn, Nvda databases are struggling. Um, 34 00:02:08,863 --> 00:02:13,153 S1: it's causing gaps and inaccuracies. So somebody needs to fix that. Uh, 35 00:02:13,153 --> 00:02:16,663 S1: criminals in Montreal are using Apple's AirTags to track and 36 00:02:16,663 --> 00:02:21,523 S1: steal cars. And police are using Apple's AirTags to track criminals. 37 00:02:21,523 --> 00:02:25,033 S1: So a lot of AirTag use on both sides of 38 00:02:25,033 --> 00:02:27,493 S1: the fence there. And this one's super cool. This sponsor 39 00:02:27,493 --> 00:02:29,203 S1: I want to talk about this one because I'm actually 40 00:02:29,203 --> 00:02:33,583 S1: advising for them the first AI SoC analyst that autonomously 41 00:02:33,583 --> 00:02:37,393 S1: investigates alerts. This thing will pull alerts. I've seen this 42 00:02:37,393 --> 00:02:40,903 S1: live pull alerts. And this is a sponsor, by the way. Sponsor. 43 00:02:40,903 --> 00:02:43,633 S1: And I'm an advisor for them. So I'm just excited 44 00:02:43,633 --> 00:02:46,633 S1: about them. It's called drop zone AI. So you take 45 00:02:46,633 --> 00:02:49,093 S1: an alert from anywhere in your stack, some tool that 46 00:02:49,093 --> 00:02:52,363 S1: you have that produces an alert like an endpoint cloud 47 00:02:52,363 --> 00:02:56,863 S1: doesn't matter. And it takes alerts from your environment and 48 00:02:56,863 --> 00:03:01,093 S1: performs autonomous. It basically goes and starts researching exactly like 49 00:03:01,093 --> 00:03:03,643 S1: a human. It's like starts digging in, boom, get some 50 00:03:03,643 --> 00:03:07,633 S1: results from that lookup and that investigation takes that goes 51 00:03:07,633 --> 00:03:10,573 S1: to the next step and just starts doing multiple steps, 52 00:03:10,573 --> 00:03:13,393 S1: just like a regular SoC analyst, just like a human. 53 00:03:13,393 --> 00:03:16,003 S1: And once it gets done, it puts that all together 54 00:03:16,003 --> 00:03:20,803 S1: and generates a decision ready report. So you could basically 55 00:03:20,803 --> 00:03:24,223 S1: use that. And it's basically exactly as if you had 56 00:03:24,223 --> 00:03:26,293 S1: got that from a SoC analyst. And you can make 57 00:03:26,293 --> 00:03:29,863 S1: a decision based on that. And it's no playbooks, no code, 58 00:03:29,863 --> 00:03:33,523 S1: no prompts required. It just you feed it alerts and 59 00:03:33,523 --> 00:03:36,283 S1: it goes and does investigations and brings back a report. 60 00:03:36,283 --> 00:03:39,823 S1: It's absolutely insane. It's exactly what I've been talking about 61 00:03:39,823 --> 00:03:45,013 S1: as being like how AI is most going to affect security. 62 00:03:45,013 --> 00:03:47,473 S1: And I've been saying for the longest time it's all 63 00:03:47,473 --> 00:03:50,023 S1: going to be agent based. And that's exactly what this is. 64 00:03:50,023 --> 00:03:52,153 S1: And like I said, it's so good. I just became 65 00:03:52,153 --> 00:03:55,273 S1: an advisor and strop sonar AI and you could actually 66 00:03:55,273 --> 00:03:57,493 S1: go and see a demo of it. Working on a 67 00:03:57,493 --> 00:04:02,053 S1: real alert. Panera's week long incident was, in fact, a 68 00:04:02,053 --> 00:04:06,283 S1: ransomware attack. Cece's new High Risk Communities webpage offers a 69 00:04:06,283 --> 00:04:10,183 S1: bunch of guides and volunteer support and discounted tools, thanks 70 00:04:10,183 --> 00:04:13,813 S1: to Defender Fee for sponsoring as well. Israel's military used 71 00:04:13,813 --> 00:04:18,583 S1: an AI named lavender to pinpoint 37,000 potential Hamas targets, 72 00:04:18,583 --> 00:04:22,303 S1: and that's definitely getting people pretty upset about that, because 73 00:04:22,303 --> 00:04:25,423 S1: it's one thing to identify someone and do further investigation. 74 00:04:25,423 --> 00:04:28,453 S1: It's another thing if you're identifying and then like targeting. 75 00:04:28,453 --> 00:04:30,853 S1: And again, I don't know if that's actually happening. I 76 00:04:30,853 --> 00:04:33,943 S1: don't know the degree they're trusting this targeting that's coming 77 00:04:33,943 --> 00:04:37,543 S1: up from lavender. I don't know if they're going directly 78 00:04:37,543 --> 00:04:41,203 S1: to attacks or killing or anything like that. But the 79 00:04:41,203 --> 00:04:43,903 S1: point is it's worth having a conversation about what they're 80 00:04:43,903 --> 00:04:47,383 S1: doing with it. Given the fact that I can be 81 00:04:47,383 --> 00:04:49,933 S1: super fast and super accurate, but it can also have 82 00:04:49,933 --> 00:04:52,303 S1: a lot of flaws, and the more serious you're taking 83 00:04:52,303 --> 00:04:55,693 S1: the results, the more important those flaws are, right? Technology. 