Pleasantly Whimsical

[JM]: Today, we're going to jump right into some follow-up.

[JM]: And we're going to talk about the age verification legislation that we talked about a little while ago, where governments across the world are introducing legislation to verify the age of people who are using online services and...

[JM]: It seems that popular chat platform Discord has been hacked.

[JM]: And this is being described as every user's worst nightmare and reminds me of a similar hack that was done against the Tea app that we talked about before.

[JM]: But this one appears to be a lot bigger.

[JM]: And among other data, it seems that Discord said that around 70,000 users may have had their government photo IDs exposed.

[JM]: And the hackers posted example images of some of these government IDs that they had stolen.

[JM]: And to me, this news just underscores what a bad idea this kind of age verification legislation is.

[JM]: Because the consequences are very significant.

[JM]: I mean, I have been asked — not by Discord because I don't use Discord — but I have signed up for services in the past, sometimes financial institutions that require you to do this whole take a selfie with a government ID thing.

[JM]: And it felt really invasive and uncomfortable to me when I did it

[JM]: the very first time and really every time I've done it.

[JM]: Thankfully, it's been a while since I've had to do that.

[JM]: And I think the next time that it comes up, I'm going to feel even less comfortable given that now we have very real-world, high-profile cases where this information has been breached and is now in the hands of people who probably are going to do bad things with it.

[JM]: So in addition to the dire consequences of this legislation, it also to me is just obviously ineffective because people who want to get around the age verification, because it is location-specific, right?

[JM]: These are governments that are passing these laws.

[JM]: So people who want to get around it

[JM]: will almost always just use a VPN to connect to a part of the world that is sane enough not to pass this kind of government surveillance legislation.

[JM]: In addition to it being ineffective, there are other more secure and privacy-focused ways of safeguarding minors.

[JM]: The device-level controls that are built into iPhones, and presumably Android devices, this to me feels like the most effective way

[JM]: to manage children's internet access.

[JM]: And yet instead, we have this flawed legislation that is causing people to do things like prompt users to take selfies with government IDs that again, are getting breached and shared and abused.

[JM]: And my hope — but certainly not my expectation — is that certain governments will take a look at breaches like this and think, "You know what, maybe we should not do this."

[DJ]: I hope that too.

[DJ]: And yet, it's a faint hope.

[DJ]: And my frustration with this is that you said the legislation is ineffective, and probably pretty ineffective at actually protecting any human beings from material harms, which is nominally its purpose.

[DJ]: But I think I can't help looking at it as like, oh, no, no, no.

[DJ]: This legislation is effective at, you know, shoring up people's political careers and making the broad mass of society feel as though something, *something* is being done about these terrible harms that we keep hearing so much about, whether they're actually substantiated or not.

[DJ]: They're generally not.

[DJ]: And that's what's frustrating to me is just this notion that like the system gets proposed, people who actually know anything about it go, "That's a really bad idea."

[DJ]: "It's going to have these consequences and it's not really going to accomplish what you want it to."

[DJ]: It gets implemented and it turns out to be a really bad idea that has these predictable consequences and probably isn't going to accomplish what we want it to.

[DJ]: And what's going to happen?

[DJ]: Is there going to be this like governments are going to go like, "Oh, man, we were wrong."

[DJ]: "We're going to repeal this legislation because it's such a bad idea."

[DJ]: No.

[DJ]: Instead, what's going to happen is Discord is going to offer the affected users a year of credit-protection services.

[DJ]: And then there will be a class action lawsuit.

[DJ]: And a few years from now, several people will receive $2.15 in compensation in return for all of their personal identification being stolen looking forward.

[DJ]: And that's going to happen again and again and again and again and again.

[DJ]: Infuriatingly, we, society writ large, and our institutions keep doing these stupid things,

[DJ]: crap things, and there's essentially no consequences.

[DJ]: I mean, there are bad consequences for *us*.

[DJ]: There's essentially no consequences for the people who impose this stuff.

[JM]: You're right that this legislation is indeed effective — not at its stated purpose of protecting children,

[JM]: but it's effective at the government surveillance that it is actually trying behind the scenes to implement, right?

[JM]: Like if you can identify a user with government ID, well, then when you subpoena the company that's collected this, well, everything you've done on that platform is now identifiable directly to you, like with receipts.

[JM]: So it is a authoritarian government's wet dream to have

[JM]: this kind of data on individual users that use a platform.

[JM]: So yeah, of course they're going to want to see it everywhere, and using it to protect children is just the usual "Do it for the children" smoke-screen.

[DJ]: Some companies, like at least when a company makes a bad decision about like banning third party hard drives, they might actually suffer some sort of reputational damage.

[DJ]: But it's really weird to me that this kind of thing, like the companies demanding more and more and more personal information, in this case because governments are telling them to.

[DJ]: I'm sure the companies are not like losing any sleep over "we get to gather more potentially very valuable information about our users".

[DJ]: "That sucks."

[DJ]: "We hate that idea."

[DJ]: "We're definitely not going to do it."

[DJ]: "We're definitely not going to store people's personal ID in, I don't know, a non-reversible format so that it can't be stolen."

[DJ]: Incidentally, what?

[DJ]: How in a world where we've been salting password hashes for like 25 years, is there a database of people's driver's licenses that you can just yank?

