Great Dog Uprising

[JM]: You know that phrase, "Nobody knows you're a dog on the Internet?"

[DJ]: I'm not sure I do, actually.

[DJ]: Is that the dog sitting in a burning house with a cup of coffee saying, this is fine, or something else?

[JM]: No, not that one.

[JM]: This is a cartoon from 1993 published in The New Yorker by Peter Steiner.

[JM]: And it's a dog sitting at a desk in front of a computer talking to another dog nearby saying, "On the Internet, nobody knows you're a dog."

[JM]: Because in 1993, the Internet was very much new, to most people, anyway.

[JM]: And the idea, of course, is that when you're on the Internet, you can pretend to be someone else.

[JM]: And this author was having fun with this concept and the idea that a dog could be on the Internet and other people wouldn't know it's a dog.

[JM]: And I'll make this comic the chapter art here just so you can check it out.

[DJ]: This comic raises so many more questions than it answers.

[DJ]: So one dog was talking to the other dog while using the Internet, but the dog was saying, on the Internet, no one knows you're a dog.

[DJ]: So I assume they're talking to people who are not dogs, i.e.

[DJ]: humans, and sort of tricking them into thinking that they're also a human, even though they're a dog.

[DJ]: And then they're sharing this with their dog companion.

[DJ]: So like, does that dog also want to get on the Internet and pretend to be a human?

[DJ]: Like, is this something dogs have been wanting for centuries and only now with the rise of the Internet, they can finally trick us into thinking that they're people?

[JM]: It's all part of the inevitable Great Dog Uprising of 2038 that's coming.

[JM]: So yeah, yes, they've long wanted to do this.

[DJ]: I mean, I guess I'd ask them why it's taking them so long if that comic is from 1993.

[DJ]: Like, come on, guys.

[JM]: Yeah, they're dogs.

[JM]: Takes a little longer, apparently.

[DJ]: Well, they're pretty distracted by chew toys and stuff like that.

[DJ]: Dogs are not great project managers is something I've noticed, even if they can use computers.

[JM]: That is very true.

[JM]: I mentioned this comic because we now live in an era in which it's not a question of whether or not the people you're interacting with on the Internet are dogs or not, but the question of whether they are humans or not is very much an open question these days.

[DJ]: Right.

[DJ]: There's a lot of parakeets using the Internet now.

[DJ]: The occasional iguana.

[DJ]: It's troubling.

[JM]: It's a problem, yes, but less from the animal kingdom and more from software bots powered by large language models.

[JM]: And we have talked a little bit about the concept of potentially creating a web of trust and vouching for other humans as being human.

[JM]: And a few days ago, I came across a project that I wanted to mention called human.json.

[DJ]: Because there's no better representation of a human being than the JavaScript Object Notation.

[DJ]: There's no better tool for capturing all the multifaceted dimensions of humanity than a text-based data representation language with squiggly braces and equal signs.

[DJ]: Or no, sorry, colons, not equal signs.

[DJ]: Screwed up the JSON bit.

[JM]: You're right that using an obscure data structure format is a strange way to represent humans.

[JM]: It's a good point.

[DJ]: I mean, I think the bottom line is XML would have been better, but here we are.

[JM]: So the idea here is that there's an open source project that I will link to in the show notes, and it describes it this way.

[JM]: "Human.json is a lightweight protocol for humans to assert authorship of their site content and vouch for the humanity of others.

[JM]: It uses URL ownership as identity and trust propagates through a crawlable web of vouchers between sites.

[JM]: One of the problems with the Internet today is that a lot of the content is AI generated.

[JM]: There's no way to know for sure if a site is maintained by a real human or if it's just slop.

[JM]: The only way to know for sure is by getting to know the authors, which usually takes time and requires developing a relationship with them through other channels like email or social media.

[JM]: But what if we could expand that trust by building a web of vouchers between sites?

[JM]: Human.json is a protocol based on that trust.

[JM]: It works the following way.

[JM]: You declare yourself to be a human and your commitment to avoid AI generated content by posting a human.json file somewhere on your site and linking to it.

[JM]: The presence of this file indicates that the content of your site was generated by you.

[JM]: It's fine to use AI tools that assist you with spell checking, grammar, formatting, etc.

[JM]: In the same human.json file, you can also vouch for the humanity of other sites that you trust.

[JM]: This is optional, but it creates a network of trusted sites that allows people to discover other real humans by following that web of vouchers.

[JM]: Now, if someone trusts you, they can expand the scope of that trust to the people that you're vouching for."

[JM]: And then there's also browser extensions provided for Chromium and Firefox that allow you to simplify the management of this trust and identifying if a site that you're on is maintained by a human.

[JM]: So for example, if you visit the site of someone who's published a human.json file, you will see a blue dot on the extension in your menu bar that indicates the presence of this human.json file.