84 00:04:55,693 --> 00:04:58,303 S1: There's a rumor that Sam Altman and Johnny IV are 85 00:04:58,303 --> 00:05:02,203 S1: building some sort of handheld device through a new secret company. 86 00:05:02,203 --> 00:05:05,053 S1: Who knows if that's a real or not. It sounds 87 00:05:05,053 --> 00:05:07,213 S1: real to me because we know that Johnny IV is 88 00:05:07,213 --> 00:05:10,183 S1: actually working with them. We know he's a design person. 89 00:05:10,183 --> 00:05:12,973 S1: We know we need these types of devices. So it 90 00:05:12,973 --> 00:05:16,663 S1: makes sense to me. OpenAI released improved ways of doing 91 00:05:16,663 --> 00:05:20,023 S1: fine tuning. Uh, new paper shows that adding more agents 92 00:05:20,023 --> 00:05:23,773 S1: to large language models boosts their performance. How could it not? 93 00:05:23,773 --> 00:05:26,623 S1: How could it not? I mean, like like I've been 94 00:05:26,623 --> 00:05:29,743 S1: telling everyone agents is the way, agents is the way 95 00:05:29,743 --> 00:05:31,963 S1: this is all going to move forward. And then each 96 00:05:31,963 --> 00:05:35,983 S1: individual agent when it gets smarter because the AI models improve, 97 00:05:35,983 --> 00:05:38,023 S1: that just makes the whole thing smarter. But the way 98 00:05:38,023 --> 00:05:41,803 S1: to get ultimately smart and actually pass humans is to 99 00:05:41,803 --> 00:05:45,763 S1: have that combination of agents working together. That system is 100 00:05:45,763 --> 00:05:47,863 S1: what's smart. It's the same as our brain is no 101 00:05:47,863 --> 00:05:49,933 S1: different than our brain. Our brain has a whole bunch 102 00:05:49,933 --> 00:05:54,133 S1: of areas, each individually only do kind of one thing, 103 00:05:54,133 --> 00:05:57,733 S1: and they're not very smart by themselves. The whole thing 104 00:05:57,733 --> 00:06:00,283 S1: that's made it smart over all these millions of years 105 00:06:00,283 --> 00:06:03,583 S1: is the fact that they work together and they're sharing information, 106 00:06:03,583 --> 00:06:06,103 S1: and it's like it's a system. The system is what 107 00:06:06,103 --> 00:06:08,773 S1: makes us smart. And it's going to be no different 108 00:06:08,773 --> 00:06:12,523 S1: with agents. Now, eventually you might have an AI that's 109 00:06:12,523 --> 00:06:16,453 S1: just one agent or one model or one agent or whatever. 110 00:06:16,453 --> 00:06:18,943 S1: And it's so smart. It's just smart by itself because 111 00:06:18,943 --> 00:06:21,463 S1: it's different than a human. But the way we're first 112 00:06:21,463 --> 00:06:24,073 S1: going to get to AGI, mark my words, is going 113 00:06:24,073 --> 00:06:27,403 S1: to be through a system of AI components working together. 114 00:06:27,403 --> 00:06:29,323 S1: And I've got a blog on that somewhere. New paper 115 00:06:29,323 --> 00:06:33,133 S1: explores how I might be leading us towards knowledge collapse 116 00:06:33,133 --> 00:06:37,273 S1: by oversimplifying complex information. I like that, but I don't 117 00:06:37,273 --> 00:06:40,933 S1: think the complex information goes away. Just because somebody is 118 00:06:40,933 --> 00:06:45,973 S1: simplifying or summarizing complex information, it doesn't mean we have 119 00:06:45,973 --> 00:06:50,383 S1: to stop feeding the AI the original raw form. One 120 00:06:50,383 --> 00:06:52,963 S1: doesn't have to supplant the other, and we could just 121 00:06:52,963 --> 00:06:55,033 S1: ask for the long form. We can ask for the 122 00:06:55,033 --> 00:06:57,073 S1: short form. I don't see that as a trade off 123 00:06:57,103 --> 00:06:59,953 S1: us is trying to get South Korea to stop exporting 124 00:06:59,953 --> 00:07:05,113 S1: chipmaking tools to China, US testing, energy storage and heated sand. 125 00:07:05,113 --> 00:07:08,143 S1: This tech is like trippy to me. Energy storage and 126 00:07:08,143 --> 00:07:14,953 