[DJ]: How do you put regulation in place in 2025 that doesn't have some sort of technical details in it about how, by the way, you must try to store this information in a way that it isn't trivially stealosifiable?

[DJ]: It's really frustrating.

[DJ]: So anyway, I'm much happier talking about companies that reverse bad policies after their sales plummet.

[JM]: Indeed, because it seems that Synology, who we talked about before, making a decision to restrict the kinds of hard drives and solid state drives that you can put inside their popular line of network-attached storage has apparently reversed its policy

[JM]: and is now allowing third-party unapproved drives, reportedly because sales of their network-attached storage devices plummeted after implementing this policy.

[JM]: There's a part of me that wants to go, "Yay, people voted with their dollars!"

[JM]: "And a company that implemented a really customer hostile policy was forced by some definition of forced to reverse this policy and allow people to do the kinds of things that they have always expected to be able to do, which is to go out and buy a drive and put it into the enclosure that they bought..."

[JM]: "... and use it without scary warnings, or other indications in their software that 'Danger! Danger! This is an unapproved drive!', etc."

[JM]: But I'm not really sure that this is the full-throated reversal that perhaps it seems like it is by reading a headline. I haven't really delved into the details to see if it is a complete reversal.

[JM]: I will put a link to Synology's drive compatibility policies in the show notes.

[JM]: You can read it and thumb through it and see if it makes sense to you.

[JM]: It looks like a big wall of "I don't know what they're trying to communicate here."

[JM]: But for the sake of argument, let's just assume that this is a complete reversal.

[JM]: I don't really believe that it is, but let's just say that it is.

[JM]: If Synology and the folks that run it can restore functionality that they took away via a software update,

[JM]: then they can just take it away again, right?

[JM]: Like, why would anyone trust Synology not to pull something like this again in the future?

[JM]: Why would anyone trust Synology not to try other potentially-worse encrappification strategies in the future?

[JM]: So for me personally, their response does not restore the trust that they burned.

[JM]: They lit that goodwill on fire.

[JM]: And this response for me does nothing to earn it back.

[DJ]: Yeah, that makes a lot of sense.

[DJ]: There's an aphorism I really like that goes something like, "Trust takes years to build and seconds to destroy." The greedy grasping customer-hostile nature of the original decision.

[DJ]: It creates more than just an aversion to the fact of the decision, right?

[DJ]: It's not just like, "Oh,"

[DJ]: "I do not want to buy a Synology if it will show me a warning when I put a Western Digital hard drive in it."

[DJ]: "Oh, it will no longer show me that warning."

[DJ]: "Therefore, I shall buy a Synology again."

[DJ]: No, that's not the thing.

[DJ]: Like, instead, we're sitting here going like, "Oh, wow, you guys are a company that would do something like this?"

[DJ]: Right.

[DJ]: And then to your point, like, therefore, "What else will you do that I'm going to hate?"

[DJ]: "Maybe I'm just not going to buy your products anymore."

[DJ]: I don't know.

[DJ]: I think it comes back to that.

[DJ]: Like trust takes a long time to build and it's easy to lose.

[DJ]: And so companies should try not to lose it.

[DJ]: Companies like Apple, Apple and their operating systems.

[DJ]: Sorry, I'm all about the segues today, Justin.

[DJ]: I can't help it.

[JM]: On that note, why don't you tell us about your experiences with MacOS and iOS 26, which I have not used in any way, shape, or form.

[JM]: I only dunk on it by seeing screenshots and videos that other people have taken.

[JM]: Since you have had real-world, hands-on experience with these operating systems, please share with us your two cents.

[DJ]: I don't like them.

[DJ]: The thing is, there were all these takes when they came out, or even were first announced about the whole "Liquid Glass" user interface design thingy.

[DJ]: And so I almost feel like I'm just jumping on a bandwagon when I complain about it.

[DJ]: But no, I'm complaining about it because I don't like it.

[DJ]: And I will grant you, maybe it's one of those things where in a couple of years, I don't even notice.

[DJ]: And in fact, if you brought me back to the way the user interface used to look, I'd be like, ugh.

[DJ]: But...

[DJ]: I doubt it.

[DJ]: There are just so many things about the weird, like glassy effects where they are taken completely in isolation.

[DJ]: They are impressive user interface effects.

[DJ]: Don't get me wrong, but they're bad and I don't like them.

[DJ]: And especially the way they were pitched.

[DJ]: So this kind of goes to the thing about trust, right?

[DJ]: Where it's not just that a company, like a little warning in your Synology software that says your hard drive is scary, you should be scared.

[DJ]: That itself is a bad experience, but what it implies about the people who made it is worse, right?

[DJ]: "Oh no, I'm using a computer made by charlatans."

[DJ]: Like that's the real reaction.

[DJ]: I will talk about a couple of very specific things I dislike about the Liquid Glass interface in a minute,

[DJ]: But behind it all is the way it was pitched by its creators, which was this whole rigmarole about like, elevating your content and bringing the focus to your content and oh, the user interface should vanish so that you can pay attention to the content, which doesn't really make it clear how you operate your computer since you know, the user interface is has to be present for you to do that, but nevermind.

