1 00:00:17,267 --> 00:00:20,267 S1: All right. Welcome to episode 45. Posted a bunch of 2 00:00:20,267 --> 00:00:23,347 S1: fabric extractions of stuff that I've been watching and reading 3 00:00:23,347 --> 00:00:27,227 S1: and stuff like that. So like YouTube videos, blog posts, 4 00:00:27,627 --> 00:00:30,107 S1: academic papers, and I think I'm going to do more 5 00:00:30,107 --> 00:00:33,107 S1: of this because I'm actually building this into my Pi, 6 00:00:33,147 --> 00:00:37,107 S1: which is pi personal AI infrastructure. And this is the 7 00:00:37,107 --> 00:00:40,307 S1: infrastructure you've heard me talking about in the augmented class. 8 00:00:40,987 --> 00:00:43,987 S1: And just like all over X and in the newsletter 9 00:00:43,987 --> 00:00:46,987 S1: over the years, this is basically the back end core 10 00:00:46,987 --> 00:00:51,347 S1: infrastructure that I basically runs for my life. So it's 11 00:00:51,346 --> 00:00:56,987 S1: basically finding all my content. It's categorizing all my content, labeling, sorting, 12 00:00:57,027 --> 00:01:02,867 S1: doing things like that, extracting summaries, um, you know, characterizing. 13 00:01:02,867 --> 00:01:06,147 S1: And I'm basically going to use that to continue enhancing 14 00:01:06,147 --> 00:01:08,467 S1: my life. Right. So I'm always adding to this thing. 15 00:01:09,267 --> 00:01:11,947 S1: So one of the core components of that is actually 16 00:01:13,507 --> 00:01:17,607 S1: finding new content, um, in an automated way. So basically 17 00:01:17,607 --> 00:01:21,167 S1: there's I'll give you a good example. There's on arXiv. 18 00:01:21,207 --> 00:01:22,967 S1: I think that's the way you pronounce this. I've heard 19 00:01:22,967 --> 00:01:27,207 S1: people say this, I used to call it arXiv because whatever, 20 00:01:27,207 --> 00:01:32,767 S1: I'm silly arXiv, I guess it's called arXiv. It's a 21 00:01:32,767 --> 00:01:36,727 S1: cool name. It's a better name than arXiv, although that's 22 00:01:36,727 --> 00:01:40,206 S1: how it's spelled. But whatever on arXiv there is a 23 00:01:40,607 --> 00:01:48,287 S1: RSS feed for. CS I so it's papers coming out 24 00:01:48,287 --> 00:01:52,647 S1: of arXiv that are both CS related and AI related. 25 00:01:52,687 --> 00:01:55,447 S1: And obviously this is a sweet spot for me. So 26 00:01:56,007 --> 00:01:57,967 S1: I want to be able to go and get those papers. 27 00:01:57,967 --> 00:02:01,247 S1: But I also want to know how high quality they are, like, 28 00:02:01,247 --> 00:02:04,487 S1: what are the interesting claims that they're making? Like how 29 00:02:04,487 --> 00:02:09,326 S1: good was their methodology, their testing? Like, was it a metastudy? 30 00:02:09,367 --> 00:02:12,727 S1: Did they do their own independent research for the paper, 31 00:02:13,367 --> 00:02:16,587 S1: like all those combinations. And there's already a fabric pattern 32 00:02:16,587 --> 00:02:18,667 S1: for this, which is analyze paper. And I think I 33 00:02:18,667 --> 00:02:21,627 S1: have a new version of it as well. But anyway, 34 00:02:21,627 --> 00:02:24,587 S1: I want to do that type of automated analysis on 35 00:02:24,787 --> 00:02:27,867 S1: every paper I might want to read. Okay. This is 36 00:02:27,867 --> 00:02:31,026 S1: kind of giving you hints of like what my personal 37 00:02:31,026 --> 00:02:34,227 S1: AI infrastructure does. And by the way, I'm aware of 38 00:02:34,227 --> 00:02:36,627 S1: the fact that I'm saying like a little bit too much. 39 00:02:36,626 --> 00:02:41,707 S1: So I'm trying to tamper that down. I like the word. 40 00:02:41,707 --> 00:02:43,987 S1: I think it's a cool transition word, but I don't 41 00:02:43,987 --> 00:02:47,547 S1: want to use it 40 times in every sentence. Anyway, 42 00:02:48,186 --> 00:02:51,626 S1: this is the infrastructure that I'm actually building, and this 43 00:02:51,626 --> 00:02:54,107 S1: is one of the modules that's really, really important to 44 00:02:54,107 --> 00:02:56,667 S1: me because it's not just papers. I want to see 45 00:02:56,667 --> 00:02:58,827 S1: this for open source intelligence. I want to see this 46 00:02:58,827 --> 00:03:01,267 S1: for all the different stuff. So these are all the 47 00:03:01,267 --> 00:03:05,346 S1: independent modules that I'm building. And actually those essentially turn 48 00:03:05,346 --> 00:03:08,546 S1: into products for me. Right. So my Intel product is 49 00:03:08,547 --> 00:03:11,787 S1: going to be the output of one of these. Threshold 50 00:03:11,786 --> 00:03:17,007 S1: is already the output of several of these, right? So anyway, 51 00:03:17,047 --> 00:03:21,647 S1: that's just a brief little thing on my Pi. Pi. 52 00:03:23,966 --> 00:03:27,727 S1: All that to say that these fabric extractions are essentially 53 00:03:27,727 --> 00:03:31,647 S1: outputs from these modules where it's found a piece of 54 00:03:31,647 --> 00:03:34,406 S1: really unique content, or I manually found it and I 55 00:03:34,406 --> 00:03:39,807 S1: just manually ran, uh, fabric against it, and it outputted 56 00:03:39,807 --> 00:03:42,087 S1: this output so you could check it on X, you 57 00:03:42,087 --> 00:03:44,407 S1: can see it on, you know, I do a few 58 00:03:44,407 --> 00:03:46,527 S1: of these in the newsletter every once in a while, 59 00:03:46,527 --> 00:03:50,407 S1: but I find it really useful to go and extract predictions, 60 00:03:50,607 --> 00:03:54,487 S1: to extract wisdom, to extract insights, or is another one 61 00:03:54,487 --> 00:03:58,167 S1: of my favorite. There's also one that just extracts quotes, 62 00:03:58,567 --> 00:04:02,567 S1: which can also be really interesting. There's um, I think 63 00:04:02,567 --> 00:04:07,207 S1: there's a one that just extracts like data or recommendations. 64 00:04:07,207 --> 00:04:11,267 S1: I think there's definitely a recommendations. One, actually, now that 65 00:04:11,267 --> 00:04:13,467 S1: I think about it, I need to have one that 66 00:04:13,467 --> 00:04:16,827 S1: extracts like a book list or a watch list. Um, 67 00:04:16,827 --> 00:04:18,587 S1: I'm not sure if I already made that one or 68 00:04:18,587 --> 00:04:21,467 S1: if I need to go add that to fabric. Anyway, 69 00:04:21,467 --> 00:04:27,227 S1: all this to say, I want my pi Pi always 70 00:04:27,227 --> 00:04:31,587 S1: running in the background, always collecting, always finding, always discovering, 71 00:04:31,587 --> 00:04:36,267 S1: always rotating. And before you think, well, hold on, I 72 00:04:36,307 --> 00:04:39,627 S1: is doing too much work for you. No, it is 73 00:04:39,627 --> 00:04:42,387 S1: not doing too much work for me. And here's why. 74 00:04:43,707 --> 00:04:48,347 S1: Part of this entire thing is the determination of what 75 00:04:48,347 --> 00:04:52,226 S1: I should go and do slowly. Okay? Anything that makes 76 00:04:52,227 --> 00:04:56,547 S1: it into that bucket, I must sit down and manually 77 00:04:56,547 --> 00:05:00,827 S1: enjoy and listen to and process and use my pen 78 00:05:00,827 --> 00:05:04,426 S1: and my index cards and write down notes or Apple 79 00:05:04,427 --> 00:05:06,347 S1: notes or whatever it is if I have to pause 80 00:05:06,347 --> 00:05:09,727 S1: this thing. But I still think the absolute best way 81 00:05:09,727 --> 00:05:14,367 S1: to experience anything is in the slow form, right? What's 82 00:05:14,367 --> 00:05:17,367 S1: still the best possible way to spend an evening is 83 00:05:17,407 --> 00:05:22,087 S1: with friends and family and smart people or whatever, and 84 00:05:22,087 --> 00:05:26,847 S1: you're basically having conversation with actual humans. So to me, 85 00:05:26,847 --> 00:05:29,807 S1: all this AI stuff, the whole purpose of it is 86 00:05:29,807 --> 00:05:33,367 S1: to enhance your ability to do those things. And this 87 00:05:33,367 --> 00:05:36,327 S1: is still this is all part of human 3.0. It's 88 00:05:36,327 --> 00:05:38,967 S1: all part of like, what should we actually be doing? 