[JM]: And from that extension, you can trust a particular site, you can remove that trust, you can block a site, and the color of the dot indicator will vary depending on how many hops are needed

[JM]: to reach the site you're on from someone that you trust.

[JM]: So if it's just zero to one hops, it's green, if it's two hops, it's yellow, three plus orange, and so on.

[JM]: And as I said before, I'm intrigued by the idea of creating webs of trust that indicate whether something that you're reading is

[JM]: has a high confidence level of being from a human.

[JM]: And assuming that you are careful with whom you trust, because the people that they mark as trusted affects your web of trust.

[JM]: So as long as you are selective about who you choose to trust, then it seems like a pretty decent solution.

[JM]: Provided, of course, that more than a handful of people ever implement this, which given how the world works, the probability of that happening is pretty low.

[JM]: But of course, I wanted to be the change that I want to see in the world.

[JM]: So of course, the first thing I did was create a plugin for a Pelican that allows you to

[JM]: easily generate a human.json file for your site.

[JM]: I started out by using a TOML file as the source for the people or the sites that I wanted to add, in part because the human.json spec really only has two bits of information for each site that you're trying to vouch for the link and the date that you're vouching for them.

[JM]: But there's a very real possibility that when I am adding links to my source list, that I may not necessarily remember that a particular link is associated with a particular human.

[JM]: And that is, after all, generally speaking, the point here.

[JM]: So I wanted to have another field for a name or at least a description of what this link is, because sometimes it's just not immediately discernible from just looking at a URL.

[JM]: And that's the only extra field at the moment that I have in my source TOML file.

[JM]: But I wasn't content to just have a source TOML file as the only way to add new items, because that could be kind of cumbersome and might add friction in terms of vouching for new links.

[JM]: So when you install this human plugin for Pelican, I also included a command line tool that you can run. It's just: `vouchfor`

[JM]: And when you run `vouchfor` and then

[JM]: one or more arguments, you can easily add new links and names for those links to your source TOML file.

[JM]: So I assume Dan that you rushed right out, given that you also have a Pelican site, and installed this plugin and generated your own human.json file.

[DJ]: Well, brace yourself for disappointment, my friend, because I didn't and I'm not going to because I don't really like this idea, for starters.

[DJ]: There are a couple of things I don't like about it.

[DJ]: I don't like the optics of it because there are no good precedents in our history for trying to declare which of us is and is not a human.

[DJ]: It's just I don't like the language.

[DJ]: Obviously, the person who created this is not saying like, we must hunt among the replicants for who is actually a human being.

[DJ]: I just watched Blade Runner 2049 the other day.

[DJ]: So I'm thinking about replicants.

[DJ]: Great movie.

[DJ]: But like what this is really about is, is your website full of what we call AI slop?

[DJ]: Or is it not?

[DJ]: Right?

[DJ]: Like, I think that's what this is really about.

[DJ]: So then the idea is take some of the load off of me as I go around the Internet trying to like scrutinize every piece of content that I look at to see whether or not I think someone actually put any care into its creation or whether it's merely the output of a large language model.

[DJ]: And to take some of that pressure off my back, we can create a little network of sites where someone else already did that and said, it's cool, it's cool.

[DJ]: Dan's website has content on it that he actually put effort into writing and isn't just the output of a large language model.

[DJ]: So I kind of wish we came up with a different way of asserting that than to say Dan is a human, because there is no alternative.

[DJ]: I can use an OpenClaw bot to probably register a domain name and create a website all by itself, and then post content that is just the output of a large language model.

[DJ]: But a human was still involved in that, right?

[DJ]: There's still some impetus that sets the wheel in motion.

[DJ]: That's more of a philosophical disagreement with the language being used in this project, but also a technological disagreement.

[DJ]: which is back in the day when blogs existed, but of course they still do exist, and these are the sorts of websites we're mostly talking about, right?

[DJ]: Like Engadget or the New York Times is not going to have a human.json file ever, sorry.

[DJ]: But you have a little website where you write things, and I have a little website where I write things, and I presume the creator of this project also does.

[DJ]: So, you know, those are basically what we used to call blogs way back when that on the Internet, nobody knows you're a dog comic was first illustrated.

[DJ]: And one thing that blogs used to do a lot of the time is they would have a set of links on them.

[DJ]: It was often known as a blog role where the author of the website would say, here are other websites I enjoy reading.

[JM]: I actually still have that at the very end of my homepage.

[JM]: I have usually three featured sites that are other sites that I, that I like and read and enjoy.

[DJ]: Nice.

[DJ]: And so I guess I'm saying that the aims of this project would be better realized instead of having this sort of abstruse technological solution that requires people to like build plugins into their website and install browser extensions and

[DJ]: by simply bringing back or making more popular that sharing of links from one author to their audience.

[DJ]: So just like, hey, you like reading my stuff?

[DJ]: Well, here are people I like to read.