[DJ]: There are just so many little things in this that, again, maybe I'll love them in six months, maybe, but I doubt it.

[DJ]: Because you get these user interface elements, and let's say the dock is a good example, and it sits there at the bottom of the screen.

[DJ]: I don't want to think about it ever, right?

[DJ]: But I've had it overlap a window like Slack.

[DJ]: And in the bottom left of the Slack window, there's my list of channels and direct message conversations and things.

[DJ]: And one of them will be highlighted with a highlight that in my color scheme is reddish.

[DJ]: So there's this reddish bar in the lower left-hand part of the screen.

[DJ]: And when I scroll the channel list, the bar moves towards the top or bottom as I scroll through the channels.

[DJ]: The bar is the selected channel.

[DJ]: And now...

[DJ]: as of MacOS Tahoe when the bar slides behind the dock, the dock suddenly has this huge warped red-like shape appear in it, as though the bar was sliding behind like a piece of glass and

[DJ]: every time this happens, my eyeballs, which are trying to focus on the actual content that's roughly in the center of my screen, are irresistibly drawn towards the movement of this stupid effect.

[DJ]: So not only do I not like the effect, but I'm also forced to rhetorically go to the people who made it and say, "This is bulls**t, guys, because you are distracting me from the thing that you told me you made this design to help me focus on."

[DJ]: So the thing I dislike the most about these operating systems is not just the visual effects themselves, which, sorry, I might not have made this clear, I don't like, but what I like even less is what it's doing to how I feel about Apple, because I've been a fan of their products, including their software, for a long, long time, and that's ceasing to be the case.

[DJ]: And that sucks.

[JM]: Every screenshot and video that I've seen of the 26-series of operating systems comes across as this somewhat cartoonish Fisher-Price-looking toy operating system for kids.

[JM]: Like it doesn't look like it's made for grownups who want to get actual productive things done.

[JM]: And I agree with you that one of the most galling things is the completely made-up rationale that they pitch this with, right?

[JM]: Saying this allows you to focus on the underlying content, which it by all accounts does not do and instead actively distracts you from the content that you were trying to focus on in the first place.

[JM]: Rather than comment on this further, again, as someone who's not even used it, I would like to direct you to a fantastic article called "Benjamin Button Reviews MacOS", in which this genius author starts out by saying Apple's first desktop operating system was Tahoe.

[JM]: "Like any first version, it had a lot of issues."

[JM]: "Users and critics flooded the web with negative reviews."

[JM]: And I'm not going to obviously read the entire article to you, but you read this and it goes from Tahoe to Sequoia to all the way to Mac OS X Cheetah.

[JM]: Cheetah is the latest refined version of this.

[JM]: Started out with Tahoe and then went all the way down to Cheetah as the refined, polished version of the operating system.

[JM]: And this to me is just pure gold.

[JM]: I love everything about it.

[DJ]: It's extremely wonderful.

[DJ]: You should definitely read it.

[DJ]: I want to ask you, Justin, because you made an interesting observation, which was about the unseriousness.

[DJ]: Like, as you said, it feels like childish, right, was your criticism of the current design scheme of MacOS and iOS and stuff too, right?

[DJ]: But I feel like and kind of as this article points out, like MacOS somehow is where this feels the worst.

[DJ]: I think if I had to say why it is perhaps because there's a couple of reasons, actually.

[DJ]: So, sorry, I'm wandering away from the question I was going to ask you.

[DJ]: I'll come back to it.

[DJ]: I do actually think that this design language works a little better on the touch-based smaller screen of an iPhone.

[DJ]: I still find the things distracting there.

[DJ]: One of the first places I noticed the distraction was actually in the Mail app on my new work iPhone.

[DJ]: I have these new operating systems on my work devices.

[DJ]: That's the difference, right?

[DJ]: Like neither Justin nor I have upgraded our personal devices to these operating systems because they give us big yikes.

[DJ]: But I don't have that choice at work.

[DJ]: Work decides when I upgrade my computer for better or worse.

[DJ]: So that's why I've gotten to try these.

[DJ]: But on my work phone, I go into the mail app and there's some...

[DJ]: I don't even remember what it is.

[DJ]: I think it's when you go back from a message to the message list that the buttons reshape themselves with an animation because there's different buttons on the two screens.

[DJ]: And again, like from a pure technology perspective, that's neat.

[DJ]: But what it made me do is I'm suddenly hyper-aware of the user interface of the Mail app, which I have never been before and have no desire to be.

[DJ]: Like I just expect to tap a button and jump back to the list of email messages.

[DJ]: And instead my eyes go, what was that?

[DJ]: Because there's this weird unexpected motion.

[DJ]: But I will say the elements do feel and look a little more natural on this screen, especially when you're touching them anyway, like with your finger.

[DJ]: They just feel weirder and more obtrusive on a computer that you're probably operating with a keyboard and a mouse.

[DJ]: So that's one of them.

[DJ]: And then I think the other one is...

[DJ]: a conceit that at least you and I share is that a computer is kind of where you do work.

[DJ]: I want to be like careful with the language here, but generally, yeah, like I try to accomplish things with my computer, like writing or programming or recording a podcast or what have you.

[DJ]: Whereas I think my phone really is more of a, this is a cliche at this point, but like consumption device.

