This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
In support of mistake theorists and against outgroup reification
There's even more drama in the AI art sphere than before. Actually multiple dramas, all around Stable Diffusion, as is customary now. Artists on Twitter are threatening class action, half-seriously rallying around Greg "by greg rutkowsky, trending on artstation, 8K" Rutkowsky, Palmers and others; Danbooru is being slowly purged of content on takedown requests, in the wake of NovelAI SD-based image generator release (drastically superior to earlier stuff, and allegedly tracing over whole pictures); StabilityAI Discord has banned Automatic111, a hero and pillar of the community, the developer of the leading UI, due to him
being an asocial Russian assholeimplementing means to handle the recently leaked (by a third party) NovelAI models and allegedly stealing like 20 lines from their proprietary code, apparently to placate NAI and send a message to future pirates and business partners about Emad's priorities; Voldy denies the theft and counter-accuses NAI of copying his prompt-weights code without license; Stability staff/mods, it is claimed, have taken over a community subreddit by guile, to no comment from higher-ups (update: Emad magnanimously agreed to concessions); Emad keeps postponing the release of an improved 1.5 citing great responsibility and «tweaks to handle extreme cases» which is taken to mean «castration on dataset level». It looks like another Open-for-profit company has been revealed as an embrace-extend-extinguish scheme, and we will have to learn, after all, to pool resources on our own. Or maybe it's all catastrophizing by excitable FOSS fanatics. The situation develops rapidly.…But I'd rather leave the drama discussion to @Porean, seeing as he's posted it first, and talk about something less spicy. After interactions with Hlynka here, I want to share an observation about conflict theoretic lens and unhelpful labeling of the outgroup – such as artists on 4chan /ic board calling AI users «pajeet».
This has to do with Emad's origins and the racism-tinged contempt for «soulless tech bro nerds» pervasive among the 4chan creative intelligentsia, of course (the Twitterati attack soulless tech bro nerds as such, without racial qualifiers). No equivalent prejudice against South Asians exists in Russia. So, there AI users and people arguing in favor of this tech are labeled «neuroschizo». I wonder about other sectors of the Internet.
Yes, singular «pajeet», singular «schizo». It's not just a meme. They (well, many of them) report it to mods as such, they whine about a sleep-deprived obsessive fanatic who keeps pestering them in their strongholds (a not-so-implausible scenario, but clearly wrong in this case). And I posit that this is Indicative Of A General Failure-Prone Tendency Of Conflict Theorists, who have a powerful presence here. I have ridiculed Scott's Conflict-Mistake framing myself. But the core objection – namely that the Conflict model is evidently true, as well as people's willingness to lie in service of their terminal goals – may be a cognitive poison pill.
Anonymous imageboards have been a mighty forge of internet culture. What is often underappreciated is how strong they are pound for pound, in terms of active user or post count – even the biggest 4chan boards are like a middling subreddit or Discord/Telegram chat. Why is that? Freedom of expression, lack of reputation and all that jazz, you know it. But I think that they're a bit similar to latent diffusion models: they are more efficient, due to compressing the trappings of a social network into a lower-dimensionality space. By stripping identification and the complexity it can carry, they allow – nay, force – individual voices to be associated instead with archetypes of groups, in every individual interaction. You cannot be a token, like on Facebook or what have you: everyone is a type, but which type, depends on the mood and the topic. This inflates the effective population size to parity with a big society where full connectedness is impossible and knowledge about others must spread by rumors and vague stereotypes. It makes boards a self-running, accelerated social experiment. (admittedly this theory needs more polish)
Anons know, of course, that they are few in number (although a Legion). And they can see that people are pretty stable in their quirks. And they are aware that people can lie when it serves them. So they overcorrect into reifying recognizable opinions as marks of an individual or at most a small coherent group. Someone sneering at «Chuds» comes, in /pol/ mythology, from some specific discord – and needs to be reminded that he «will never be a woman». On /ic/, someone explaining how latent diffusion works is… «cool story pajeet».
