1 00:00:00,880 --> 00:00:05,040 S1: Unsupervised Learning is a podcast about trends and ideas in cybersecurity, 2 00:00:05,080 --> 00:00:09,960 S1: national security, AI, technology and society, and how best to 3 00:00:10,000 --> 00:00:18,000 S1: upgrade ourselves to be ready for what's coming. All right, 4 00:00:18,000 --> 00:00:21,919 S1: starting off this week. So I completely reset my email 5 00:00:21,920 --> 00:00:26,280 S1: labels and filters this week. So cruft going back to like, 6 00:00:26,480 --> 00:00:29,000 S1: I don't know, 20 years old or whatever from Gmail. 7 00:00:29,440 --> 00:00:32,120 S1: Basically all those filters got rid of them because I'm 8 00:00:32,120 --> 00:00:34,680 S1: going to start filtering with AI. I think at some 9 00:00:34,680 --> 00:00:37,400 S1: point instead, I'm already doing a little bit of that 10 00:00:37,400 --> 00:00:42,239 S1: with superhuman. I just I'm worried I'm missing things, and 11 00:00:42,240 --> 00:00:45,240 S1: I've already found a few things, like different bills for 12 00:00:45,240 --> 00:00:48,240 S1: things I thought I had turned off or whatever. Um, 13 00:00:48,240 --> 00:00:51,360 S1: but anyway, very clean feeling. I mean, that was probably 14 00:00:51,760 --> 00:00:55,400 S1: a couple of hundred filters that were in there and 15 00:00:55,440 --> 00:01:00,160 S1: like weird labels and subfolders. So I basically consolidated folders, 16 00:01:00,410 --> 00:01:05,050 S1: Labels rules, and I only have like less than a 17 00:01:05,050 --> 00:01:09,410 S1: dozen now of each. So very nice, clean feeling. Highly recommended. 18 00:01:10,209 --> 00:01:13,289 S1: Going to be trying out Karpathy's idea. This is Andrej 19 00:01:13,330 --> 00:01:16,929 S1: Karpathy has an idea of using a single Apple note 20 00:01:16,930 --> 00:01:20,090 S1: instead of actually using, uh, you know, a million. I 21 00:01:20,130 --> 00:01:24,130 S1: have currently around 2900, and it's not really a nice 22 00:01:24,130 --> 00:01:28,450 S1: situation to be in. So I'm going to try this 23 00:01:28,490 --> 00:01:30,330 S1: single note thing. It's pretty cool. And I got a 24 00:01:30,330 --> 00:01:33,170 S1: link here in the show and, uh, watch your API 25 00:01:33,170 --> 00:01:37,850 S1: key and I'd agents because I've got a $2,000 bill, 26 00:01:37,850 --> 00:01:41,250 S1: which is the limit that I set because I did 27 00:01:41,250 --> 00:01:45,130 S1: this weird web documentation lookup. And some for some reason 28 00:01:45,130 --> 00:01:47,610 S1: it was calling the LLM every time it did it. 29 00:01:48,090 --> 00:01:50,250 S1: And I was watching it run because I thought it 30 00:01:50,250 --> 00:01:53,290 S1: was just a pure HTTP query. I don't know why 31 00:01:53,290 --> 00:01:56,250 S1: I was using an agent to process results each time. 32 00:01:56,250 --> 00:02:00,330 S1: It's very silly. But anyway, um, yeah, that's why you 33 00:02:00,430 --> 00:02:03,030 S1: should have limits set. I'm glad I had mine set, 34 00:02:03,830 --> 00:02:06,510 S1: but now I've broken out multiple keys and lowered some 35 00:02:06,510 --> 00:02:10,550 S1: limits to make it, uh, even more of a backstop. 36 00:02:10,950 --> 00:02:14,870 S1: Go delete your 23 andme data. They are selling out. 37 00:02:15,110 --> 00:02:18,710 S1: They are going bankrupt and selling their data to whoever. 38 00:02:19,350 --> 00:02:21,750 S1: So go delete that if you have it. Uh, new 39 00:02:21,750 --> 00:02:27,950 S1: obscure book recommendation. Fanged noumena. Fanged noumena or fanged noumena. 