89 00:05:39,327 --> 00:05:42,967 S1: And what we should actually be doing is magnifying human experience. 90 00:05:43,407 --> 00:05:48,607 S1: Slow experience. Right. All this speeding up and all this 91 00:05:48,607 --> 00:05:52,127 S1: collection and all this, I should be in service of that, 92 00:05:52,527 --> 00:05:57,807 S1: not in opposition to that. All right. That's how we 93 00:05:57,807 --> 00:06:00,287 S1: get stuck on the first item. Not even getting into 94 00:06:00,287 --> 00:06:04,207 S1: the newsletter. Okay. Yeah. Go sign up for the podcast again, 95 00:06:04,207 --> 00:06:06,807 S1: which is silly because you're already listening to it. Uh, 96 00:06:06,807 --> 00:06:11,237 S1: let's see here. My friend Tracy Talbot is presenting Tell 97 00:06:11,237 --> 00:06:17,437 S1: No Lies. Teaching AI to know when it doesn't know. Yeah, 98 00:06:17,437 --> 00:06:19,737 S1: to know when it doesn't know. And this is June 99 00:06:19,737 --> 00:06:22,877 S1: 26th in Austin. And she's got a LinkedIn post that 100 00:06:22,877 --> 00:06:25,957 S1: I linked in the newsletter. I worked with her at Apple. 101 00:06:25,957 --> 00:06:31,676 S1: She was extraordinary there at Apple, she was doing data quality, uh, 102 00:06:31,677 --> 00:06:37,557 S1: on our team. And yeah, just really, really great person and, um, 103 00:06:37,557 --> 00:06:42,197 S1: really thoughtful about data quality and data security and all 104 00:06:42,197 --> 00:06:48,197 S1: these different aspects of data and basically data management at 105 00:06:48,197 --> 00:06:51,557 S1: scale and thinking about what it means to trust data. 106 00:06:51,997 --> 00:06:56,197 S1: So now she's getting into AI, and she's been doing 107 00:06:56,197 --> 00:06:57,957 S1: a whole lot of really cool stuff with that. So 108 00:06:57,957 --> 00:07:01,837 S1: definitely recommend going to see her talk again. That's June 109 00:07:01,837 --> 00:07:05,937 S1: 26th in Austin. All right. Places I'm going to be 110 00:07:05,937 --> 00:07:09,737 S1: speaking in 2025, and I realized I missed an item 111 00:07:09,737 --> 00:07:13,657 S1: on the list. Various talks and panels. Blackhat USA in 112 00:07:13,657 --> 00:07:17,617 S1: Vegas speaking at the Swiss Cyberstorm in Bern, Switzerland. It's 113 00:07:17,617 --> 00:07:20,897 S1: going to be an AI workshop. Actually going to do 114 00:07:20,897 --> 00:07:23,057 S1: a workshop for people, which is going to be a 115 00:07:23,057 --> 00:07:27,937 S1: combination of like augmented fabric slash, essentially like personal AI 116 00:07:27,937 --> 00:07:30,737 S1: infrastructure is kind of what that thing's going to be about. 117 00:07:31,537 --> 00:07:35,897 S1: Keynoting Appsec USA in DC in November and speaking at Blackhat, 118 00:07:35,937 --> 00:07:41,057 S1: Mia in Saudi Arabia in December. And the one that 119 00:07:41,057 --> 00:07:44,217 S1: I missed is I'm probably going to be doing something 120 00:07:44,217 --> 00:07:48,657 S1: in Sweden as well. Um, I can't remember the exact 121 00:07:48,657 --> 00:07:51,737 S1: date for that one, but, uh, yeah, I'll probably add 122 00:07:51,737 --> 00:07:56,337 S1: that to the list for this next coming newsletter. On Monday, 123 00:07:57,497 --> 00:08:00,737 S1: personal tech Stack updates I'm all in on cloud code. 124 00:08:01,017 --> 00:08:03,697 S1: I'm not going to go into why, but ultimately it's 125 00:08:03,837 --> 00:08:09,797 S1: because here I am going into why anthropic uses this 126 00:08:09,797 --> 00:08:14,677 S1: as their Agentic coding agent. And they've gone all in 127 00:08:14,677 --> 00:08:19,197 S1: on Agentic coding, which means this thing isn't like hacking 128 00:08:19,237 --> 00:08:24,397 S1: together agents, it's actually using agents to execute the various tasks. 129 00:08:24,397 --> 00:08:27,517 S1: And as of today, I just updated cloud code and 130 00:08:27,517 --> 00:08:30,797 S1: now it has planning built in. So as I've been 131 00:08:30,797 --> 00:08:34,717 S1: talking about pretty much everywhere, I think scaffolding around AI 132 00:08:35,117 --> 00:08:39,916 S1: is the superpower that's about to unlock this whole thing, 133 00:08:40,717 --> 00:08:43,796 S1: because the models are already smart. The problem is they 134 00:08:43,797 --> 00:08:47,037 S1: can't think for long. They have short memories. So we 135 00:08:47,037 --> 00:08:52,117 S1: have all these context problems. So any any person, any 136 00:08:52,117 --> 00:08:56,957 S1: intelligent being gets really stupid when it can't remember longer than, 137 00:08:56,997 --> 00:09:01,276 S1: you know, a short conversation and then you have to 138 00:09:01,276 --> 00:09:04,737 S1: go back into this, you know, stored memory bank. And 139 00:09:04,737 --> 00:09:08,497 S1: it's not perfect. I'm talking about Rag, for example. Or 140 00:09:08,497 --> 00:09:11,256 S1: if you have actual humans and they struggle with long 141 00:09:11,296 --> 00:09:15,256 S1: term memory or even short term memory, obviously they're going 142 00:09:15,296 --> 00:09:17,217 S1: to be less functional, right? They're going to be less 143 00:09:17,217 --> 00:09:20,297 S1: capable than they used to be before they lost those abilities. 144 00:09:20,737 --> 00:09:27,217 S1: So the ability to um, as Dwarkesh talks about, learn 145 00:09:27,217 --> 00:09:30,297 S1: on the job. Okay. You think about a knowledge worker. 146 00:09:30,617 --> 00:09:33,576 S1: They go in, they do orientation, they learn on the job. Okay. 147 00:09:33,577 --> 00:09:36,657 S1: Guess what? After that first week, they're way better than 148 00:09:36,656 --> 00:09:39,737 S1: when they started because now they have some knowledge. Well, 149 00:09:39,737 --> 00:09:42,897 S1: what happens after six months? What happens after a year? 150 00:09:42,896 --> 00:09:46,737 S1: What happens after they've been there for 20 years. They 151 00:09:46,737 --> 00:09:50,617 S1: now have all this stored knowledge about this company, which 152 00:09:50,617 --> 00:09:54,177 S1: is now part of their context. But the difference is 153 00:09:54,176 --> 00:09:58,617 S1: with a human, they are naturally, fluidly. Every time they 154 00:09:58,617 --> 00:10:03,077 S1: make a decision, they are thinking of their stored knowledge. 