S1: heated sand 135MW of power output for five days straight 127 00:07:14,953 --> 00:07:18,583 S1: and Aura's rolling out symptom radar. They're not calling it 128 00:07:18,583 --> 00:07:21,613 S1: illness detection for obvious reasons. I haven't looked to my 129 00:07:21,613 --> 00:07:23,983 S1: app to see if I have that yet. I do 130 00:07:23,983 --> 00:07:27,433 S1: have an aura though. Okay, Amazon is ditching its Cashierless 131 00:07:27,433 --> 00:07:30,763 S1: walkout technology is kind of sad. I guess it's too early, 132 00:07:30,763 --> 00:07:33,523 S1: but they're switching to something a little more normal, like 133 00:07:33,523 --> 00:07:36,673 S1: a hybrid humans. New studies are showing the wealthy are 134 00:07:36,673 --> 00:07:39,793 S1: starting to have more kids than the poor. Again, need 135 00:07:39,793 --> 00:07:43,093 S1: lots of studies to show that's actually true, but interesting trend. 136 00:07:43,093 --> 00:07:45,733 S1: NASA is doing live streaming the eclipse. I watched it, 137 00:07:45,733 --> 00:07:48,913 S1: it was super cool. It was actually moving through multiple 138 00:07:48,913 --> 00:07:52,273 S1: cities and you would see each person, each city as 139 00:07:52,273 --> 00:07:55,313 S1: it was moving. It would go to the full totality 140 00:07:55,313 --> 00:07:57,683 S1: and everyone would freak out. It was quite cool. It 141 00:07:57,683 --> 00:08:00,713 S1: was a very cool stream. TSMC did not take much 142 00:08:00,713 --> 00:08:03,173 S1: damage at all. They were actually only down for like 143 00:08:03,173 --> 00:08:06,293 S1: a day and they went back to start full production. 144 00:08:06,293 --> 00:08:09,743 S1: And it's because their buildings have some really cool stabilization 145 00:08:09,743 --> 00:08:12,533 S1: tech that allows them to not get too messed up 146 00:08:12,533 --> 00:08:15,713 S1: from earthquakes. However, the earthquake was on the other side 147 00:08:15,713 --> 00:08:18,593 S1: of the island, so who knows if it's a bigger 148 00:08:18,593 --> 00:08:20,873 S1: like it was an eight and it was nearby. I'm 149 00:08:20,873 --> 00:08:23,393 S1: not sure the buildings would be able to handle that, 150 00:08:23,393 --> 00:08:27,113 S1: but who knows. Israeli military dismissed two senior officers and 151 00:08:27,113 --> 00:08:31,163 S1: reprimanded three others for an airstrike that mistakenly killed World 152 00:08:31,163 --> 00:08:36,023 S1: Central Kitchen volunteers in Gaza. The UK's exporting workers to 153 00:08:36,023 --> 00:08:40,673 S1: fill higher paying US jobs and US venture capital investments 154 00:08:40,673 --> 00:08:45,593 S1: went down $36.6 billion in 2024. Had a really cool 155 00:08:45,593 --> 00:08:48,893 S1: conversation with Mike private from Return on Security about these 156 00:08:48,893 --> 00:08:53,303 S1: overall economic trends, especially around cybersecurity, and that will go 157 00:08:53,303 --> 00:08:57,473 S1: up soon. McKinsey is offering UK employee UK employees nine 158 00:08:57,473 --> 00:09:00,893 S1: months of pay to voluntarily leave McKinsey. Gen Z is 159 00:09:00,893 --> 00:09:03,983 S1: going for trades like welding and plumbing over college and 160 00:09:03,983 --> 00:09:07,313 S1: student debt, and home insurers are now using aerial images 161 00:09:07,313 --> 00:09:11,843 S1: from satellites to decide who gets dropped from coverage. I 162 00:09:11,843 --> 00:09:15,863 S1: assume that's because maybe somebody's building something in their backyard, 163 00:09:15,863 --> 00:09:18,473 S1: or they have too many cars. Like, I'm not sure 164 00:09:18,473 --> 00:09:20,423 S1: what would be on the policy that they'd be able 165 00:09:20,423 --> 00:09:22,913 S1: to see in a satellite photo, maybe add ons to 166 00:09:22,913 --> 00:09:25,343 S1: the house or something, I don't know. And all right, 167 00:09:25,343 --> 00:09:29,183 S1: got a couple cool ideas here. Another view of imposter syndrome. 168 00:09:29,183 --> 00:09:33,053 S1: So somebody said outwork your imposter syndrome. That was the 169 00:09:33,053 --> 00:09:36,653 S1: post outwork your imposter syndrome. And I'm, like, working harder 170 00:09:36,653 --> 00:09:39,533 S1: isn't the solution. In my opinion, the solution is to 171 00:09:39,533 --> 00:09:43,403 S1: work on bigger problems that are super important to solve. 