[DJ]: And then those people can do the same.

[DJ]: I feel like that's a better solution to the problem trying to be grappled with here than this one.

[JM]: Regarding the philosophical point you raised, there is a certain tone in the project README that is clearly taking a stance.

[JM]: And I hear you with regards to that.

[JM]: And I don't necessarily agree with that.

[JM]: Like to me, I don't really care if someone uses a large language model to generate text

[JM]: and post it on their site as long as they stand by it.

[JM]: Like if they read it and they're like, "Yeah, this is representative of me.

[JM]: Like I would have written this myself more or less.

[JM]: And I stand by it."

[JM]: I personally don't have a problem

[JM]: if that's what people do.

[JM]: I think the author of this project feels otherwise, and that's fine.

[JM]: And yes, you could spawn your own OpenClaw instance and generate a whole bunch of slop.

[JM]: But if you did that, and I knew that you did that, I would remove you from my list of vouched sites because...

[DJ]: You would make a statement on the Internet that I do not belong among the body of humanity. I am not human, is what you would be saying.

[JM]: No, I would not remove you from the list.

[JM]: I misspoke.

[JM]: I would not remove you from the list.

[JM]: I would remove your OpenClaw-powered slop machine from the list.

[JM]: Your personal site that presumably you continue to write yourself, or at least mostly, or at least you stand by it, I would leave.

[JM]: So that's the point.

[JM]: You're not necessarily representing a human.

[JM]: You're saying that this *link* is written by a human, at least for the most part.

[DJ]: I know.

[DJ]: That's why I think the title of the project is a bad choice.

[JM]: I think that's a bit of a nitpick, but I hear you.

[DJ]: Well, it is definitionally a nitpick, but because there are multiple things I don't like about it, I get to gang that one along with them, right?

[DJ]: I agree with you that the other... My other main complaint is that...

[DJ]: The project, like a lot of such projects sort of take it as a fundamental premise that like using a large language model is bad.

[DJ]: It's like, it's like by putting this on your website, you assert that like you don't use AI or whatever.

[DJ]: And it's like, well, I may or may not like assert that I, uh,

[DJ]: It feels like the intent of the project is to approximate something that's better approximated some other way.

[DJ]: Because I guess one argument you could make is that the stuff we call AI slop, like the problem with it is not fundamentally that it was generated by a large language model.

[DJ]: The problem with it is it sucks.

[DJ]: And in order to make it not suck, a human being has to spend time.

[DJ]: Like even if you, let's say you want to write a post about something and you use a large language model to draft it.

[DJ]: If you just take what the large language model put out and posted it, it's not worth anyone's time to read because as we have by now all experienced in many times, the raw output of large language models is not interesting, is rarely insightful, is often misleading, et cetera, et cetera.

[DJ]: So it requires a human to go in there and be like, all right, I have a bunch of words now.

[DJ]: How do I shape them into the thing that I actually intended to say?

[DJ]: That has always been the craft of writing.

[DJ]: So I think what we're really complaining about is now there's lots of...

[DJ]: really low value written content on the Internet.

[DJ]: And we want to focus on the stuff that's actually high value, frankly, regardless of how it was created.

[DJ]: It just so happens that a lot of that, the low value stuff is just the raw output of an AI.

[DJ]: I don't know.

[DJ]: Maybe I am.

[DJ]: Maybe I'm just nitpicking.

[DJ]: I just think that this project has no hope of "solving" the "problem" that it hopes to.

[DJ]: So I'm not going to install it on my website.

[JM]: Which is indeed your prerogative, but I feel like there's a multitude of arguments being made here.

[JM]: The idea that ultimately we just want stuff that's good to me, I think is misguided.

[JM]: Like that is not the point here.

[JM]: If there are plenty of humans that write terrible things that I don't have any desire to read because it's bad, either because it's bad on some moral or ethical level or

[JM]: there's bad opinions or the writing is bad.

[JM]: There's tons of stuff that humans write that I have no interest in reading for a variety of reasons.

[JM]: But those people, according to this project, are still marked as having published something on a link that is made by a human.

[JM]: And that is the point.

[JM]: It's not about whether it's good.

[JM]: If...

[JM]: LLMs get to the point where everything they produce is amazing and better than most humans can write and more interesting and more pleasurable for us to read.

[JM]: That is also not the point.

[JM]: It is not the point of this project to distinguish between output that's good and bad.

[JM]: The point is for you to have some indication as to whether the thing you're reading was written by a human.

[JM]: And I think that has value.

[DJ]: I think you chose a curious way of arguing for that.

[DJ]: But I guess the point is that I guess the point you're making is that it's arbitrary.

[DJ]: Like what you care about is a human asserted they wrote this or a human neglected to assert that they wrote this.