40 00:02:27,990 --> 00:02:30,830 S1: Maybe my friend Joel Parrish recommended that to me. So 41 00:02:30,830 --> 00:02:33,510 S1: I'm going to read that. And I was made emotionally 42 00:02:33,510 --> 00:02:38,510 S1: leaky last night from this, uh, video of this pianist, um, 43 00:02:38,510 --> 00:02:41,829 S1: on this YouTube channel called Great Measures. And he did 44 00:02:41,830 --> 00:02:45,550 S1: a cover of Fade to Black by Metallica, which is 45 00:02:45,590 --> 00:02:47,910 S1: kind of an emotional song for like, a personal reason 46 00:02:47,910 --> 00:02:50,670 S1: for me, uh, related to, like, a friend in high school, 47 00:02:50,830 --> 00:02:53,550 S1: but I don't know, I don't really listen to the 48 00:02:53,550 --> 00:02:56,510 S1: song that much, and I don't know, musically, it's not 49 00:02:56,510 --> 00:03:00,230 S1: that exciting to me, but it has some emotional meaning 50 00:03:00,400 --> 00:03:02,280 S1: to me. And then so I'm watching him play it 51 00:03:02,280 --> 00:03:05,880 S1: on piano and it is stunning. I mean, it is 52 00:03:05,880 --> 00:03:10,200 S1: absolutely just gorgeous because he's adding like so much more 53 00:03:10,200 --> 00:03:13,200 S1: depth and everything to it. Plus it's on piano and 54 00:03:13,200 --> 00:03:16,240 S1: it's like, yeah, it, uh, it got me. Definitely worth 55 00:03:16,240 --> 00:03:19,080 S1: checking out. And I love the channel. The channel is 56 00:03:19,080 --> 00:03:23,400 S1: called Great Measures. So it's basically, uh, a metal guy 57 00:03:23,440 --> 00:03:28,880 S1: showing his pianist friend, so classically trained musician, showing him 58 00:03:28,880 --> 00:03:31,960 S1: a bunch of metal. It's quite awesome. So I'm joining 59 00:03:31,960 --> 00:03:35,080 S1: Caleb Sima and Edward Wu for a panel on Drop 60 00:03:35,120 --> 00:03:40,400 S1: Zone AI Security Frontiers conference, and that is on March 27th, 61 00:03:40,400 --> 00:03:42,760 S1: which is well, it will have already happened by the 62 00:03:42,760 --> 00:03:44,800 S1: time you hear this. But, um, yeah, we're going to 63 00:03:44,840 --> 00:03:48,600 S1: talk about where JNI stands in security today and where 64 00:03:48,600 --> 00:03:51,600 S1: it's headed. And it's a virtual free and worth it. 65 00:03:51,600 --> 00:03:54,080 S1: So you can go watch the thing even though it's 66 00:03:54,080 --> 00:03:58,680 S1: already recorded. Cybersecurity y white box red teaming makes me 67 00:03:58,680 --> 00:04:03,370 S1: feel weird. So this guy is like strand Nexus, which 68 00:04:03,370 --> 00:04:07,810 S1: are how to pronounce that. But Zygi shares unsettling experiences 69 00:04:07,810 --> 00:04:13,610 S1: with models appearing to express distress during advanced LM safety techniques. 70 00:04:13,610 --> 00:04:16,210 S1: And the quote here is it just doesn't feel good 71 00:04:16,210 --> 00:04:20,210 S1: to be responsible for making models scream. It distracts me 72 00:04:20,210 --> 00:04:23,690 S1: from doing research and makes me write rambling blog posts. 73 00:04:23,690 --> 00:04:27,210 S1: That's uh, yeah, that's a problem. It feels like Black 74 00:04:27,210 --> 00:04:30,010 S1: Mirror to me. Like I'm not worried that it's actually conscious, 75 00:04:30,010 --> 00:04:31,970 S1: but at some point I will be and it'll be 76 00:04:31,970 --> 00:04:34,969 S1: hard to know the difference. And like he said, it's 77 00:04:34,970 --> 00:04:38,450 S1: still disturbing. If the thing is making all the sounds 78 00:04:38,450 --> 00:04:42,130 S1: of it hurting, even though you you think you know 79 00:04:42,170 --> 00:04:44,970 S1: that it doesn't have the capability to do so. Next 80 00:04:44,970 --> 00:04:48,930 S1: one here, right? White House OpSec fail. White House accidentally 81 00:04:48,930 --> 00:04:53,130 S1: revealed top secret Houthi bombing plans to. The editor of 82 00:04:53,130 --> 00:04:56,610 S1: The Atlantic magazine shared the plans in a signal group 83 00:04:56,610 --> 00:05:00,789 S1: that didn't realize the reporter was in there. And the 84 00:05:00,790 --> 00:05:05,070 S1: worst part is the message before they started sharing the stuff. 