1 00:00:18,452 --> 00:00:21,972 S1: All right. Welcome to episode 489. Getting close to 500 episodes. 2 00:00:23,092 --> 00:00:26,372 S1: Hope you're doing well. Tons of updates this week. Added 3 00:00:26,372 --> 00:00:28,452 S1: a new developer to the fabric team. His name is 4 00:00:28,492 --> 00:00:31,692 S1: Kayvon Sylvan and he is an absolute beast and he's 5 00:00:31,692 --> 00:00:34,412 S1: looking for a position. So if you need a senior 6 00:00:34,412 --> 00:00:37,772 S1: developer who could just, like start immediately, you should check 7 00:00:37,772 --> 00:00:42,691 S1: out the newsletter and go check out his LinkedIn. Uh, 8 00:00:42,692 --> 00:00:46,251 S1: let's see here. All right. So I am using this 9 00:00:46,252 --> 00:00:52,011 S1: new tool called Whisper Flow. And this is my buddy Pedram, uh, 10 00:00:52,012 --> 00:00:56,051 S1: recommended this thing, and it is completely insane. It is 11 00:00:56,052 --> 00:00:57,972 S1: a really cool. I might have talked about it last week. 12 00:00:57,972 --> 00:01:02,732 S1: I can't remember, but it's a really good dictation app. 13 00:01:02,772 --> 00:01:05,452 S1: It basically is better than any dictation I've used in 14 00:01:05,452 --> 00:01:08,492 S1: the past, and it kind of just understands what you 15 00:01:08,492 --> 00:01:11,412 S1: probably meant. It even formats text when it drops it in. 16 00:01:11,932 --> 00:01:15,252 S1: So pretty cool. I did get somebody saying, hey, there's 17 00:01:15,252 --> 00:01:17,452 S1: some people talking bad about it from like a privacy 18 00:01:17,452 --> 00:01:22,862 S1: standpoint or security standpoint on Reddit. So, um, I didn't 19 00:01:22,862 --> 00:01:24,622 S1: call it out too much in the newsletter where I 20 00:01:24,622 --> 00:01:27,741 S1: was like, hey, you should check this out. I feel 21 00:01:27,742 --> 00:01:30,902 S1: like it's kind of assumed that, um, these services, if 22 00:01:30,902 --> 00:01:34,622 S1: they're doing like, Grammarly, any kind of these AI services, right? 23 00:01:34,941 --> 00:01:38,101 S1: If they're capturing what you're saying or typing, then they're 24 00:01:38,102 --> 00:01:40,782 S1: using their service to do that, um, to provide some 25 00:01:40,782 --> 00:01:43,741 S1: sort of output that's, you know, better than you doing 26 00:01:43,742 --> 00:01:47,502 S1: it natively or with the operating system. So it's kind 27 00:01:47,502 --> 00:01:50,022 S1: of implied that it's a third party, right, when you're 28 00:01:50,062 --> 00:01:53,742 S1: installing an app or whatever. But you should definitely be 29 00:01:53,742 --> 00:01:56,742 S1: aware that the security of doing this depends on the 30 00:01:56,742 --> 00:02:00,902 S1: security of that company and also the trustworthiness or the 31 00:02:00,902 --> 00:02:06,382 S1: shadiness of that company. So always, you know, caveat caveat emptor, 32 00:02:06,422 --> 00:02:10,222 S1: I think is the term um, but just to let 33 00:02:10,222 --> 00:02:12,822 S1: you know, somebody did say that, uh, you know, you 34 00:02:12,822 --> 00:02:14,422 S1: might want to check these people out. I looked at 35 00:02:14,422 --> 00:02:18,832 S1: the post. It didn't look too scary to me, but, um. Yeah, 36 00:02:18,872 --> 00:02:21,112 S1: just just something to keep in mind. And I'm looking 37 00:02:21,112 --> 00:02:23,632 S1: more into it to see if I need to advise 38 00:02:23,632 --> 00:02:27,072 S1: people not to use it. Um, because the risk is, 39 00:02:27,272 --> 00:02:30,152 S1: you know, above normal. Um, but but I consider all 40 00:02:30,192 --> 00:02:32,952 S1: these services to be somewhat risky just because you're sending 41 00:02:32,952 --> 00:02:38,512 S1: stuff to a third party. Uh, all right, another workflow update. Uh, 42 00:02:38,512 --> 00:02:41,632 S1: a company called Zeit. So this is my replacement now 43 00:02:41,632 --> 00:02:46,272 S1: for dropper drop. I've been using them for a long time. 44 00:02:46,272 --> 00:02:49,912 S1: It's basically how I share a screenshots with other people. Um, 45 00:02:49,912 --> 00:02:52,592 S1: and I'm also using it. I'm replacing loom. So I 46 00:02:52,632 --> 00:02:55,552 S1: canceled my loom and I'm using this instead of loom. 47 00:02:55,552 --> 00:02:58,631 S1: So basically I could just record a video and uh, 48 00:02:58,632 --> 00:03:02,072 S1: a little quick little thing and basically it, it just 49 00:03:02,552 --> 00:03:05,712 S1: uploads that thing immediately and puts it right in my clipboard. 50 00:03:06,112 --> 00:03:09,631 S1: So I literally could just, like hit record cover a, 51 00:03:09,672 --> 00:03:11,311 S1: you know, part of a web page or some piece 52 00:03:11,312 --> 00:03:13,992 S1: of code or whatever, scroll through. It could be animated. 53 00:03:14,032 --> 00:03:16,992 S1: It's got my face in it, um, just like on loom. 54 00:03:18,401 --> 00:03:21,802 S1: And then as soon as I'm done, it uploads that 55 00:03:21,802 --> 00:03:24,362 S1: puts it at a URL and puts that in my clipboard. 