It's an experiment that represents, at small scale, the superstitious nature of network age paranoia. In larger communities, the same assumptions are applied on group level. Everyone who disagrees with me is a Russian bot! Except if I'm a Putin loyalist, then it's gotta be ЦІПСО, Ukrainian propaganda division (that's what I am these days, according to many of my compatriots). If you're an American Right-Winger, it's some sort of GloboHomo WEF talking head. If you're a good progressive, it's probably a Fascist representing a unified anti-Lib front.
This is psychologically comforting for a few simple reasons.
First, such a problem is very legible and simple. There's no structural reason for the thing that oppresses you to exist and be the way it is, no grandiose system of incentives, just malign will of a finite set of human actors, fallible flesh and blood with a self-serving narrative.
Second, it's surmountable. Calling the enemy out is half the battle; getting him banned is another one fourth, after that you just gather up the boys and attack his turf. The hostile group is not representing the majority of the population (usually is puny), is easily identifiable and kinda ridiculous. Maybe just one weirdo, a «pajeet» or «chud» at that.
Third, and most importantly, it excuses ignorance. You can plug bananas in your ears because conflict theory predicts that the enemy will lie, or try to deceive you while not technically lying, to demoralize you. And why would he keep investing effort into that, coming up with arguments tailor-made for you? Of course because his onslaught isn't really going according to plan, in part, precisely because people are not falling for it! That's what those artists think too. AI proponents are lying collectively to break their spirit; they just need to wait it out while the pajeet runs out of steam; they don't need to adapt.
They're wrong.
It's unhelpful to have a dead wrong model of the conflict they really are in. One big and obvious reason: because it precludes communication with people who are different but not the enemy and are trying to help – or at least negotiate. In trying to not be a 0HPLovecraft-style quokka, such a conflict theorist ends up being simply a myopic rat, destined for being made marginal and obsolete. The great discovery that counteragents might lie is a point of a wholly unreasonable pride of a certain brand of reactionaries. It's also a source of a delusion as damaging as the inability to conceive of bad faith.
Tl;DR: Conflict theories have a failure mode or reifying the opposition, that can lead to cognitive closure and lack of response to arguments. The very assumption of bad faith, without extra precaution, bootstraps the development of theories on how bad faith propaganda is being delivered to you, for what reason and by whom. From then on, truth is ever harder to find. People should remember that, when assuming they can see through the opponent and dismissing opponents out of hand.
I very likely wrote some of the posts on /ic/ you’re referring to.
My mental model of the developers/proponents of AI art (and AI in general) is that they believe that they’re genuinely making the world a better place, at least by the measure of their own terminal values. I just happen to sharply disagree with them.
Obviously, posts written on 4chan to blow off steam and commiserate with people in your own camp do not always reflect the nuance and complexity of one’s actual views.
EDIT: Well, since I just brought up the subject of having nuanced views, I should acknowledge that I don’t think the motives of AI developers are entirely pure-hearted in all cases. If you read the /sdg/ and /hdg/ threads, hardly a thread goes by without someone saying “fuck artists” or “it’s over for artcels”. There’s clearly some amount of resentment there for people who possessed a skill that they wanted, but were not able to obtain for whatever reason. As for a broader UN/WEF conspiracy to reduce the global population by replacing workers with automation - obviously I don’t have any concrete evidence of an intentional conspiracy, but I do fear that a future like that is possible, even if no one is consciously intending to bring it about.
Excellent!
Incidentally, I do not post there. 4chan is banned in Turkey, my VPN is banned on 4chan and bying a passcode or investing effort into cheaper workarounds feels not worth it. Still, I've written an angry response for a yet another maddeningly arrogant thread – struggling to fit it into 2000 characters – the other week, before finding out that no two-bit hack works. This is as good an opportunity to share it as will ever come.