85 00:05:05,630 --> 00:05:08,790 S1: We are currently clean on OpSec, and I believe that 86 00:05:08,790 --> 00:05:11,630 S1: was like the Secretary of State saying that we are 87 00:05:11,630 --> 00:05:15,110 S1: currently clean on OpSec. And of course, the person also 88 00:05:15,710 --> 00:05:19,110 S1: receiving that message is a civilian and the head of 89 00:05:19,110 --> 00:05:23,390 S1: a magazine. My goodness, my agent, security and companies like Microsoft. 90 00:05:23,390 --> 00:05:25,670 S1: So I got invited to a Microsoft media event last 91 00:05:25,670 --> 00:05:28,150 S1: week in SF where they showed off all the AI 92 00:05:28,310 --> 00:05:31,830 S1: agent stuff and copilot that they're talking about this week. Basically, 93 00:05:31,830 --> 00:05:34,469 S1: they're adding agents to tons of products under the banner 94 00:05:34,470 --> 00:05:37,630 S1: of copilot. And I had a single thought, well, spending 95 00:05:37,630 --> 00:05:40,550 S1: like three hours talking to everyone there from like red 96 00:05:40,589 --> 00:05:44,830 S1: team to threat intelligence, incident response or whatever. But basically 97 00:05:44,830 --> 00:05:47,349 S1: startups need to hurry up, because what I saw in 98 00:05:47,390 --> 00:05:51,029 S1: that room, I think is like the future and definitely 99 00:05:51,029 --> 00:05:54,430 S1: not sponsored by Microsoft. I'm not even like a Microsoft fanboy, 100 00:05:54,589 --> 00:05:58,110 S1: but many of the agents in the room could talk 101 00:05:58,110 --> 00:06:03,680 S1: to all of the other Microsoft services, right? Microsoft vulnerability management. 102 00:06:03,680 --> 00:06:07,280 S1: Identity and access management, asset management. They could talk to 103 00:06:07,320 --> 00:06:10,880 S1: all those different systems. They could pull context from across 104 00:06:10,880 --> 00:06:17,920 S1: the entire organization like HR, like asset management endpoints, uh, 105 00:06:17,920 --> 00:06:23,240 S1: cloud vulnerability data, local vulnerability data on the actual workstations, 106 00:06:23,240 --> 00:06:27,560 S1: vulnerability management ticketing systems. What I'm saying is the companies 107 00:06:27,560 --> 00:06:30,920 S1: that are going to win this AI security game are 108 00:06:30,920 --> 00:06:34,039 S1: not necessarily the ones with the best AI or agent tech, 109 00:06:34,400 --> 00:06:38,800 S1: but the ones that can best leverage customer company context 110 00:06:39,160 --> 00:06:42,640 S1: for their AI or agents. So at first it'll be 111 00:06:42,640 --> 00:06:45,559 S1: startups because they can move the fastest, right? But soon 112 00:06:45,560 --> 00:06:49,280 S1: they're going to have a major disadvantage compared to like 113 00:06:49,320 --> 00:06:54,920 S1: Microsoft and companies that give access, or companies where the 114 00:06:54,920 --> 00:06:59,490 S1: AI has the ability to access unified company context UCC. 115 00:06:59,730 --> 00:07:03,370 S1: So other companies, I think, like Amazon and Databricks and stuff, 116 00:07:03,370 --> 00:07:09,090 S1: will also try to build this UCC unified customer context 117 00:07:09,890 --> 00:07:13,250 S1: so that they'll be able to create that data store 118 00:07:13,250 --> 00:07:15,970 S1: for people to use AI with. So it won't only 119 00:07:15,970 --> 00:07:19,490 S1: be Microsoft, right, or Google or Amazon, because people like 120 00:07:19,490 --> 00:07:22,370 S1: Databricks and other companies are going to come in and say, hey, 121 00:07:22,370 --> 00:07:24,610 S1: let me just collect all your data and put it 122 00:07:24,610 --> 00:07:27,410 S1: into a UCC. And then any AI that you have 123 00:07:27,410 --> 00:07:29,490 S1: from these different vendors, they could all tap into the 124 00:07:29,490 --> 00:07:32,330 S1: same thing. So that will also be a thing. It 125 00:07:32,330 --> 00:07:35,929 S1: already is a thing. Again, companies like Databricks, I believe 126 00:07:35,930 --> 00:07:38,330 S1: they're playing in that space. But you don't want to 127 00:07:38,330 --> 00:07:41,730 S1: be a startup trying to implement AI in a customer's company. 