172 00:09:43,403 --> 00:09:47,213 S1: That way your focus isn't internal, it's actually external. And 173 00:09:47,213 --> 00:09:49,193 S1: I actually go into this, so I'm going to go 174 00:09:49,193 --> 00:09:52,163 S1: ahead and open this up. So I say framed this way, 175 00:09:52,163 --> 00:09:55,253 S1: imposter syndrome is ultimately a problem of thinking too much 176 00:09:55,253 --> 00:09:58,943 S1: internally versus externally. Because you're thinking how do I compare 177 00:09:58,943 --> 00:10:01,403 S1: to them? What do they think of me? Right. And 178 00:10:01,403 --> 00:10:08,013 S1: so the solution I think is to. Focus on. Something 179 00:10:08,013 --> 00:10:11,823 S1: outside of you, right? Stop putting the focus on yourself. 180 00:10:12,533 --> 00:10:14,783 S1: How do I compare? What do they think of me? 181 00:10:15,263 --> 00:10:19,103 S1: That's thinking about yourself. Instead, focus on the problem. And 182 00:10:19,103 --> 00:10:22,613 S1: now you won't really have imposter syndrome because you're thinking about, like, 183 00:10:22,613 --> 00:10:25,373 S1: what am I doing that's useful to help me with 184 00:10:25,373 --> 00:10:27,893 S1: this really big problem, which has nothing to do with me. 185 00:10:27,893 --> 00:10:31,763 S1: It's about me working on that thing. So instead, focus 186 00:10:31,763 --> 00:10:33,893 S1: on the problem and how to fix it. And this 187 00:10:33,893 --> 00:10:36,413 S1: also works for happiness. It doesn't come from focusing on self, 188 00:10:36,413 --> 00:10:39,053 S1: it comes from focusing on other. So that was that 189 00:10:39,053 --> 00:10:42,623 S1: piece there. And this one's really interesting. It's the first 190 00:10:42,623 --> 00:10:45,053 S1: time I've ever seen Tyler Cohen be wrong. It's like 191 00:10:45,053 --> 00:10:47,303 S1: one of the smartest people that I know of on 192 00:10:47,303 --> 00:10:52,103 S1: the planet. And he had Jonathan Height on and Jonathan 193 00:10:52,103 --> 00:10:55,733 S1: High was talking about how bad social media is for 194 00:10:55,733 --> 00:11:00,053 S1: young girls and how I could potentially make that worse 195 00:11:00,053 --> 00:11:02,933 S1: and everything. And Tyler Cohen is like, don't worry about it. 196 00:11:02,933 --> 00:11:05,513 S1: AI is going to solve this because I is going 197 00:11:05,513 --> 00:11:10,373 S1: to summarize everything, and the summarization of of social media 198 00:11:10,373 --> 00:11:13,073 S1: is going to solve this problem and fix it. And 199 00:11:13,163 --> 00:11:16,613 S1: I'm like, how are you not seeing this now? Jonathan 200 00:11:16,613 --> 00:11:18,833 S1: didn't have this point, but I'm going to make this 201 00:11:18,833 --> 00:11:22,403 S1: point right now. That is how Tyler Cohen is going 202 00:11:22,403 --> 00:11:25,883 S1: to use AI, right? It's how I'm using AI right now, 203 00:11:25,883 --> 00:11:28,133 S1: and it's how, you know, Jonathan's probably going to use 204 00:11:28,133 --> 00:11:31,253 S1: it as well. Summarization of content. So you're just getting 205 00:11:31,253 --> 00:11:33,803 S1: the thing. And that might actually take you away from 206 00:11:33,803 --> 00:11:36,353 S1: being caught up in like, oh he said she said 207 00:11:36,353 --> 00:11:39,593 S1: and all the drama or toxicity or whatever. Well that's fine. 208 00:11:39,593 --> 00:11:41,843 S1: And that again, that's how they're going to use it. 209 00:11:41,843 --> 00:11:44,453 S1: That's how a whole bunch of intellectual people who have 210 00:11:44,453 --> 00:11:47,063 S1: made it in life and are older are probably going 211 00:11:47,063 --> 00:11:50,513 S1: to use it, but not for young people consuming viral 212 00:11:50,513 --> 00:11:53,603 S1: and toxic content, because for them, the content itself is 213 00:11:53,603 --> 00:11:57,653 S1: the point, not the summary. And here's my example of this. 214 00:11:57,653 --> 00:12:01,223 S1: Does Tyler think I will send people who love stand 215 00:12:01,223 --> 00:12:04,463 S1: up comedy a summary of the jokes made in a 216 00:12:04,463 --> 00:12:08,243 S1: given standup as a substitute for going to actual comedy shows. 