[DJ]: Like I usually pull out my phone to look at something, not to make something.

[DJ]: And so this new design pattern feels really obtrusive on the computer, both for the sort of ergonomic reasons, but also to your point, it's like, this feels kind of like

[DJ]: yeah a toy or childish or whatever but if you do think of your computer as "I just want to come here and make stuff and get things done — please stop distracting me with buttons changing shape and stuff like that." That does feel bad now, that does lead me to my question which is if you do go back especially back quite a long way to the first half a dozen versions of MacOS,

[DJ]: I would argue that their user interface was a little childish-looking in certain ways, too.

[DJ]: They used to have these progress bars that were diagonally striped with bright colors, sort of like a candy cane, if you were using it back then.

[DJ]: How did the user interface strike you at that point in its evolution?

[JM]: I don't recall personally feeling nor do I recall anyone commenting that this interface felt like a child's toy in terms of look, feel, how one uses it, etc.

[JM]: I remember it was very innovative for its time in terms of the usage of translucency

[JM]: and effects that made buttons appear spherical.

[JM]: It's easy to look back at the pinstriping that you pointed out and say, okay, well, it's understandable why in subsequent versions of the operating system, they pulled out the pinstriping, they dialed down some of the aqua, as it was called, effects from their operating system.

[JM]: But when I look back at it, sure, like anything else, because it was released a long time ago, yeah, it feels a little dated.

[JM]: But at no point do I look at it and think, if I were to use that today, I would find it unpleasant or it wouldn't feel like a serious operating system.

[JM]: That's something I would still use today and would enjoy using it.

[JM]: I don't know that that's how I'm going to feel about Tahoe.

[JM]: But again, I have to reserve judgment because I haven't used it.

[DJ]: That's fair.

[DJ]: I find myself interested in digging into the *why*, like if we can even find it, and not because I'm trying to refute a viewpoint.

[DJ]: If we just wanted a five-second soundbite on the new operating systems: I don't like them.

[DJ]: It's that simple.

[DJ]: I don't like them, but...

[DJ]: I am curious because I agree with you that the early versions from like from 20 years ago was like the first time I used a modern Mac, like running like Mac OS X, as opposed to this so-called like classic Mac operating system that came before it.

[DJ]: And when I think about it, it always was kind of playful, or dare I say, maybe a little childish.

[DJ]: I mean, the Mac's fundamental icon is like this little computer with a smiley face.

[DJ]: There is something fundamentally sort of Fisher-Price-y about it, frankly.

[DJ]: But at the same time, that's never really turned me off until now.

[DJ]: But it is a little weird, though, that it started off very...

[DJ]: pinstripey and brushed metal.

[DJ]: Apple went through these various phases of having an operating system user interface that didn't exactly fade into the background.

[DJ]: I think at the beginning it actually felt novel to me because it was such an interesting departure from like Windows.

[DJ]: The first time I minimized a window on MacOS and it does the so-called "genie effect" where it's sort of the window like warps its shape and appears to be like sucked down into a little thing in the dock.

[DJ]: I love that.

[DJ]: Actually, I still love it.

[DJ]: That I would describe as novel.

[DJ]: It's fun, but also it's not obtrusive.

[DJ]: And something about this latest paradigm just feels so much more...

[DJ]: It feels like the designers are saying...

[DJ]: "Huh?"

[DJ]: "What are you doing on your Mac?"

[DJ]: "I don't care about that."

[DJ]: "Look, look at the user interface."

[DJ]: And I'm like, "No, I know the point of the user interface is for it to disappear so I can get my work done."

[DJ]: And even then, like over the past few years, Apple went through a period where they did too much disappearing, where they started being like, well, "We don't have buttons anymore."

[DJ]: Everything's hidden until you mouse over it.

[DJ]: And again, it's like, no, that's bad too.

[DJ]: Guys, it's just and this is what I meant about how it's starting to sour my feelings about just the company as a whole, because it just feels like they've lost the plot with how to give me a system that I actually am happy to use every day.

[JM]: I think the difference is that the early versions of Mac OS X felt whimsical in ways that were pleasing.

[JM]: Whether it was the design of the Finder icon or whether the genie window minimization effect into the dock, whatever those things were, they didn't feel childish.

[JM]: They didn't feel like Fisher-Price toys.

[JM]: They felt fun and whimsical in a very boring beige Windows world.

[JM]: And that's what attracted so many people to the platform.

[JM]: What makes this round of design changes different in my inexperienced mind — again, as it relates to not having used these new operating systems — are these heavily-rounded corners, colors that don't seem to go together, inconsistency in the user interface design,

[JM]: the lower information density, and just the overly flat nature of the design as it's evolved.

[JM]: Just to use the window buttons as an example, I mentioned these glossy transparent spheres in red, yellow, and green

[JM]: that appeared in the Aqua interface at the upper left hand corner of windows.

[JM]: And over the years, all of that depth and character has been removed, to the point where you just have these very nondescript flat red, yellow, and green circles.

[JM]: They're just not interesting.

[JM]: And that's been around for a while.

[JM]: That's not specific to Tahoe.

[JM]: But to me, that's the kind of changes that I'm seeing in these screenshots that feels different to me than the early versions of Mac OS X.