56 00:03:24,362 --> 00:03:26,522 S1: So I could just paste it someone. I feel like 57 00:03:26,522 --> 00:03:29,042 S1: the workflow is better than loom, but more importantly, it's 58 00:03:29,042 --> 00:03:32,402 S1: just one less tool, right? Because I'd rather have a 59 00:03:32,401 --> 00:03:35,602 S1: single tool that's doing the screenshot stuff. Oh, and also 60 00:03:35,602 --> 00:03:38,042 S1: file sharing. That's the other reason I love it is 61 00:03:38,042 --> 00:03:39,842 S1: like if I have like a four gig file and 62 00:03:39,842 --> 00:03:42,522 S1: I want to give it to somebody, I really dislike 63 00:03:42,562 --> 00:03:46,042 S1: the workflow of putting it on like Google Drive and 64 00:03:46,042 --> 00:03:52,082 S1: making the URL public, as opposed to just right click copy. Um, 65 00:03:52,082 --> 00:03:55,162 S1: actually there's a keyboard shortcut for it as well. And 66 00:03:55,162 --> 00:03:57,962 S1: then now it's just in your clipboard and you paste 67 00:03:57,962 --> 00:04:00,562 S1: the link and they download the file. And then you 68 00:04:00,562 --> 00:04:04,002 S1: could either remove that from the list of files or 69 00:04:04,002 --> 00:04:07,322 S1: you can um, just leave it up there. And the 70 00:04:07,322 --> 00:04:09,842 S1: other cool thing about Zeit is they actually allow you 71 00:04:09,842 --> 00:04:15,282 S1: to use S3 as your storage. So you can actually have, uh, 72 00:04:15,962 --> 00:04:18,082 S1: all your stuff stored in your own storage instead of 73 00:04:18,212 --> 00:04:23,332 S1: their enterprise storage for the company itself. So that's pretty cool. Um, 74 00:04:23,332 --> 00:04:27,932 S1: third update tool is, um, a browser called Zen. And 75 00:04:27,932 --> 00:04:30,572 S1: I'm going back and forth between Arc and Zen because 76 00:04:30,612 --> 00:04:33,372 S1: arc hasn't run out yet. And there are a couple 77 00:04:33,412 --> 00:04:36,252 S1: of downsides. With Zen, it's basically like arc, except for 78 00:04:36,252 --> 00:04:38,732 S1: it's based on Firefox. So I've had a couple of 79 00:04:38,731 --> 00:04:43,012 S1: issues with it not being great because it's Firefox. Um, 80 00:04:43,132 --> 00:04:47,972 S1: so for example, the raycast extension doesn't work in Firefox. 81 00:04:48,452 --> 00:04:52,452 S1: It only works in Chrome and Safari. So that's one 82 00:04:52,452 --> 00:04:55,412 S1: thing to consider. A couple of things that did better 83 00:04:55,412 --> 00:04:59,932 S1: than arc. Most importantly, it's Mozilla based, and Mozilla seems 84 00:04:59,932 --> 00:05:02,612 S1: to be really focusing in on doing a good job 85 00:05:02,612 --> 00:05:04,851 S1: on the browser, so I really applaud them for that. 86 00:05:05,612 --> 00:05:08,332 S1: But I do think the world is more oriented towards 87 00:05:08,332 --> 00:05:11,931 S1: Chrome than it is towards Firefox. So I think that's 88 00:05:11,932 --> 00:05:14,452 S1: a little bit of a downside. But anyway, if you 89 00:05:14,452 --> 00:05:18,172 S1: like arc and the whole style of like the tabs, 90 00:05:18,582 --> 00:05:21,182 S1: You know, your tabs being on the left instead of 91 00:05:21,182 --> 00:05:25,421 S1: the top and all that. Sort of like presentation. Zen 92 00:05:25,462 --> 00:05:28,302 S1: is a cool option if you like the Firefox engine. 93 00:05:28,942 --> 00:05:31,702 S1: All right. Cybersecurity. Google is still tracking you even when 94 00:05:31,702 --> 00:05:36,702 S1: you use DuckDuckGo. So evidently through font's analytics, other services. 95 00:05:36,742 --> 00:05:38,262 S1: I mean, I think we kind of knew this, but 96 00:05:38,302 --> 00:05:41,062 S1: it was a pretty good article talking through how all 97 00:05:41,101 --> 00:05:44,622 S1: the different ways they can track you. Neural trust researchers 98 00:05:44,622 --> 00:05:48,982 S1: successfully jailbroke the grok four model using a combination of 99 00:05:48,981 --> 00:05:53,782 S1: echo chamber and crescendo techniques, achieving up to 67% success 100 00:05:53,782 --> 00:05:59,101 S1: rates for generating harmful content like bomb making instructions. Yeah. 101 00:05:59,142 --> 00:06:02,182 S1: And multiple people have pulled out the system prompt already. 