217 00:12:08,243 --> 00:12:10,973 S1: And I've got I've got an I summary here. Here's 218 00:12:10,973 --> 00:12:13,793 S1: your summary of this standup. There were three jokes on 219 00:12:13,793 --> 00:12:17,633 S1: women in stereotypes, four jokes on how clumsy the comedian 220 00:12:17,633 --> 00:12:22,073 S1: is to playful racist jokes. Two hecklers were addressed. Applause 221 00:12:22,073 --> 00:12:25,073 S1: was three out of five compared to other performers. We 222 00:12:25,073 --> 00:12:29,123 S1: hope you've enjoyed this hilarious I summary from comics I. 223 00:12:29,153 --> 00:12:32,513 S1: Does that work for comedy? No, that doesn't work for comedy. 224 00:12:32,513 --> 00:12:35,393 S1: And it won't work for young kids consuming viral or 225 00:12:35,393 --> 00:12:40,043 S1: toxic content unless you have some sort of like draconic like, 226 00:12:40,043 --> 00:12:44,633 S1: I guess, massive control where they weren't actually allowed to 227 00:12:44,633 --> 00:12:47,723 S1: go to the real thing. They couldn't go to TikTok, 228 00:12:47,723 --> 00:12:50,303 S1: they can't go to Instagram, and all they get is 229 00:12:50,303 --> 00:12:53,183 S1: the stupid summary. Nobody would. They they wouldn't even care 230 00:12:53,183 --> 00:12:55,193 S1: about the summary. I mean, that's not even going to 231 00:12:55,193 --> 00:12:59,003 S1: be a thing. Nobody cares. And if it did, they 232 00:12:59,003 --> 00:13:01,283 S1: wouldn't use it. And if they had the option or 233 00:13:01,283 --> 00:13:04,133 S1: if they had both, they wouldn't choose the summaries. They 234 00:13:04,133 --> 00:13:05,843 S1: would not use the summaries at all, and they would 235 00:13:05,843 --> 00:13:08,453 S1: just go to the original thing. Now that being said, 236 00:13:08,723 --> 00:13:11,543 S1: it is Tyler Cohen. So there's a chance that I 237 00:13:11,543 --> 00:13:14,573 S1: didn't understand what he was saying. And also Jonathan definitely 238 00:13:14,573 --> 00:13:17,783 S1: didn't either. So he could be misunderstanding Tyler's point. So 239 00:13:17,783 --> 00:13:20,513 S1: I want to offer him that. There's also the other thing, 240 00:13:20,513 --> 00:13:24,023 S1: which is you might just be right, and I'm just wrong. Now, 241 00:13:24,023 --> 00:13:26,873 S1: I wouldn't normally say that because I think fairly highly 242 00:13:26,873 --> 00:13:30,713 S1: of my thinking capabilities, but it's Tyler Cohen, so I'm 243 00:13:30,713 --> 00:13:33,713 S1: leaving that window a little more open than usual. All right. 244 00:13:33,713 --> 00:13:37,163 S1: Deep faked content summaries. Oh, yeah. This is crazy. So 245 00:13:37,163 --> 00:13:38,423 S1: I don't know. I don't know if I woke up 246 00:13:38,423 --> 00:13:41,333 S1: with this idea. I think the main interface that we're 247 00:13:41,333 --> 00:13:45,293 S1: about to have for content, let's say, okay, I got 248 00:13:45,293 --> 00:13:49,343 S1: buddies who make content. John Hammond makes content. Jason Haddix 249 00:13:49,343 --> 00:13:53,183 S1: makes content. Clint Gibbler makes content. So they're putting out, 250 00:13:53,183 --> 00:13:56,033 S1: let's say they're putting out videos, they're putting out text, 251 00:13:56,033 --> 00:13:58,613 S1: they're doing live talks. They're doing all these different things, 252 00:13:58,613 --> 00:14:03,383 S1: different mediums, different formats. I think what's going to be 253 00:14:03,383 --> 00:14:07,433 S1: happening very soon is let's do the R version, even 254 00:14:07,433 --> 00:14:09,923 S1: though we don't have R yet. The earlier version will 255 00:14:09,923 --> 00:14:12,353 S1: just be little videos on your phone. But let's do 256 00:14:12,353 --> 00:14:15,443 S1: the R version or the R version combined with a 257 00:14:15,443 --> 00:14:18,293 S1: digital assistant, which is AI in your head or in 258 00:14:18,293 --> 00:14:21,143 S1: your mobile device. So it's essentially you say, hey, look, 259 00:14:21,143 --> 00:14:23,873 S1: what is Jason been up to? What is John been 260 00:14:23,873 --> 00:14:26,273 S1: up to? What is Clint been up to? And it 261 00:14:26,273 --> 00:14:29,573 S1: will actually deepfake them okay. Because it knows how much 262 00:14:29,573 --> 00:14:32,453 S1: time I have. Let's say I want a two minute summary. 