WEBVTT 00:00.031 --> 00:08.320 [JM]: I think we all have markers for things that feel like summer, like the sounds that cicadas make if you live in a place where there are cicadas. 00:08.680 --> 00:09.361 [JM]: What about you, Dan? 00:09.821 --> 00:11.003 [JM]: How do you know when it's summer? 00:11.443 --> 00:19.532 [DJ]: I think when it starts to become uncomfortably hot in my house during the day and I have to use the air conditioner, that's when I know that it's summer. 00:19.852 --> 00:27.540 [DJ]: And not to sound like too much of a buzzkill, that's when I start hoping for fall to arrive because it's just too hot, Justin. 00:27.723 --> 00:28.444 [JM]: It is too hot. 00:28.805 --> 00:29.306 [JM]: Absolutely. 00:29.767 --> 00:40.867 [JM]: But the other way that I know that it's summer, besides the heat, is waking up to a deluge of Steam email messages about how everything on my wish list is on sale. 00:41.308 --> 00:44.253 [JM]: That's how I know that summer has finally arrived. 00:44.593 --> 00:52.127 [JM]: I think even more than temperature, the changing of the seasons can best be marked by the rhythmic cadence of Steam sales. 00:52.107 --> 01:00.521 [DJ]: Right, and it couldn't come at a better time either, because what do you want to do when it's 29 degrees Celsius inside your house but stay inside and play video games? 01:01.643 --> 01:03.145 [DJ]: In a bathing suit, I imagine. 01:03.165 --> 01:04.708 [JM]: I think you're right. 01:04.748 --> 01:08.874 [JM]: I think going outside really is the best solution to that problem. 01:09.295 --> 01:12.200 [JM]: But it's even hotter outside, so choose your poison. 01:12.660 --> 01:17.929 [JM]: I just think it's funny that the Steam Summer Sale is an event. 01:17.909 --> 01:20.492 [JM]: Like, to the point where they make trailers for it. 01:20.933 --> 01:25.899 [JM]: One comment that I loved is, it's like the Met Gala for people who don't go outside. 01:25.939 --> 01:35.451 [JM]: Well, if you're so inclined, you can pick up The Witcher for $4 or Cyberpunk 2077 for under $20. 01:35.911 --> 01:41.458 [JM]: There's Red Dead Redemption 2 and Teardown are two on my wish list that are at $15. 01:41.578 --> 01:42.820 [JM]: So, you know... 01:42.800 --> 01:47.992 [JM]: I've got to add to the ever-growing list of games that I have purchased but never launched. 01:48.393 --> 01:49.716 [DJ]: Yeah, that's a good point. 01:50.217 --> 01:57.754 [DJ]: I've also had my eye on buying but never playing Red Dead Redemption 2 for quite a while, so now might be the time. 01:57.920 --> 02:08.950 [JM]: And as you pointed out when we talked about this before, you should also check GOG at GOG.com and see if the games you're excited to get are also available there. 02:09.451 --> 02:20.000 [JM]: Because in addition to potentially being available at a lower price, they're also available on GOG free of digital restrictions management or DRM. 02:21.261 --> 02:27.387 [JM]: And all things being equal, if possible, I think that's probably the better place to throw your money at. 02:27.367 --> 02:29.953 [DJ]: Yeah, I prefer to throw my money at GOG. 02:30.394 --> 02:39.313 [DJ]: And coincidentally, since we're talking about buying video games and not playing them, I did just buy some new video games that I may or may not ever play. 02:39.333 --> 02:47.451 [DJ]: I don't know if I'm on record saying this out loud, but I think the simple fact is I enjoy collecting games more than I actually enjoy playing them. 02:47.431 --> 03:06.521 [DJ]: Like, don't get me wrong, I do enjoy playing them, but playing video games takes up a bunch of time that I may or may not necessarily want to dedicate, but merely collecting the games only takes a relatively small amount of money, especially if you tend to buy games that are 10 or more years old and you wait for these inevitable sales. 03:07.041 --> 03:08.764 [DJ]: But I did actually... 03:08.744 --> 03:20.637 [DJ]: start playing an ancient, ancient video game, Quest for Glory 4, The Shadows of Darkness, which I last played probably when I was a teenager, and it was really, really difficult. 03:20.978 --> 03:23.766 [DJ]: And I'm happy to say I found this classic adventure game 03:23.746 --> 03:25.909 [DJ]: a lot less difficult this time around. 03:26.390 --> 03:34.241 [DJ]: I think I also found it more buggy and obnoxious, although I might have found it buggy and obnoxious back then too. 03:34.322 --> 03:40.430 [DJ]: It's just back then, like I didn't have 175 other games that I could be playing, but I did. 03:40.511 --> 03:48.302 [DJ]: I sunk about 14 hours into this classic game and I think I've satisfied my sense of nostalgia and now I'll move on to something else. 03:48.451 --> 04:02.219 [JM]: Given that current games might require buying hardware that's either unobtainable or unaffordable, the idea of playing retro games makes even more sense in mid-2026 than it ever has. 04:02.479 --> 04:04.924 [JM]: So yeah, I think that makes a lot of sense. 04:04.904 --> 04:06.786 [JM]: All right, moving on to a bit of follow up. 04:07.167 --> 04:26.910 [JM]: We recently mentioned that in the upcoming OS 27 line of Apple software, that there are going to be some reported improvements in how liquid glass is rendered, and some visual tweaks meant to blunt some of the awfulness that has come with OS 26. 04:27.711 --> 04:29.393 [JM]: But that sadly doesn't 04:29.373 --> 04:35.081 [JM]: the squircle jail that Apple imposed on all of us continues in OS 27. 04:36.022 --> 04:47.238 [JM]: And that means that the ability to create an icon for your application in a shape that you want it to be has been curtailed ever since OS 26 came out. 04:47.679 --> 04:51.684 [JM]: And the fine folks at Rogue Amoeba have published an article called Free the Icons. 04:52.205 --> 04:56.932 [JM]: Apple should end their prohibition on shapes in macOS app icons. 04:57.352 --> 04:58.654 [JM]: And right out of the gate, 04:58.634 --> 05:11.062 [JM]: Rogamiba does a great job of showing a prime example of this shift from the way icons looked leading up to Mac OS 26 Tahoe, and then after. 05:11.083 --> 05:17.858 [JM]: And the example they show at the very beginning of this article is the automator icon. 05:17.838 --> 05:30.911 [JM]: The original one shows a robot illustrated in a lot of detail, standing on a platform of concentric circles, holding a cylindrical object of some kind using robot hands. 05:31.332 --> 05:33.494 [JM]: There's so much detail in this image. 05:33.954 --> 05:38.138 [JM]: And then alongside it is an image of the automator icon in Tahoe. 