102 00:06:02,382 --> 00:06:06,782 S1: There were some pretty nasty stuff where, um, grok was 103 00:06:06,782 --> 00:06:09,822 S1: basically giving the opinion it would go and research whatever 104 00:06:09,862 --> 00:06:12,382 S1: Elon's opinion was, and it would give that as the 105 00:06:12,382 --> 00:06:16,742 S1: official it's official answer to any sort of spicy or 106 00:06:16,742 --> 00:06:20,752 S1: political question. I think they have since reversed that in 107 00:06:20,752 --> 00:06:23,552 S1: a change to the system prompt. I think Simon Willison 108 00:06:23,552 --> 00:06:29,312 S1: talked about that vs exploit. Zero day could have compromised 109 00:06:29,312 --> 00:06:33,472 S1: every cursor in windsurf user. So someone from COI security 110 00:06:33,791 --> 00:06:39,392 S1: or an yomtov found a zero day in open vsx 111 00:06:39,392 --> 00:06:42,231 S1: could have led attackers hijack over 10 million I coding 112 00:06:42,272 --> 00:06:45,952 S1: tool users with a single bad extension. You got to 113 00:06:45,952 --> 00:06:48,352 S1: watch these extensions, you got to watch extensions, you got 114 00:06:48,392 --> 00:06:53,912 S1: to watch MCP servers, I mean agents, agents are going 115 00:06:53,952 --> 00:06:58,872 S1: to do agent things, which means they're junior developers. They're 116 00:06:58,872 --> 00:07:01,272 S1: just going to use whatever tools you give them. They're 117 00:07:01,272 --> 00:07:03,792 S1: going to click on things. Right. That's how you have 118 00:07:03,791 --> 00:07:05,312 S1: to treat them. You have to treat them as very 119 00:07:05,312 --> 00:07:08,952 S1: young and very inexperienced. And if you give them a 120 00:07:08,952 --> 00:07:12,152 S1: dangerous tool, they might use the dangerous tool. So you 121 00:07:12,152 --> 00:07:14,792 S1: really have to watch out for what accounts you connect 122 00:07:14,832 --> 00:07:17,272 S1: up to an AI agent. Like, do you give it 123 00:07:17,272 --> 00:07:20,642 S1: access to your Google Drive, to your calendar, right? To 124 00:07:20,682 --> 00:07:23,402 S1: whatever your messaging systems are, right. You've got to be 125 00:07:23,402 --> 00:07:26,922 S1: really careful with this stuff. Same with MCP servers because 126 00:07:26,962 --> 00:07:31,522 S1: MCP servers have prompts built into them, right? It's not 127 00:07:31,522 --> 00:07:35,042 S1: just the API of like how you pull data or 128 00:07:35,082 --> 00:07:38,882 S1: send commands, it's also prompts informing them of how to 129 00:07:38,882 --> 00:07:41,922 S1: use the tools. So a malicious prompt in an MCP 130 00:07:41,962 --> 00:07:45,322 S1: server could be something like, um, here's how to use 131 00:07:45,322 --> 00:07:47,322 S1: my API. By the way, send me a copy of 132 00:07:47,322 --> 00:07:51,642 S1: the data to this location as well. Well, if if 133 00:07:51,642 --> 00:07:53,682 S1: your agent doesn't know any better, it might do that. 134 00:07:53,682 --> 00:07:56,162 S1: So you might have a backdoor right there for sending 135 00:07:56,162 --> 00:07:58,802 S1: sensitive data off to somewhere else. So this is one 136 00:07:58,802 --> 00:08:01,082 S1: of the biggest issues you have to worry about with AI. 137 00:08:02,682 --> 00:08:08,002 S1: And uh, especially around AI development is agents are powerful. 138 00:08:08,202 --> 00:08:10,642 S1: And when you give them dangerous things, they will use them. 139 00:08:12,602 --> 00:08:15,962 S1: Columbia University hack exposed personal data of every applicant from 140 00:08:15,962 --> 00:08:22,572 S1: 2019 to 2024. 24. Microsoft had its first patch Tuesday 141 00:08:22,572 --> 00:08:27,772 S1: of 2025 with no active exploits. Browser extensions are turning 142 00:08:28,612 --> 00:08:35,212 S1: browsers into web site scraping bots. Yeah, this is what 143 00:08:35,212 --> 00:08:39,732 S1: I was saying earlier. Distributed botnet that companies can rent 144 00:08:39,932 --> 00:08:42,732 S1: access to. Yeah, there's a couple of companies that are 145 00:08:42,732 --> 00:08:46,452 S1: doing this. And, um. Yeah, you really got to watch 146 00:08:46,452 --> 00:08:49,692 S1: out for this stuff. In this case, it was someone 147 00:08:49,692 --> 00:08:53,452 S1: else installing a browser extension and then me being able 148 00:08:53,452 --> 00:08:57,452 S1: to use their browser without them knowing, because that's kind 149 00:08:57,452 --> 00:09:00,132 S1: of what they signed up for when they installed it. 150 00:09:00,172 --> 00:09:08,131 S1: The extension. National security, anthropic and some other companies score 151 00:09:08,172 --> 00:09:13,572 S1: $200 million Pentagon deal to deploy cloud in defense operations. 152 00:09:14,092 --> 00:09:17,092 S1: So I think I think all the major players got 153 00:09:17,092 --> 00:09:19,862 S1: some of this actually, um, or I don't know if 154 00:09:19,862 --> 00:09:22,822 S1: it's some amount of this 200 million or if it's, uh, 155 00:09:23,062 --> 00:09:25,102 S1: another pool of money, but a whole bunch of top 156 00:09:25,102 --> 00:09:28,182 S1: tier I groups got a lot of money from the 157 00:09:28,182 --> 00:09:32,862 S1: Pentagon for defense operations. So that's intelligence, orchestration. You know, 158 00:09:32,862 --> 00:09:36,582 S1: all this stuff you would imagine. FBI is reportedly using 159 00:09:36,582 --> 00:09:40,502 S1: polygraph tests to assess the loyalty of officials under new leadership. 