WEBVTT - UL NO. 431: Companies are Graphs of Algorithms 0:00:00.080 --> 0:00:02.540 When you go through airport security, there's a line where 0:00:02.540 --> 0:00:05.600 the TSA agent checks your ID, and another line where 0:00:05.600 --> 0:00:08.030 a machine scans your bag. The same thing happens in 0:00:08.030 --> 0:00:11.540 enterprise security, but instead of passengers and luggage, it's end 0:00:11.539 --> 0:00:14.720 users and their devices. These days, most companies are pretty 0:00:14.720 --> 0:00:16.700 good at the first part of the equation where they 0:00:16.700 --> 0:00:19.790 check user identity, but user devices can roll right through 0:00:19.790 --> 0:00:23.780 authentication without getting inspected at all. In fact, 47% of 0:00:23.780 --> 0:00:28.190 companies allow unmanaged, untrusted devices to access their data. That 0:00:28.190 --> 0:00:30.320 means an employee can log in from a laptop that 0:00:30.320 --> 0:00:33.320 has its firewall turned off and hasn't been updated in 0:00:33.320 --> 0:00:36.380 six months. Or worse, that laptop might be a bad 0:00:36.380 --> 0:00:40.519 actor using employee credentials. One password finally solves the device 0:00:40.520 --> 0:00:43.700 trust problem. One password ensures that no device can log 0:00:43.700 --> 0:00:47.839 into your Okta protected apps unless it passes your security checks. Plus, 0:00:47.840 --> 0:00:51.409 you can use one password on devices without MDM, like 0:00:51.409 --> 0:00:55.760 your Linux fleet, contractor devices, and every BYoD phone and 0:00:55.760 --> 0:00:59.600 laptop in your company. Visit one password comm slash unsupervised 0:00:59.600 --> 0:01:02.270 learning to watch a demo and see how it works. 0:01:02.270 --> 0:01:12.140 That's one password.com/unsupervised learning. Welcome to Unsupervised Learning, a security 0:01:12.170 --> 0:01:14.869 I and meaning focused podcast that looks at how best 0:01:14.870 --> 0:01:18.230 to thrive as humans in a post AI world. It 0:01:18.230 --> 0:01:22.250 combines original ideas, analysis, and mental models to bring not 0:01:22.250 --> 0:01:25.369 just the news, but why it matters and how to respond. 0:01:29.540 --> 0:01:32.660 All right. Welcome to unsupervised learning. This is Daniel Meisler 0:01:32.660 --> 0:01:36.500 and this is episode 431. This is RSA week. Got 0:01:36.500 --> 0:01:38.990 grok support coming to fabric. If you've not messed with 0:01:38.990 --> 0:01:41.959 grok yet you absolutely need to go do this. It 0:01:41.959 --> 0:01:46.520 is insanely cool to use. It is so fast! I 0:01:46.520 --> 0:01:48.260 would love to do a demo, but we need to 0:01:48.260 --> 0:01:51.320 get through the show, so I'll probably do a demo 0:01:51.320 --> 0:01:55.250 in a separate talk or uh, little video soon. My 0:01:55.250 --> 0:01:59.480 buddy Clint Gibler and also Caleb Simon gave awesome talks 0:01:59.480 --> 0:02:02.300 at Bsides, so that's super cool. Be sure to check 0:02:02.300 --> 0:02:04.309 those out when they come out. I got a new 0:02:04.310 --> 0:02:07.880 essay here on how consultancies are about to move into 0:02:07.880 --> 0:02:11.600 departments and companies, and basically break them into pieces and 0:02:11.600 --> 0:02:14.000 apply AI to them. I've been wanting to write this 0:02:14.000 --> 0:02:16.370 one for a very long time. It's called companies are 0:02:16.370 --> 0:02:19.550 a just a graph of algorithms, and I'm probably going 0:02:19.550 --> 0:02:22.549 to record this as a standalone episode as well. Second 0:02:22.550 --> 0:02:25.730 new essay this week is how I think prompting is 0:02:25.730 --> 0:02:28.820 kind of the center of AI, and even though there's 0:02:28.820 --> 0:02:31.280 lots of cool stuff you could do with like fine 0:02:31.280 --> 0:02:33.830 tuning and training your own models and stuff, I think 0:02:33.830 --> 0:02:36.500 prompting is still where it's at, and this is kind 0:02:36.500 --> 0:02:39.650 of like an important walk through, or at least a 0:02:39.650 --> 0:02:44.210 decent walk through. On why I believe that, security wise, 0:02:44.210 --> 0:02:47.720 Biden administration is changing it so that you can they 0:02:47.720 --> 0:02:51.830 can basically hire within it, and especially security. They can 0:02:51.830 --> 0:02:55.130 hire essentially people who don't have a degree, which has 0:02:55.130 --> 0:02:58.070 been a huge limiting factor for them to get talent. 