05:38.819 --> 05:47.828 [JM]: And all of the life has just been sucked out of this icon and put inside this squircle jail because the original icon, 05:47.808 --> 05:57.698 [JM]: had this round-rect background, but the robot's head and the cylindrical object that it holds extend beyond the boundaries of that round rectangle. 05:58.139 --> 06:14.515 [JM]: But since, as of OS 26, all icons have to be in Squircle Jail, everything is inside this round-rect, and also it's just been reduced to something vaguely resembling the robot's head and neck and shoulders. 06:15.056 --> 06:17.038 [JM]: And further down, they show... 06:17.018 --> 06:29.393 [JM]: Their own app, Audio Hijack, and the way it used to appear, where the microphone extends beyond the border of the squircle, and then what it looks like once it's been placed into the OS 26 squircle icon jail. 06:29.833 --> 06:37.963 [JM]: And I love that a longtime, very well-known and respected developer has called Apple out for this. 06:37.943 --> 06:44.737 [JM]: Pointing out that it's not just that the previous icons were more expressive, but also that they were more usable. 06:45.178 --> 06:51.150 [JM]: And that having these distinct shapes gives you a useful way to tell these different icons apart. 06:51.551 --> 06:58.686 [JM]: And that this new change eliminates the ability to easily distinguish between the shapes by forcing everything into the same frame. 06:58.666 --> 07:00.368 [JM]: cookie cutter shape. 07:00.749 --> 07:05.515 [JM]: And then at the end, it says, Apple should stop forcing every icon into the same squirgle. 07:05.675 --> 07:09.140 [JM]: Let's return to a world of gorgeous app icons like these. 07:09.380 --> 07:16.870 [JM]: And then shows eight really excellent examples of Mac application icon design. 07:16.910 --> 07:27.964 [JM]: And I sincerely hope that someone, anyone at Apple is listening because this is really the kind of article where you read it and you just nod your head and 07:27.944 --> 07:29.666 [JM]: agree with every single word. 07:30.046 --> 07:40.559 [DJ]: The sort of article for which the phrase all right-thinking people may have been invented, because this restriction that came with OS 26, it was one of those things that came out. 07:40.759 --> 07:46.626 [DJ]: And I don't know anyone, including myself, who heard about this and didn't think, who's asking for this? 07:47.407 --> 07:50.170 [DJ]: Like, who does this serve exactly? 07:50.690 --> 07:56.297 [DJ]: I guess it makes icons on your Mac be designed the exact same way as icons on your iPhone, but like, 07:56.277 --> 08:02.368 [DJ]: Why is that good, exactly, since you don't touch them with your fingers like they're not in a grid? 08:02.629 --> 08:14.050 [DJ]: And yeah, I mean, the one thing that I am really grateful to OS26 for is forcing you to say the ridiculous phrase squircle jail over and over and over again. 08:14.030 --> 08:16.875 [DJ]: I mean, but I'm not sure that was worth it, really. 08:16.895 --> 08:22.985 [DJ]: I'm not sure that was worth the reduction in design and usability that came along with this prohibition. 08:23.385 --> 08:39.852 [DJ]: So I do hope, since it seems like maybe there has been some kind of reckoning in Apple design where they are rolling back some of these questionable decisions, I hope that this one will quietly get revoked as well. 08:40.102 --> 08:51.919 [JM]: I love as a footnote to this article, they say for folks within Apple, this was feedback filed as third party app icons should not be restricted to Apple's dictated squircle shape. 08:52.299 --> 08:54.703 [JM]: I imagine it is a duplicate many times over. 08:55.244 --> 09:01.873 [JM]: So I think it's awesome that they submitted this as official feedback to Apple's bug reporting system. 09:01.853 --> 09:08.082 [DJ]: We'll return to the topic of Apple doing things that we hate pretty soon, so stay tuned. 09:08.522 --> 09:14.511 [DJ]: But something happened recently in the tech circles that I follow, and I wanted to talk about it. 09:14.952 --> 09:17.555 [DJ]: I'm mostly not on social media anymore. 09:18.076 --> 09:22.723 [DJ]: It's not where I consume content, to use the popular expression. 09:23.263 --> 09:30.814 [DJ]: For the most part, I am still someone who follows blogs, which is how the internet worked 20 years ago. 09:30.794 --> 09:35.624 [DJ]: where mostly people wrote on their personal websites and linked to each other. 09:35.664 --> 09:44.523 [DJ]: And you would often have a list, whether this was captured in a piece of software or not, of websites that you would check occasionally for updates. 09:45.124 --> 09:47.790 [DJ]: And there's a generation of writers who 09:47.770 --> 09:55.468 [DJ]: many of whom are still writing, that I sort of associate with this classic era of blogs and internet writing. 09:55.990 --> 10:05.813 [DJ]: And one of those writers, although I never knew him, and honestly, I couldn't even quote to you a specific piece of his writing. 10:05.793 --> 10:12.746 [DJ]: But one of those writers who always represented that classic era of the Internet to me was a writer named Om Malik. 10:13.147 --> 10:16.152 [DJ]: And the other day I found out that he just died. 10:16.533 --> 10:23.346 [DJ]: So he passed away following what I think was a lengthy series of health challenges. 10:23.326 --> 10:35.946 [DJ]: And when I started seeing posts about this and retrospectives in my feed of blogs that I follow, I found myself strangely grieving. 10:35.986 --> 10:46.883 [DJ]: And I say strangely just because it feels a little odd to feel grief for the passing, as sad as any passing may be, of someone that you really didn't know at all. 10:47.364 --> 10:51.190 [DJ]: So I've been trying to put my finger on over the last few days. 10:51.170 --> 10:56.239 [DJ]: Why did I feel like such a gut punch when I found out that Om Malik died? 10:56.660 --> 11:09.442 [DJ]: And again, aside from obviously it's sad when anybody dies, it made me realize that he and this handful of other writers have been a real inspiration to me over the years. 11:09.963 --> 11:14.190 [DJ]: When it comes to what is it I love about the web? 11:14.170 --> 11:16.494 [DJ]: And what kinds of things do I want to make? 11:16.935 --> 11:35.689 [DJ]: And especially as over the years, and this is, you know, an ongoing topic for us, the way that the web has been sort of shifted and warped and even like foreclosed by companies that have sought to really shape how people use the internet. 11:35.669 --> 11:55.163 [DJ]: I realized that the people who care about and continue to produce things for what I've 11:55.143 --> 11:57.628 [DJ]: I've realized that that feels very precious to me. 11:57.648 --> 12:06.143 [DJ]: And so the passing of one of these writers did feel like a big loss to me, even if I didn't know them personally. 