128 00:07:41,730 --> 00:07:45,370 S1: When you don't have access to their UCC and you're 129 00:07:45,370 --> 00:07:49,450 S1: competing against someone else who does like Microsoft, right? That 130 00:07:49,450 --> 00:07:52,210 S1: is a bad place to be. So the main game 131 00:07:52,210 --> 00:07:56,490 S1: for making AI useful or powerful, especially in security and 132 00:07:56,490 --> 00:08:00,980 S1: security startups, will be gaining access to unified customer context. 133 00:08:01,020 --> 00:08:04,660 S1: This is all especially relevant to cybersecurity because security use 134 00:08:04,660 --> 00:08:09,340 S1: cases really, really benefit from context. Their identity, actions, history 135 00:08:09,340 --> 00:08:12,940 S1: of the thing you're investigating right across multiple systems. Also, 136 00:08:12,980 --> 00:08:17,180 S1: there's the issue of securing the UCC, since it will 137 00:08:17,220 --> 00:08:20,660 S1: be the absolute most sensitive data store in the entire company, 138 00:08:21,220 --> 00:08:24,860 S1: all the juiciest bits of the company all in one place, 139 00:08:25,340 --> 00:08:30,380 S1: which is like a red teamer or attackers dream. Cloudflare 140 00:08:30,420 --> 00:08:35,100 S1: launched an AI labyrinth feature that messes with unauthorized AI 141 00:08:35,140 --> 00:08:38,980 S1: scrapers by feeding them endless pages of irrelevant but real 142 00:08:39,020 --> 00:08:42,300 S1: looking content instead of actually blocking them. So it's like 143 00:08:42,300 --> 00:08:47,100 S1: a classic, like honeypot slash deception thing. Uh, to counter 144 00:08:47,100 --> 00:08:51,500 S1: rogue AI that won't honor the robots.txt file. A rush 145 00:08:51,500 --> 00:08:56,140 S1: release of JFK assassination files exposed 400 Social Security numbers 146 00:08:56,140 --> 00:09:01,160 S1: and other sensitive data belonging to former congressional staffers, many 147 00:09:01,160 --> 00:09:04,920 S1: of which of whom are actually high ranking officials now. 148 00:09:05,520 --> 00:09:10,000 S1: But we just doxed their Social Security numbers in this release. 149 00:09:10,000 --> 00:09:15,160 S1: New cybersecurity compensation research shows high six figure salaries are 150 00:09:15,160 --> 00:09:19,480 S1: not stopping 60% of security professionals from thinking about leaving 151 00:09:19,480 --> 00:09:23,360 S1: their job within a year. National security OpenAI is pressuring 152 00:09:23,360 --> 00:09:27,240 S1: the Trump administration to allow copyright scraping for AI training, 153 00:09:27,400 --> 00:09:31,280 S1: claiming America will lose the AI race to China without 154 00:09:31,280 --> 00:09:34,040 S1: having full access to scrape. And a lot of people 155 00:09:34,040 --> 00:09:37,240 S1: see this as like, corporate bullshit, trying to use security 156 00:09:37,240 --> 00:09:42,480 S1: to give them a corporate advantage. But unfortunately, it's actually true. 157 00:09:42,520 --> 00:09:46,240 S1: This is actually true because China has no limitations whatsoever 158 00:09:46,280 --> 00:09:49,160 S1: on what it trains on, on what it crawls. They 159 00:09:49,160 --> 00:09:54,920 S1: steal whatever, consume whatever with 100% free reign. And that 160 00:09:54,920 --> 00:09:58,600 S1: is a path or an accelerator to get to AGI 161 00:09:58,690 --> 00:10:01,290 S1: or ASI or whatever. So the question is who do 162 00:10:01,290 --> 00:10:03,930 S1: you want to have AGI or ASI first? The US 163 00:10:03,970 --> 00:10:06,650 S1: or China? And for me, Trump makes that