Please don't take it too personally. Ahem:
...now, in a less 4channy mood, I need to admit that artists do have a point. Making them obsolete using their own work is a scummy move, regardless of artists' character flaws. And mundane considerations of fairness aside, we need a framework for the coming AI era where all jobs will yield to machine learning at some point. One option would be to institute a sort of collective entitlement, monetary and cultural, for members of guilds whose jobs have become automated, because the sweat of their brow, examples of their product, has enabled the creation of that automation to no lesser extent than the work of developers. A derivative «coom artist» can go pound sand on Twitter. Greg or Clive should have a stake in the pie of machine-produced visual content, and maybe some Prometheus Medal for their contribution.
But of course even their habits and egos are not worth hampering the greatest celebration in history.
@2rafa: no idea how the law will decide, it seems that purely by the letter of the law artists have no case, but the law may be changing soon. The logic above is why I tentatively approve of artists and other IP holders getting their wish come true and excising their stuff from training data. Spamming tags associated with generically high-quality pictures (which is all there is to «by greg rutkowski») is tasteless, and training models on Greg to learn his style along with his themes is silly and inefficient, but it's such an easy shortcut it may clog our creativity for an arbitrary span of time. With fat-soaked American IP like Mickey Mouse it's even worse, why keep clinging to it in the age of visual post-scarcity? It only ever got anywhere due to starting early. We need a clean break and an Aesthetic Space Browser, and the surest way to develop it is to begin with taking away the easy and cheap option. Now that the potential is known, the way there won't take as long.
Below is a translation from a Russian /ic/ equivalent, one of a series of texts that I consider to be some of the best writing on this subject available anywhere, from a man much smarter and a better writer than myself, and an obvious inspiration. (if anyone wants other texts, let me know).
Man, that's so
funnyfucked up.«Artist growth». «Income». What income now.
The basic model understands Euclidean 3D in classical perspective. Why this particular space? Because this abstract model is the most efficient way to explain the presented set of data. Why does it understand «objects» and the orthogonal group of transformations? Because it is the most convenient way to compress the data through a hierarchical decomposition of the visual field. Why does it understand lighting? Because, just as in the previous cases, the model was able to reverse-engineer the scene and decompose it into surfaces, light sources, and color space. Why does it understand «style»? Because «style» is a microscopic speck in terms of information volume, a cosmetic superstructure overlay for the underlying geometry, textures and lighting.
And yes, «composition» is also an element of style. So are different kinds of perspective. For a mangaka, «fish-eye» means years of wrestling with built-in visual cognitive biases, painstaking fiddling in awkward 2D, selecting, combining and projecting primitives, generating and tracking pencil trajectories, iterating a loop of deliberately memorized algorithms with elements of pseudorandom – assembling a cognitive chimera from the elements of thought amenable to control. For the model, it is only a vector weighing a few kilobytes that directs the hierarchical blending of the base elements.
Shift by vector plus cosmetic rendering. Click – an orange grows a coat of feathers. Click – feathers become made of glass. Click – fisheye. Click – reverse perspective. Click – Kuvshinov. Click-click – Vrubel. Gumroads have never taught anyone anything. Creative work is a pipeline. 99% in the technical tricks – and a bit of lucky pseudorandom.
And it's not even the beginning, it's just a precursor. Text, pictures, music, video, speech. Sculpture, architecture, engineering. What even is genuinely complicated in our culture? Mathematics? But proofs of theorems are isomorphic to programs – and coding, too, is kind of, well.
Only meat and quale will remain for the market. You can't carry quale to the market. Hence, only meat.
...I'm just complaining. We've all been blackpilled to the point of being stiff a long time ago.
That's not the essence of it. Creativity is now officially a closed topic. Well, you know, not very long ago, among the linguistically oriented highly intelligent public (philologists, writers, language philosophers, mathematicians...) it was fashionable to do topology, play word golf, admire Hofstadter, and subscribe to "Word ways" magazine. All we have left from those glorious times are scraps, fragments, and vestiges in the form of ubiquitous arroword and crossword puzzles.