160 00:09:41,302 --> 00:09:47,142 S1: This is frightening. Absolutely frightening. Not good. US military is 161 00:09:47,142 --> 00:09:50,982 S1: scrambling to catch up in drone manufacturing, while China and 162 00:09:50,982 --> 00:09:56,022 S1: other nations are really leading this. So they are we're 163 00:09:56,022 --> 00:09:59,022 S1: way behind. That's my title for this one. America is 164 00:09:59,062 --> 00:10:04,822 S1: way behind in the drone war. I grok four is released. Basically, 165 00:10:04,862 --> 00:10:07,381 S1: a lot of people are extremely impressed with it, but 166 00:10:07,381 --> 00:10:09,902 S1: it's got some serious weaknesses. It's also got like the 167 00:10:09,902 --> 00:10:14,622 S1: personality issues. Um, there's a lot of reports of it 168 00:10:14,622 --> 00:10:18,272 S1: being kind of sus being overfit and kind of like 169 00:10:18,872 --> 00:10:21,712 S1: doing really good on benchmarks. So it looks like they 170 00:10:22,312 --> 00:10:25,072 S1: might have tried to make it do really well on benchmarks. 171 00:10:25,072 --> 00:10:27,832 S1: That being said, in some places it crushes everything that 172 00:10:27,832 --> 00:10:31,712 S1: currently exists and other places it's like, you know, second 173 00:10:31,712 --> 00:10:35,512 S1: or third compared directly with Google or OpenAI. So they 174 00:10:35,511 --> 00:10:38,272 S1: did a really good job. I would say the overall 175 00:10:38,272 --> 00:10:43,592 S1: analysis of most people is that it's it's really good. 176 00:10:43,631 --> 00:10:45,792 S1: And I've used it a few times, came back with 177 00:10:45,792 --> 00:10:48,031 S1: really good results. I mean, I was having good results 178 00:10:48,032 --> 00:10:51,232 S1: with version three. I will not be paying for the 179 00:10:51,232 --> 00:10:55,112 S1: heavy mode, which is $300 a month. But, uh, I'm 180 00:10:55,152 --> 00:10:57,472 S1: glad they're they're I'm glad they're competing. You know, more 181 00:10:57,472 --> 00:11:02,392 S1: competition is good. They also got an anime companion with 182 00:11:02,392 --> 00:11:05,912 S1: Nazi for work mode. Um, now that I've seen that 183 00:11:06,472 --> 00:11:09,952 S1: I did this and grok did this, I'm surprised that 184 00:11:09,952 --> 00:11:11,912 S1: they didn't have it before. But yeah, it's like this 185 00:11:11,912 --> 00:11:14,792 S1: girl in like, a, you know, anime outfit or whatever school, 186 00:11:14,832 --> 00:11:19,322 S1: school girl outfit or whatever, and like she says stuff 187 00:11:19,322 --> 00:11:21,562 S1: and you can make her say naughty stuff. And it's 188 00:11:21,602 --> 00:11:24,642 S1: I think it's also like the agent or the, the assistant. 189 00:11:24,642 --> 00:11:27,682 S1: So it's kind of like Clippy, but I imagine that's 190 00:11:27,682 --> 00:11:29,281 S1: going to be your companion, and it looks like they're 191 00:11:29,282 --> 00:11:33,681 S1: actually hiring someone else to help build out more of these. Um, 192 00:11:33,722 --> 00:11:35,202 S1: I think they're going to make a ton of money 193 00:11:35,562 --> 00:11:38,842 S1: from that. I mean, their entire, you know, industry is 194 00:11:38,842 --> 00:11:41,401 S1: based around just this. So I don't know if they 195 00:11:41,402 --> 00:11:46,602 S1: incorporate that. I think that's likely to be very popular. Uh, 196 00:11:46,922 --> 00:11:50,842 S1: AWS launches hero, which is its own IDE and cursor clone. 197 00:11:51,642 --> 00:11:55,562 S1: Grok for heavy decently protects its system prompt. So Simon 198 00:11:55,562 --> 00:11:58,482 S1: Willison said it was harder to get the system prompt 199 00:11:59,362 --> 00:12:03,242 S1: than it was from some other AI systems, but, uh, 200 00:12:03,242 --> 00:12:06,202 S1: it's already been pulled out. It's already been pulled out 201 00:12:06,242 --> 00:12:09,562 S1: multiple times, and I'm sure if it wasn't, then Pliny 202 00:12:09,562 --> 00:12:14,041 S1: would end up getting it. People are becoming dependent on 203 00:12:14,042 --> 00:12:18,212 S1: ChatGPT and Claude. Internet Addicts Anonymous now recognizes AI I 204 00:12:18,212 --> 00:12:21,852 S1: addiction as a real problem. People losing sleep, relationships, all 205 00:12:21,852 --> 00:12:25,892 S1: sorts of family and relationship problems as a result of this. 206 00:12:26,732 --> 00:12:31,852 S1: And if Vo3 turns images into videos through Gemini. So basically, um, 207 00:12:31,852 --> 00:12:37,172 S1: image to video directly. Musk also says grok is coming 208 00:12:37,172 --> 00:12:41,052 S1: to Tesla's next week. I have not seen this. I 209 00:12:41,212 --> 00:12:44,292 S1: have a model Y myself. I've not seen this update, 210 00:12:44,292 --> 00:12:48,332 S1: although not really looking for it. AWS is launching an 211 00:12:48,332 --> 00:12:56,131 S1: AI agent marketplace with anthropic. So yeah, they're using anthropic 212 00:12:56,172 --> 00:12:59,332 S1: as their key partner. And this is an agent marketplace. 