[DJ]: And that's what it comes down because, as you said, like there's lots of stuff that humans write that is garbage that isn't worth reading,

[DJ]: that's offensive, or at the very least boring and pointless and repetitive and plagiaristic and etc.

[DJ]: But all that really matters is that they said, no, no, no, I didn't use a large language model to write this.

[DJ]: I guess I just don't get why that's actually valuable.

[DJ]: But, I mean, it's okay if you do think that's valuable.

[DJ]: I just don't see it.

[JM]: I think implicit in what you are saying is that you, Dan, personally don't care whether something that you're reading on some site is written by a human or not.

[JM]: Or maybe what you're saying is you don't care whether you know or not.

[JM]: Because if you did, presumably this project, if it actually got any traction and if it actually delivered on its premise, would provide you with value.

[DJ]: I think the latter thing that you mentioned is the case that like, I don't really care if I know or not.

[DJ]: I guess for me, yeah, what I'm saying is that when I read a piece of writing, I intend to evaluate it based on its inherent merits, which like emerge from the arrangement of the words.

[DJ]: How those words got arranged that way, does it really matter?

[DJ]: Now, right now I'm inclined to say no.

[DJ]: Of course, it's slightly easier to do that because we can see that basically the raw output of a large language model is low quality, right?

[DJ]: But okay, let's imagine a hypothetical future where there's a lot more like bots that at some point some humans set them in motion, but now like they created more bots that created more bots that created more websites.

[DJ]: And so there are things that like no human ever really intended to have created and yet it exists.

[DJ]: But it wasn't made by a person.

[DJ]: It was made by a software.

[DJ]: So then, like, is it important that I know that?

[DJ]: I don't know.

[DJ]: Like, for what reason is it important?

[DJ]: I guess that's kind of what I come back to with this is if I think about consuming a piece of writing, am I consuming it to be entertained?

[DJ]: Am I consuming it to be informed?

[DJ]: Am I consuming it to be changed in some way, which is what both of those prior things do to you ultimately?

[DJ]: And if so, like if in fact I am entertained, if in fact I am informed by a piece of writing, does it matter that it was written by a robot that was made by a robot that was made by a robot that was made by a robot?

[DJ]: I should have only done three of those.

[DJ]: That was made by a human?

[DJ]: Or does it only matter that it was enlightening or entertaining or etc.?

[DJ]: I don't actually know the answer to that question.

[DJ]: That's an interesting one, I guess.

[DJ]: But I guess at least at this point in time, I'm leaning more towards it doesn't really matter if the output came from a robot or a person if it was in fact entertaining or enlightening.

[JM]: Yeah, I think for you, it's a little bit like watching a magic trick where as long as you don't know how the illusion is done, you don't really care because you can just enjoy it as it is.

[JM]: So the not-knowing part for you is in some ways a feature and not a bug.

[JM]: And I think for me and for a lot of other people, knowing is kind of important.

[JM]: Like we have talked about contributions to open source projects and knowing

[JM]: you made this argument, if I am characterizing it fairly, that to some degree, we shouldn't care whether a pull request is written by a human or not, as long as it's good.

[JM]: And I think what I have argued is, no, the information is useful.

[JM]: What you choose to do with it as the person on the other side of this exchange is entirely up to you, but the information still has value.

[JM]: So to bring it back to this topic, if I'm reading some content on the Internet,

[JM]: to me, there's value in knowing whether a human wrote it, period.

[JM]: But that's just me.

[DJ]: Yeah, I get it.

[DJ]: I think that another thing that's come to my mind, which reminds me of maybe why I have this objection, is I care a lot about what's usually called the ad hominem fallacy, which is like a fallacy of reasoning that—let's see how badly I butcher this—

[DJ]: It refers to a form of argument where you attack the character, motive, or some other attribute of a person making the argument rather than the substance of their argument.

[DJ]: And I guess one of my interpretations of that fallacy has been an idea's value is inherent in the idea and not where it came from.

[DJ]: If a bad person makes a good point, it's still a good point.

[DJ]: It doesn't stop being a good point just because they're a bad person.

[DJ]: And one of the challenges of being

[DJ]: a reasoning being is to try to be able to hold contradictions like that in mind at the same time, right, is that like I want to try to evaluate this idea regardless of where it came from, and I suppose that's what's coming up for me here is that is why I'm making that argument that, like

[DJ]: fundamentally, if I find a piece of writing valuable, the fact that someone generated it using a large language model instead of carefully writing and editing it themselves, I think I would probably find that disappointing.

[DJ]: But I would still want to obtain the value of the actual idea.

[JM]: And you can, it sounds to me like your chief objection, or not really objection, but the thing that chaps your hide is the tone of the README.

[JM]: Because the tone of the README does have this ever-so-faint, "if you use generative software, you're a bad person, and you should feel bad" type of vibe.

[JM]: It's not explicit.

[JM]: But I get that anytime someone feels like they are criticizing people who use the tools,

[JM]: I think that raises your hackles a bit, understandably.