question a 164 00:10:06,650 --> 00:10:09,050 S1: little bit harder to answer. But my answer is still 165 00:10:09,050 --> 00:10:13,809 S1: the US. Americans are buying overseas residency and citizenship as 166 00:10:13,809 --> 00:10:17,730 S1: a hedge against uncertainty in the US. China unveiled a 167 00:10:17,730 --> 00:10:22,450 S1: deep sea cable cutting device capable of severing undersea communications 168 00:10:22,850 --> 00:10:27,569 S1: at depths twice beyond where existing infrastructure operates, so they 169 00:10:27,570 --> 00:10:30,490 S1: can find deeper cables that are further down in a 170 00:10:30,490 --> 00:10:34,250 S1: trench or whatever it says unveiled. I wrote that word 171 00:10:34,610 --> 00:10:37,209 S1: based on the story, but it's kind of weird to like, 172 00:10:37,450 --> 00:10:39,290 S1: but I'm like, is this a press release? Or are 173 00:10:39,290 --> 00:10:42,650 S1: they like, you know, showing off the cool toy that 174 00:10:42,650 --> 00:10:45,370 S1: they have? And I don't know. I'm thinking of like CES, 175 00:10:45,410 --> 00:10:48,410 S1: they're like, come check out the cable cutter. London's Heathrow 176 00:10:48,410 --> 00:10:51,570 S1: Airport announced a full day shutdown after a significant fire 177 00:10:51,570 --> 00:10:54,650 S1: at a nearby electrical substation knocked out power to the 178 00:10:54,650 --> 00:10:56,770 S1: entire facility. A lot of questions are going to be 179 00:10:56,770 --> 00:11:00,660 S1: asked about that. I French soir sur les arc Prise 180 00:11:00,660 --> 00:11:04,100 S1: Foundation released a new AI intelligence test. I think this 181 00:11:04,100 --> 00:11:07,339 S1: is just another version of Arc, but the best AI 182 00:11:07,380 --> 00:11:11,900 S1: models are currently only scoring 1% while humans get 60%. 183 00:11:12,420 --> 00:11:18,380 S1: That is fantastic. Anthropic's Claude, not Anthropic's Anthropic's cloud has 184 00:11:18,380 --> 00:11:22,260 S1: finally added I hate saying finally seems rude and trite, 185 00:11:22,260 --> 00:11:26,860 S1: but whatever. Added web search to its AI chatbot catching 186 00:11:26,860 --> 00:11:30,860 S1: up to ChatGPT with clickable citations. I really want this 187 00:11:30,900 --> 00:11:33,460 S1: in the API. It's not there yet, and they're apparently 188 00:11:33,460 --> 00:11:38,059 S1: using Brave Search to power that, uh, search capability. Gmail 189 00:11:38,059 --> 00:11:40,740 S1: is rolling out an AI powered search that ranks results 190 00:11:40,740 --> 00:11:44,140 S1: based on relevance, instead of just showing the newest emails first. 191 00:11:44,179 --> 00:11:46,900 S1: This is cool, but I want AI filtering and AI 192 00:11:46,940 --> 00:11:50,900 S1: auto drafts. Like why don't they get more Gemini into Gmail? 193 00:11:51,540 --> 00:11:54,020 S1: It seems like they have the ability. Can't wait for 194 00:11:54,020 --> 00:11:58,140 S1: that to happen. Technology. Apple is updating AirPods Max next 195 00:11:58,280 --> 00:12:02,000 S1: month to add lossless and ultra low latency audio. Matt 196 00:12:02,000 --> 00:12:06,200 S1: Claude says that while type flags make sense for terminal commands, 197 00:12:06,200 --> 00:12:09,520 S1: you should use the dash dash force style option, the 198 00:12:09,520 --> 00:12:14,960 S1: long option instead for better readability, someone wrote. Seth Larson wrote. 199 00:12:14,960 --> 00:12:17,920 S1: This thing I fear for the Unauthenticated Web argues that 200 00:12:17,920 --> 00:12:21,840 S1: the increasingly common sign in to continue message on websites 201 00:12:22,160 --> 00:12:25,360 S1: is destroying the open promise of the web. Think we 202 00:12:25,400 --> 00:12:27,920 S1: talked about that last week as well. Nvidia says they 203 00:12:27,920 --> 00:12:30,680 S1: are investing hundreds of billions of dollars in the US 204 00:12:30,679 --> 00:12:34,839 S1: manufactured chips over the next four years, shifting away from 205 00:12:34,840 --> 00:12:38,960 S1: Asia and Trump's tariff threats. This is exactly what Trump 206 00:12:38,960 --> 00:12:41,439 S1: was trying to do with his policies, and it's positive. 