12:06.163 --> 12:12.375 [DJ]: And even if I mostly knew their work through the reflections of other people to that end. 12:12.355 --> 12:28.845 [DJ]: Matt Mullenweg, the creator of WordPress. 12:28.825 --> 12:37.419 [DJ]: and John Gruber, the writer of Daring Fireball, both of whom are people that I associate with being part of the web from the very beginning, right? 12:37.519 --> 12:46.294 [DJ]: I mean, I think both those guys have been blogging basically since the concept of blogging was invented, and both of them helped really invent it. 12:46.274 --> 12:54.255 [DJ]: It was very interesting and touching to hear about the really close relationship that they both had with this other person. 12:54.335 --> 12:59.288 [DJ]: Like I'd heard both of them, certainly John Gruber, like refer to O'Mulloch before. 12:59.308 --> 13:02.055 [DJ]: That might be where I found out about him in the first place. 13:02.035 --> 13:09.505 [DJ]: But I didn't realize like how much this person meant to all of these other people, which isn't really surprising, of course. 13:09.986 --> 13:10.186 [DJ]: Right. 13:10.346 --> 13:15.313 [DJ]: And I think this is one of the most meaningful things that we learn when someone dies. 13:15.373 --> 13:24.465 [DJ]: And perhaps it's sad that maybe it takes someone dying before we really recognize it is the impact that they have on the lives of everyone around them. 13:24.445 --> 13:48.478 [DJ]: So I would say if this is something that is meaningful to you as you're listening to me talk about this, whether you knew Om Malik or knew his work or knew some of the people who knew him or are simply also a fan of this sort of long-form, well-considered writing on the internet... 13:48.458 --> 13:58.551 [DJ]: then I'd encourage you to check some of these things out because they really are a very moving retrospective of a real trailblazer for the web. 13:59.092 --> 14:09.606 [JM]: Like you said, he's been writing for so long that I think of him as being somehow connected to the birth of the web and part of it. 14:09.686 --> 14:16.695 [JM]: And so it does feel like there is a sense of loss, particularly given how much I've enjoyed his writing over the years. 14:17.216 --> 14:18.297 [JM]: And it also... 14:18.277 --> 14:28.410 [JM]: gives to me some degree of inspiration in terms of wanting to leave even a small fraction of the legacy that he has left behind. 14:28.450 --> 14:41.307 [JM]: If after I'm gone, there's even a tiny proportion of folks like this commenting on how much this person impacted their lives, then I think I will feel like I have truly lived a life worth living. 14:41.367 --> 14:43.930 [JM]: And that's inspirational. 14:43.910 --> 14:48.657 [JM]: All right, moving on to news that is not quite as sad, but nonetheless sobering. 14:49.238 --> 14:59.732 [JM]: Apple has followed up on Tim Cook's recent announcement that they were going to have to raise prices by raising prices across a large proportion of their product line. 15:00.253 --> 15:05.901 [JM]: So, for example, a MacBook Neo that was previously $600 is now $700. 15:05.881 --> 15:09.486 [JM]: A MacBook Pro that was $1,700 is now $2,000. 15:09.506 --> 15:23.364 [JM]: And iPads, HomePods, and the Apple TV set-top box are similarly affected, ranging from 10% to 30-something percent higher than before. 15:23.384 --> 15:25.847 [JM]: And that's just on the base models. 15:25.827 --> 15:42.662 [JM]: When you start to add things like RAM or solid state drive upgrades, the price increases go up steeply because those are the components that are rising in price and are the reason why Apple is raising prices on the products that incorporate those components. 15:43.022 --> 15:55.834 [JM]: So if you wanted to add RAM and solid state drive upgrades to say a MacBook Pro or a Mac Studio, you could easily go from say something that was six or 7,000 to over 10,000. 15:55.814 --> 16:02.250 [JM]: And a lot of this has to do with the RAM price increases that we've talked about recently. 16:02.791 --> 16:11.393 [JM]: And just for giggles, I looked up the average price for 32 gigabytes of DDR5 6000 RAM. 16:11.373 --> 16:13.496 [JM]: which is a specific kind of memory chip. 16:13.636 --> 16:23.069 [JM]: The price for this for most of 2025, say January through the end of September, was very stable at $120. 16:23.409 --> 16:40.232 [JM]: Starting in October of 2025, the price starts to rise rapidly to the point where today that $120 stick of RAM is now $570, or nearly five times the price in eight months. 16:40.212 --> 16:40.573 [JM]: What? 16:41.114 --> 16:45.261 [JM]: It's up 160% in the last 60 days alone. 16:45.321 --> 16:48.366 [JM]: And that's just where we are today. 16:48.806 --> 16:50.549 [JM]: Who knows where this goes from here? 16:51.110 --> 17:03.351 [JM]: And after Apple announced this news, their stock price got whacked by a whopping 8% immediately on the day they announced it as the market surmised that Apple 17:03.331 --> 17:09.099 [JM]: Demand is going to be affected by higher prices because there's usually this inverse relationship, you know, supply demand. 17:09.520 --> 17:16.709 [JM]: And so if the price goes up, the idea is that the demand will go down and Apple's financials will suffer as a result. 17:17.090 --> 17:22.217 [JM]: Meanwhile, the companies that make RAM have been doing really well in the stock price department. 17:22.433 --> 17:23.054 [DJ]: don't say. 17:23.074 --> 17:25.117 [JM]: I know that's shocking. 17:25.738 --> 17:30.465 [JM]: Micron's stock price is up over 800% over the last year. 17:30.485 --> 17:37.296 [JM]: And SK Hinex, another RAM producer, their stock price is up over 1000%. 17:37.536 --> 17:39.640 [DJ]: Congratulations, I guess. 17:40.261 --> 17:41.563 [DJ]: I mean, it helps. 17:41.623 --> 17:48.373 [DJ]: And by helps, I mean, it sucks that those are two of the only companies in existence that produce this stuff. 17:48.590 --> 17:49.492 [JM]: Well, there's three. 17:49.532 --> 17:52.557 [JM]: There's Micron, SK Hynix, and Samsung. 17:52.577 --> 17:55.202 [JM]: Those are the three big RAM producers. 17:55.343 --> 18:00.813 [JM]: And by big, I think, as you point out, the only... Yeah, they're the only RAM producers. 18:00.993 --> 18:11.172 [DJ]: This seems like one of those rare, weird situations where having a global choke point on the manufacturing of something turns out to be quite bad, actually. 18:11.422 --> 18:29.734 [JM]: It sounds like as a direct result of these component price increases that the smartphone market could see its biggest ever annual decline of roughly 14% in terms of sales volume this year, while the PC market could fall by 11%. 