[JM]: Well, because it's – yes.

[DJ]: Like so many things in online discourse, it is the form of like a purity test essentially, which I'm just – I'm generally not a fan of.

[DJ]: Like it's one more way that people on the Internet can say, are you on the right team?

[DJ]: Because if you're not, you're bad.

[JM]: I'm just setting that whole messy thing aside in this discussion personally, because I just don't care that much about that part of it.

[JM]: If that's how the author of this feels, that people are bad for using these bad tools, I don't care.

[JM]: They can feel that way all day long.

[JM]: I don't feel that way about other people, generally speaking.

[JM]: But bringing it back to the topic at hand, to me, again, there's value in knowing whether something I'm consuming was crafted by a human or not.

[JM]: And I don't care about the whole stance that some people have.

[JM]: Because it works both ways, right?

[JM]: There's people that say that, "If you don't use these tools, you're some Luddite technophobe dinosaur who's going to be left in the dustbin of history.

[JM]: So have fun with that."

[JM]: I'm not interested in either side of these.

[JM]: Like they're just boring to me.

[JM]: There's plenty of interesting ethical questions, you know, about these tools, but I don't know.

[JM]: Neither of those really interest me.

[JM]: So in the end, I find this concept interesting and I find the implementation intriguing and

[JM]: and it'll probably go nowhere.

[JM]: But I enjoyed my attempt to manifest the success of this concept so that there can be people who, like me, potentially someday, who can determine whether something that they're reading or watching or whatever was written by a fellow human.

[DJ]: What you just said also is helping me articulate, I think, another reason that this doesn't appeal to me.

[DJ]: Because you said, like, you know, I want to know whether the thing I wrote was written by a human or not.

[DJ]: And I guess I can't get away from the notion that, like, I'm sorry, the only way you can really know that is to use your own reasoning, is to, like, read it and go, do I see any, do I sense any humanity in here?

[DJ]: Because, like...

[DJ]: The only thing the plugin really tells you is the person implemented the plugin on their website and enough other people who also implemented the plugin know about them to add them to the web, the web of trust, right?

[JM]: Yes, but all you're really saying is that you have to choose wisely.

[JM]: You have to choose the people that you're vouching for wisely because if you don't, then it defeats the purpose and it can be gamed.

[DJ]: No, I guess I'm saying, I guess I'm saying the problem, I guess it's about like, is it about people or is it about work?

[DJ]: Like, is it about content?

[DJ]: Because the web of trust is, asserts this person has relationships with other people.

[DJ]: It's like, okay, good to know.

[DJ]: But at the same time, like the problem is if you totally depend on that, which I realize is not what you're intending to do.

[DJ]: That's a little, so it's a little bit of a straw man, but like,

[DJ]: if you depend completely on the web of trust, then if you ever encounter some piece of writing in the wild whose author does not participate in this web, how will you know whether it was written by a human or not?

[DJ]: And I'm saying you need to be able to evaluate it like on its own and draw your own conclusions.

[JM]: Okay, but that presumes, as we've discussed, that the technology doesn't advance to a point where it's indistinguishable.

[JM]: So the point of this project is to foretell an era in which you cannot reasonably tell the difference by yourself and thus have this as a tool that you can use, should it provide that value for you.

[JM]: In any case, if there are listeners out there who have any interest in knowing whether or not the thing that they are interacting with is a human, you can check out some of the links that I will put in the show notes regarding this project and the plugin that I created.

[DJ]: But if you use WordPress, you're SOL.

[DJ]: Sorry.

[JM]: Oh, I am sure someone's already delivered that plugin already.

[JM]: All right, moving on.

[JM]: In other news, Nvidia has announced DLSS 5.

[JM]: And I don't know what DLSS stands for, but I know what it is.

[JM]: And it is a way for...

[JM]: Nvidia powered graphics cards to insert frames in between actual rendered frames based on its guess as to what those frames should be.

[JM]: So in other words, instead of cranking out ever faster and faster graphics cards, which is not easy to do.

[JM]: The idea is to, from some people's perspective, cheat and interpolate frames, thereby juicing the frame rate without having to actually render all of those frames per second.

[JM]: And DLSS 5 is in the news because in addition to...

[JM]: merely rendering the difference between say one frame and another and guessing as to what that frame would look like DLSS five seeks to improve the thing that it's rendering.

[JM]: So not just filling in the gap between one frame and another, but changing

[JM]: what is being rendered to make it say look more detailed and in this announcement Nvidia showed a video of a game in which a person in one frame is with DLSS 5 off and then in the next frame they show it with it on and in the frame with it on the person indeed

[JM]: looks much more photorealistic than the un-DLSS-ified person looks.

[JM]: And the Internet's resounding response to this announcement is a rabid collection of disgust, derision, and a lot of lulls.