[JM]: It just feels like a lot of the whimsy has been drained out of these user interface designs.

[DJ]: I will say that pleasantly whimsical is an excellent way to describe what was good about the early MacOS operating system.

[DJ]: And I also think it's a really good description of our podcast.

[JM]: Indeed it is.

[DJ]: Listen to abstractions.

[DJ]: It's pleasantly whimsical.

[DJ]: Something that sticks out to me about what you're saying is it almost feels like we're trying to make the opposite points at the same time about how it really sucks that they've drained all the joy and interest, but also I hate the fact that they're doing all this crazy new stuff.

[DJ]: I think the way that comes together is they've started adding, I don't know about whimsy, but something

[DJ]: back into their user interface, like the way that the colors warp and slide behind the magical Liquid Glass, but it actually doesn't really feel whimsical.

[DJ]: It's just obnoxious.

[DJ]: Like, that seems to be the difference.

[DJ]: Instead of taking the stoplight icons and making them look like little gemstones again, which might be whimsical,

[DJ]: I'm just distracted all the time now.

[DJ]: Like, why is the entire upper user interface of Safari randomly changing color, Justin?

[DJ]: Why?

[DJ]: Well, I'll tell you why.

[DJ]: It's because it's trying to match whatever arbitrary color appears at the top of the web site you're looking at.

[DJ]: Guess what?

[DJ]: That does not actually help you focus on the web site.

[DJ]: It makes you go, huh?

[DJ]: Why did my user interface suddenly change color?

[DJ]: I want someone to sit down with Alan Dye, who is, as I understand it, the head of ruining MacOS at Apple.

[DJ]: That might not be his official job title.

[DJ]: And just ask him all the questions that the consumer side is asking, which is just like, "Why do you think this is a good idea?"

[JM]: I think there's a reason we've never seen him on some talk show or interview, because that...

[JM]: probably wouldn't go the way he wants it to go.

[JM]: Unsurprisingly, I'm going to stick with the pre-26 versions of these operating systems for a while.

[JM]: And perhaps I'll change my mind at some point, but for now I think I'm good.

[DJ]: No, it's fair.

[DJ]: I think you'll end up upgrading by accident when they like switch the, cause they do that now, right?

[DJ]: Where you tap to update to like 15.7.1 and first the screen slides over and goes like, "OS 26!"

[DJ]: And you're like, "Whoa, wait a minute."

[DJ]: "I almost tapped on that."

[JM]: I am so mad about this.

[JM]: Just the other day, I went to update from 15.7 to 15.7.1 and I tap on the little info icon next to that Sequoia update, which is below

[JM]: the big advertisement for Tahoe OS 26, which I am, as I've said, not interested in at the moment.

[JM]: And so I tap on the tiny "i" next to the update that I actually want.

[JM]: And on the next screen where I can select which of the related updates I want, the thing that's selected is the Tahoe OS 26 thing that I definitely did not tap next to.

[JM]: "No", I said, "Show me the info about the 15.7.1 update."

[JM]: And it shows me a screen where if I just tap the button to proceed, it would go ahead and install Tahoe.

[JM]: Not the thing I selected.

[JM]: It is this really evil dark pattern that is beneath Apple.

[JM]: It is just beneath them.

[JM]: And this is not a bug.

[JM]: This behavior has been around for a while.

[JM]: This is not new.

[JM]: And they're doing this intentionally.

[JM]: And it makes me really angry.

[DJ]: Same.

[DJ]: I'm glad you used the phrase "dark patterns" because that came to me too.

[DJ]: It's interesting that it's not that recent because I hadn't really noticed it before.

[DJ]: And I was going to say it's like someone or a group of people at Apple found a web site describing dark patterns and they thought, "Oh, can we do all of these?"

[DJ]: "How do we get all of these into the next release?"

[JM]: I'm pretty sure that at no point in the Apple HIG or Human Interface Guidelines, I'm pretty sure there's not a Dark Patterns chapter that someone's following.

[JM]: "Hmm."

[JM]: "You know what we should do when they select this one update?"

[DJ]: Are you sure that's still true?

[DJ]: They really might've added one.

[DJ]: Seems like.

[JM]: Yes, well, in other news, lots of companies are investing lots of money in generative software and this investment seems to be accelerating at a very, very rapid pace.

[JM]: Two weeks ago, Nvidia agreed to invest as much as $100 billion in OpenAI to help them fund a data center build-out.

[JM]: And in turn, OpenAI committed to filling these data centers with unholy numbers of Nvidia chips.

[JM]: The day after that announcement, OpenAI confirmed that it made a separate $300 billion deal with Oracle to build out data centers in the United States.

[JM]: Oracle in turn is spending billions on Nvidia chips to put into servers in those data centers, sending money back to Nvidia.

[JM]: And just this week, OpenAI signed a partnership with Nvidia rival

[JM]: AMD to deploy tens of billions of dollars worth of its chips.

[JM]: And as part of that deal, OpenAI will become one of AMD's largest shareholders.

[JM]: All of this activity is causing people to take a really hard look at what's going on right now

[JM]: and ask the question: "Are we in a generative software bubble?"

[JM]: So let's talk a little bit about what the evidence shows for or against this idea that we might be in some so-called "AI" investment bubble.