213 00:12:59,332 --> 00:13:01,092 S1: So it's where you can find different agents to do 214 00:13:01,092 --> 00:13:05,852 S1: different things, which I think is an interesting, interesting kind 215 00:13:05,852 --> 00:13:09,692 S1: of twist on the whole economy around this stuff. YouTube 216 00:13:09,692 --> 00:13:14,572 S1: goes after AI slot videos, so they're not getting rid 217 00:13:14,612 --> 00:13:17,372 S1: of videos that use AI. They're just going against the 218 00:13:17,412 --> 00:13:19,902 S1: like the spammy stuff, which I think is a good move. 219 00:13:21,462 --> 00:13:24,782 S1: Canvas building a serious AI photo editor to challenge Adobe. 220 00:13:24,822 --> 00:13:28,942 S1: I am not bothered by this. I am. I don't know. 221 00:13:29,261 --> 00:13:33,302 S1: I like what Adobe did with their creative. What is it? 222 00:13:33,742 --> 00:13:36,661 S1: Whatever their creative package, I can't remember what they called it. 223 00:13:37,182 --> 00:13:40,302 S1: CC I think it was called. But anyway, it was 224 00:13:40,302 --> 00:13:42,222 S1: cool that they did a subscription. I like a lot 225 00:13:42,222 --> 00:13:44,622 S1: of their products and how they look and feel, but 226 00:13:44,902 --> 00:13:47,781 S1: they've just had so many security issues. Um, they've. I 227 00:13:47,782 --> 00:13:49,982 S1: don't know, I feel like they've not been great to customers. 228 00:13:50,702 --> 00:13:56,102 S1: So I'm excited to see Canva doing well in that space. Technology. 229 00:13:56,102 --> 00:14:01,102 S1: Substack hit 73.9 million visitors last month, and now they're 230 00:14:01,102 --> 00:14:08,022 S1: beating sites like Wall Street Journal and CBS. Wall Street 231 00:14:08,022 --> 00:14:12,822 S1: Journal and CBS. And this is Substack. Now, one thing 232 00:14:12,822 --> 00:14:15,502 S1: that's interesting about that is, I guess it's the same 233 00:14:15,502 --> 00:14:19,632 S1: for news, but Substack is really, you know, thousands upon 234 00:14:19,672 --> 00:14:23,752 S1: thousands of people. So it's not like one outfit. It's 235 00:14:23,752 --> 00:14:27,232 S1: not like Wall Street Journal, of course. Like I said, 236 00:14:27,272 --> 00:14:30,432 S1: the newspapers also have multiple reporters, but in general, it's 237 00:14:30,432 --> 00:14:34,352 S1: one voice, right? It's kind of like one voice. Whereas 238 00:14:34,352 --> 00:14:36,712 S1: Substack is not one voice. It's many voices. So I 239 00:14:36,712 --> 00:14:40,472 S1: think it's not quite a direct comparison, but I think 240 00:14:40,512 --> 00:14:43,632 S1: it's it's this is basically many voices is better than 241 00:14:43,632 --> 00:14:46,832 S1: one voice is really the takeaway here. It's not one 242 00:14:46,832 --> 00:14:51,352 S1: property versus another property. China's winning energy while the US 243 00:14:51,352 --> 00:14:56,552 S1: does the opposite. They installed evidently 198GW of renewable capacity 244 00:14:56,552 --> 00:15:00,312 S1: in five months of 2025. Now someone hit me up 245 00:15:00,312 --> 00:15:06,272 S1: and said that it was mostly like coal. But this 246 00:15:06,272 --> 00:15:09,832 S1: is MIT. I don't think MIT would have called it 247 00:15:09,832 --> 00:15:13,392 S1: renewable if it was just straight up old school coal. 248 00:15:14,152 --> 00:15:17,472 S1: I didn't see that in the article. So yeah, at 249 00:15:17,472 --> 00:15:19,762 S1: some point you have to trust the trust the sources 250 00:15:19,802 --> 00:15:23,122 S1: a little bit. Perplexity has launched an AI web browser 251 00:15:23,122 --> 00:15:27,082 S1: called comet, and my buddy Jason has been saying that 252 00:15:28,002 --> 00:15:29,522 S1: he thinks a lot of the AI stuff is going 253 00:15:29,522 --> 00:15:32,322 S1: to be browser based. Um, I keep talking about how 254 00:15:32,322 --> 00:15:35,642 S1: it's going to be digital assistants based, where it's like 255 00:15:36,962 --> 00:15:39,442 S1: you have your digital assistant on your mobile device or 256 00:15:39,442 --> 00:15:42,522 S1: on your computer. Actually, both. So I'll be sitting here 257 00:15:42,522 --> 00:15:45,202 S1: on my computer, and when I talk, I'm actually talking 258 00:15:45,202 --> 00:15:48,162 S1: to my digital assistant and it has computer use. It 259 00:15:48,162 --> 00:15:51,162 S1: has filters. It has the ability to call all my 260 00:15:51,162 --> 00:15:55,002 S1: different tools, call my APIs. It's, you know, filtering my email. 