207 00:12:41,480 --> 00:12:44,480 S1: But I'm worried that the damage will actually be worse 208 00:12:44,480 --> 00:12:48,440 S1: than the benefit. NYPD has dramatically expanded its drone program, 209 00:12:48,440 --> 00:12:55,600 S1: sending them to thousands of 911 911 911 calls. While 210 00:12:55,600 --> 00:12:59,610 S1: privacy advocates are worried about the lack of transparency, and 211 00:12:59,610 --> 00:13:02,610 S1: they're basically saying this could be used for surveillance humans. 212 00:13:02,650 --> 00:13:06,490 S1: New research from Aalto University suggests Earth has way more 213 00:13:06,490 --> 00:13:11,410 S1: people than the official 8.2 billion, due to major undercounting 214 00:13:11,410 --> 00:13:14,730 S1: in rural areas. I wonder how major they're talking about. 215 00:13:14,730 --> 00:13:17,130 S1: Are they talking about, like, up to 9 billion? Like 216 00:13:17,170 --> 00:13:20,570 S1: that would an extra billion? Are we losing an extra billion? 217 00:13:20,610 --> 00:13:24,130 S1: Curious what they I didn't find the actual number that 218 00:13:24,130 --> 00:13:28,050 S1: they're re-estimating. It looked like they just said it was bad. 219 00:13:28,090 --> 00:13:31,969 S1: Tyler Cowen shares insights from his conversation with Ezra Klein 220 00:13:32,010 --> 00:13:35,730 S1: about the new book abundance. Okay, so check this out. 221 00:13:35,730 --> 00:13:38,290 S1: You have to go check out this book. I just 222 00:13:38,290 --> 00:13:43,010 S1: listened to the, um, podcast with Lex and the co-author 223 00:13:43,010 --> 00:13:46,890 S1: of the book with him, with, uh, Ezra Klein. And 224 00:13:46,890 --> 00:13:53,130 S1: it is fantastic, in fact, that Technological Republic by Alex 225 00:13:53,130 --> 00:13:58,500 S1: Karp combined with this is, um, these are very centrist books. 226 00:13:58,980 --> 00:14:02,340 S1: It's Alex Karp being a lot more liberal because he's 227 00:14:02,340 --> 00:14:05,260 S1: considered kind of right. It's being a lot more liberal. 228 00:14:05,460 --> 00:14:08,820 S1: And then this book is from two liberals, but they're 229 00:14:08,820 --> 00:14:11,900 S1: talking a lot more centrist, a lot more kind of 230 00:14:11,940 --> 00:14:16,380 S1: Republican about the government and how to move forward and 231 00:14:16,380 --> 00:14:19,940 S1: all the mistakes that the liberals have made that enabled 232 00:14:19,940 --> 00:14:23,660 S1: Trump to come to power. So basically, if you are 233 00:14:23,700 --> 00:14:28,979 S1: a centrist or you like first principle thinking, uh, you 234 00:14:29,020 --> 00:14:31,580 S1: like to figure out, like why the government is broken, 235 00:14:31,620 --> 00:14:34,420 S1: like what we could do to fix it. Read this book, 236 00:14:34,460 --> 00:14:38,540 S1: abundance and read the book The Technological Republic. They are 237 00:14:38,580 --> 00:14:41,660 S1: fantastic books. And to be clear, I haven't read the 238 00:14:41,660 --> 00:14:45,540 S1: book actually, but I've listened to hours of analysis on it, 239 00:14:45,740 --> 00:14:48,180 S1: and I've they've kind of already covered the whole thing, 240 00:14:48,180 --> 00:14:50,500 S1: but I'm still going to read it, uh, cover to cover, 241 00:14:50,740 --> 00:14:53,020 S1: telling staff that the way to get ahead is not 242 00:14:53,020 --> 00:14:56,140 S1: to accumulate a giant fiefdom, and AI is going to 243 00:14:56,140 --> 00:14:58,280 S1: do the same. It's going to clean all that out. 244 00:14:58,760 --> 00:15:02,240 S1: David Kellogg explains the essential differences between a manager, director 245 00:15:02,240 --> 00:15:05,520 S1: and a VP, with the VP being accountable for results 246 00:15:05,520 --> 00:15:09,080 S1: regardless of who approved the plan. Isn't that the case 247 00:15:09,080 --> 00:15:14,880 S1: for director as well? Jonathan Kipnis and his team discovered 248 00:15:14,880 --> 00:15:19,280 S1: that rejuvenating the brain's lymphatic vessels improves memory in old 249 00:15:19,280 --> 00:15:24,160 S1: mice by helping clear waste that contributes to cognitive decline. 