The end of the glorious era came almost immediately by historical standards. And not because the fashion has changed and people got bored with something there. Just because there appeared a cheap microchip, capable of closing all open problems and finding all the interesting structures in about an hour – even without any algorithms, just by brute force.
The topic was closed. And no one even noticed how it happened. It went down quickly, almost immediately. Flip-flop, and that was it.
Poets have been constructing palindromes for centuries. A microchip came along and found them all. Aristocrats played word ladders for decades. A microchip came and built them all. The thicket of texts was once home to the most fascinating pokémons – but a microchip came and collected them all.
Dickgirls, dickgirls on dickgirls, vegetables... Does it hurt, doesn't it hurt, is that really what it's about now? [Skill] ceilings don't decide anything anymore. There are no ceilings. That's it. Flip-flop on the scale of an entire culture. I haven't made it. I was too late. And there was only a little bit left to go. Oh, man.
I'll take this as an invitation to translate a couple more of his posts. @FCfromSSC, @Primaprimaprima, @VelveteenAmbush, @07mk may find it interesting.
Blah hlah. «A neural network is just a tool». Instrumentality is not an essential characteristic, not a set of immutable and stable invariants – it is just a position in a net of relationships. Photoshop is your tool. You are a tool in the hands of the customer. The customer is a tool in the hands of media corporations clogging up the information space with hyperstimuli, pussytriggers, freakcodes, miyazaki-glitches, and other visual-advertising slop. In this hierarchy of tools, you're just a gasket. You are tolerated only for the time being, being used as a cow to milk for datasets. While corporations cling to their «know-how» with a deadly grip, the goyim are being sold on «sharing is caring» out of every newspaper. So, how'd it go, guys? You like it? You want some more, or you got it figured out already?
Judging by the millet you've chewed up here, you know nothing about either. Even worse, you don't even know anything about yourself, and your doctrine about man is hundreds of years out of date. You see, the future is very unevenly distributed over mankind. Conditional modernity is an integral sum of anachronisms, the civilizational clock within different social strata is at times lagging, at times gaining by thousands of years, and while the fiery sparks of futurism are rising into the sky, somewhere deep under the silver amalgam of modernity the tectonic plates of archaicism, which roots dig into the Paleolithic, are heavily tumbling. I have seen [fields, Neo, vast fields] whole institutions filled with former-people, unable to grasp the moral obsolescence of their own discipline. They live, and sometimes even live very well – but their lives are as relevant to modernity as the life of a bushman poking the earth with a stick.
You are a former-person, a representative of the Paleolithic. You vastly overestimate both your own complexity and that of the culture that has produced you. You wave your pokestick menacingly and seriously believe that your so-called «academia» is something more than the mere escheated craft of projecting probability distributions learned by the visual cortex into a two-dimensional peephole. Hail to the great academia! Hail the ball of feces molded by the calloused hands of industrious dung beetles, a combination of kolkhoz hunches, try-anews and rules of thumb, carefully handed down from holey teachers to their equally penetrated students!
It doesn't even occur to you that instead of «building a nose» as a perspective projection of 3D into 2D, which some 3D humans force others to learn doing for years, you can «build a face» as a generalized object in 1000000D-space – and get all potentially possible, that is, all fitting into model-learned probability distributions, human faces as projections of this generalized object into the plane. The neural network does not need to «build a nose», it is already built within it as an element of some hyper-object in hyperspace – and, unlike you fool, it can project this nose into a subspace of any dimension: into a line, into a plane, into three-dimensional volume, into ten-dimensional volume, anywhere.
«Constructing» is not your forte, it's your bottleneck. The obligatory decomposition of the image into hundreds of layers of Photoshop, priming and underpainting is another bottleneck, a phoney and lifeless detour where a neural network can cut a straight path. The local, single-channel application of the picture using hand as the manipulator, slow and semi-random coverage of the plane with monotonous brushstroke patterns, all that stuffy, mothballed, chthonic horror, is just a creative disability in comparison to the neural network that builds an image globally, multithreaded over the entire plane of the canvas, and which, unlike you, is just as globally and multithreadedly able to build objects of any dimensionality.