263 00:14:32,453 --> 00:14:36,533 S1: It will take the long presentation that Jason is going 264 00:14:36,533 --> 00:14:38,453 S1: to do. Jason is getting ready to do an AI 265 00:14:38,453 --> 00:14:41,633 S1: talk called Red purple, blue AI. Let's say it's actually 266 00:14:41,633 --> 00:14:43,343 S1: it is like it's a class. Oh, by the way, 267 00:14:43,343 --> 00:14:45,413 S1: you should go sign up for this class. His class 268 00:14:45,413 --> 00:14:48,533 S1: is called red purple blue I it's going to be 269 00:14:48,533 --> 00:14:50,513 S1: an amazing class. I've heard a lot about it. I'm 270 00:14:50,513 --> 00:14:52,853 S1: going to be in it as well. But for example, 271 00:14:52,853 --> 00:14:55,403 S1: let's say he gives a talk about that a public talk. 272 00:14:55,403 --> 00:14:57,713 S1: The other one's not public. But let's say he gives 273 00:14:57,713 --> 00:15:00,263 S1: a talk about that. Actually he's doing a space con 274 00:15:00,263 --> 00:15:02,363 S1: coming up soon. So that's a good example. Let's say 275 00:15:02,363 --> 00:15:04,553 S1: I don't have the one hour or the two hours 276 00:15:04,553 --> 00:15:06,173 S1: to watch that. I only have two minutes. I'm about 277 00:15:06,173 --> 00:15:08,153 S1: to get on a train where I don't have any connection. 278 00:15:08,153 --> 00:15:11,933 S1: Whatever it's going to make, Jason and Jason is going to. 279 00:15:12,073 --> 00:15:14,503 S1: Teach me what he covered, but he's going to do 280 00:15:14,503 --> 00:15:17,233 S1: it in 30s or he's going to do it in 60s. 281 00:15:17,233 --> 00:15:20,533 S1: It's going to be a deepfake of Jason doing Jason's 282 00:15:20,533 --> 00:15:23,383 S1: own content. Same for John. John Hammond has a thing 283 00:15:23,383 --> 00:15:25,873 S1: about this new piece of malware. Clint has this new 284 00:15:25,873 --> 00:15:28,243 S1: thing about this new tool that he made or that 285 00:15:28,243 --> 00:15:31,573 S1: he saw, and he's talking about deepfakes are going to 286 00:15:31,573 --> 00:15:36,523 S1: allow the actual creator to scale their delivery of their 287 00:15:36,523 --> 00:15:41,263 S1: content as a video to any size chunk that the 288 00:15:41,263 --> 00:15:43,933 S1: consumer wants to see. Give me a ten second summary. 289 00:15:43,933 --> 00:15:47,053 S1: Give me a 32nd summary. Give me a one minute summary, 290 00:15:47,053 --> 00:15:49,873 S1: a two minute summary, a ten minute summary, whatever. But 291 00:15:49,873 --> 00:15:53,683 S1: it will dynamically write the content, which is a summary 292 00:15:53,683 --> 00:15:57,133 S1: of the content, which smashes it down to that size, 293 00:15:57,133 --> 00:16:01,633 S1: and then it will perfectly deepfake that creator and the 294 00:16:01,633 --> 00:16:05,653 S1: mouth will match, the mouth will perfectly match. It'll look 295 00:16:05,653 --> 00:16:10,933 S1: exactly or almost exactly and eventually exactly like the creator. Now, 296 00:16:10,933 --> 00:16:13,003 S1: why would the creator want to do that? Because it's 297 00:16:13,003 --> 00:16:15,493 S1: still their content, okay? It's still their content. It's not 298 00:16:15,493 --> 00:16:18,043 S1: the original raw form, but in a lot of ways 299 00:16:18,043 --> 00:16:19,603 S1: it's going to be better because it's not going to 300 00:16:19,603 --> 00:16:22,213 S1: have ums and ors. It's going to be very crisp 301 00:16:22,213 --> 00:16:26,413 S1: and concise, and most importantly, it won't be some third party, right? 302 00:16:26,413 --> 00:16:30,073 S1: Because my digital assistant can also render it using their face. 303 00:16:30,073 --> 00:16:32,713 S1: But it'll be really cool for the for the actual 304 00:16:32,713 --> 00:16:36,523 S1: creator to be still the one that's delivering it. And 305 00:16:36,523 --> 00:16:39,643 S1: what's cool about that? So here's what's really cool about that. 306 00:16:39,643 --> 00:16:43,543 S1: It'll do multiple languages. So now it's still Clint. It's 307 00:16:43,543 --> 00:16:46,843 S1: still John. It's still Jason. Jason is still saying the 308 00:16:46,843 --> 00:16:47,623 S1: thing to me. 309 00:16:47,623 --> 00:16:50,743 S2: Pedro diciendo la content. 310 00:16:50,743 --> 00:16:51,073 S1: In. 311 00:16:51,073 --> 00:16:53,263 S2: Espanol. Entonces puedo escuchar. 312 00:16:53,263 --> 00:16:56,413 S1: En espanol and I won't know the difference. It'll look 313 00:16:56,413 --> 00:17:00,073 S1: exactly as if Jason is speaking in Spanish instead of English. 