250 00:15:24,200 --> 00:15:26,280 S1: So my question is, how do I do this for me? 251 00:15:26,320 --> 00:15:30,160 S1: Because I'm not a mouse. Yeah. How do you rejuvenate 252 00:15:30,160 --> 00:15:34,880 S1: the brain's lymphatic vessels? Is there like a pill? Is 253 00:15:34,880 --> 00:15:38,600 S1: that like a workout? What do you do? All right. Ideas? Hi. Agency. 254 00:15:38,720 --> 00:15:41,200 S1: I've been hearing this concept a lot in the last 255 00:15:41,200 --> 00:15:43,520 S1: few months, and there are people arguing. It's one of 256 00:15:43,520 --> 00:15:46,200 S1: the most important ideas out there right now. It's also 257 00:15:46,200 --> 00:15:50,200 S1: highly related to my work on H3 human 3.0. So 258 00:15:50,200 --> 00:15:52,040 S1: I'm going to do a deep dive on it. But 259 00:15:52,040 --> 00:15:55,640 S1: it's roughly the ability to solve problems by believing they're 260 00:15:55,640 --> 00:16:00,210 S1: not unsolvable Solvable if they don't defy physics. So it's 261 00:16:00,210 --> 00:16:03,770 S1: like this very I want to say Ayn Randian. Honestly, 262 00:16:03,770 --> 00:16:07,810 S1: like John Galt, kind of like everything is possible. Like, 263 00:16:07,810 --> 00:16:11,130 S1: just go do it. Um, you know, you could just 264 00:16:11,130 --> 00:16:14,530 S1: make things. I feel like it's very elani in that way. 265 00:16:14,650 --> 00:16:19,210 S1: In a good Elon way. So I like the concept. 266 00:16:19,210 --> 00:16:21,370 S1: I think it's a powerful concept. Oh, here's another way 267 00:16:21,370 --> 00:16:23,890 S1: to say it. A sense that the story given to 268 00:16:23,890 --> 00:16:27,450 S1: you by other people about what you want or can 269 00:16:27,490 --> 00:16:32,250 S1: or cannot do is just that a story that is powerful. 270 00:16:32,290 --> 00:16:34,610 S1: How much do flaws and traumas enhance us? I worry 271 00:16:34,610 --> 00:16:38,650 S1: a lot about making life too easy as a society 272 00:16:38,690 --> 00:16:43,610 S1: or as parents. So there's this timeless struggle that parents have, 273 00:16:43,730 --> 00:16:45,730 S1: especially when they're like immigrants. And they went through a 274 00:16:45,730 --> 00:16:47,730 S1: lot of stuff, and they want to make sure their 275 00:16:47,730 --> 00:16:49,770 S1: children don't have to suffer in that way, but then 276 00:16:49,770 --> 00:16:52,730 S1: they end up making what they deem to be like 277 00:16:52,770 --> 00:16:56,530 S1: lesser adults, like adults that are less capable of taking 278 00:16:56,530 --> 00:16:59,820 S1: on the world than they were. So I love this quote, 279 00:16:59,820 --> 00:17:03,180 S1: which I saw about this. Uh, from Jason Liu. I 280 00:17:03,220 --> 00:17:05,219 S1: worked a lot of my mental health and now I 281 00:17:05,260 --> 00:17:07,699 S1: am no longer ambitious. I worked a lot on my 282 00:17:07,700 --> 00:17:11,740 S1: mental health, and now I am no longer ambitious. Great discovery. 283 00:17:11,780 --> 00:17:16,139 S1: Most bitter people you will ever meet. Gut punching, three 284 00:17:16,140 --> 00:17:18,220 S1: paragraph essay. I'm not going to read the whole thing. 285 00:17:18,260 --> 00:17:21,899 S1: Go check it out. Delphi AI new platform lets you 286 00:17:21,940 --> 00:17:27,580 S1: create and share a digital clone of yourself. Lange, Magnus. No. Lange, Magnus. 