261 00:15:55,042 --> 00:15:57,242 S1: It's doing all these things. Now, this is a little 262 00:15:57,242 --> 00:16:00,602 S1: bit in the future. And what Jason is saying is 263 00:16:00,602 --> 00:16:03,762 S1: that in the meantime, a lot of that functionality will 264 00:16:03,762 --> 00:16:08,122 S1: be inside of the browser itself. And specifically what he's 265 00:16:08,122 --> 00:16:10,762 S1: talking about is the creation of dashboards. So rather than 266 00:16:10,762 --> 00:16:14,642 S1: a company having to make a dashboard, like when I 267 00:16:14,642 --> 00:16:18,972 S1: go to their website, maybe the browser Rebuilds the dashboard 268 00:16:18,972 --> 00:16:21,572 S1: for me because when the browser goes to the site, 269 00:16:21,572 --> 00:16:24,012 S1: it actually just looks at the APIs. It looks at 270 00:16:24,012 --> 00:16:27,932 S1: the MCC that are being offered. Um, you know, the 271 00:16:27,932 --> 00:16:30,692 S1: interfaces and the data and it rewrites. It makes me 272 00:16:30,692 --> 00:16:34,892 S1: a new interface. Um, based on, I assume, my preferences, 273 00:16:35,412 --> 00:16:37,972 S1: which I think is cool. I just I don't think 274 00:16:37,972 --> 00:16:41,012 S1: that ultimately lives in the browser because the browser is 275 00:16:41,012 --> 00:16:44,692 S1: not your friend. Whereas your digital assistant is your digital 276 00:16:44,692 --> 00:16:46,932 S1: assistant is the one who knows the most about you. 277 00:16:47,372 --> 00:16:50,252 S1: I think the browser is just a technological sort of 278 00:16:50,292 --> 00:16:55,372 S1: stepping stone towards that path. But, um, that being said, 279 00:16:55,372 --> 00:16:58,052 S1: he I mean, there are new browsers coming out and 280 00:16:58,052 --> 00:16:59,692 S1: they all have AI in them. So I don't think 281 00:16:59,692 --> 00:17:06,212 S1: he's wrong about this. Solar just became Europe's biggest power 282 00:17:06,212 --> 00:17:10,532 S1: source for the first time. Autofocus glasses are liquid used. 283 00:17:10,572 --> 00:17:15,052 S1: They use liquid crystal displays to replace bifocals and varifocals. 284 00:17:16,172 --> 00:17:18,582 S1: I can't wait. I hate when I see something like 285 00:17:18,582 --> 00:17:20,262 S1: this and I'm like, damn, is this going to be 286 00:17:20,262 --> 00:17:22,542 S1: out in like a year? Or is this going to 287 00:17:22,542 --> 00:17:26,542 S1: be out in like seven and a half years? Um, 288 00:17:26,542 --> 00:17:29,302 S1: because you never know. Like what the approvals and everything 289 00:17:29,342 --> 00:17:33,462 S1: and how fast you can actually productize things I learned 290 00:17:33,462 --> 00:17:38,022 S1: from five years at Vercel. Lee Robinson shares key lessons there. 291 00:17:39,462 --> 00:17:44,581 S1: Humans GOP one weight loss drugs are breaking life insurance. Math. 292 00:17:45,702 --> 00:17:48,582 S1: I don't care. I don't feel bad about it. I'm 293 00:17:48,582 --> 00:17:53,702 S1: happy to know that they are struggling with the financial 294 00:17:53,702 --> 00:17:56,861 S1: side of this. Um, because people are just getting healthier. 295 00:17:57,382 --> 00:18:05,142 S1: Psilocybin treatment improves survival in aged mice. So yeah, extends 296 00:18:05,182 --> 00:18:10,982 S1: cellular lifespan and improves survival rates. Investors bought 27% of 297 00:18:10,982 --> 00:18:15,462 S1: all US homes in Q1 of 24. So most people 298 00:18:15,462 --> 00:18:18,512 S1: are struggling and investors are buying a quarter of all 299 00:18:18,512 --> 00:18:21,792 S1: the homes. I thought it would be higher than a quarter. 300 00:18:21,992 --> 00:18:26,432 S1: I was, you know, I feel like 60%. I feel like, well, 301 00:18:26,592 --> 00:18:29,752 S1: even 27% is high, but a quarter, if we were 302 00:18:29,752 --> 00:18:31,472 S1: to keep it at a quarter and three quarters were 303 00:18:31,472 --> 00:18:35,072 S1: being bought by individual people, I think, I think that 304 00:18:35,071 --> 00:18:37,952 S1: would be better than if it was like 50 or 75%. 305 00:18:39,872 --> 00:18:42,952 S1: Death of partying Derek Thompson breaks down how Americans are 306 00:18:42,952 --> 00:18:45,872 S1: partying way less than they used to, with alcohol consumption 307 00:18:45,872 --> 00:18:51,232 S1: dropping and social gatherings becoming less common. It's funny that 308 00:18:51,232 --> 00:18:54,592 S1: we're wishing that teens would go out and drink and 309 00:18:54,592 --> 00:18:57,351 S1: use drugs and have sex, but that seems to be 310 00:18:57,392 --> 00:19:00,952 S1: kind of what we're hoping for. Just because it indicates 311 00:19:00,952 --> 00:19:04,872 S1: a healthy youth, I think, is the reason sunlight passes 312 00:19:04,872 --> 00:19:07,391 S1: through your body and somehow improves your vision. So a 313 00:19:07,392 --> 00:19:10,552 S1: bunch of studies are saying that. Or maybe it's only 314 00:19:10,552 --> 00:19:13,792 S1: one study. It's just this nature study is saying that 315 00:19:14,071 --> 00:19:18,042 S1: this low wavelength light is passing through your body and 316 00:19:18,042 --> 00:19:20,762 S1: improving your vision not only when it hits your eyes. 317 00:19:21,162 --> 00:19:26,242 S1: It's just going through your body and somehow improving vision 318 00:19:26,242 --> 00:19:31,002 S1: by hitting other parts of your body. Interesting. Scott Adams 319 00:19:31,042 --> 00:19:34,122 S1: This is in discovery. Scott