[JM]: One analyst who believes that we are in said bubble said that this bubble is 17 times the size of the dot-com frenzy and four times bigger than the subprime real estate meltdown in 2008.

[JM]: Going back to OpenAI for a moment, who again is not the only company pouring unholy amounts of money into generative software, chips, and data centers.

[JM]: But OpenAI alone has now struck these deals with Nvidia, AMD, Oracle, that in total could easily top $1 trillion.

[JM]: And you might be forgiven for thinking that a startup that spends a trillion dollars is perhaps a lot of money.

[JM]: I can't think of too many startups that have spent a trillion dollars.

[DJ]: If you came to me and you're just like, "Dan, I'm sorry, but I think that a startup spending a trillion dollars is an awful lot of money."

[DJ]: My response would be: "I forgive you."

[DJ]: "Also, how much money did you say again?"

[DJ]: ... would also be part of my response.

[JM]: Right.

[JM]: But even if you thought that a trillion dollars was a lot of money to raise and theoretically spend, Sam Altman, the CEO of OpenAI, has said publicly that he expects OpenAI to invest trillions, plural, to build data centers to run their generative software models.

[JM]: And practically speaking, I don't know that this has ever been done before.

[JM]: I don't know that any startup has tried that, has made it part of their platform that we're going to raise trillions of dollars.

[JM]: And even if you did do that, the chance of you going to investors and saying, okay,

[JM]: "We want you to give us trillions of dollars," most of the time, investors are going to pause and perhaps have cold feet about the idea of writing checks of that magnitude.

[JM]: And that's where I feel like these circular deals come in, because it's hard to borrow trillions of dollars from investors.

[JM]: So if you make deals instead in which OpenAI gets, say, 10% of AMD, and that announcement drives AMD's stock price up, which means your 10% share of the stock just also went up a whole lot,

[JM]: effectively paying for the graphic processing units, the chips that OpenAI would otherwise have had to purchase from AMD.

[JM]: Nvidia invests $100 billion into OpenAI, which OpenAI uses to then buy those chips that they would otherwise have had to just outright pay cash for.

[JM]: So in some ways, it feels a bit like an end-run around the traditional, "I'm a startup, I go out and I get investors to give me money that I invest in the business."

[JM]: Well, now you don't have to do that as much if you can get a company to give you the money, invest in you, so that you can then spend it on their products.

[JM]: As the chapter art for this chapter, I'm going to put an image of

[JM]: a little diagram of all of these deals that I just mentioned and how the flows go from one company to another because it's somewhat entertaining to look at.

[DJ]: Have you considered putting an image of a black hole as the chapter art?

[DJ]: I realize our show's art already kind of looks like a black hole, but...

[DJ]: It's just occurring to me as you're describing it that it's almost like OpenAI has kind of become like the investment vehicle version of a black hole where like its mass is now so great that it's just pulling in money from all the companies around it.

[JM]: Yeah.

[JM]: And some of us are old enough to remember the dot-com bubble, and the similar kinds of circular deals that were a hallmark of that financial market collapse.

[JM]: So to me, this bubble definitely feels like it is an order of magnitude bigger.

[JM]: Once again, 17 times bigger, according to one analyst.

[JM]: And it's all predicated on there being huge returns, right?

[JM]: Because otherwise, why would you invest trillions of dollars into one company?

[JM]: Mind you, not just the entire industry.

[JM]: One company wants to raise trillions of dollars.

[JM]: So if you look at it as an industry in the whole, it's many trillions of dollars, supposedly.

[JM]: And so you need to get many trillions of dollars back in terms of returns.

[JM]: And the only way you can do that is if you have these very significant productivity gains.

[JM]: And I don't know that the industry is going to get there.

[JM]: Like if you look at the trajectory, ChatGPT 3 comes out, wows everyone, and it costs $50 million to ship.

[JM]: ChatGPT 4 comes out, people are similarly impressed with the gains it's made,

[JM]: relative to the previous version.

[JM]: But instead of $50 million, now it's $500 million to produce. ChatGPT 5 comes out and costs $5 billion to produce, but it was delayed.

[JM]: And when it was released, people were like, "I guess it's a little better."

[JM]: It wasn't like a huge win over the previous version, but it still cost an entire order of magnitude more in terms of investment.

[DJ]: Are those really the numbers?

[DJ]: Like it was like $50 mil, $500 mil, and $5 billion for the major versions of GPT?

[DJ]: That is my understanding.

[DJ]: Wow.

[JM]: Okay.

[JM]: So if there's this huge amount of investment required, and it seems like we're getting diminishing returns in terms of the advancement

[JM]: of those models over time.

[JM]: That doesn't bode well.

[JM]: And in addition to that, it doesn't seem like there's much of a moat either.

[JM]: It doesn't seem like there's a lot that prevents other companies from producing models that are of similar caliber, whether it's DeepSeek or others.

[JM]: It doesn't seem like there's enough of a competitive advantage after spending potentially trillions of dollars relative to other competitors.

[JM]: And even if all of that weren't true, then there's the question, again, returning to productivity, there's a study done by METR that looked at employees actual work output.

[JM]: And they did the study.

[JM]: And at the end, they noticed that developers had completed tasks 20% slower using these tools than when working without them.