[DJ]: Yeah, and most importantly, memes.

[DJ]: Don't forget the memes.

[JM]: So many memes.

[DJ]: That's actually what DLSS stands for.

[DJ]: So many memes.

[JM]: I didn't even know this whole DLSS5 thing was going on until I started seeing these memes appearing in my feed.

[JM]: And I felt like I was very much out of the loop.

[JM]: And over time, once you see enough of them, you sort of connect the dots and it became somewhat clear what people were talking about.

[JM]: And I'll put a link to one of these meme collections because some of them are really good.

[JM]: I think one of my favorites is Pac-Man, which is just a yellow circle with a piece of pie cut out of it.

[JM]: And the DLSS...

[DJ]: You know, Justin, I can't help but notice that you're struggling to describe Pac-Man.

[DJ]: But I think the other side of that is, frankly, if any of our listeners are not familiar with Pac-Man, they are beyond our help.

[JM]: Perhaps true.

[JM]: But so that's the one with DLSS5 off as your everyday, the Pac-Man that you probably can envision.

[JM]: And then next to it is the rendering with DLSS5 on where it's now a sphere replete with lighting and shadows.

[JM]: And the Pac-Man has human lips.

[DJ]: Oh no.

[JM]: And beard stubble.

[JM]: Genius.

[JM]: So good.

[DJ]: So are you going to run out and buy one of these video cards or are they not available on the market because OpenAI bought all of them ahead of time?

[JM]: No and yes, respectively.

[JM]: I have no interest in buying them.

[JM]: And no, even if I wanted to, probably can't because #generative-software-sucks-up-all-the-demand.

[JM]: I think this topic is interesting in part because it relates to our previous discussion about human-crafted content and art versus stuff churned out by some generative software.

[JM]: Because games are, like good writing...

[JM]: art crafted by humans, often painstakingly, often with emotion, often designed to elicit emotion.

[JM]: And as you pointed out, to have that emotion leave us changed.

[JM]: And I don't think that so-called upscaling some art that a human created,

[JM]: crafted by hand, and rendering it in a photorealistic way that really saps a decent amount of character from the thing that the human created...

[JM]: I don't know.

[JM]: This doesn't really interest me very much.

[JM]: I can see why a lot of people are up in arms about it and I can definitely see why people are dunking on it.

[DJ]: Yeah, totally.

[DJ]: I'm excited to announce my new open source project, which is called Human.VideoCard.

[DJ]: And what it does is it allows video cards to assert that they are human and that they are just rendering the frames exactly as the underlying data tells them to, as opposed to generating new frames.

[DJ]: And then what they can do is also assert which other video cards have also asserted that they're doing that, forming a web of trust between video cards.

[DJ]: Of course, it doesn't really matter because no one can have video cards anymore because OpenAI bought all of them.

[DJ]: But inside OpenAI's data centers, these video cards will be able to reassure each other that they are, in fact, human.

[DJ]: To be serious for like two seconds, though.

[DJ]: I apologize in advance.

[DJ]: There really is, I mean I'm very much with you, even like you know, setting aside our disagreements about Human.json, but like I'm very much with you that it's so clear to me that human beings are touched by you know what we call creativity, like other human beings' relationship with the world and our own. The writer Brandon Sanderson recently did a speech, might have been a TED talk, I can't remember...

[DJ]: But he made a great point where he said, like of being a writer or an artist of any kind, he said, the thing that matters about making art is that it changes you.

[DJ]: The thing you made is just kind of a happy side effect.

[DJ]: Like it's nice that there's a book there now.

[DJ]: But what really matters is the what it did to you as a person to write the book, and I love that point.

[DJ]: I think it's very true.

[DJ]: And it completely works on the other side of it, too.

[DJ]: Like the thing that matters about reading a book is that it changes you.

[DJ]: And I think we collectively are missing so much of the plot in this sort of rush to, like, to generate more.

[DJ]: Like, we've talked about this in the generative coding context, which, like, has kind of, I mean, for me, it's kind of taken over my job.

[DJ]: Not even in a bad way.

[DJ]: I just mean it's become this fundamental part of...

[DJ]: the way I do things.

[DJ]: But I've already experienced the fact that like, it doesn't take very long for me to be like, I'm generating so much code so quickly, that fact ceases to matter, because I can only actually provide the humanity to a small part of the time, right?

[DJ]: Like the bottleneck is always me.

[DJ]: And I think some people are like, oh, we got to get rid of that bottleneck somehow.

[DJ]: And I would argue, no, we don't.

[DJ]: Being human with all of its various limitations and capabilities is the point.

[DJ]: And so I do keep coming back to you sort of touched on it with this video game, this video game thing.

[DJ]: I sound like my dad right now.

[DJ]: Sorry, dad.

[DJ]: But this thing about, guys, now your video card will make every video game look more awesome.