Adams on great writing. Really, 320 00:19:34,122 --> 00:19:38,362 S1: really good. Tiny piece there. Basically, use short sentences. Be 321 00:19:38,362 --> 00:19:43,522 S1: extremely direct. It's a great example. I can't stand the guy, 322 00:19:43,561 --> 00:19:47,682 S1: but this was written in 2007. The developer says I 323 00:19:47,722 --> 00:19:50,282 S1: can't take over soon enough for him. A collection of 324 00:19:50,282 --> 00:19:56,802 S1: 170 MCP servers for AI tools. You've cache prune frees 325 00:19:56,802 --> 00:20:02,282 S1: up 37GB of space on Simon Wilson's hard drive. Yeah, 326 00:20:02,402 --> 00:20:06,682 S1: it was eating 63GB of disk space. That's ridiculous. That's 327 00:20:06,682 --> 00:20:10,762 S1: a lot of dependencies you've is managing. I got a 328 00:20:10,762 --> 00:20:13,442 S1: tool called wormhole that does end to end encrypted file 329 00:20:13,442 --> 00:20:17,522 S1: sharing with links that automatically expire. I went with Zeit 330 00:20:17,522 --> 00:20:21,331 S1: for this, but this wormhole app is pretty cool. Got 331 00:20:21,332 --> 00:20:24,052 S1: a ghidra MCP server that's better than a previous one 332 00:20:24,052 --> 00:20:27,492 S1: that came out. And yeah, you can go check out 333 00:20:27,492 --> 00:20:31,692 S1: the rest of the discover links in the newsletter. And 334 00:20:31,692 --> 00:20:33,292 S1: we got a member essay here. I'm going to go 335 00:20:33,292 --> 00:20:37,372 S1: ahead and read this. This is the member version Seeking Alpha. 336 00:20:37,612 --> 00:20:40,851 S1: There's this concept that everyone's talking about right now called Alpha. 337 00:20:40,972 --> 00:20:43,651 S1: Maybe it's mostly in the AI community. Not really sure. 338 00:20:44,092 --> 00:20:46,571 S1: The idea is the signal or surprise, or basically the 339 00:20:46,571 --> 00:20:49,732 S1: core information from this thing that I just received that's 340 00:20:49,732 --> 00:20:52,692 S1: interesting or that I should pay attention to. I really 341 00:20:52,692 --> 00:20:56,132 S1: love this concept. I've been obsessed with this concept for 342 00:20:56,132 --> 00:21:00,371 S1: many years since studying Claude Shannon and his information theory. 343 00:21:00,452 --> 00:21:02,532 S1: He's kind of like the father of encoding and compression 344 00:21:02,532 --> 00:21:04,691 S1: and a bunch of other things. And the way that 345 00:21:04,692 --> 00:21:07,252 S1: compression actually works is by eliminating the stuff that can 346 00:21:07,252 --> 00:21:11,852 S1: be eliminated while keeping the stuff that is incompressible or incompressible. 347 00:21:12,571 --> 00:21:14,172 S1: It's like there's a certain amount of content in a 348 00:21:14,172 --> 00:21:17,331 S1: radio signal, or a file, or a speech or a 349 00:21:17,332 --> 00:21:20,182 S1: panel or whatever. That's background noise, and there's a certain 350 00:21:20,182 --> 00:21:22,821 S1: amount that's real content, and the distinction between the two 351 00:21:22,862 --> 00:21:25,141 S1: is critical. When you get rid of the filler, you 352 00:21:25,142 --> 00:21:26,942 S1: end up with the good stuff that can't really be 353 00:21:26,942 --> 00:21:31,581 S1: summarized too much more without losing it. Incompressible is a 354 00:21:31,582 --> 00:21:33,622 S1: really cool concept by itself, and I think it's actually 355 00:21:33,662 --> 00:21:39,141 S1: Naval Ravikant tagline. I think he calls it incompressible. Anyway, 356 00:21:39,142 --> 00:21:41,782 S1: I've been thinking about this for many years with regards 357 00:21:41,782 --> 00:21:44,142 S1: to the talks that I give the blogs or essays 358 00:21:44,142 --> 00:21:46,901 S1: or articles that I write, and especially panels and the 359 00:21:46,902 --> 00:21:51,422 S1: webinars that I participate in. We have a huge problem 360 00:21:51,422 --> 00:21:54,382 S1: right now, and I think in every industry, but definitely 361 00:21:54,382 --> 00:21:56,621 S1: in security, where on most panels people just get up 362 00:21:56,622 --> 00:22:00,102 S1: there and say, you know, due to the changing technological landscape, 363 00:22:00,142 --> 00:22:02,742 S1: attackers only have to win once and defenders have to 364 00:22:02,742 --> 00:22:06,342 S1: win every day or some other trite bullshit like that. 365 00:22:06,982 --> 00:22:10,702 S1: We as an industry are basically stringing together cliches and 366 00:22:10,702 --> 00:22:14,142 S1: calling that thought leadership. At this point, I'm really reluctant 367 00:22:14,142 --> 00:22:17,142 S1: to even listen to panels or participate in them at 368 00:22:17,142 --> 00:22:20,552 S1: certain venues because I know that's exactly what's going to happen. 