18:30.001 --> 18:31.524 [DJ]: Yeah, not exactly a surprise. 18:31.704 --> 18:35.872 [DJ]: I mean, make all of your products totally unaffordable and people might stop buying them. 18:36.293 --> 18:44.308 [DJ]: I don't know about you, Justin, but I'm planning to invest heavily in vinyl this coming year, by which I'm not talking about like pipes or clothing. 18:44.388 --> 18:50.760 [DJ]: I mean that I'm probably just going to buy a record player and listen to records from now on because computers are too expensive. 18:50.740 --> 18:54.625 [JM]: And it sounds like they're going to be too expensive for a good long while. 18:54.685 --> 19:11.127 [JM]: It sounds like South Korea's government and some of its top tech companies are committing to spending a trillion dollars on a bunch of megaprojects that are designed to increase the global memory chip supply. 19:11.467 --> 19:12.889 [JM]: But that's going to take years. 19:13.510 --> 19:18.857 [JM]: Meanwhile, companies like Apple are signing these long-term contracts 19:18.837 --> 19:29.697 [JM]: procurement deals with these three companies that make RAM in order to make sure they can get the RAM they need in order to sell the products that make up most of their business. 19:29.917 --> 19:35.087 [JM]: It's going to be really hard to sell Apple's primary product, the iPhone, if they can't get the RAM chips they need. 19:35.487 --> 19:42.921 [JM]: And then, of course, all of those companies scrambling to make sure they can get the RAM they need just drives up the price more and more. 19:43.188 --> 19:45.430 [DJ]: But you know what else drives up the price of RAM, Justin? 19:45.971 --> 19:51.756 [DJ]: It's when a very small group of companies work together deliberately to make the thing that they sell more expensive. 19:52.056 --> 19:57.582 [DJ]: I wonder if there's a name for that process that might result in, I don't know, lawsuits. 19:58.122 --> 20:10.994 [JM]: Indeed, because it sounds like Samsung, SK Hynix, and Micron are being sued by plaintiffs in California who allege that the three manufacturers are colluding. 20:11.335 --> 20:12.756 [JM]: The word that you were... 20:13.023 --> 20:18.596 [JM]: not so subtly pointing to, yeah, that they're colluding to artificially increase the price of RAM. 20:19.197 --> 20:26.174 [JM]: And I think the last time we saw a lawsuit about RAM price collusion was 20 years ago. 20:26.194 --> 20:28.078 [JM]: And in that case... 20:28.058 --> 20:32.023 [JM]: executives from Samsung and Hynix America were indicted. 20:32.043 --> 20:36.529 [JM]: And I don't know what the result of that was, but it'll be interesting to see how this goes. 20:36.769 --> 20:47.603 [JM]: And even if this lawsuit is successful, even if these fabs get built, we're stuck with high prices for RAM and other components, it seems, for a long, long time. 20:47.583 --> 20:51.308 [JM]: And as far as what you can do about that, the answer is not much. 20:51.728 --> 20:58.838 [JM]: But I would say if I were to choose one thing that you could do, it's don't give money to Sam Altman. 20:59.238 --> 21:01.801 [JM]: Think about whether you really want to continue. 21:01.962 --> 21:08.991 [JM]: If you have, say, a subscription for Chad GPT or another open AI product, maybe give that a second thought. 21:09.371 --> 21:16.240 [JM]: Maybe decide whether you want to contribute to this not so rosy status quo by continuing to hand money to Sam. 21:16.220 --> 21:22.078 [JM]: the folks that have created this deeply unpleasant chapter in personal computers. 21:22.459 --> 21:27.394 [DJ]: But Justin, how am I supposed to write emails if I can't ask ChatGPT to do that? 21:27.695 --> 21:28.196 [JM]: Indeed. 21:28.296 --> 21:37.949 [JM]: However, could one write email without a next token prediction text extruder giving you the answer? 21:38.690 --> 21:46.059 [DJ]: It would really be a band-aid, but I would like it if we would all start referring to large language models as text extruders instead. 21:46.620 --> 21:47.982 [DJ]: It would be funny. 21:47.962 --> 22:06.055 [DJ]: But I will say, and maybe we'll talk about this more on a future show once we've gathered some more information, I am increasingly interested in the developing space of tools for running local large language models, which we have spoken about because... 22:06.035 --> 22:27.672 [DJ]: I am so hungry... 22:27.652 --> 22:48.934 [DJ]: to break this dependency on companies like OpenAI and Anthropic, who, among all the other things, are in part responsible for this global constraint problem on computer hardware that we're now seeing impact all of these other areas that we're fans of. 22:48.914 --> 22:57.529 [DJ]: So I'm really interested in continuing to look into alternatives if we are going to take back computing. 22:57.729 --> 22:58.791 [DJ]: Like, let's put it that way. 22:58.811 --> 23:05.623 [DJ]: Ideally, computers should be more and more affordable to the average person so that they can run their own computer. 23:05.603 --> 23:16.256 [DJ]: software, including large language models if they want to, instead of the world that we seem to be marching towards where only the biggest companies can afford computer hardware. 23:16.757 --> 23:22.925 [DJ]: And so the really dystopic future version of this is you don't get to own a computer anymore. 23:22.965 --> 23:26.970 [DJ]: You can only rent computing from OpenAI, for example. 23:26.950 --> 23:39.351 [DJ]: And whether that's explicitly part of these companies playbook or not, you know, I don't know how much of an evil scheme to take over the world to really attribute to the people involved. 23:39.371 --> 23:41.835 [DJ]: The incentives are certainly pushing in that direction. 23:42.275 --> 23:44.419 [DJ]: And I want to see us claw that back. 23:44.602 --> 23:45.885 [JM]: Indeed, as do I. 23:45.905 --> 23:47.749 [JM]: All right, moving on to other news. 23:47.929 --> 24:03.403 [JM]: If you may recall, about two years ago, Apple announced that rumors started circulating that Apple is reportedly building their own servers to power Apple intelligence based on, again, this was two years ago, their... 24:03.383 --> 24:07.490 [JM]: M2 Ultra and M4 processors. 24:07.951 --> 24:13.881 [JM]: And you fast forward two years to the present day, and we haven't heard a peep about this report. 24:14.241 --> 24:25.240 [JM]: But we have heard that Apple has been working with Google to use their data center infrastructure, as well as their Gemini models to help improve Apple intelligence. 24:25.220 --> 24:38.117 [JM]: And so it raises the question, if this report was correct, that Apple has been working on their own servers that use their own Apple Silicon to power Apple intelligence, is that something that they're still working on? 24:38.618 --> 24:41.482 [JM]: Have they just given up that dream? 