[JM]: And the researchers were stunned because this is not the result they expected.

[JM]: Now, this is one study.

[JM]: Who knows what the results of further studies will show?

[JM]: And I'm sure there will be a lot of them.

[JM]: But it does point to something that folks call a capability reliability gap.

[JM]: The old CRG.

[JM]: Where the capability improves, but the reliability isn't keeping up.

[JM]: And that gap might explain why...

[JM]: these generative software tools have failed to deliver tangible results for the businesses that use it.

[JM]: It seems that some researchers at MIT looked at the results of 300 publicly disclosed generative software projects, and they found that 95% of them failed to deliver

[JM]: any increase in profitability.

[JM]: A separate report from McKinsey and company found that more than 80% of companies reporting having used generative software tools said that the technology had no tangible impact on their earnings.

[JM]: So it's possible that we've entered into a trough of disillusionment phase of the development of this technology.

[JM]: And there is the possibility, like a lot of technologies, that there is a productivity J curve where we're in that trough of disillusionment, but we are approaching the

[JM]: upward part of the J, right?

[JM]: The hockey stick part where all of a sudden productivity just skyrockets and goes through the roof, but it may not.

[JM]: And what happens if a whole bunch of managers push for generative software use at their organizations, even though it could hurt productivity instead of helping it, which is the point of them pushing it into their organizations.

[JM]: If enough companies do this, we could be looking at very serious harm to productivity over time.

[JM]: One person said, quote, email was one of those technologies that made us feel more productive, but actually did the opposite.

[JM]: I worry we may be headed down the same path with AI, end quote.

[JM]: And I have very similar concerns that this could very well be the case that

[JM]: As more and more people are handed these generative software tools and told, "Here, use these."

[JM]: And it actually hurts their productivity and their morale because some folks don't really enjoy using them, in part because it might make them less productive.

[JM]: I do have concerns that you start to combine all of these things together.

[JM]: And it doesn't look like this will end well, because as I said before, the amount of money

[JM]: that the industry would have to make to justify all the spending is just astronomical.

[JM]: You're talking about trillions of dollars.

[JM]: I just don't see how this money gets made back.

[DJ]: Yes.

[DJ]: And the focus on just this very small set of companies feels brittle.

[DJ]: I mean, I'm not an investment professional.

[DJ]: I would not be foolish enough to give investment advice on a podcast if I were an investment professional, but it just, it's really unclear to me with these, these giant companies that are moving so much capital around.

[DJ]: I just want to be like,

[DJ]: "What do you guys actually really think the upside of this is?"

[DJ]: "What outcome are you actually betting on?"

[DJ]: Because I don't hear answers to that question that fill me with hope.

[DJ]: One of the articles that we were looking through when we were kind of researching this topic, there was a quote from an analyst.

[DJ]: And this analyst was saying about Sam Altman, the CEO of OpenAI,

[DJ]: "Altman is either leading us to a massive economic downturn or to the promised land, but it's not clear which is which."

[DJ]: They said something like that.

[DJ]: And I couldn't help reacting to this quote by saying, well, yeah, it's not clear which is which, but the problem I have with that dichotomy is,

[DJ]: is that one of those things has some kind of basis in material reality.

[DJ]: Like, we know what an economic downturn looks like, and we've been there before, but their use of the term "the promised land" is both unfortunate and extremely on-point for this whole discussion, because there really does seem to be this notion that, like, so much of what's driving the hype around large language model powered AI and etc.,

[DJ]: is this notion that there's some kind of vaguely specified utopian future awaiting all of humanity because we all know how investment in large-scale technology tends to benefit everyone and not just a tiny handful of people disproportionately.

[DJ]: That's what kind of gives me the heebie-jeebies about this whole thing is like all these companies investing so much in each other.

[DJ]: Yeah, feels kind of fundamentally broken.

[DJ]: But then also it's like, "Wait, sorry, I didn't get it."

[DJ]: "What was the upside you were hoping for?"

[DJ]: "No one will have to write emails anymore."

[DJ]: "That doesn't seem like a $100 trillion payoff."

[DJ]: "Oh, it's actually the next phase of human evolution."

[DJ]: Well, that's even if that is possible, and I strongly doubt it, but what do I know?

[DJ]: It doesn't really seem like the kind of thing that should be run by a single publicly-traded American technology company.

[DJ]: Just saying.

[DJ]: So... "Yikes", I guess, is my response to all of this.

[DJ]: I think to your point, like it just, yeah, seems bad.

[DJ]: It seems like a bad sitch.

[JM]: It seems to me the only way out of this is for there to be some big breakthrough.

[JM]: Because right now it looks like they're running up against a wall, this wall of diminishing returns where they're not making big gains, despite pouring gigantic sums of money

[JM]: into data centers and chips and research and development.

[JM]: And all we're getting is very slightly better stochastic parrots where they continue to confabulate.

[JM]: And unless there is some wildly unexpected breakthrough, I just don't see how they get out of this seemingly inevitable future that I don't think looks very rosy.

[JM]: Particularly when you think about how these things are trained, right?

[JM]: They're trained on all of our output.

[JM]: They're trained on all of human knowledge, or at least most of it.

[JM]: But increasingly, the things they're going to be training on are the output of generative models.