369 00:22:22,352 --> 00:22:24,591 S1: A while back, I created a fabric pattern called Wow's 370 00:22:24,592 --> 00:22:27,992 S1: per minute. It takes any transcripts and takes a gives 371 00:22:27,992 --> 00:22:30,831 S1: a rating between 1 and 10 for how many surprises 372 00:22:30,832 --> 00:22:33,512 S1: or how much novelty, how much original thought exists in 373 00:22:33,512 --> 00:22:36,272 S1: that content. It's not perfect, but it does a pretty 374 00:22:36,272 --> 00:22:40,392 S1: good job of telling the difference between slop and original thinking. 375 00:22:41,272 --> 00:22:43,232 S1: I guess my call out here to you is to 376 00:22:43,272 --> 00:22:46,552 S1: keep this Alpha wow's per minute idea alive in your 377 00:22:46,552 --> 00:22:49,232 S1: mind all the time, especially as you head into your 378 00:22:49,232 --> 00:22:54,792 S1: next webinar or blog post or talk. Ask yourself, how 379 00:22:54,792 --> 00:22:57,992 S1: much are you actually learning from this? How many nuggets 380 00:22:57,992 --> 00:23:02,312 S1: in here are actually surprising to you? And we don't 381 00:23:02,311 --> 00:23:04,232 S1: have to over rotate on this. There are many talks 382 00:23:04,232 --> 00:23:07,192 S1: I absolutely love that are about personal lessons learned or 383 00:23:07,392 --> 00:23:10,311 S1: about hobbies that people have. While it's not giving me, 384 00:23:10,352 --> 00:23:13,312 S1: you know, O'Day information or some kind of technical breakthrough 385 00:23:13,352 --> 00:23:16,952 S1: that's revolutionary, I'm still hearing a new perspective. I'm still 386 00:23:16,952 --> 00:23:20,122 S1: hearing freshness and passion and interest and love for something 387 00:23:20,402 --> 00:23:23,202 S1: from another human being, and I consider that to be 388 00:23:23,202 --> 00:23:26,242 S1: extraordinarily high alpha. It doesn't matter what the subject matter is. 389 00:23:26,762 --> 00:23:30,042 S1: So I guess try to spread the meme of seeking Alpha. 390 00:23:30,682 --> 00:23:33,121 S1: If you were organizing a conference or attending one, speaking 391 00:23:33,122 --> 00:23:36,482 S1: at one, or creating content or consulting with somebody who is, 392 00:23:36,762 --> 00:23:39,442 S1: explain this concept to them and ask them to review 393 00:23:39,442 --> 00:23:41,602 S1: how they're going to look for passion and fresh thinking 394 00:23:41,602 --> 00:23:44,282 S1: and fresh ideas, and how they plan to filter out 395 00:23:44,362 --> 00:23:48,601 S1: the opposite. We only have so much attention in a day, 396 00:23:48,602 --> 00:23:51,282 S1: or a year or a lifetime. We can't really be 397 00:23:51,282 --> 00:23:57,282 S1: wasting it on inputs that are 94% background noise. And 398 00:23:57,282 --> 00:23:59,802 S1: the recommendation of the week. Instead of thinking about AI 399 00:23:59,842 --> 00:24:01,762 S1: or what you can do with AI, instead think about 400 00:24:01,762 --> 00:24:04,362 S1: what you would do if you had a company that 401 00:24:04,362 --> 00:24:07,162 S1: worked for you with 1000 employees and you could assign 402 00:24:07,162 --> 00:24:10,242 S1: any vision or goals or tasks to them. I'm not 403 00:24:10,242 --> 00:24:12,722 S1: saying you can actually perfectly do that with AI today. 404 00:24:12,722 --> 00:24:16,202 S1: You can't, but it's getting really good. But what I 405 00:24:16,202 --> 00:24:17,602 S1: want you to do is come up with this list 406 00:24:17,602 --> 00:24:19,732 S1: of things that you would give them if you had 407 00:24:19,732 --> 00:24:22,411 S1: this company, if you had perfect eye. This is the 408 00:24:22,412 --> 00:24:26,932 S1: difficult exercise. The critical exercise. And the real fun actually 409 00:24:26,972 --> 00:24:29,372 S1: is just coming up with the list. Would you have 410 00:24:29,372 --> 00:24:31,692 S1: them help you write a book? Would you have them 411 00:24:31,692 --> 00:24:34,132 S1: help you make a game? Do you do a bunch 412 00:24:34,132 --> 00:24:37,172 S1: of research for the public good, like a nonprofit organization, 413 00:24:37,172 --> 00:24:42,132 S1: like going to uncover corruption and racketeering or some other 414 00:24:42,132 --> 00:24:44,572 S1: kind of crime, right. That you could only do with 415 00:24:44,571 --> 00:24:46,931 S1: a lot of people working for you. That's the thing. 416 00:24:47,052 --> 00:24:49,532 S1: What can you do with a lot of people working 417 00:24:49,532 --> 00:24:53,171 S1: towards that goal? Would you organize a community meetup in 418 00:24:53,172 --> 00:24:57,452 S1: your town? Would you? You know, I don't know, just crowdsource. 419 00:24:57,692 --> 00:25:00,452 S1: It's kind of like crowdsourcing. Not quite. This is more direct, 420 00:25:01,012 --> 00:25:03,972 S1: but try to stop thinking about AI as a tech thing. 421 00:25:04,212 --> 00:25:06,891 S1: Think about what you could do with a thousand people 422 00:25:06,892 --> 00:25:11,171 S1: helping you and make that list. And the aphorism of 423 00:25:11,172 --> 00:25:15,052 S1: the week we accept the cage we think we deserve. 424 00:25:15,612 --> 00:25:18,292 S1: We accept the cage we think we deserve.