155 00:10:03,237 --> 00:10:07,557 S1: They incorporate all of their decisions with their stored knowledge 156 00:10:07,557 --> 00:10:10,957 S1: every time they make any decision. That is a thing 157 00:10:10,957 --> 00:10:15,237 S1: that I currently cannot do. It is struggling. It's like, oh, 158 00:10:15,276 --> 00:10:18,436 S1: I've got to go grab all this context and stuff 159 00:10:18,437 --> 00:10:21,957 S1: it into the prompt. How do you grab context of 160 00:10:21,957 --> 00:10:25,636 S1: 25 years of learning at a job? How do you 161 00:10:25,636 --> 00:10:28,117 S1: know what that even means? How do you even know 162 00:10:28,877 --> 00:10:33,836 S1: what to query to stuff into your decision making process? 163 00:10:33,957 --> 00:10:38,117 S1: This is completely transparent with humans. When we are like, oh, 164 00:10:38,156 --> 00:10:41,477 S1: let me approach this problem from a holistic standpoint. Let 165 00:10:41,477 --> 00:10:45,357 S1: me talk to all these disparate teams across the org, 166 00:10:45,516 --> 00:10:47,117 S1: and I'm going to gather notes and I'm going to 167 00:10:47,117 --> 00:10:48,997 S1: take notes. I'm going to write them down or whatever. 168 00:10:49,237 --> 00:10:52,477 S1: Your brain is going crazy with all of its knowledge 169 00:10:52,477 --> 00:10:56,477 S1: that you've gathered over these 25 years, and it's making 170 00:10:56,477 --> 00:11:00,857 S1: its own queries and pulling back that context, and these 171 00:11:00,857 --> 00:11:02,977 S1: are coming back to you as kind of like insights 172 00:11:02,977 --> 00:11:07,256 S1: or thoughts or creativity. You're not getting to see exactly 173 00:11:07,256 --> 00:11:09,776 S1: what the queries look like or even what the returned 174 00:11:09,776 --> 00:11:13,697 S1: results look like. We do not have this with AI. 175 00:11:14,097 --> 00:11:19,377 S1: We must manually go and make manual queries into context, 176 00:11:19,416 --> 00:11:23,697 S1: like with rag or like with stuffing other context or 177 00:11:23,697 --> 00:11:28,497 S1: context summaries into the prompts themselves, right? This is the 178 00:11:28,497 --> 00:11:32,296 S1: scaffolding that we need. We need to emulate the process 179 00:11:32,296 --> 00:11:37,776 S1: that humans execute naturally every day, you know, hundreds or 180 00:11:37,817 --> 00:11:40,416 S1: thousands or millions of times. However many times it is 181 00:11:40,416 --> 00:11:44,097 S1: that we're making decisions and little tiny, you know, nano 182 00:11:44,097 --> 00:11:49,577 S1: decisions and micro decisions. So I don't remember how I 183 00:11:49,577 --> 00:11:56,217 S1: got onto this, but scaffolding massively, massively important. And, uh, 184 00:11:56,217 --> 00:11:59,117 S1: that's why I think context is is so important to 185 00:11:59,156 --> 00:12:03,516 S1: all of this stuff. Um. Oh, I remember how Claude Cote. 186 00:12:03,516 --> 00:12:07,156 S1: So they are all in on the agent stuff and 187 00:12:07,156 --> 00:12:10,557 S1: they are all in on the planning component. Okay, this 188 00:12:10,557 --> 00:12:13,517 S1: planning component is yet another piece of what I just 189 00:12:13,516 --> 00:12:16,836 S1: talked about. It is, you know, keeping in mind what 190 00:12:16,837 --> 00:12:20,117 S1: we're trying to do overall and then checking off little 191 00:12:20,156 --> 00:12:22,796 S1: boxes of like, yeah, we've done this. Yeah, we've done this. 192 00:12:22,957 --> 00:12:28,636 S1: It's all about, uh, context management. Now context is a 193 00:12:28,636 --> 00:12:34,797 S1: little bit, um, overloaded here. I would say context planning management. Um, 194 00:12:35,117 --> 00:12:40,197 S1: learning on the job management, long term task management. These 195 00:12:40,197 --> 00:12:45,156 S1: are the types of things that really good knowledge workers, 196 00:12:45,437 --> 00:12:49,877 S1: like a principle engineer who's been somewhere for 25 years. 197 00:12:50,557 --> 00:12:52,316 S1: These are the types of things that they are really 198 00:12:52,317 --> 00:12:55,537 S1: good at. And these are the things in my opinion, 199 00:12:55,536 --> 00:12:58,896 S1: this is a prediction and this is just my prediction 200 00:12:59,497 --> 00:13:04,217 S1: is that we're going to see major advances in those areas, 201 00:13:04,217 --> 00:13:06,737 S1: which are actually kind of like tricks. This is what 202 00:13:06,737 --> 00:13:10,497 S1: I've been telling my friend, uh, Joel ever since 23, 203 00:13:10,536 --> 00:13:12,896 S1: that these types of tricks are the things that are 204 00:13:12,896 --> 00:13:15,057 S1: going to push us over the edge. These are the 205 00:13:15,057 --> 00:13:19,016 S1: types of tricks that are going to hyper jump. I 206 00:13:19,136 --> 00:13:24,217 S1: way further ahead. And I think what I think my 207 00:13:24,217 --> 00:13:29,057 S1: theory has been borne out through things like Post-training post-training 208 00:13:29,057 --> 00:13:34,617 S1: is not fundamentally better neural nets. It's better ways of 209 00:13:34,737 --> 00:13:39,217 S1: shaping and getting neural nets to do what we actually 210 00:13:39,217 --> 00:13:44,377 S1: want them to do. Right. So just zooming out, I 211 00:13:44,377 --> 00:13:47,817 S1: think Cloud Code has this figured out. I think anthropic 212 00:13:48,097 --> 00:13:53,017 S1: really deeply understands this. And they are building agents and 213 00:13:53,016 --> 00:13:57,317 S1: planning management and context management. They're doing really, really well 214 00:13:57,317 --> 00:14:00,997 S1: with it in a way that I'm just really excited 215 00:14:00,997 --> 00:14:03,556 S1: to build things with cloud code in a way that's 216 00:14:03,557 --> 00:14:07,037 S1: not really hitting for me. Uh, on the other platforms, 217 00:14:07,036 --> 00:14:12,317 S1: my secondary backup for cloud code is still cursor. Um, 218 00:14:12,317 --> 00:14:15,676 S1: I was messing with, uh, windsurf for a while and 219 00:14:15,676 --> 00:14:18,197 S1: some other tools, and I still dabble with all the 220 00:14:18,197 --> 00:14:21,517 S1: other tools when I see something, uh, come out or new, 221 00:14:21,517 --> 00:14:24,957 S1: I go and tinker with it. But right now, cloud 222 00:14:24,957 --> 00:14:29,717 S1: code is my main, with cursor being secondary. I'm also 223 00:14:29,717 --> 00:14:33,196 S1: switched over to using Dia as my main daily driver 224 00:14:33,197 --> 00:14:37,597 S1: for my browser. And um, I'm on all the Mac 225 00:14:37,597 --> 00:14:43,237 S1: OS 26 betas. They are fantastic. And uh, specifically having 226 00:14:43,237 --> 00:14:47,836 S1: your phone app on the desktop is fantastic. And, uh, 227 00:14:47,837 --> 00:14:51,977 S1: the podcast app is much improved. You can actually go 228 00:14:51,977 --> 00:14:53,617 S1: above the two x limit. You can go up to 229 00:14:53,657 --> 00:14:57,177 S1: three x now. And yeah, overall I would say Apple 230 00:14:57,177 --> 00:15:01,976 S1: just really did a lot of quality updates. For example, 231 00:15:02,017 --> 00:15:07,137 S1: AirPod handoffs are much better. So that's it for updates 232 00:15:07,137 --> 00:15:11,057 S1: I think. And let's get into cybersecurity. Apple quietly fixed 233 00:15:11,057 --> 00:15:16,377 S1: an iPhone Zero Day that was used against journalists. Echo 234 00:15:16,377 --> 00:15:22,856 S1: leak using uses markdown syntax to bypass Microsoft 365 Copilot security. 