0:02:58.070 --> 0:03:01.970 So that's really, really good news. The CEO of UnitedHealth 0:03:01.970 --> 0:03:06.050 took personal responsibility for paying the $22 million ransom to 0:03:06.050 --> 0:03:09.950 get business back running, and that is really interesting that 0:03:09.950 --> 0:03:12.920 he did that. We'll see what the fallout is. I 0:03:12.919 --> 0:03:15.170 do worry a little bit about the signaling that it's 0:03:15.169 --> 0:03:17.600 basically saying, hey, it's okay to pay ransoms, which of 0:03:17.600 --> 0:03:20.180 course kind of propagates it. It's like, oh, we never 0:03:20.180 --> 0:03:23.780 negotiate with terrorists type of deal. That policy makes it 0:03:23.780 --> 0:03:27.350 so that people are less likely to become terrorists. So 0:03:27.350 --> 0:03:30.619 we'll see how that plays out. Satya Nadella sent out 0:03:30.620 --> 0:03:33.770 a Bill gates type memo saying that security was the 0:03:33.770 --> 0:03:36.890 top priority, which is cool to see history repeat itself there. 0:03:36.890 --> 0:03:41.660 Cybersecurity consultancy got busted for trying to well, no, a 0:03:41.660 --> 0:03:45.170 consultant got busted for trying to extort an IT firm 0:03:45.170 --> 0:03:50.270 for $1.5 million by threatening to leak their secrets. Not 0:03:50.270 --> 0:03:53.300 the way to get what you want. Or maybe it is, 0:03:53.300 --> 0:03:56.750 I don't know. Verizon, AT&T, T-Mobile and sprint got hit 0:03:56.750 --> 0:04:01.550 with a $200 million fine for selling customer location data. Oh, 0:04:01.550 --> 0:04:04.820 this is insane. A team trained a robot dog in 0:04:04.820 --> 0:04:08.810 a simulation. Completely in a simulation how to walk on 0:04:08.810 --> 0:04:11.330 a ball. And then they took that code and put 0:04:11.330 --> 0:04:13.580 it in an actual robot dog and put it on 0:04:13.580 --> 0:04:17.029 an actual ball. And it just worked. I mean, think 0:04:17.029 --> 0:04:19.430 about how that's going to transfer to lots of different 0:04:19.430 --> 0:04:24.710 human problems. Google is pushing for a change to immigration policies. 0:04:24.710 --> 0:04:27.710 They basically say that we're losing AI and security talent 0:04:27.710 --> 0:04:30.800 because they're going to other countries. This is a tweet 0:04:30.800 --> 0:04:34.010 of mine about the difference between a company that uses 0:04:34.010 --> 0:04:37.969 back end models versus the whole company is actually the 0:04:37.970 --> 0:04:41.179 back end model. So it's like, what's your vulnerability to 0:04:41.180 --> 0:04:46.219 getting sherlocked by someone like OpenAI or anthropic? And this 0:04:46.220 --> 0:04:49.880 is basically a breakdown for how you should actually build 0:04:49.880 --> 0:04:52.159 your company so that it just gets better when those 0:04:52.160 --> 0:04:55.280 things improve, it doesn't actually get replaced. I got a 0:04:55.279 --> 0:04:58.310 doctor buddy who loves AI note taking. She said she 0:04:58.310 --> 0:05:01.310 was basically going to give up and just get out 0:05:01.310 --> 0:05:05.420 of practicing medicine. And basically I note taking saved her 0:05:05.420 --> 0:05:09.170 from doing that. So that's cool. Someone's criticizing Sam Altman's 0:05:09.170 --> 0:05:12.740 approach as a blend of fear, ignoring uncertainties and riding 0:05:12.740 --> 0:05:16.820 the hype wave. Apple is supposedly working on a big 0:05:16.820 --> 0:05:20.510 AI team, pulling people from Google and building up some 0:05:20.510 --> 0:05:23.900 sort of lab in Zurich. My buddy Joseph Thacker just 0:05:23.900 --> 0:05:28.550 wrote a great post on assumptions in Lm's assumptions made. 0:05:28.550 --> 0:05:31.520 This is the article you want to go check it out? It's, uh, 0:05:31.520 --> 0:05:36.289 really good. Somebody automated a YouTube shorts channel entirely entirely 0:05:36.290 --> 0:05:40.880 with free tools. So it's a short book. Summaries is 0:05:40.880 --> 0:05:43.820 what it is, and it's just releasing