Oh so that's it, pockets on her jeans. A veiled attempt to appeal to the informality of pragmatism. The most flaccid argumentum ad props I've ever encountered. By the way, do you remember what year Ackoff's On Purposeful Systems came out? Was it '72 or something? Trapped in the shut-dead vault of their own stupidity and ignorance, members of the creative class still find even long-solved problems cognitively impenetrable.
Ten million «consciously done» entangled and crooked and forsaken paths, thousands of tons of glucose burnt in the cracked craniums of the creative class – and we end up with the same ten standard plots and a couple hundred classic archetypes. This is how functional-structural convergence works. This is how power law distribution works. This is how ethical and cognitive specialization of sapiens works. A finite number of basic emotions, a finite number of their variations, a finite number of forms of group interaction, a finite number of social conflicts, a finite number of characters in the actor network – «love» here, «treason» there, »family» to the right, «war» to the left. Locally, you're all incredibly individual and monstrously creative – but when you go out into the vastness of the Artstation or even the most obscure booru, one clutches his head. A woman with tits, a woman with a dick, a woman strong and independent – a foreign, eggplant flavor girl. What is obscured in small samples, becomes evident in big data. It's all the same shit.
Fractal laughters. Sneers within sneers.
The deal is in the multimodality, in the virtually implemented possibility to reconstruct the latent model of «the world» from its two qualitatively dissimilar discrete projections. Instead of the natural language (or rather, its stripped-down version used for tagging) and a piece of N^n characterizing geometric and color space, you can take any two models of anything fixed in any arbitrary languages – and build a proxy-manyamirok [delusion-world] with them, guaranteeing a coherent inference in each half of this bicameral. This is the same «Geometric Machine» (?), only built on self-learning, providing x1000 in power, and in addition capable of running on any cardboard plug, not just a distant mainframe of the eggheads.
Multimodality enables you to avoid local roadblocks by changing the representation, allowing you to bypass the problem area along a parallel model, either by rising to another level of abstraction, or by decomposing the problem into subtasks within a more concrete and detailed representation. Connecting a hundred bicamerals on different pairs of models into an orgraph, we get the very thing the fathers of cybernetics dreamed of and about which St. Academician Ivakhnenko directly wrote half a century back: a multilayered hypersurface of layer-models connected by forward and backward connections of information archivation and inflation.
Considering the fact that such structures of models docked into polyads keep popping up in various fields related to reasoning, from mathematics to cognitive sciences and from Lotman's culturological essays to pedagogical articles, one may suggest that they are as universal as the g factor, cortical columns, deep learning models and other abstract motes like atomic theory in physics, set theory in mathematics or finite element method in engineering. In any of the local domains, sooner or later a universal methodological wunderwaffe capable of completely covering the problem field is discovered.
Understanding is a transformation of information within a network of models representing other models and themselves, a long branching tail-cyclic graph, a rhizome of algorithmic tunnels conveying through the system a long echo of mutual interpretations, a reverberation of meanings in autoreferential algorithmic corridors. Did you understand what I just said? I didn't either. But this is how it is. We implement this transformation every day, every second, within hundreds of parallel threads – and use its results in the form of object icons on the screen and a dull linguistic constructor, with which we desperately try to fight reality.
Attacking Vasya [who doesn't value true art] won't help you. It's not about bydlo, it's about the fact that «meaning» has become an engineering problem, literally just now, before your very eyes. And instead of assessing existential risks and opportunities to hedge them, you've taken a couple of dumb verbal devices and are trying to assemble either a butt plug or a virtual inner politruk out of them. Repent, simpleton – you'll get a discount.
cont.
This is exceedingly relevant to my interests.
Hmm.