[JM]: And so you're going to get this feedback loop where they train on...

[JM]: the output of generative models, which goes back into the model, and then they train on that.

[JM]: And I don't see how this doesn't cause some rather serious degree of degradation over time.

[DJ]: Maybe.

[DJ]: I'd love to see some more details on that because I've heard a lot of people make that assumption that just like we see certain problems emerge from, you know, what's known as human inbreeding, right?

[DJ]: Where if you don't keep a gene pool somewhat refreshed with new genetic information, there are

[DJ]: what we would call various kinds of degradation.

[DJ]: Everyone seems to just sort of take it as a given that if you pump the output of an LLM into the training data of an LLM, therefore it's bad.

[DJ]: I mean, yeah, that sounds right, but I guess show me some proof that that's actually true.

[DJ]: I'd be interested in seeing.

[DJ]: However, it does also seem to be the case that, like, what freaks me out a bit is that it's not clear what the breakthrough is, especially because, like, I've seen, like, GPT-3, 3.5, 4...

[DJ]: And for the kinds of capabilities we were trying to get out of these large language models, you could see like a real noticeable increase in capability.

[DJ]: But as that seems to tail off, the curve either has to gain a steeper slope again, so that like GPT-6 is actually way, way, way, way, way better at writing emails.

[DJ]: I think that might be part of it too, is that like when people talk about productivity and

[DJ]: it's not always clear what they mean, right?

[DJ]: Like a lot of the focus is on knowledge work and stuff like it can write your emails for you or whatever.

[DJ]: But like, I don't see how writing emails faster leads to companies massively increasing their revenue.

[DJ]: I don't understand what the narratives.

[DJ]: Yeah.

[DJ]: Like the, the money doesn't have narratives that accompany it that go like, right.

[DJ]: Yeah.

[DJ]: Okay.

[DJ]: I can see why you're pouring all of these resources into this.

[DJ]: I can't see it.

[DJ]: Can't see it, which is probably why I'm not going to be a billionaire.

[JM]: To be clear, I'm not saying that I have evidence that there is this generative slop training feedback loop degradation.

[JM]: I'm mainly just going by feels at this point.

[JM]: And also the idea that the slop generation scales in a way that it couldn't before, right?

[JM]: So if you look at human knowledge that the initial models were trained on...

[JM]: There's a limit to the degree that humans could just put up a bunch of stuff that's made-up, right?

[JM]: There's plenty of hoaxes and conspiracy theorists that post screeds on the interwebs that aren't true, right?

[JM]: But there's a limit to the degree that humans can just sit there all day long and type out stuff that's made up.

[JM]: There's no limit...

[JM]: on the amount of made up confabulated slop that generative software can output.

[JM]: And so I think that to me is why just logically from a reasoning perspective, I have concerns about how this will work unless they magically can

[JM]: figure out ways of detecting confabulation when they do their training.

[JM]: But if they could do that, then the models themselves wouldn't confabulate in the first place.

[JM]: So I'm going to go back to square one, which is, yeah, it's feels for me, but I'm making my prediction:

[JM]: I think this is going to be a problem.

[DJ]: Yeah, that's fair.

[DJ]: I'm sorry, I poked the bear on that one, because I actually think the models are the whole thing where it's like, well, "models aren't going to get better because they're trained on AI slop" is not really even the most interesting/horrifying part of this story.

[DJ]: To me, the more horrifying thing is, again, it's just like, guys, we got to invest trillions.

[DJ]: And I'm just not hearing any version of for what?

[DJ]: To what end, other than to let Sam Altman buy 15 houses, I guess, that like, makes it clear why we would want to do this.

[DJ]: Like if you look at you talked about moats, and one of that is from an investment perspective, right?

[DJ]: That's a concern.

[DJ]: It's like, yeah, okay, OpenAI, we're giving you all this money, but like,

[DJ]: How are you going to make it back when your competitors can just do exactly what you did?

[DJ]: And so OpenAI is like putting out products, which is interesting.

[DJ]: The fact that they're putting out products is interesting, but then you look at the products they're putting out and they put out this Sora thing, which as far as I can tell is just Facebook where all of the content is generated by AI instead of almost all of it.

[DJ]: And it's like that, that's the epicenter of human achievement, Sam, that's what you've been working towards for all these decades.

[DJ]: And I'm sure he would be like, no, of course not.

[DJ]: You don't understand my vision, dah, dah, dah, dah, whatever.

[DJ]: But it's like,

[DJ]: just the stuff that we're getting on the way to this hypothetical promised land is like, yeah, I don't know.

[DJ]: It does not fill me with a sense of wonder and amazement.

[JM]: It will be interesting to see where the market goes because every time I've thought, okay, the market's got to be near its apex, it just keeps going and going and going up.

[JM]: And obviously that just cannot continue forever, particularly if some of the concerns that we've raised here are valid.

[JM]: So I guess… Grab a bucket of popcorn and strap in and enjoy what should be a really interesting ride.

[JM]: All right, everybody.

[JM]: Thanks for listening.

[JM]: We hope you enjoyed the show.

[JM]: You can find me on the web at justinmayer.com and you can find Dan on the web at danj.ca.

[JM]: Share your thoughts about this episode via the Fediverse at justin.ramble.space.