314 00:17:00,073 --> 00:17:02,983 S1: And I think that is wonderful. Same with Chinese, same 315 00:17:02,983 --> 00:17:06,493 S1: with every language. Not any more difficult. So now your 316 00:17:06,493 --> 00:17:10,663 S1: deepfake content is going out to every language all at once, 317 00:17:10,663 --> 00:17:12,703 S1: as soon as you drop a piece of content. So 318 00:17:12,703 --> 00:17:15,763 S1: watch this. You drop a piece of content. It's one 319 00:17:15,763 --> 00:17:18,973 S1: hour long. Okay, perfect example. I just released a video. 320 00:17:18,973 --> 00:17:21,643 S1: It's one hour and ten minutes long. I only did 321 00:17:21,643 --> 00:17:23,713 S1: it in English because I'm only capable of doing it 322 00:17:23,713 --> 00:17:26,053 S1: in English. I could do it partially in Spanish, but 323 00:17:26,053 --> 00:17:29,803 S1: it would be kind of crappy. So that one hour 324 00:17:29,803 --> 00:17:32,773 S1: and ten minute piece of content can now be put 325 00:17:32,773 --> 00:17:38,593 S1: in Hindi, Spanish, Mandarin, Cantonese, Filipino, like all these different languages, 326 00:17:38,593 --> 00:17:42,553 S1: but not just converted one for one to the language. 327 00:17:42,553 --> 00:17:46,153 S1: Also converted to a ten second version, a 32nd version, 328 00:17:46,153 --> 00:17:49,093 S1: a one minute version. And it's still me. It's still 329 00:17:49,093 --> 00:17:52,483 S1: my face, it's still my whatever gestures, all this, it's 330 00:17:52,483 --> 00:17:56,803 S1: still me delivering the content. That is a game changer. 331 00:17:56,803 --> 00:18:01,393 S1: An absolute game changer because it's scalable content delivery in 332 00:18:01,393 --> 00:18:04,483 S1: every language, at every bite sized chunk. So now when 333 00:18:04,483 --> 00:18:07,543 S1: I'm on a train or I'm exercising, I could have 334 00:18:07,543 --> 00:18:10,603 S1: my AR interface. I see the person is delivering the 335 00:18:10,603 --> 00:18:13,543 S1: content and it's just like, I'm watching that now. Maybe 336 00:18:13,543 --> 00:18:15,313 S1: there will be some sort of tag that says, there's 337 00:18:15,313 --> 00:18:18,253 S1: this roar, or is this an AI deepfake of it? 338 00:18:18,253 --> 00:18:22,183 S1: But it's not even quite a deepfake. It's using deepfake technology. 339 00:18:22,183 --> 00:18:25,453 S1: But a deepfake is like almost unauthorized. This will be 340 00:18:25,453 --> 00:18:28,273 S1: completely authorized. Um, and of course, there will be versions 341 00:18:28,273 --> 00:18:30,553 S1: of this that are not authorized, and that will be 342 00:18:30,553 --> 00:18:33,253 S1: an actual deepfake. But anyway, you get the idea. This 343 00:18:33,253 --> 00:18:37,063 S1: is going to massively change how we consume content, because 344 00:18:37,063 --> 00:18:40,513 S1: we want that content in different forms, different languages, but 345 00:18:40,513 --> 00:18:42,823 S1: we still want to see the creator doing it. All right. 346 00:18:42,853 --> 00:18:45,613 S1: AI and music. AI is not going to ruin music. 347 00:18:45,913 --> 00:18:49,243 S1: Have we forgot about pop? Pop is, you know, a 348 00:18:49,243 --> 00:18:51,673 S1: few chords and a hook. It's, you know, a lot 349 00:18:51,673 --> 00:18:54,823 S1: of people would say it's low quality music, like low 350 00:18:54,823 --> 00:18:58,423 S1: quality food. But it's the same with doing customer service calls, 351 00:18:58,423 --> 00:19:02,413 S1: sales calls. We forget how low the bar is for 352 00:19:02,413 --> 00:19:05,263 S1: being better than an average human, and that's why I 353 00:19:05,263 --> 00:19:07,873 S1: music is going to be pretty good. And actually I 354 00:19:07,873 --> 00:19:11,293 S1: think I have a link to one here anyway. Discovery 355 00:19:11,293 --> 00:19:14,863 S1: section Luke Stefan's hack Luke put out an amazing blog 356 00:19:14,863 --> 00:19:18,583 S1: talking about his evolution of bug bounty automation. Talks about 357 00:19:18,583 --> 00:19:21,313 S1: going from Bash to Python to Golang, and how he 358 00:19:21,313 --> 00:19:25,843 S1: eventually ended up at Cloud Native. Really great piece, Thomas wrote. 359 00:19:26,113 --> 00:19:29,113 S1: He's super awesome and he wrote a piece about applying 360 00:19:29,113 --> 00:19:32,083 S1: LMS to Threat Intelligence. Really cool. It's got a Jupyter 361 00:19:32,083 --> 00:19:36,043 S1: notebook here for the code as well. SWE agent autonomously 362 00:19:36,043 --> 00:19:39,793 S1: fixes bugs in GitHub repos BR is a Python framework. 