[DJ]: It's like, but why?

[DJ]: But why actually?

[DJ]: In fact, does it matter that a video game's graphics look more photorealistic?

[DJ]: How often does that actually make the game more fun?

[DJ]: Which presumably is the point.

[DJ]: Right.

[DJ]: And the same thing being like now with this large language model, I can churn out 450,000 words of prose in like a day and then you could read it.

[DJ]: And it's like, I mean, I guess I could read it, but what's the point if it has no impact on me?

[DJ]: And what was the point of generating it if it had no impact on you?

[DJ]: I think the fundamental conflict between what human beings dare to call art and generative software is that the generation is not the important part of either the creation or the consumption of art.

[JM]: Absolutely.

[JM]: And I think it's interesting, as a side note, the way that the Nvidia CEO responded to this collective derision and outrage.

[DJ]: Did he agree with it?

[DJ]: I'm on the edge of my seat trying to imagine how he might have responded.

[JM]: He responded by saying, "First of all, they're completely wrong."

[JM]: He said, "The reason, as I have explained very carefully, is that DLSS5 fuses controllability of the geometry and textures accurately,

[JM]: and everything about the game, with generative AI."

[JM]: He adds that developers can still fine-tune the generative AI to make it match their style, adding that DLSS 5 adds generative capability to the existing geometry of the game, but that "It doesn't change the artistic control. It's not post-processing at the frame level — it's generative control at the geometry level."

[JM]: And to me, that's just a bunch of gibberish, in the end.

[JM]: It's totally just misses the point. Because just looking at the frame provided, right...

[JM]: Yes:

[JM]: This person looks more photorealistic with DLSS 5 on.

[JM]: But in some ways, that makes it look worse.

[JM]: Because anyone who knows anything about the Uncanny Valley understands that unless it looks precisely the way something would look in real life, it's going to look just ever-so-slightly off enough that it causes this...

[JM]: usually unconscious, but sometimes fully conscious, revulsion in the part of the human that's experiencing it.

[JM]: Because part of you is just looking at something and noticing on some level of your being that it doesn't look right.

[JM]: And so the artistic rendering is usually more interesting.

[JM]: That's why if you look at cartoons, they still more or less look like cartoons.

[JM]: They're not trying to make these fantastical creatures, by and large, look real, because that actually takes away from the experience instead of adding to it.

[JM]: And I think that that's what is being missed here.

[JM]: And going back to a similar concept that we just talked about a minute ago.

[JM]: So if we take this other argument, okay, well, let's say that DLSS 6...

[JM]: fully nails the photorealism and makes it look like something you would see with your own eyes.

[JM]: Does that change your opinion of it?

[JM]: For me, I'm not sure it does.

[JM]: It changes the Uncanny Valley problem, but the other issues still remain.

[DJ]: I personally much prefer playing Stardew Valley to Uncanny Valley, but...

[DJ]: You're right.

[DJ]: And there's more to it than just the Uncanny Valley effect, too.

[DJ]: So there's a writer and cartoonist named Scott McCloud who wrote a legendary book, legendary in comics circles anyway, called Understanding Comics.

[DJ]: And among the things that he eloquently demonstrates in that book is that an image that is more iconic can actually have greater emotional impact than an image that is more realistic.

[DJ]: And the way he demonstrates that is by, among other things, demonstrating an effect that I think I recently learned is called "pareidolia", which is the tendency to see like patterns where there are no patterns.

[DJ]: The most common example of this is that we tend to see human faces even where there are no human faces, which is why you think your car is in a good or bad mood based on the shape of its headlights and front grille.

[DJ]: Yeah, like a more iconic image is often more affecting.

[DJ]: And because of the, you know, the way our brains engage with fictional things, there's that notion of like suspension of disbelief, right?

[DJ]: So like you can be very, very invested on whether or not Mario is going to successfully land on the shell of that Koopa, even though like none of that has any relationship to reality, right?

[DJ]: Like Mario vaguely resembles what a person might kind of look like if seen from the side, but yeah,

[DJ]: In your head, it's not hard to go like, yeah, like this is the character, it matters.

[DJ]: You don't need photorealism to enjoy a game.

[DJ]: And indeed, there's been, especially in the indie gaming sphere over the past several years, like there's been a huge upswell of this so-called 8-bit style, where even though people are making video games with current technology to be used on current video game hardware...

[DJ]: They are choosing graphics that resemble the heavily pixelated graphics that you saw on like the Nintendo and Super Nintendo video game consoles 30 years ago because it's easier for a game studio with fewer resources to create graphics that are more simplistic like that.

[DJ]: And those games are still full of character, even though their graphics are nothing like photorealistic.

[DJ]: So it does feel like a bit of speaking of fallacies like we were earlier.

[DJ]: I think that so often happens in the tech industry.