235 00:15:23,337 --> 00:15:25,897 S1: So some researchers found a smart way to steal data 236 00:15:25,897 --> 00:15:31,417 S1: from Copilot by using obscure markdown link formats that Microsoft 237 00:15:31,417 --> 00:15:37,737 S1: did not filter. Researchers turn 2 a.m. Tokyo hotel room 238 00:15:37,737 --> 00:15:43,217 S1: chat into a Netflix RC. So Chubbs, who is a 239 00:15:43,897 --> 00:15:51,446 S1: friend of mine and very, very good, um, attacker. Uh, Tucker. Um, 240 00:15:51,847 --> 00:15:56,207 S1: what are we saying? Researcher. Bug. Bounty guy. Um, security 241 00:15:56,247 --> 00:16:00,647 S1: guy focused on offensive security and automated testing. He's also 242 00:16:00,687 --> 00:16:04,327 S1: the creator of Asset note. So he's just one of 243 00:16:04,327 --> 00:16:08,127 S1: the top, you know, top 1% or 1% of the 1%. 244 00:16:08,487 --> 00:16:10,847 S1: I would put him in Sam Curry like up in 245 00:16:10,887 --> 00:16:15,927 S1: that top, top echelon. And, uh, basically they found, uh, 246 00:16:17,767 --> 00:16:21,247 S1: the way they found a way to attack an unclaimed 247 00:16:21,247 --> 00:16:25,686 S1: internal package name called NFC logger. And they use that 248 00:16:25,687 --> 00:16:30,527 S1: to get Archie actually on a Netflix, uh, system or 249 00:16:30,527 --> 00:16:34,767 S1: within Netflix itself. And I'm sure they fixed that. Otherwise 250 00:16:34,767 --> 00:16:39,007 S1: we wouldn't be talking about it. All right. Going to 251 00:16:39,007 --> 00:16:43,446 S1: skip over the tedious sessions, uh, offer that was in 252 00:16:43,447 --> 00:16:46,767 S1: the thing. Uh, I don't really want to talk about that. Um, 253 00:16:47,167 --> 00:16:51,427 S1: HTML spec now escapes angle brackets and attributes. So Chrome 254 00:16:51,427 --> 00:16:55,107 S1: and other browsers are now going to start escaping. Left 255 00:16:55,507 --> 00:17:00,547 S1: and right caret characters in HTML attributes to prevent mutation 256 00:17:01,267 --> 00:17:07,427 S1: XSS attacks, Interpol dismantles 20,000 malicious IPS linked to 69 257 00:17:07,427 --> 00:17:14,347 S1: malware variants. Cursor Security Rules project tackles unsafe AI generated code. 258 00:17:14,787 --> 00:17:20,667 S1: Europol says Europol yeah. Europol says stolen data has become 259 00:17:20,667 --> 00:17:25,107 S1: the new underground currency. Phishing messages. This is a quote 260 00:17:25,347 --> 00:17:28,387 S1: created by Llms have higher success rate than those written 261 00:17:28,387 --> 00:17:33,867 S1: by humans. And this is a euro pool. Uh io 262 00:17:33,867 --> 00:17:41,786 S1: CTA report. US airlines quietly selling flight data to DHS. Yeah. 263 00:17:41,827 --> 00:17:45,507 S1: Government agencies can search 39 months of flight data, including 264 00:17:45,507 --> 00:17:50,286 S1: passenger names, itineraries, travel dates and credit card numbers, and 265 00:17:50,287 --> 00:17:54,167 S1: other people buying this data are include like the Secret Service, SEC, 266 00:17:54,206 --> 00:18:00,927 S1: DEA and US Marshal Service. And that's in addition to 267 00:18:01,247 --> 00:18:06,087 S1: the DHS and all its various services. Amazon's AI agents 268 00:18:06,087 --> 00:18:11,726 S1: can build cyber defense signatures in minutes. National security thoughts 269 00:18:11,726 --> 00:18:15,847 S1: on Israel's action against Iran. I'm actually not going to 270 00:18:15,847 --> 00:18:21,167 S1: go super into this one. It's an evolving situation. As 271 00:18:21,167 --> 00:18:24,686 S1: everyone knows, new stuff is happening like every few hours 272 00:18:24,686 --> 00:18:28,086 S1: on this, uh, go, go read the newsletter if you 273 00:18:28,087 --> 00:18:32,247 S1: want to see my take on this. Army is commissioning 274 00:18:32,247 --> 00:18:36,526 S1: Big Tech executives as lieutenant colonel. So for Silicon Valley 275 00:18:36,527 --> 00:18:42,007 S1: executives from meta, Palantir and OpenAI are being brought in 276 00:18:42,047 --> 00:18:46,186 S1: as lieutenant colonels to speed up the government's adoption of technology, 277 00:18:46,946 --> 00:18:51,587 S1: and I'm overall for this, but I also have my 278 00:18:51,587 --> 00:18:58,627 S1: spider senses going off in terms of like dystopia type situation. Again, I, 279 00:18:59,387 --> 00:19:03,946 S1: I see the value of Palantir. I see the value 280 00:19:03,946 --> 00:19:10,067 S1: of winning at almost any cost against China. Right? That 281 00:19:10,067 --> 00:19:12,427 S1: is why I support this type of move from the 282 00:19:12,427 --> 00:19:17,867 S1: military and why I support overall. In some ways, companies 283 00:19:17,867 --> 00:19:22,627 S1: like Palantir and and drill and you know why these 284 00:19:22,627 --> 00:19:26,907 S1: companies exist? Yes, we need to move super fast. Unfortunately, 285 00:19:27,507 --> 00:19:33,186 S1: the cultures and the ambitions and a whole bunch of 286 00:19:33,186 --> 00:19:38,147 S1: externalities often come with moving really fast and in really 287 00:19:38,147 --> 00:19:42,706 S1: ambitious ways in with technology companies where you're making tons 288 00:19:42,706 --> 00:19:46,727 S1: of money supporting the military. I don't think I need 289 00:19:46,726 --> 00:19:50,807 S1: to spell it out. There's possibility for harm here, right? 290 00:19:51,247 --> 00:19:56,127 S1: There's possibility for harm here. And I think we have 291 00:19:56,127 --> 00:19:58,487 S1: to take that risk. I think this is the right 292 00:19:58,486 --> 00:20:02,326 S1: move to take that risk. But holy crap, is it scary? Yes. 293 00:20:02,686 --> 00:20:05,486 S1: And do we need oversight? Yes. Do we need to 294 00:20:05,486 --> 00:20:08,527 S1: be extremely careful? Yes. Do we need to not turn 295 00:20:08,567 --> 00:20:12,526 S1: our eyes away and just hope it all works out? Absolutely. 296 00:20:14,927 --> 00:20:19,687 S1: Cheap drones will massively disrupt current large military dominance. Yeah. 297 00:20:19,726 --> 00:20:24,046 S1: Israel and Ukraine are showing that 500 drones can destroy 298 00:20:24,047 --> 00:20:28,207 S1: extremely expensive military gear, which could upend the advantage of 299 00:20:28,767 --> 00:20:32,647 S1: countries like, you know, Russia, the US. We're already seeing 300 00:20:32,647 --> 00:20:40,407 S1: it with Russia in Ukraine. Uh, big scope aside on 301 00:20:40,407 --> 00:20:42,587 S1: this one. I'm curious about how this is going to 302 00:20:42,587 --> 00:20:46,187 S1: start trickling into consumer and personal security. Like when will 303 00:20:46,226 --> 00:20:50,627 S1: executive protection teams all need to have like Anti-drone tech 304 00:20:50,667 --> 00:20:54,307 S1: as part of their package? I think probably sooner rather 305 00:20:54,307 --> 00:20:59,467 S1: than later. I that apple paper saying AIS don't reason 306 00:20:59,466 --> 00:21:04,507 S1: was highly flawed, and someone named Alex Lawson says that 307 00:21:04,507 --> 00:21:08,307 S1: the paper basically didn't do a good job and it 308 00:21:08,307 --> 00:21:13,667 S1: was easy to make better examples. Although I've heard someone 309 00:21:13,667 --> 00:21:17,267 S1: did a paper like this, this might have been Alex Lawson. Actually, 310 00:21:17,307 --> 00:21:19,627 S1: I don't know if it was this paper. I heard 311 00:21:19,627 --> 00:21:24,266 S1: somebody say that that whole paper, this whole takedown paper 312 00:21:24,267 --> 00:21:28,507 S1: was actually Claude generated, and it wasn't a real paper. 