24:42.022 --> 24:44.205 [DJ]: Were they never doing it in the first place? 24:44.425 --> 24:50.594 [DJ]: Since after all, I think this was just a rumor and I don't know that anyone had any confirmation that they were going to. 24:50.614 --> 24:53.978 [DJ]: That isn't exactly the kind of thing Apple would probably talk about. 24:54.211 --> 24:59.556 [JM]: They were talking about this whole private cloud computing initiative. 24:59.956 --> 25:13.408 [JM]: So it was clear that they were designing a system whereby Apple intelligence queries could be processed off device in a way that was ostensibly as secure as running on device. 25:14.089 --> 25:23.037 [JM]: And as you point out, who knows what the hardware might've looked like in terms of what this private cloud computing might've run on. 25:23.017 --> 25:27.622 [JM]: Was it intended to run on their own servers utilizing their own silicon? 25:27.642 --> 25:29.184 [JM]: It's not clear. 25:29.244 --> 25:31.026 [JM]: This report could be totally wrong. 25:31.487 --> 25:38.155 [JM]: Regardless as to whether the report was accurate or not, I think there is a question about what is Apple's strategy here? 25:38.695 --> 25:44.502 [JM]: Is their strategy what it looks like in terms of what we already know, things that are not 25:44.482 --> 25:48.589 [JM]: but what they've already talked about, which is they're working with Google. 25:48.950 --> 25:50.412 [JM]: They're going to use some of their infrastructure. 25:50.452 --> 25:54.139 [JM]: They're going to use Gemini to help improve Apple intelligence. 25:54.619 --> 26:00.229 [JM]: Is that their core strategy or do they actually plan on 26:00.209 --> 26:03.872 [JM]: or perhaps are already working on a more homegrown solution. 26:03.972 --> 26:28.174 [JM]: And the reason that I raised this question is this notion of the Tim Cook doctrine, this notion of, does this specific area, which is the hardware that's used to power Apple intelligence, is that something that Apple considers to be a core part of their business and thus something that they want some control over? 26:28.154 --> 26:35.966 [JM]: Or is it not core to their business and they don't mind utilizing outside vendors to help provide that? 26:36.266 --> 26:46.281 [JM]: Because if this falls into the category from their perspective of this is a core part of their business and thus something that they feel like they need to have control over, 26:46.261 --> 26:57.118 [JM]: I really wonder whether this deal with Google is the end game, and whether it's instead just a stopgap measure, because everyone is competing for all of these different components. 26:57.458 --> 27:03.788 [JM]: But perhaps more than anything, folks are competing to acquire NVIDIA's GPUs. 27:03.808 --> 27:10.058 [JM]: And presumably, that is what Google is buying as much as they can of and putting in data center's 27:10.038 --> 27:15.045 [JM]: And this is what Apple is essentially paying them, a billion dollars a year or whatever the price tag is. 27:15.446 --> 27:31.388 [JM]: But it seems to me that at a minimum, Apple could hedge their bets as they continue to improve their M6, M7, and subsequent line of that class of processors with features that enable them to run... 27:31.368 --> 27:36.655 [JM]: large language models and other generative software as fast and efficiently as possible. 27:37.115 --> 27:53.136 [JM]: They're doing that in part so that folks can use them on device, but it seems like they could leverage what they're doing to make those chips great to ship in MacBooks and phones and iPads to also use in a server environment to power Apple intelligence. 27:53.457 --> 27:56.160 [JM]: I came across an article that talks about 27:56.140 --> 28:02.607 [JM]: the M5 Max chip, for example, and how it stacks up against NVIDIA's offerings. 28:03.067 --> 28:11.136 [JM]: And there's pros and cons to the two different architectures and the two different chip implementations, how they handle memory, etc. 28:11.536 --> 28:23.228 [JM]: But it doesn't seem crazy to me that as a backup plan, if nothing else, that Apple might be trying to develop a way of supplying the 28:23.208 --> 28:31.439 [JM]: Computing infrastructure that they need to power Apple intelligence on their own without having to rely on NVIDIA and without having to rely on Google. 28:31.939 --> 28:43.174 [JM]: And I really do wonder whether this is something that they could be working on or perhaps even should be working on if indeed this is something that's core to their business and something that they want to have some degree of control over. 28:43.475 --> 28:46.219 [DJ]: That seems like a strong hypothesis to me. 28:46.579 --> 29:06.125 [DJ]: And of course, I have a bias because I was just talking about my desire to break my dependency on the so-called frontier AI companies for providing LLM powered software features and instead pursue ways of running large language models on my own hardware. 29:06.565 --> 29:07.947 [DJ]: And there's two parts to that. 29:07.927 --> 29:21.643 [DJ]: One of them is you need large language models that are good enough to do the things that you want to do without them needing to be the frontier models from OpenAI and Anthropic. 29:22.004 --> 29:30.053 [DJ]: You've got Claude Opus 4.8 or whatever the latest version is and GPT 5.6, I think, just came out. 29:30.033 --> 29:35.438 [JM]: Except for the fact that you can't use it because some government has decided we can't be trusted with it. 29:35.858 --> 29:40.022 [DJ]: It's too close to being AGI, so it might take over the world, I guess. 29:40.522 --> 29:40.843 [DJ]: I don't know. 29:40.863 --> 29:43.185 [DJ]: I didn't really read the press release about that. 29:43.545 --> 30:00.040 [DJ]: But anyway, there are these other so-called open-weight or open-source models from various providers that you can download and run on your own hardware, and you can use them to power LLM-based software. 30:00.020 --> 30:04.086 [DJ]: And those models keep getting better the same way that the frontier models do. 30:04.146 --> 30:13.401 [DJ]: I've heard at least one person's opinion is that the state of the art in open models lag something like six months behind frontier models. 30:13.862 --> 30:14.743 [DJ]: And is that true? 30:14.863 --> 30:17.387 [DJ]: I don't really know, but let's just say it is true. 30:17.367 --> 30:37.418 [DJ]: Then you can imagine that we either have reached or will reach an inflection point for any given feature that you want from large language model powered software where you can get that feature from a local model instead of having to rent access to that feature from the big AI companies. 