[DJ]: There is a sort of optimization fallacy, which is that if you can turn a dial from three to four, then the best outcome possible is necessarily to get that dial all the way up to 100.

[DJ]: When with a lot of things that is not necessarily true, trying to provide some version of like so-called photo realism to the graphics of every video game.

[DJ]: Well, for starters, it doesn't really make sense.

[DJ]: Like if I'm playing an indie game with that sort of so-called pixel art graphics on my Nvidia DLSS five, what's it going to do to the graphics, right?

[DJ]: Like that's a game where the, the graphics fidelity is based on these very specific stories.

[DJ]: sprites being rendered at very specific times and places.

[DJ]: So like, what's it going to do there?

[DJ]: I mean, I guess the CEO of Nvidia would probably be like, no, see, in a case like that, the developer would just turn it off or something.

[DJ]: But even in the case where you have a game that is trying to look like a sort of photorealistic three-dimensional environment, you have these various other problems with realism, right?

[DJ]: There's a lot of video games that include a lot of physical violence, like a video game where you hit someone with a sword.

[DJ]: And if you have a really primitive or sort of iconically rendered video game, the way that works, the violence exists, but it's very divorced from reality.

[DJ]: And I guess I'm going to argue that that's better than the alternative where the violence exists, but is as realistic as possible.

[DJ]: And so one of the things I wonder is if you have this video game where like you hit people with a sword and it's a 3D world and there's like blood and stuff, but like the blood is represented as like red stuff,

[DJ]: objects that essentially spatter on the screen or whatever, like, that's pretty graphically violent, but like when the DLSS 5 makes the face of the person that you're hitting with a sword look more realistic, it's presumably not going to make other stuff more realistic, too, right? Like, it's not going to make it so their viscera fall out when you disembowel them, if you'll pardon the example, but

[DJ]: again, like, what's the aim of this? How does it actually change the experience for the better? To sort of isolate this one notion... Like what if the character's face looked more like a photograph and impose that on on your video games and so again, like, sorry Jensen, like maybe I just didn't pay enough attention to when you exhaustively explained this in a way that should have totally satisfied me — I guess that's on me and everyone else on the Internet but

[DJ]: it doesn't feel like it makes a lot of sense.

[DJ]: It feels like a silly gimmick, frankly.

[JM]: I imagine that DLSS 5 does more than just make human faces look more photorealistic.

[JM]: I imagine that it is intended to improve other aspects of the rendering, either by making lighting more realistic, making shadows more realistic, improving depth, detail.

[JM]: I don't really know exactly all the various things that this

[JM]: so-called neural function is performing.

[JM]: But in the end, I just don't think it matters.

[JM]: Like you said, it's not going to change how much I'm enjoying the game.

[JM]: The graphics capabilities available to us without this feature are already phenomenal and can create really incredibly detailed, gorgeous rendered scenes.

[JM]: And my main concern is in addition to not adding any value in terms of the experience for people,

[JM]: my main concern is the potential for homogenization, just like large language models can emit slop that is largely indistinguishable from other LLM-generated slop.

[JM]: I think there is the potential for game developers to use this capability as a relatively broad brush and kind of just swipe it all over the place and produce this kind of bland look that's the same from one first-person game to the next first-person game.

[JM]: And I don't play a lot of games in general and those kinds of games in particular.

[JM]: So it doesn't really affect me a whole lot.

[JM]: I just think in general, it sounds like a net-negative for the craft of games.

[JM]: And I guess we'll find out.

[JM]: But so far, it seems like a lot of other people are similarly unexcited about this new development.

[DJ]: It does seem like if the next time Stephen King or choose your author of choice published a book, the publisher excitedly declared that they had used a large language model to alter the text thereof before or like as you're reading the book.

[DJ]: As you're reading this on your e-reader, the LLM is adaptively altering the story to be more exciting.

[DJ]: And I'm really curious, like how many readers, let alone how many writers would think that was awesome.

[DJ]: Like so many things, it does kind of feel like, again, an entity that's sort of in a gatekeeping position, right?

[DJ]: Like in this case, a video card is literally stands between the software of a video game and your ability to interact with it.

[DJ]: And that like that gatekeeper is kind of going like, oh, we know best.

[DJ]: So don't worry.

[DJ]: Like we're going to take this thing that someone else worked really hard to make and that you're looking forward to enjoying and we're going to screw around with it.

[DJ]: It'll be better.

[DJ]: Are you sure it'll be better?

[DJ]: It doesn't seem like it will.

[DJ]: Hey, we patiently explained to you how it would be better.

[DJ]: Weren't you paying attention?

[DJ]: Anyway, I'm looking forward to my photorealistic Pac-Man.

[DJ]: Replete with beard stubble.

[DJ]: Does he still eat dots or are they like sandwiches now?

[JM]: Who knows?

[JM]: All right, that's all for this episode.

[JM]: Thanks everyone for listening.

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

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