313 00:21:28,507 --> 00:21:31,987 S1: And it was just like, I don't know, fake or 314 00:21:31,986 --> 00:21:35,227 S1: hype or whatever, and that they didn't expect it to 315 00:21:35,267 --> 00:21:37,507 S1: blow up so much. I don't know if it was 316 00:21:37,507 --> 00:21:41,927 S1: this one, but interesting to note. Also interesting to note, 317 00:21:41,927 --> 00:21:43,887 S1: and I'm going to beat up on myself here for 318 00:21:43,887 --> 00:21:48,607 S1: a second. I don't know if this was the paper 319 00:21:48,607 --> 00:21:52,327 S1: that said that it doesn't actually matter if I know, 320 00:21:53,407 --> 00:21:56,007 S1: or it doesn't actually matter if it was the paper 321 00:21:56,007 --> 00:21:58,887 S1: or not. The fact that I put it in the 322 00:21:58,887 --> 00:22:04,887 S1: newsletter is potentially a strike against me. Think about this. 323 00:22:06,087 --> 00:22:11,047 S1: I counted as a strike against myself. Um, the thing is, 324 00:22:11,167 --> 00:22:13,967 S1: I'm not sure if I would do it differently. It's 325 00:22:13,966 --> 00:22:17,807 S1: just maybe I, I don't know. This is weird. Okay. 326 00:22:17,847 --> 00:22:19,967 S1: Because when I put something in the newsletter and I 327 00:22:19,966 --> 00:22:23,807 S1: talk about it, how much of a high. How much 328 00:22:23,807 --> 00:22:27,287 S1: have I endorsed this thing and said for sure that 329 00:22:27,287 --> 00:22:29,807 S1: this is a real paper? For sure. These are real 330 00:22:29,807 --> 00:22:34,447 S1: results because I read the paper fully. Not only that, 331 00:22:34,446 --> 00:22:38,466 S1: but I looked at the all the testing methodologies And 332 00:22:38,466 --> 00:22:42,787 S1: I evaluated the data and I confirmed that their testing 333 00:22:42,787 --> 00:22:45,787 S1: was not a lie. That is a level of depth 334 00:22:45,787 --> 00:22:48,907 S1: that I cannot do for the amount of papers and 335 00:22:48,907 --> 00:22:52,587 S1: stories that I'm doing. So what I'm saying is it 336 00:22:52,587 --> 00:22:56,267 S1: is possible for me to be duped. Um, and this 337 00:22:56,267 --> 00:22:59,506 S1: one very well could be, uh, in fact, you know what? 338 00:22:59,507 --> 00:23:00,907 S1: I'm going to click on a link here and see 339 00:23:00,907 --> 00:23:08,306 S1: if there's possible, like updates. I don't think so. I 340 00:23:08,307 --> 00:23:10,507 S1: don't think so. I don't think it applied. It doesn't 341 00:23:10,507 --> 00:23:13,786 S1: actually matter if it applied. This paper that I included 342 00:23:13,827 --> 00:23:16,747 S1: might be completely legit. And I think it is, but 343 00:23:16,747 --> 00:23:21,387 S1: it doesn't actually matter because this strike against me still counts, right? 344 00:23:21,427 --> 00:23:24,467 S1: So this is a question from me to myself and 345 00:23:24,466 --> 00:23:28,547 S1: also from you to me and from me to you, 346 00:23:28,547 --> 00:23:33,867 S1: to me. You got to watch what people are posting, 347 00:23:33,867 --> 00:23:36,446 S1: and you got to watch with how sure that they 348 00:23:36,446 --> 00:23:40,647 S1: are that that thing is actually real, and whether or 349 00:23:40,647 --> 00:23:44,246 S1: not they're just using something that sounds real to support 350 00:23:44,247 --> 00:23:48,686 S1: their own opinion. This one happened to go with my opinion. Oh, 351 00:23:48,686 --> 00:23:51,486 S1: so what do you know? Magically, I just included in 352 00:23:51,486 --> 00:23:55,286 S1: the newsletter as support for my opinion. Well, what if 353 00:23:55,287 --> 00:23:58,607 S1: it turns out that was complete garbage? Well, then I'm 354 00:23:58,607 --> 00:24:02,647 S1: just like anyone else who is posting a random garbage. 355 00:24:03,567 --> 00:24:06,127 S1: I would argue I'm not like that, but I'm saying 356 00:24:06,167 --> 00:24:08,246 S1: it looks kind of the same, so I don't really 357 00:24:08,247 --> 00:24:12,006 S1: care how much the difference is. So just keep in mind. 358 00:24:12,007 --> 00:24:15,047 S1: And not just for me, but for yourself. You got 359 00:24:15,047 --> 00:24:18,607 S1: to watch for whether or not evidence that's coming in 360 00:24:18,847 --> 00:24:22,967 S1: is only air quote evidence, because how much time have 361 00:24:22,966 --> 00:24:27,486 S1: you actually spent looking at the evidence, redoing the paper yourself? 362 00:24:28,206 --> 00:24:33,167 S1: You know. And so how much are we using external 363 00:24:33,547 --> 00:24:40,387 S1: Validation or citations, pointing to a paper as support for 364 00:24:40,387 --> 00:24:43,946 S1: the paper being legit. And I think it gets more 365 00:24:43,946 --> 00:24:46,907 S1: and more serious the further we go. Um, the good 366 00:24:46,907 --> 00:24:51,107 S1: news is, a big part of my personal AI structure 367 00:24:51,107 --> 00:24:53,827 S1: is going to be to find stuff like this. I 368 00:24:54,147 --> 00:24:55,947 S1: want to be able to find stuff like this with 369 00:24:55,946 --> 00:24:59,706 S1: my AI infrastructure, where it's like, I don't know, seems 370 00:24:59,706 --> 00:25:02,907 S1: like they use a lot of catchy AI terms in here, 371 00:25:02,907 --> 00:25:06,787 S1: and it looks like they were really hand-wavy with their examples. Like, 372 00:25:06,787 --> 00:25:08,867 S1: I want to be able to catch this stuff for myself. 373 00:25:09,907 --> 00:25:13,106 S1: Just something to keep in mind. OpenAI's O3 Pro is 374 00:25:13,107 --> 00:25:18,266 S1: very smart, but context hungry. Agreed. Man killed by police 375 00:25:18,267 --> 00:25:23,387 S1: after spiraling into a ChatGPT driven psychosis. Google plans to 376 00:25:23,427 --> 00:25:27,507 S1: cut ties with scale AI because basically meta bot like 377 00:25:27,507 --> 00:25:31,387 S1: half of them or something. CIO wants to clone staff 378 00:25:31,387 --> 00:25:36,127 S1: as digital twins and AI agents. UC San Diego CIO 379 00:25:36,167 --> 00:25:40,246 S1: wants to extract knowledge from experienced IT staff and create 380 00:25:40,247 --> 00:25:46,967 S1: digital twins that handle routine problems. Well, yeah, obviously, you know, 381 00:25:47,007 --> 00:25:50,686 S1: you got somebody, uh, gray beard, you know, Chris Myers, 382 00:25:51,127 --> 00:25:54,686 S1: who's been at the company for 41 years. And they 383 00:25:54,686 --> 00:25:58,287 S1: know everything. And anyone who, you know, finds something broken 384 00:25:58,287 --> 00:26:01,927 S1: that nobody could fix, they all call Chris Myers. So 385 00:26:01,966 --> 00:26:05,607 S1: the CEO is like, let's go find Chris Myers. Let's 386 00:26:05,607 --> 00:26:10,087 S1: go interview him for seven weeks in a dark, padded 387 00:26:10,087 --> 00:26:12,967 S1: room and pull out all of his knowledge and put 388 00:26:12,966 --> 00:26:15,686 S1: that into an AI agent. And now you could just 389 00:26:15,726 --> 00:26:18,967 S1: ask a virtual Chris Myers. Well, obviously we need this 390 00:26:18,966 --> 00:26:21,327 S1: for everything, right? We need this for the best doctors. 