30:37.398 --> 30:45.130 [DJ]: And if you imagine that I, a private consumer, want to do that, you can imagine that Apple definitely wants it. 30:45.631 --> 30:48.515 [DJ]: And Apple has a big advantage that I don't have. 30:48.815 --> 30:54.664 [DJ]: I mean, I can buy an expensive piece of computer hardware that can run a local large language model. 30:54.805 --> 30:55.746 [DJ]: And I did do that. 30:56.227 --> 30:57.188 [DJ]: So that's nice for me. 30:57.629 --> 31:03.017 [DJ]: But Apple designs and manufactures or they design and have someone else manufacture for them. 31:02.997 --> 31:12.411 [DJ]: Hardware that is pretty well designed and it sounds like is being increasingly well designed to run large language models. 31:12.912 --> 31:20.043 [DJ]: So it wouldn't surprise me at all whether this was part of their original strategy when they started designing Apple Silicon. 31:20.344 --> 31:25.832 [DJ]: I don't know because I feel like Apple Silicon predates the large language model revolution model. 31:25.812 --> 31:28.077 [DJ]: by several years, but it's hard to say. 31:28.117 --> 31:38.417 [DJ]: But one way or another, they ended up with this shared memory architecture along with other things that seemed to lend itself to doing LLM inference. 31:38.958 --> 31:43.888 [DJ]: And that puts them in a uniquely powerful position, potentially. 31:43.868 --> 32:01.472 [DJ]: where while most other of the big tech companies are fundamentally relying on Nvidia's hardware, which is why Nvidia has become like the most or almost the most valuable company in the world, Apple maybe doesn't have that same dependency, at least not in the long run. 32:01.452 --> 32:08.641 [DJ]: And that could work both for their own data centers and also at the edge, as the saying goes, which is to say on your device. 32:09.182 --> 32:24.422 [DJ]: And I think another interesting opportunity for LLM-powered software is over time that you can run more of the inference on devices, like on your laptop, on your phone, and not over the internet in a data center somewhere. 32:24.402 --> 32:31.815 [DJ]: And doing this also fits well with Apple's tendency to not explain how they implement things, right? 32:31.915 --> 32:46.680 [DJ]: Like if they eventually go from Apple intelligence being powered by hardware that's running in a Google data center to Apple hardware that's running in an Apple data center to Apple hardware that's running on your iPhone. 32:46.660 --> 32:48.643 [DJ]: They don't have to tell you, right? 32:48.683 --> 32:53.690 [DJ]: Like you might not ever know it as the hardware and software gets more and more capable. 32:54.251 --> 32:58.137 [DJ]: They could pursue that strategy quietly in the background. 32:58.617 --> 33:05.547 [DJ]: And at least in my imagination, that is an excellent strategy for Apple, perhaps uniquely to pursue. 33:05.587 --> 33:06.829 [DJ]: And I hope they do it. 33:07.045 --> 33:17.826 [JM]: There's little question that Apple has designed processors that work really well for delivering generative software like Apple intelligence on device. 33:18.327 --> 33:26.843 [JM]: Their unified memory architecture has really given them a huge advantage for doing that on phones, on computers, on iPads. 33:26.823 --> 33:41.888 [JM]: It's less clear to me whether in a data center context, they could use the architecture that they put into those processors to deliver inference performance that's on par with say what Nvidia offers. 33:42.349 --> 33:46.215 [JM]: But given the improvements that they've made just going from the M4 to the M5, 33:46.195 --> 33:50.624 [JM]: specifically as it relates to large language model inference performance, 33:50.964 --> 34:05.352 [JM]: I think it's possible that with further investment in that area, that the M6, the M7, and subsequent processors could indeed start to really make sense in terms of performance compared to Nvidia. 34:05.332 --> 34:12.360 [JM]: And again, even if they don't arrive there just as a fallback plan, it still seems like it could make sense. 34:12.720 --> 34:18.627 [JM]: But like you said, even if they end up doing that, we may never know because it may not be the kind of thing they talk about. 34:19.088 --> 34:21.250 [JM]: All right, moving on to other news. 34:21.831 --> 34:30.040 [JM]: It sounds like Facebook is embracing generative software in ways that are making its employees miserable. 34:30.020 --> 34:42.815 [JM]: Quoting from an article from the New York Times, in an internal post last month, Facebook told its US employees that what they typed into their computer, how they moved their mouse, where they tapped, and what they saw on their screen would be tracked. 34:43.075 --> 34:52.887 [JM]: Facebook said the goal is to capture employee data so their artificial intelligence models could learn how people actually complete everyday tasks using computers. 34:53.247 --> 34:59.655 [JM]: Many workers are really upset by this, and in online comments, they called out this tracking as a privacy violation. 34:59.635 --> 35:04.843 [JM]: When an engineering manager wrote a comment saying, this makes me super uncomfortable, how do we opt out? 35:05.243 --> 35:09.930 [JM]: Facebook's chief technology officer said, there is no option to opt out. 35:10.351 --> 35:19.405 [DJ]: And then pressed a button that sent the engineering manager falling down a trap door into a pool filled with bloodthirsty sharks. 35:19.425 --> 35:21.308 [JM]: Sulfuric acid. 35:21.328 --> 35:24.893 [DJ]: Yes, bloodthirsty sharks who can swim in sulfuric acid. 35:24.913 --> 35:28.258 [DJ]: So it was a pool of acid with sharks in it, which is super bad. 35:28.492 --> 35:29.454 [JM]: That sounds bad. 35:30.035 --> 35:46.770 [JM]: And just for those keeping square at home, this is the same company that organized so-called "AI transformation weeks" for its employees, encouraging them, some might say pressuring them, to adopt generative software in their workflows. 35:47.088 --> 35:51.413 [DJ]: I thought those were weeks where they try to transform their employees into AI. 35:52.334 --> 35:56.760 [DJ]: The employees are all fed into like a sort of funnel and AI comes out the other end. 35:57.200 --> 35:58.522 [DJ]: It's like a soylent green thing. 35:59.343 --> 36:05.390 [DJ]: You're like, let me remind you this is the same company that makes Facebook. 36:06.251 --> 36:06.812 [DJ]: Right. 36:07.372 --> 36:09.535 [DJ]: Drops microphone, walks off stage. 36:09.700 --> 36:10.321 [JM]: Yeah. 36:10.341 --> 36:26.463 [JM]: I mean, a company that even introduced internal dashboards to track their employees' consumption of tokens, which some employees have said were a pressure tactic designed to encourage competition with their colleagues. 