You've read Library of Babel, I presume? That library also skipped straight to the generalized object in 100000000D-space and got all possible distributions. Generators have the advantage of a search function, which is a big advantage, but the search space is also very, very large.
The end-state we're looking for is to see "the correct image". we can search coarse, or we can search fine. Coarse search was available before AI: we've all done it when we browse large art repositories, pinterest, image boards, boorus. Fine search is what a mature artist (meaning an artist who has gotten past fighting their hand and eye) does, and it's so much faster and more satisfying that rough search because you are completely in control. you make the image as you please, in all the details, you change what you want and keep what you want, and it all comes very cleanly together. It's a marvelous feeling.
It's frustrating to fuck up a nose or a hand and have to try again. It's likewise frustrating for the generator to not give you exactly what you want. It's the same frustration, in fact: the picture's coming out wrong. And the solution in both cases is to fix it, to push fine control down all the relevant pathways, whether with brush and pencil or ai-driven editing tools. More pathways means more complexity, and either we're back at a hundred photoshop layers, or we're at photoshop without the hundred layers, fine search without the pain.
but it's still fine search, and fine search is art, and the people doing it are therefore still artists!
Like, there's two possible futures. In one, the AI reads your mind so perfectly that you don't even need to ask it for what you want, and all you do is consume. That sounds pretty horrible to me, but I concede that there's probably a lot of people who'd happily take it. In this future, Artists are in fact dead. So are humans, in my view. If this is the desired future... well, hookers and heroin are already available, so what's keeping you?
In the other, the AI can't read your mind, so you have to use it to get what you want. and if you are discriminating in what you want, if you want to tweak the pose, the lighting, sharper in the face, make the hair flow this way, adjust the composition... if you're not satisfied with a image, but want the image, then you're an artist, and the AI is your tool. And you're probably going to have some analog to underpainting and photoshop layers and so on, because control is always going to involve difficulty, discipline, and the resulting disparities in outcome.
This reminds me of nothing so much as the friend who once argued that since there's a limited number of notes available, we're going to run out of new music to make. Well, we haven't yet, and I'd bet we never will... and that's with the relatively constrained palette of tone and rythm. And he thinks that the entire range of thought and experience will produce less variety? Bullshit.
I think he's wrong here. And not a little wrong, but fundamentally misunderstanding what art is and how people interact with it.
Art is recorded. Because it's recorded, it can be reused: one piece of art can satisfy multiple consumers. So if human creativity and human desire are as constrained as he claims, If the artists are making all the same shit, and the consumers just want all the same shit, then presumably there's a cap to the amount of art we actually need, right? Like, once we've written more than a lifetime's worth of books, there should be no demand for more? ...Only no, because people don't want GENERIC EXTRUDED BOOK PRODUCT, they want specific kinds of books. Murder mysteries, romance novels, techno-thrillers, historical fiction...
Okay, so maybe once we've written a lifetime's worth of murder mysteries, we can happily close the genre and rest from our labours forever more? Only again, observably, no. It seems inarguable that desire for art is like the plastic crazy straw design community: It's fractal, there is no bottom. Humans observably throttle their desire for novelty up and down depending on the environment, and I see no reason to believe that there's any practical upper limit to this mechanism. Increase supply by several orders of magnitude, as we have in the modern era, and people simply raise their expectations, grow more discriminating. We have much more art and much better art now than we ever have, and the demand is higher than it's ever been.
90% of everything is crap. 90% of everything has always been crap. 90% of everything always will be crap, because it's not the stuff, it's our assessment of relative value given the environment. That's how humans actually work, how art actually works. Until you get the AI that reads minds, we're still going to be hunting for glimpses of the supernal, on both the producer and consumer ends.
I hang out on Artstation, and I get where he's coming from. A lot of the art there is truly quite repetitive. Likewise, when I read web fiction or manga or watch movies, it's easy to find patterns, even ruts. But that repetition is a consequence of artistic failure on the part of the producers, not a limitation to art itself. They're all making the same shit not because there's nothing else to make, but because their creativity has grown morbidly incestuous. They're all just copying each other's homework, and so it all comes out the same. Then an actual creator with an actual vision comes along and blows them all away... and then maybe they get lazy and start with the nutrient fluid. Humans are weak. But the fact remains that the path is there to those who seek it.