391 00:26:21,327 --> 00:26:24,407 S1: We need this for the best psychiatrists. We need this 392 00:26:24,407 --> 00:26:27,647 S1: for everything. This is you know, this is the problem. When, 393 00:26:28,127 --> 00:26:32,077 S1: you know, extremely experienced people leave a company, you lose 394 00:26:32,077 --> 00:26:36,957 S1: all the knowledge that they took with them. And unfortunately, 395 00:26:37,597 --> 00:26:42,716 S1: also for AI, it's also knowledge that you can't possibly 396 00:26:43,357 --> 00:26:47,917 S1: extract from them efficiently, right? You can't possibly ask enough 397 00:26:47,917 --> 00:26:52,716 S1: questions to to gather their knowledge. Their knowledge is, let's 398 00:26:52,716 --> 00:26:57,157 S1: call it infinite. 41 years of knowledge essentially baked into 399 00:26:57,196 --> 00:27:02,316 S1: an AI model. That's a great analogy. Actually it. How 400 00:27:02,317 --> 00:27:06,157 S1: many questions do you have to ask? You know, uh, 401 00:27:06,637 --> 00:27:11,357 S1: O3 to get all of O3 knowledge out of it? 402 00:27:11,677 --> 00:27:15,036 S1: The answer is an infinite number of questions. So it 403 00:27:15,037 --> 00:27:18,357 S1: doesn't matter how much you interview Chris Myers, who's been 404 00:27:18,357 --> 00:27:21,677 S1: there for 41 years, you're not going to get enough. Right? 405 00:27:21,716 --> 00:27:28,317 S1: So so this is a serious challenge. ChatGPT dominates LLM 406 00:27:28,357 --> 00:27:32,057 S1: usage at 80% 86% market share. I did not know 407 00:27:32,057 --> 00:27:35,177 S1: it was that high. Technology fabric. Summary of the massive 408 00:27:35,216 --> 00:27:39,217 S1: Google outage. Yeah. This is this is really cool. Um, 409 00:27:40,297 --> 00:27:42,737 S1: so they put out the, you know, Google put out 410 00:27:42,736 --> 00:27:47,617 S1: their their giant paper saying what happened. And when I see, 411 00:27:47,657 --> 00:27:50,857 S1: you know, a 14 page paper or whatever. And it's 412 00:27:50,857 --> 00:27:53,816 S1: about an outage and describing an outage. It's not like 413 00:27:53,817 --> 00:27:57,377 S1: deep human knowledge that I'm getting. That's where I run Extraction's. 414 00:27:57,696 --> 00:28:03,216 S1: So I've got this, um, pattern called create story explanation, 415 00:28:03,696 --> 00:28:09,496 S1: create underscore story underscore explanation. It is one of my 416 00:28:09,497 --> 00:28:14,617 S1: favorite patterns. It explains anything at like a ninth grade 417 00:28:14,617 --> 00:28:17,897 S1: level in kind of a story format. So this happened, 418 00:28:17,897 --> 00:28:22,336 S1: then this happened, then this happened. And it's just very clear. 419 00:28:22,456 --> 00:28:26,057 S1: So I recommend checking it out if you care about 420 00:28:26,057 --> 00:28:28,997 S1: that outage, but more importantly, using it against other stuff. 421 00:28:29,597 --> 00:28:33,797 S1: YouTube officially beats all other streaming platforms in US viewership. 422 00:28:34,157 --> 00:28:40,437 S1: They have 13% of all US TV viewership. Oh, wow. Okay. 423 00:28:40,477 --> 00:28:44,277 S1: I thought it was 13% of streaming. Okay, I read 424 00:28:44,277 --> 00:28:52,117 S1: that wrong. 13% of all US TV viewership. That's insane. 425 00:28:55,117 --> 00:29:00,877 S1: That's insane. Okay. YouTube by itself, 13% of TV. Yeah, 426 00:29:00,917 --> 00:29:03,037 S1: it seems very high. I bet you that goes up 427 00:29:03,037 --> 00:29:07,437 S1: crazy amounts in the next few years. It's basically the 428 00:29:07,437 --> 00:29:09,796 S1: only TV that I watched. I was actually at a 429 00:29:09,797 --> 00:29:14,437 S1: resort yesterday in Napa, and they only had regular TV, 430 00:29:14,437 --> 00:29:18,036 S1: and it was like. It was like a commercial rectangle. 431 00:29:19,197 --> 00:29:22,197 S1: All I did was watch commercials and occasionally they would 432 00:29:22,197 --> 00:29:26,336 S1: show like part of a show. Anyway, Amazon joins the 433 00:29:26,337 --> 00:29:34,857 S1: big nuclear party, buying 1.2GW of AWS. tvOS 26 hints 434 00:29:34,857 --> 00:29:37,817 S1: at built in camera for a new Apple TV 4K. 435 00:29:38,537 --> 00:29:40,697 S1: I wish there was an Apple TV 8-K and I 436 00:29:40,697 --> 00:29:43,977 S1: wish it had ten Gigabit Ethernet. That's what I wish. 437 00:29:44,097 --> 00:29:45,897 S1: But I would like to see a camera because I'd 438 00:29:45,897 --> 00:29:49,737 S1: like to do FaceTime with, um, in the living room 439 00:29:49,737 --> 00:29:52,417 S1: sitting on the couch. That would be nice. I also 440 00:29:52,417 --> 00:29:56,537 S1: don't want a camera facing the living room, so I 441 00:29:56,537 --> 00:30:01,417 S1: would need one of those flip dongle cover things. Apple's 442 00:30:01,417 --> 00:30:05,177 S1: iPad OS 26 finally makes the iPad a real computer. 443 00:30:05,177 --> 00:30:10,017 S1: Everyone is talking about this. People are saying that this 444 00:30:10,057 --> 00:30:14,897 S1: actually allows you to use an iPad as a laptop, 445 00:30:15,417 --> 00:30:19,257 S1: almost like very closely. And it's kind of been universal 446 00:30:19,257 --> 00:30:22,097 S1: that people are saying, Holy crap, this is really, really good. 447 00:30:22,697 --> 00:30:26,117 S1: I have not. I don't use my iPads much and 448 00:30:26,117 --> 00:30:29,877 S1: the reason is I just didn't like the OS. The 449 00:30:29,877 --> 00:30:32,316 S1: only thing that I use it for is reading and 450 00:30:32,317 --> 00:30:36,157 S1: drawing and sketching and doing stuff like that with like freeform. 451 00:30:36,637 --> 00:30:39,197 S1: And for that I really love it. And for reading, 452 00:30:39,197 --> 00:30:41,197 S1: I really love it. But I don't use it as 453 00:30:41,197 --> 00:30:44,797 S1: a computer because I prefer laptops. Maybe that will change 454 00:30:46,117 --> 00:30:48,677 S1: the argument that it's time to kill Siri. I did 455 00:30:48,677 --> 00:30:52,157 S1: not know that I believe this, but I definitely do. 456 00:30:52,197 --> 00:30:57,277 S1: I think once Apple fixes its AI interface, its universal 457 00:30:57,277 --> 00:31:00,717 S1: AI to life OS that I've been talking about, they 458 00:31:00,717 --> 00:31:04,397 S1: need to call it something new. Siri has too much baggage. 459 00:31:05,517 --> 00:31:08,797 S2: I can show them if you ask. 460 00:31:09,277 --> 00:31:15,077 S1: Hey Siri, stop. The one time Siri actually listens when 461 00:31:15,077 --> 00:31:21,117 S1: I'm trying to record it. Essentially, that system does not 462 00:31:21,117 --> 00:31:25,096 S1: turn on and turn off lights reliably. It has so 463 00:31:25,097 --> 00:31:29,257 S1: many problems. It's been over ten years. It had its attempt, 464 00:31:29,297 --> 00:31:32,097 S1: it had its chances, and now it needs a new name. 465 00:31:32,137 --> 00:31:35,697 S1: I thoroughly believe this. It's just a marketing concept. When 466 00:31:35,697 --> 00:31:39,217 S1: you have enough bad baggage associated with something, you got 467 00:31:39,217 --> 00:31:42,497 S1: to flip it over. You got to retire it. Nvidia 468 00:31:42,497 --> 00:31:46,337 S1: writes off China revenue in company forecasts because, you know 469 00:31:46,377 --> 00:31:50,497 S1: it's too unpredictable. With the current administration, Waymo rides cost 470 00:31:50,497 --> 00:31:53,377 S1: more than Uber or Lyft, but people are happy paying 471 00:31:53,377 --> 00:31:57,217 S1: it so often 5 or $10 more for a ride 472 00:31:57,697 --> 