363 00:19:39,793 --> 00:19:44,923 S1: Simplifies building an AI apps using building blocks. Open source 364 00:19:44,923 --> 00:19:49,723 S1: textbook makes the art of mathematics accessible. Jim Graham turns 365 00:19:49,723 --> 00:19:53,533 S1: threat modeling into a self-hosted web app. This one is super, 366 00:19:53,533 --> 00:19:56,893 S1: super cool. Chelsea now lets you tweak images so you 367 00:19:56,893 --> 00:19:59,653 S1: can basically have an image that was created with Dall-E 368 00:19:59,653 --> 00:20:01,723 S1: and you could just like rub a like the face 369 00:20:01,723 --> 00:20:03,973 S1: or whatever and say, redo that part and it will 370 00:20:03,973 --> 00:20:06,823 S1: redo it, but integrate it with everything around it. So 371 00:20:06,823 --> 00:20:10,033 S1: that is something that Midjourney had that that Dall-E didn't, 372 00:20:10,033 --> 00:20:13,983 S1: and now Dall-E has it as well. Claude's API now 373 00:20:13,983 --> 00:20:16,893 S1: has a new tools feature that allows you to, like, 374 00:20:16,893 --> 00:20:18,933 S1: browse the web and do all kinds of stuff. I 375 00:20:18,933 --> 00:20:22,353 S1: kind of feel like agent functionality is going to blend 376 00:20:22,353 --> 00:20:25,713 S1: right into the models, so I'm not quite sure how 377 00:20:25,713 --> 00:20:30,933 S1: long we're going to have, like these elaborate agent frameworks, 378 00:20:30,933 --> 00:20:35,283 S1: because that might just be part of using AI. You 379 00:20:35,283 --> 00:20:37,833 S1: might just say exactly what you want to do in 380 00:20:37,833 --> 00:20:40,773 S1: the prompt, and that will actually be the agent framework. 381 00:20:40,773 --> 00:20:43,473 S1: It'll actually spin up how many agents are needed to 382 00:20:43,473 --> 00:20:46,113 S1: do that. And maybe you have some parameters right there 383 00:20:46,113 --> 00:20:47,973 S1: in the prompt that does it. But I kind of 384 00:20:47,973 --> 00:20:50,613 S1: feel like that's going to just blend right into the 385 00:20:50,613 --> 00:20:54,003 S1: language of the of the models themselves. And kids are 386 00:20:54,003 --> 00:20:57,273 S1: learning math from deep fakes. Okay. This is another example 387 00:20:57,273 --> 00:20:59,943 S1: of this learning math from deep fakes of Taylor Swift 388 00:20:59,943 --> 00:21:03,093 S1: and Drake on TikTok. And it's going well. I'm excited 389 00:21:03,093 --> 00:21:06,273 S1: for this. I am super excited for this look. There 390 00:21:06,273 --> 00:21:09,273 S1: are negatives that are going to come from defects. Everyone 391 00:21:09,273 --> 00:21:12,273 S1: knows that. Let's find the positives. The positives are if 392 00:21:12,273 --> 00:21:15,963 S1: people love Taylor Swift, let's learn math. Let's learn calculus 393 00:21:15,963 --> 00:21:18,903 S1: from Taylor Swift. I would do that. All right. Recommendation 394 00:21:18,903 --> 00:21:22,533 S1: of the week. Check out Mozart on the bass. This 395 00:21:22,533 --> 00:21:25,413 S1: is an eye track. Yeah. Go listen this tell me 396 00:21:25,413 --> 00:21:28,113 S1: that this won't be popular. It's a little bit EDM ish. 397 00:21:28,113 --> 00:21:29,973 S1: So if you don't like that, just compare it to 398 00:21:29,973 --> 00:21:32,553 S1: other EDM that you don't like, and you'll see that 399 00:21:32,553 --> 00:21:35,613 S1: I is actually quite good at making stuff you don't like. 400 00:21:35,613 --> 00:21:38,763 S1: All right, aphorism of the week don't explain your philosophy. 401 00:21:38,763 --> 00:21:45,153 S1: Embody it. Don't explain your philosophy, embody it. Epictetus. Unsupervised 402 00:21:45,153 --> 00:21:47,643 S1: learning is produced and edited by Daniel Missler on a 403 00:21:47,643 --> 00:21:52,533 S1: Neumann 87 AI microphone using Hindenburg. Intro and outro. Music 404 00:21:52,533 --> 00:21:55,623 S1: is by zombie with the Y, and to get the 405 00:21:55,623 --> 00:21:57,693 S1: text and links from this episode, sign up for the 406 00:21:57,693 --> 00:22:03,343 S1: newsletter version of the show at Daniel missler.com/newsletter. We'll see 407 00:22:03,343 --> 00:22:03,943 S1: you next time.