episodes. It's doing 0:05:43.820 --> 0:05:47.089 the whole thing. It's just automation. Uh, really cool to 0:05:47.089 --> 0:05:49.370 go check that out, especially because it's book summaries, which 0:05:49.370 --> 0:05:54.680 are useful complexities, allure and simplicity is power. Basically, this 0:05:54.680 --> 0:05:58.700 article is arguing that simplicity is powerful, but complexity is 0:05:58.700 --> 0:06:02.240 what actually sells and gets people excited. Uh, I'm I 0:06:02.240 --> 0:06:05.570 don't agree with that for myself, but I think the 0:06:05.570 --> 0:06:09.080 article made a pretty good case for it. 30 Useful concepts. 0:06:09.080 --> 0:06:15.679 So I just found like 27,500 asteroids in old telescope photos. 0:06:15.680 --> 0:06:17.630 And this is a good example of where you need 0:06:17.660 --> 0:06:22.010 AI because there aren't enough people, professionals actually looking at 0:06:22.010 --> 0:06:25.100 the sky with telescopes. There's just not enough people. There's 0:06:25.100 --> 0:06:27.679 not enough eyes, there's not enough experts. So this is 0:06:27.680 --> 0:06:31.210 a perfect. For. I actually just having access to tons 0:06:31.210 --> 0:06:34.060 of different telescopes and then being able to just launch 0:06:34.060 --> 0:06:36.250 an alert if it sees one that looks big and 0:06:36.250 --> 0:06:38.740 it looks like it might be heading towards us, looks 0:06:38.740 --> 0:06:42.310 like we might have found a potential for extraterrestrial life 0:06:42.310 --> 0:06:46.539 by detecting a certain molecule. Looks like higher paid employees 0:06:46.540 --> 0:06:51.620 are struggling, which makes sense to me because. If you 0:06:51.620 --> 0:06:54.140 want to reduce how much you're paying in headcount, you 0:06:54.140 --> 0:06:57.680 reduce the people first. That make the most. I've seen 0:06:57.680 --> 0:07:02.330 this personally multiple times for people around me. So definitely true. 0:07:02.330 --> 0:07:06.469 Got an interesting measurement of the economy, essentially how well 0:07:06.470 --> 0:07:10.730 strippers are doing and how much they're getting tipped. Is 0:07:10.730 --> 0:07:15.630 a good indicator or at least an interesting indicator. Number 0:07:15.630 --> 0:07:20.370 one metric for longevity continues. Every study I've seen about this, 0:07:20.370 --> 0:07:24.090 it's basically VO2 max. And this is yet another study 0:07:24.090 --> 0:07:28.140 that's confirming that. Really cool essay. I didn't read it again, 0:07:28.140 --> 0:07:30.660 but I've read this like ten times. In Praise of 0:07:30.660 --> 0:07:35.340 Idleness by Bertrand Russell. And this one is really interesting. 0:07:35.340 --> 0:07:39.770 We got this woman here who is complaining about. Oh 0:07:39.770 --> 0:07:44.900 my God, they offered me. A help desk job. I 0:07:44.900 --> 0:07:47.870 keep getting these offers for help desk jobs, and then 0:07:47.870 --> 0:07:51.320 she proceeds to show off her two degrees. She's like, 0:07:51.320 --> 0:07:53.990 look at this degree and I have a second degree. 0:07:54.020 --> 0:07:56.780 Can't they see that I have these degrees? Why would 0:07:56.780 --> 0:07:59.330 they be offering me a lowly help desk job? First 0:07:59.330 --> 0:08:03.350 of all, help desk people are awesome. InfoSec Taylor Swift 0:08:03.590 --> 0:08:07.220 was a help desk person. I know so many people 0:08:07.220 --> 0:08:09.740 who started on the help desk. My buddy Jason Powell 0:08:09.740 --> 0:08:12.680 started on the help desk and now he's doing amazing 0:08:12.680 --> 0:08:17.810 things at Apple. And it's like, look, you can't. First 0:08:17.810 --> 0:08:20.840 of all, you're demeaning a group of people who work 0:08:20.840 --> 0:08:23.900 in a field. So that's a problem. Second of all, 0:08:23.900 --> 0:08:27.410 it's not above you. And third of all, those degrees 0:08:27.410 --> 0:08:30.830 don't actually mean anything for you, right? This is kind 0:08:30.830 --> 0:08:33.380 of the problem, right? She's in a heart for a 0:08:33.380 --> 0:08:38.900 hard time because. She believes that she's entitled to these things. 0:08:38.900 --> 0:08:41.059 I do want to mention, though, it's not her fault. 