287 00:17:27,980 --> 00:17:30,100 S1: New open source tool that makes it easier to build 288 00:17:30,100 --> 00:17:34,139 S1: autonomous agents using Lange chain and Lange graph. Pure clever 289 00:17:34,180 --> 00:17:37,820 S1: new browser hack lets you read any Paywalled content. It's 290 00:17:37,859 --> 00:17:41,860 S1: very much similar to AI. Rise of Agentic AI is 291 00:17:41,859 --> 00:17:44,980 S1: out which I had a chance to contribute to. Definitely 292 00:17:44,980 --> 00:17:47,940 S1: go check that out in the links. Personal best. Neat 293 00:17:47,980 --> 00:17:50,500 S1: little tool that shows you which personal blogs are most 294 00:17:50,500 --> 00:17:53,780 S1: popular on Hacker News. So I'm going to mine this 295 00:17:53,780 --> 00:17:57,110 S1: thing and actually put them all into threshold so they 296 00:17:57,109 --> 00:18:00,910 S1: will be in threshold soon. Someone says they recommend against brave. 297 00:18:01,150 --> 00:18:03,910 S1: I thought that was a pretty good arguments. I don't 298 00:18:03,910 --> 00:18:06,110 S1: use it, so it wasn't super pertinent to me, but 299 00:18:06,150 --> 00:18:10,190 S1: worth sharing. Circuit tutor. Neat little tool lets you describe 300 00:18:10,190 --> 00:18:13,430 S1: simple circuits in plain English, and get both schematics and 301 00:18:13,430 --> 00:18:20,109 S1: interactive explanations for folks who need e refreshers. Go act. 302 00:18:20,150 --> 00:18:22,790 S1: New tool that turns your text or files into a 303 00:18:22,790 --> 00:18:29,230 S1: browser based explainer. Videos and OS. New GitHub Osint tool 304 00:18:29,230 --> 00:18:34,350 S1: that scrapes public user info, including emails, organizations and repositories, 305 00:18:34,390 --> 00:18:38,149 S1: and the recommendation of the week. Ask yourself every six months, 306 00:18:38,470 --> 00:18:41,430 S1: what would you do if you weren't afraid? What would 307 00:18:41,430 --> 00:18:43,950 S1: you do right now if you weren't afraid? What's causing 308 00:18:43,950 --> 00:18:47,070 S1: the fear? Should you push through the fear and do it? 309 00:18:47,310 --> 00:18:49,590 S1: Or can you at least build a plan to reduce 310 00:18:49,590 --> 00:18:53,470 S1: the risk of doing it? Constantly challenge yourself on this. 311 00:18:53,510 --> 00:18:55,750 S1: A better life is on the other side. And the 312 00:18:55,750 --> 00:19:01,730 S1: aphorism of the week, which is insane. This thing is crazy. Crazy. 313 00:19:01,770 --> 00:19:05,090 S1: Aphorism this week you sense you should be following a 314 00:19:05,090 --> 00:19:08,410 S1: different path, a more ambitious one. You felt you were 315 00:19:08,410 --> 00:19:11,970 S1: destined for other things, but you had no idea how 316 00:19:11,970 --> 00:19:16,090 S1: to achieve them. And in your misery, you began to 317 00:19:16,130 --> 00:19:20,250 S1: hate everything around you. You sensed you should be following 318 00:19:20,250 --> 00:19:22,969 S1: a different path, a more ambitious one. You felt you 319 00:19:22,970 --> 00:19:25,210 S1: were destined for other things, but you had no idea 320 00:19:25,210 --> 00:19:28,610 S1: how to achieve them. And in your misery, you began 321 00:19:28,609 --> 00:19:34,449 S1: to hate everything around you. Fyodor Dostoyevsky. Unsupervised learning is 322 00:19:34,450 --> 00:19:39,170 S1: produced on Hindenburg Pro using an Sm7 B microphone. A 323 00:19:39,170 --> 00:19:42,050 S1: video version of the podcast is available on the Unsupervised 324 00:19:42,050 --> 00:19:45,409 S1: Learning YouTube channel, and the text version with full links 325 00:19:45,410 --> 00:19:50,649 S1: and notes is available at Daniel Miessler newsletter. We'll see 326 00:19:50,650 --> 00:19:51,410 S1: you next time.