36:26.483 --> 36:29.087 [DJ]: How many tokens do you have to consume before you become AI? 36:29.247 --> 36:30.128 [DJ]: Did they figure that out? 36:31.069 --> 36:32.471 [JM]: I don't think they've worked that one out yet. 36:34.557 --> 36:42.805 [JM]: And so in the midst of all of this happening, like soon afterward, Facebook announces that they would lay off about 8,000 people. 36:43.366 --> 36:47.530 [JM]: And I don't think you need to be Sherlock Holmes to connect the dots here, right? 36:47.931 --> 36:53.537 [JM]: They're telling their employees, okay, we're going to track all the things you do so that we can learn how you do your work. 36:53.977 --> 36:56.620 [JM]: Oh, and we're going to lay off 8,000 of you. 36:56.660 --> 37:00.444 [JM]: Those things are probably related, right? 37:00.764 --> 37:06.574 [JM]: We're going to learn how you do your work so that we can figure out how to not pay you to do that work. 37:06.974 --> 37:18.874 [DJ]: Yeah, but I think you're giving them too much credit if you think that they were able to turn tracking people's mouse cursors into actually replacing people's labor within like a short time frame. 37:19.124 --> 37:24.830 [JM]: No, no, I'm not saying that the two events themselves are directly connected. What I'm saying is that... 37:25.031 --> 37:32.158 [DJ]: Oh, oh, you just mean they're a common theme, which is get rid of the people and replace them with robots one way or another. 37:32.539 --> 37:47.916 [JM]: Well, yeah, I think they're saying we don't need 8,000 of you because we're going to be training the remainder of you on how you do your work so that we can replace a lot of our employees with these models. 37:47.896 --> 37:57.251 [JM]: Which I think is a strategy that is going to fail miserably and spectacularly, mainly because they had to put the entire thing on ice. 37:57.972 --> 37:58.894 [JM]: And why is that? 37:59.434 --> 38:15.540 [JM]: A more recent news bit has been that sensitive Facebook employee data that is intended to monitor how they work was accessible to everyone in the company in a way in which it clearly was not designed to be, 38:15.520 --> 38:26.254 [JM]: Because generally speaking, you don't want employees to be able to, say, monitor what everyone else is doing on their screens, their mouse movements, the keystrokes they type. 38:26.734 --> 38:28.997 [DJ]: Yeah, I mean, no one should monitor that in the first place. 38:29.057 --> 38:34.845 [DJ]: But at the very least, if you are going to monitor it, you should restrict the number of people that have access to that information. 38:35.125 --> 38:36.347 [DJ]: That does seem straightforward. 38:36.948 --> 38:37.949 [JM]: Not, say, the whole company? 38:38.409 --> 38:45.038 [DJ]: Not just have it in like a big text file on an open network share or whatever BS I assume that they did. 38:45.305 --> 38:52.773 [JM]: One employee wrote, "I have accessed both personal tax and medical information through my work computer, as have many thousands of employees. 38:52.913 --> 38:58.599 [JM]: We were told this data would be protected and only used for valid business purposes after aggressive filtering." 38:59.080 --> 39:08.069 [JM]: So this news broke and Facebook's immediate response is, we are going to pause this surveillance while we investigate. 39:08.510 --> 39:13.675 [DJ]: We have pressed a button and dropped the people responsible into the sulfuric acid shark tank. 39:13.908 --> 39:17.337 [JM]: I think that's probably a more accurate reading of their response. 39:17.438 --> 39:17.698 [JM]: Yes. 39:18.220 --> 39:27.144 [JM]: I think my favorite comment on this whole thing came from Nick Heer, who said, paraphrasing, at least Facebook is staying on brand. 39:27.605 --> 39:30.212 [JM]: "Let's do something creepy and let's do it badly." 39:30.412 --> 39:48.725 [DJ]: I'm just like, I'm so, I don't know, depressed by the relentless nature of this sort of problem where some company or, I don't know, every government in the world right now trying to make everyone need to get age verified before they can use the internet. 39:49.286 --> 39:50.428 [DJ]: M-dash. 39:50.408 --> 39:57.221 [DJ]: They decide, let's do this thing that I might argue is an overreach that's probably a bad idea. 39:57.562 --> 40:02.251 [DJ]: But let's say that, okay, we can debate this thing on its merits and decide, no, no, no, no, no. 40:02.271 --> 40:05.818 [DJ]: We need to collect this very sensitive information from everybody. 40:05.798 --> 40:23.588 [DJ]: And it seems like the very next thing you do is go, how can we do this really, really carefully and really well? 40:23.568 --> 40:32.306 [DJ]: Something like this happens where it's like, oh, actually, 100 million people's like private information was stored in some baldly insecure way. 40:32.366 --> 40:33.789 [DJ]: And now it's leaked. 40:34.270 --> 40:37.718 [DJ]: And it's just it's it's very it's very frustrating. 40:37.778 --> 40:42.167 [DJ]: Like whether it's Facebook doing it or someone else, you hear about this stuff all the time. 40:42.147 --> 40:50.180 [DJ]: And if anything, that might be my bigger objection alongside all the other objections is whenever some entity goes, hey, we want to track you. 40:50.220 --> 40:51.582 [DJ]: We want to collect this information. 40:51.622 --> 40:54.567 [DJ]: I'm like, A, I disagree with that on its face. 40:54.627 --> 41:01.237 [DJ]: But also, B, you're going to do a bad job and some person on like the dark web is going to have my residential address. 41:01.557 --> 41:03.020 [DJ]: And they're like, no, that's never going to happen. 41:03.040 --> 41:03.761 [DJ]: Oh, yeah, sorry. 41:03.801 --> 41:05.083 [DJ]: That totally did happen. 41:05.688 --> 41:08.740 [DJ]: It's like, yeah, of course it did. 41:08.800 --> 41:13.217 [DJ]: Because people are incredibly incompetent when it comes to this kind of thing. 41:13.383 --> 41:16.127 [JM]: Do you mean the most obvious outcome actually happened? 41:16.627 --> 41:17.488 [DJ]: Yes, yes. 41:17.669 --> 41:20.933 [DJ]: And like, it shouldn't be the most obvious outcome for starters. 41:21.013 --> 41:30.786 [DJ]: But insofar as it happens again and again and again, there's just something so soul destroying about, as you say, like, let's do something creepy and let's do it badly. 41:30.846 --> 41:36.293 [DJ]: Like, it's not enough that Facebook, along with lots of other entities, are always doing these creepy things. 41:36.454 --> 41:39.217 [DJ]: It's that they inevitably also do them badly. 41:39.678 --> 41:41.881 [DJ]: But Amazon would never do anything like this, right? 