00:32:00,977 S1: versus like Uber or a taxi. But people are still 473 00:32:00,977 --> 00:32:03,377 S1: paying it and they're still making money because it's so 474 00:32:03,377 --> 00:32:08,257 S1: much of a better experience. Humans. Trump ends protection for 475 00:32:08,257 --> 00:32:12,897 S1: Afghans as Congress scrambles to intervene. I'm really upset about this, 476 00:32:13,337 --> 00:32:18,777 S1: really upset that these Afghanis who had helped the US 477 00:32:18,777 --> 00:32:23,357 S1: military during the war and who cannot go back because 478 00:32:23,357 --> 00:32:28,877 S1: they are being hunted by the Taliban and because their 479 00:32:28,877 --> 00:32:33,437 S1: families are under like watch by the Taliban, they are 480 00:32:33,437 --> 00:32:36,677 S1: in grave danger. Their families are and they are. And 481 00:32:36,677 --> 00:32:38,477 S1: if they go back and try to link with their 482 00:32:38,477 --> 00:32:41,917 S1: families because they are forcibly kicked out of the US, 483 00:32:42,837 --> 00:32:47,997 S1: it is essentially damning them and their families, potentially to die. 484 00:32:48,837 --> 00:32:53,317 S1: And this is after they served and sacrificed essentially for 485 00:32:53,317 --> 00:32:56,836 S1: the US military. I do not believe we should treat 486 00:32:56,837 --> 00:33:03,197 S1: allies like that. Robin I system makes first autonomous scientific discovery. 487 00:33:03,197 --> 00:33:06,837 S1: So this is future houses. Robin I so I talked 488 00:33:06,837 --> 00:33:13,197 S1: about this story of automated, you know, science assistance. And 489 00:33:13,197 --> 00:33:15,997 S1: this is even more it's more end to end the 490 00:33:15,997 --> 00:33:21,777 S1: entire research process. So it, um, it discovered that Ripasudil 491 00:33:22,097 --> 00:33:28,097 S1: could treat dry macular degeneration by orchestrating multiple specialized agents 492 00:33:28,297 --> 00:33:33,337 S1: to handle the entire research process autonomously in just 2.5 months. 493 00:33:34,097 --> 00:33:37,657 S1: And who knows how much of that is like the 494 00:33:37,657 --> 00:33:41,737 S1: entire air quotes, entire research process. But supposedly it went 495 00:33:41,737 --> 00:33:46,576 S1: end to end and actually worked. So this is the 496 00:33:46,577 --> 00:33:49,377 S1: type of thing that I think is really, really exciting. 497 00:33:50,097 --> 00:33:55,097 S1: Ultra black paint might solve satellite light pollution crisis. Although 498 00:33:55,097 --> 00:33:57,897 S1: I'm curious, wouldn't you just see black dots moving through 499 00:33:57,937 --> 00:34:03,337 S1: your pictures, blocking out stars and, you know, the moon 500 00:34:03,337 --> 00:34:07,977 S1: or whatever? I guess the black dots wouldn't be as annoying. 501 00:34:09,897 --> 00:34:12,657 S1: It seems like they could still overwrite stars. If you 502 00:34:12,697 --> 00:34:14,897 S1: are trying to take a picture of the starry sky, 503 00:34:15,617 --> 00:34:18,917 S1: you might still see artifacts, but it won't be nearly 504 00:34:18,917 --> 00:34:23,116 S1: as bad as, like a bright white line drawing over 505 00:34:23,117 --> 00:34:28,517 S1: your picture. Actually, multiple bright white lines. But, uh, I 506 00:34:28,637 --> 00:34:31,837 S1: wonder how possible this is. Like, are we going to 507 00:34:31,837 --> 00:34:33,917 S1: pull all the ones that are up there down? Are 508 00:34:33,917 --> 00:34:37,637 S1: we going to paint the ones that are up there? Doubtful. 509 00:34:38,997 --> 00:34:41,317 S1: So this would involve painting all the new ones that 510 00:34:41,317 --> 00:34:46,357 S1: are going up, which unfortunately is many thousands. And the 511 00:34:46,357 --> 00:34:49,317 S1: Pentagon has evidently been pushing Americans to believe in UFOs 512 00:34:49,317 --> 00:34:53,877 S1: for decades. And evidently, according to this report, this is 513 00:34:53,877 --> 00:34:57,517 S1: to cover up actual secret weapons programs. So they basically 514 00:34:57,517 --> 00:35:00,757 S1: circulate this stuff on purpose to get people talking about 515 00:35:01,317 --> 00:35:07,957 S1: a red herring discovery fiddle. ITN brings malicious traffic detection 516 00:35:07,957 --> 00:35:10,317 S1: to man of the middle proxy. Oh, I'm not going 517 00:35:10,317 --> 00:35:11,837 S1: to actually read all of these, so I'm just going 518 00:35:11,877 --> 00:35:14,037 S1: to scroll through and look for some really cool ones, 519 00:35:14,157 --> 00:35:17,056 S1: or if any are from my friends. Let's see. Boom 520 00:35:17,057 --> 00:35:20,457 S1: boom boom boom. Oh, Sam Altman's gentle singularity. You probably 521 00:35:20,457 --> 00:35:22,377 S1: want to read this one. I've also got a fabric 522 00:35:22,377 --> 00:35:25,257 S1: extracted predictions that comes out of it in the newsletter, 523 00:35:25,257 --> 00:35:29,097 S1: so check that out. And yeah, I think that's about it. 524 00:35:29,617 --> 00:35:32,457 S1: Member essay my current I stack. So I'm going to 525 00:35:32,457 --> 00:35:36,457 S1: give you because you're members a current snapshot of all 526 00:35:36,457 --> 00:35:39,577 S1: my AI tools. So what I'm using for what scenarios 527 00:35:39,577 --> 00:35:42,817 S1: what purposes. So main models oh three Pro is my 528 00:35:42,857 --> 00:35:46,937 S1: main one, uh, latest clod like opus four or sonnet 529 00:35:46,937 --> 00:35:52,177 S1: four for writing things or artisti things. Those are really 530 00:35:52,177 --> 00:35:56,937 S1: my favorite. Um, I don't for coding, like if I'm 531 00:35:56,937 --> 00:35:59,777 S1: using cursor, um, or something like that, I tend to 532 00:35:59,777 --> 00:36:03,657 S1: use the latest clod models. Also, main workflows are browsing 533 00:36:03,657 --> 00:36:06,337 S1: in my browser, which is Dia, and then I could 534 00:36:06,337 --> 00:36:10,017 S1: do shift command C to copy the URL, and then 535 00:36:10,017 --> 00:36:12,377 S1: I could use P to put that into fabric on 536 00:36:12,377 --> 00:36:15,687 S1: the command line. I could also do shift command X 537 00:36:15,687 --> 00:36:18,327 S1: from dia, and it automatically goes and creates a bunch 538 00:36:18,327 --> 00:36:21,367 S1: of different summaries and different types of analysis of like 539 00:36:21,367 --> 00:36:24,727 S1: papers and stuff like that. That's all using my personal 540 00:36:25,527 --> 00:36:29,687 S1: AI infrastructure. So that's like the main automation because I 541 00:36:29,687 --> 00:36:31,927 S1: spend so much time in the browser. The browser is 542 00:36:31,927 --> 00:36:34,007 S1: the center of a lot of things. The other thing 543 00:36:34,007 --> 00:36:37,127 S1: that I do is, um, raycast. So if I do 544 00:36:37,167 --> 00:36:40,967 S1: command space to raycast and I just type any question, 545 00:36:40,967 --> 00:36:43,087 S1: if I hit tab instead of enter, if I hit enter, 546 00:36:43,087 --> 00:36:45,207 S1: it's going to search on Google. If I hit tab, 547 00:36:45,207 --> 00:36:49,207 S1: it's going to search on cloud for opus with reasoning. 