0:08:41.059 --> 0:08:45.200 It's the fault of the system for programming. And this 0:08:45.200 --> 0:08:47.750 is like the parents, this is the teachers, this is 0:08:47.750 --> 0:08:50.990 the whole school mechanism. And this tended to be more 0:08:50.990 --> 0:08:53.480 true in the past. Right? You put the work in, 0:08:53.480 --> 0:08:55.969 you pretty much get a job. It is just not 0:08:55.970 --> 0:08:59.000 the case anymore. And it's definitely not the case now 0:08:59.000 --> 0:09:02.810 with AI, it was already the case before. I mean, 2022. 0:09:02.840 --> 0:09:05.570 It was not the case that a credential was good enough. 0:09:05.570 --> 0:09:08.929 You already needed to be special in some sort of way. 0:09:08.929 --> 0:09:13.400 So this was already a bad take, you know, back 0:09:13.400 --> 0:09:17.600 in 2022. And now it's even worse, way worse with 0:09:17.600 --> 0:09:22.579 AI because people like this who think that their homework 0:09:22.580 --> 0:09:25.820 and their the fact that they finished these classes is 0:09:25.820 --> 0:09:29.120 actually going to be valuable to that company by itself. 0:09:29.120 --> 0:09:30.950 They're the ones who are going to get replaced by 0:09:30.950 --> 0:09:34.760 AI or just not hired at all. And this is 0:09:34.760 --> 0:09:39.830 another sort of thing around this. It's just like actually unrelated. But, um, 0:09:39.980 --> 0:09:42.830 it's basically if you're in tech and you're constantly dunking 0:09:42.830 --> 0:09:46.070 on AI, you should stop and you should think about this. 0:09:46.070 --> 0:09:49.790 AI is about to add billions or trillions of new, 0:09:49.820 --> 0:09:55.490 highly skilled and intelligent workers to the economy. And I mean, 0:09:55.490 --> 0:09:57.260 this is what's going to happen, right? And I've got 0:09:57.260 --> 0:09:59.450 a whole whole vibe here, but I'm not going to 0:09:59.450 --> 0:10:03.080 read the whole thing. Basically, the takeaway is that something 0:10:03.080 --> 0:10:05.720 about AI might be putting you off. Maybe it's like 0:10:05.720 --> 0:10:09.319 the fanboys, maybe it's the fan. The fact that it 0:10:09.320 --> 0:10:12.530 might sound to you like crypto or NFTs, you need 0:10:12.530 --> 0:10:15.559 to push that aside and like, push away the fact 0:10:15.559 --> 0:10:18.650 that you're you're turned off for some reason about this, 0:10:18.650 --> 0:10:22.250 because if you were turned off about driving or turned 0:10:22.250 --> 0:10:25.970 off about reading or writing it, just because somebody was 0:10:25.970 --> 0:10:30.320 going crazy with CrossFit around reading and writing or reading 0:10:30.320 --> 0:10:33.829 and writing, it's the best thing driving. It's the best thing. 0:10:33.830 --> 0:10:36.500 And when you saw those people talk, it annoyed you 0:10:36.500 --> 0:10:39.110 and you're like, you know what? Because of those people, 0:10:39.110 --> 0:10:41.840 I'm not going to learn how to read, write or drive. 0:10:41.840 --> 0:10:44.920 That's what you're doing with AI. That is what you're 0:10:44.920 --> 0:10:47.980 doing with AI. And guess what? It's not hurting them. 0:10:47.980 --> 0:10:51.250 It's going to hurt you. So that's why I'm saying here. 0:10:51.250 --> 0:10:53.950 That's the point of this one. And the recommendation of 0:10:53.950 --> 0:10:57.670 the week is my three minute video for how to 0:10:57.670 --> 0:11:00.460 Build a meaningful life. It's pretty cool. It's like three 0:11:00.460 --> 0:11:04.630 minutes 30s. And the aphorism of the week is one's 0:11:04.630 --> 0:11:07.330 destination is never a place, but a new way of 0:11:07.330 --> 0:11:11.440 seeing things. One's destination is never a place, but a 0:11:11.440 --> 0:11:16.540 new way of seeing things. Henry Miller Unsupervised Learning is 0:11:16.540 --> 0:11:19.689 produced and edited by Daniel Miller on a Neumann U87 0:11:19.690 --> 0:11:23.770 AI microphone using Hindenburg. Intro and outro music is by 0:11:23.770 --> 0:11:26.980 zombie with the why and to get the text and 0:11:26.980 --> 0:11:29.319 links from this episode, sign up for the newsletter version 0:11:29.320 --> 0:11:35.010 of the show at Daniel meisler.com/newsletter. We'll see you next time.