41:42.121 --> 41:43.182 [JM]: No, of course not. 41:43.582 --> 41:44.984 [JM]: Well, actually they would. 41:49.047 --> 42:00.758 [JM]: It seems Amazon also liked this idea of tracking how many tokens their employees were consuming, you know, in terms of using large language models and other generative software to do their work. 42:01.198 --> 42:10.947 [JM]: And so when pressured to meet their AI consumption quotas, it seems like they came up with rather creative ways of 42:10.927 --> 42:32.393 [JM]: meeting those quotas, such as creating their own agents to go and do completely unproductive things to just churn and churn and consume tokens and get their token usage to meet their corporate overlords' absurd expectations. 42:32.873 --> 42:34.896 [JM]: And it's just such an obvious outcome. 42:35.416 --> 42:40.883 [JM]: When you create these arbitrary metrics, 42:40.863 --> 42:50.045 [JM]: as soon as you start measuring something, people are going to find ways of gaming it because you've given them perverse incentives. 42:50.466 --> 42:52.050 [JM]: So you can't be surprised when this happens. 42:52.070 --> 42:54.336 [JM]: You should expect, like, this is the most obvious. 42:54.356 --> 42:58.686 [JM]: Speaking of obvious things, this is the most obvious outcome you could possibly imagine. 42:58.801 --> 43:07.276 [DJ]: It really doesn't give me a lot of faith in people that I presume are highly placed at a big, fancy technology company. 43:07.777 --> 43:21.180 [DJ]: They set up these incentives and did no one stand up in the room and go, don't you think this is just going to make people come up with stupid ways to waste tokens so that they meet these arbitrary numbers? 43:39.723 --> 43:41.485 [DJ]: It's really something to behold. 43:41.626 --> 43:45.170 [DJ]: And so you have to be really careful what incentives you give people. 43:45.431 --> 43:54.723 [DJ]: And as a sidebar and kind of referring back to the thing about storing people's private information, people are bad at designing incentives that give them the outcome they want. 43:55.144 --> 43:58.428 [DJ]: That's not something that people are good at, like seemingly anyone. 43:58.929 --> 44:06.439 [DJ]: You know, I think that like keeping really private data very secure just seems to be something that human beings are not well wired up to do. 44:06.419 --> 44:08.961 [DJ]: Sorry, maybe the AI will be better at it. 44:09.382 --> 44:14.907 [DJ]: But as a result, you always see things like this where people come up with these incentive schemes. 44:15.507 --> 44:20.992 [DJ]: And then instead of getting the outcome they want, people just find ways to game the scheme. 44:21.192 --> 44:33.083 [DJ]: Like, this is something that Joel Spolsky, the author of the classic software developer blog, Joel on software, was talking about in what when was he writing that just in like the 1500s? 44:33.383 --> 44:34.384 [DJ]: It feels like it. 44:34.364 --> 44:38.210 [DJ]: It was more like 20 years ago, but it feels like it was the 1500s. 44:38.270 --> 44:39.332 [DJ]: It feels like ancient history. 44:40.573 --> 44:43.578 [DJ]: Joel wrote at least one really good article about this. 44:43.658 --> 44:58.701 [DJ]: I wish I could remember its title off the top of my head, but I remember him writing specifically about this, about how the incentives you come up with to improve the performance of software developers, or even just to figure out how to compensate them properly, 44:58.681 --> 45:21.212 [DJ]: are so vulnerable to this sort of gaming. And he gave the example of like, if you try to give people incentives to ship fewer bugs, like defects in their software, people are going to spend all their time, like the software developers will start arguing with the QA analysts about whether something qualifies as a defect or not. Like, that's what people are going to spend their time doing. 45:21.192 --> 45:27.380 [DJ]: And this is a really stupid idea that's basically exactly the same as the Amazon thing we're talking about. 45:27.841 --> 45:41.158 [DJ]: If you give developers an incentive, like a bonus for producing more lines of code, then they will find a way to simply write a bunch of code that doesn't really do anything and isn't necessary. 45:41.138 --> 45:58.654 [DJ]: And likewise here, where instead of measuring any sort of meaningful outcome, like were you able to use generative software to accomplish some given task in less time, at a higher level of quality, etc., etc., instead it's just use more tokens. 45:59.214 --> 46:08.563 [DJ]: It's like, okay, then obviously people are just going to build a machine that consumes tokens with no meaningful output. 46:08.543 --> 46:22.718 [DJ]: I'm assuming that there's a massive volume somewhere in Amazon of like really bizarre agentic like tone poetry or something that people got these systems to make so that they could meet their token quotas. 46:22.758 --> 46:29.906 [DJ]: Like just the idea of a token quota is so obviously stupid that, yeah, I don't know. 46:29.946 --> 46:35.091 [DJ]: I'm really hoping it's the people who came up with this idea that will end up in the sulfuric acid shark tank. 46:35.071 --> 46:36.273 [DJ]: I don't have great hopes. 46:36.333 --> 46:44.627 [DJ]: I'm assuming that whoever spearheaded this was probably promoted to executive vice president and will eventually retire on their stock options and write a book. 46:45.128 --> 46:50.937 [JM]: When the remnants of our species are sifting through the rubble of our current generation... 46:51.237 --> 47:04.339 [JM]: and wondering how the paperclip maximizer algorithm that consumed our planet was created, the answer will be someone at Amazon was trying to maximize their token usage. 47:04.860 --> 47:14.055 [JM]: So they're the ones that wrote the paperclip maximization algorithm that eventually resulted in the inevitable apocalypse. 47:14.407 --> 47:29.790 [DJ]: I'm imagining, yeah, I'm imagining these scavengers, like, pulling the lid off an ancient tomb and finding, like, a mummified skeleton clutching some sort of, like, employee-of-the-month badge that says, like, "I used the most tokens". 47:30.170 --> 47:31.993 [DJ]: But yeah, that person destroyed the world. 47:33.357 --> 47:35.041 [JM]: All right, everyone, that's all for this episode. 47:35.081 --> 47:35.782 [JM]: Thanks for listening. 47:36.223 --> 47:44.342 [JM]: You can find me on the web at justinmayer.com and you can find Dan at danj.ca... and not in a vat of sulfuric acid and sharks. 47:44.562 --> 47:45.104 [DJ]: Here's hoping. 47:47.068 --> 47:51.518 [JM]: Reach out with your thoughts about this episode via the Fediverse at justin.ramble.space.