548 00:36:49,647 --> 00:36:54,367 S1: And that opens up inside of raycast. And any question 549 00:36:54,367 --> 00:36:55,847 S1: that I want to follow up with, I could turn 550 00:36:55,847 --> 00:36:58,127 S1: that into a full chat. And that's within the chat 551 00:36:58,167 --> 00:37:00,887 S1: window of raycast. So raycast is kind of like my 552 00:37:00,887 --> 00:37:05,847 S1: universal AI interface, and the browser is how I get 553 00:37:05,847 --> 00:37:09,527 S1: content into my personal AI infrastructure. That's on the back end. 554 00:37:09,927 --> 00:37:12,907 S1: So those are kind of like some of my main ones. 555 00:37:13,307 --> 00:37:16,547 S1: Article summaries are I could swipe right on Feedly. I 556 00:37:16,547 --> 00:37:18,947 S1: could click the read later tag in Feedly, I could 557 00:37:18,987 --> 00:37:23,587 S1: send via Dia, or I could also, um, yeah, I 558 00:37:23,587 --> 00:37:25,787 S1: could do any of those. And it actually sends into 559 00:37:25,867 --> 00:37:29,027 S1: a Google doc, multiple Google Docs, actually with different types 560 00:37:29,027 --> 00:37:36,427 S1: of analysis and sorting. So main tools fabric raycast, OpenAI, manus. Um, grok. 561 00:37:36,467 --> 00:37:38,987 S1: Actually I actually use grok for deep research. I don't 562 00:37:39,027 --> 00:37:41,907 S1: ask it normal questions, but for deep research stuff I'll 563 00:37:41,907 --> 00:37:46,147 S1: use grok and combine that with Claude and also Gemini 564 00:37:46,187 --> 00:37:50,147 S1: and also OpenAI, O3 for really, really complex tasks. Like 565 00:37:50,147 --> 00:37:52,547 S1: I did one about AI security recently for a talk 566 00:37:52,827 --> 00:37:55,346 S1: and used all of them, uh, what I'm paying for 567 00:37:55,347 --> 00:37:58,947 S1: raycast AI, OpenAI Pro, which is 200 bucks a month, 568 00:37:58,947 --> 00:38:02,587 S1: raycast is like 30 bucks a month or something like that. Uh, Claude, 569 00:38:02,627 --> 00:38:05,587 S1: at 50 bucks a month, Gemini at 50 bucks a month. Um, 570 00:38:05,587 --> 00:38:10,647 S1: I'm not using perplexity anymore because I'm using the deep 571 00:38:10,647 --> 00:38:15,807 S1: research services on all the other AI services that I'm using. Right. 572 00:38:16,167 --> 00:38:18,247 S1: I feel like deep research just kind of replaces that 573 00:38:18,247 --> 00:38:20,927 S1: for me. And plus Google now has the AI summaries 574 00:38:20,927 --> 00:38:22,847 S1: at the top. So I usually get what I need 575 00:38:22,887 --> 00:38:26,287 S1: from a Google search. And if um, I'm not going 576 00:38:26,327 --> 00:38:28,047 S1: to use Google, if I'm going to go directly to AI, 577 00:38:28,047 --> 00:38:31,407 S1: I just use raycast for that. And I'm also paying 578 00:38:31,407 --> 00:38:33,927 S1: for a million other things as part of my different businesses. 579 00:38:33,967 --> 00:38:37,927 S1: I think I mentioned before, I'm probably spending, I don't know, 580 00:38:37,967 --> 00:38:41,607 S1: 3 to 4 grand on AI a month. That's my 581 00:38:41,647 --> 00:38:44,567 S1: guess across all the different services, including the ones I 582 00:38:44,567 --> 00:38:47,767 S1: just mentioned, but also just back end processing for my 583 00:38:47,767 --> 00:38:52,087 S1: personal AI infrastructure, but most importantly for my businesses. So 584 00:38:52,607 --> 00:38:55,127 S1: I've got same page, we've got threshold, we've got different, uh, 585 00:38:55,127 --> 00:39:00,127 S1: companies that are churning through a lot of content. All right. 586 00:39:00,487 --> 00:39:04,287 S1: Recommendation of the week. Think about the sunk cost fallacy 587 00:39:04,287 --> 00:39:08,207 S1: as it relates to your current plans for education, training, career, business, etc.. 588 00:39:08,347 --> 00:39:11,107 S1: in other words, when things are changing the fastest, like 589 00:39:11,107 --> 00:39:14,387 S1: right now, you should more frequently ask yourself whether your 590 00:39:14,387 --> 00:39:16,627 S1: current trajectory is going to get you where you need 591 00:39:16,627 --> 00:39:18,787 S1: to be in, like the next two, three, or five years. 592 00:39:18,827 --> 00:39:20,307 S1: And of course, that's hard to know because things are 593 00:39:20,307 --> 00:39:23,107 S1: changing so fast. Like nobody actually knows where we're going 594 00:39:23,107 --> 00:39:26,307 S1: to be in five years. But you've got to, like, 595 00:39:26,667 --> 00:39:29,547 S1: stake your ground somewhere, right? So just imagine where you 596 00:39:29,547 --> 00:39:31,947 S1: think it's going to be. And you have to ask yourself, 597 00:39:31,947 --> 00:39:35,707 S1: is this doing it? So try not to consider all 598 00:39:35,747 --> 00:39:38,147 S1: the work that you've already put into doing X or Y, 599 00:39:38,787 --> 00:39:41,947 S1: unless you're literally, you know, like a couple of classes 600 00:39:41,947 --> 00:39:44,907 S1: or whatever away from graduating from like a four year degree, obviously, 601 00:39:44,907 --> 00:39:48,307 S1: don't just stop that and abandon all your work. But 602 00:39:48,867 --> 00:39:50,747 S1: if you're like getting ready to start a degree or 603 00:39:50,747 --> 00:39:53,867 S1: you're like, just got started with some classes, maybe start 604 00:39:53,867 --> 00:39:57,627 S1: to think four years from now, is that four year 605 00:39:57,627 --> 00:40:00,427 S1: degree really going to be valuable relative to what else 606 00:40:00,427 --> 00:40:02,547 S1: you could have spent that time on? And it doesn't 607 00:40:02,547 --> 00:40:04,507 S1: matter if you already paid some fees. It doesn't matter 608 00:40:04,507 --> 00:40:06,467 S1: if you already bought some books. You have to think 609 00:40:06,607 --> 00:40:09,207 S1: about the sunk cost fallacy, and this is really hard 610 00:40:09,207 --> 00:40:12,287 S1: to do. It's hard for me. It's hard for Daniel Kahneman, 611 00:40:12,287 --> 00:40:14,527 S1: who I think is he the one that came up 612 00:40:14,527 --> 00:40:16,487 S1: with this? I'm not sure, but he basically said he 613 00:40:16,487 --> 00:40:19,087 S1: is a victim just like everyone else. Of all the 614 00:40:19,087 --> 00:40:21,807 S1: biases that he talked about in his book, Thinking Fast 615 00:40:21,807 --> 00:40:24,847 S1: and Slow. So you really have to guard against these things. 616 00:40:24,847 --> 00:40:27,247 S1: And sunk cost fallacy is a really bad one, and 617 00:40:27,247 --> 00:40:30,927 S1: it's especially bad when you're dealing with what should you 618 00:40:30,967 --> 00:40:33,887 S1: be doing to get ready for this future that's coming up, right? 619 00:40:33,927 --> 00:40:36,767 S1: You have to think. You have to remake yourself, or 620 00:40:36,767 --> 00:40:40,647 S1: at least consider thinking about remaking yourself every month or 621 00:40:40,647 --> 00:40:43,927 S1: every six months, or every year. Not every year. Every 622 00:40:43,927 --> 00:40:46,327 S1: month or so. You got to be thinking, really? Do 623 00:40:46,327 --> 00:40:49,687 S1: I need to be spending time and money on that? Right? 624 00:40:49,727 --> 00:40:52,607 S1: Maybe there's something better. Maybe there's something more fundamental that 625 00:40:52,607 --> 00:40:55,887 S1: makes me more resilient that I should be doing instead. 626 00:40:56,367 --> 00:40:59,727 S1: And the aphorism of the week the beginning is always today. 627 00:41:00,407 --> 00:41:05,727 S1: The beginning is always today. Mary Wollstonecraft Shelley.