...honestly, I could fill two more text boxes just on this subject alone. My mind is awash with counterexamples, with the ways people find enduring novelty in the most absurdly constrained of categories.
I'll freely admit that I didn't follow most of this passage, so maybe he's just that much smarter than me. But I don't see how "meaning has become an engineering problem" is a supportable statement, or even a likely prediction. I think the guy claiming that it's another tool is strictly correct, because, again, I don't think the machine is going to be reading our minds anytime soon. Obviously if you think we're about to FOOM, your mileage may vary, but absent full-blown artificial superintelligence with no cap, tools is what we're getting.
Indeed, but what does AI have to do with it? The issue of fetishizing invested effort as the prerequisite for validation of end result, a la Marxists and PrimaX3, is wholly distinct from the issue of reward hacking. Cutting costs of procuring enjoyable art or another valuable product to almost nil is qualitatively different from cutting out products as unnecessary intermediate stage for dopamine hits. The former approach calls for a high-level conversation about our terminal values, between moralists, aestheticians, utilitarians, replicators and others. The latter one presupposes values, and resolves very quickly into the reduction of human telos to hedonic utilitarianism and human bodies to hedonium puddles, if even that.
But, that strawman aside – yes. Yes, please. Self-expression is better than raw chemical high. AI reading my mind to directly instantiate beauty that seems contained therein has been my dream since… maybe 5-8 years? As soon as my ideas outpaced my ability to express them. Rigorously imagining is a hard enough job for the artist. It was back then that I realized the impossibility of getting good in every domain that feels aesthetically pleasing, and hard limits to human skill. Were I provided some sort of Neuralink (and there finally are some successes – 1, 2) … With age, all of it shriveled and turned to dust, vague sketches of symphonies, images, dreams, whole worlds, making space for dry rasp of texts. Most likely, little of value has been lost in this case. But something is lost with every child growing up, and it adds up to a substantial sum of unrealized humanity.
It's eerie how problems predicted decades ago in science fiction are coalescing into topical conversation these days, soon to be matters of life and death. In 1964, Strugatsky Brothers have written a novelette tititled «Predatory things of the century» about the oversaturated society where people start to die after finding how to use a commodity radio detail plus some household chemicals to resonate with one's inner desires and conjure an imaginary world beyond anything one can live through. The main character, an archetypal Golden Age Sci-Fi/Social Realist hardboiled hero, takes the drug and then, in his head, debates his predecessor who's succumbed to it:
Dzhilin ends up suggesting a «hundred year plan for restoring and developing human worldview in this polity» but never elaborates. Authors themselves believed that this is an abnormally «Western» book in their oeuvre, full of horror for the future of humanity and proposing no answers. Censors and literary authorities also took issue with it, including the eventual Minister of Culture of the RSFSR Melentyev – who helped bastardize the first print edition, but ultimately deemed his failure to guide this book to an idealistic resolution the only dark spot on his conscience.
…I think we haven't made a ton of progress since then.
Anyway, regarding technicalities. The «fine search» is very fine, about as fine as one's ability to put opinion in words – assuming a smart enough model. Stable Diffusion is pretty dumb, but there are ways to refine the search that look e.g. like this now (taken from 4chan, courtesy of an anon with apparently zero prior artistic experience; I'd contend he's a real artist or at least illustrator now). And of course text-guided img2img/inpainting is trivial; if this were part of some mature editor package, we'd have convenient tools to select, guide and mix features as such, without grasping for words (maybe based on some distant relative of cross-attention) . Will it be a tool? Naturally. But a tool allowing one professional («artist») to make dozens of artisans (and hundreds of Photoshop layers) obsolete.
We do not have unlimited appetite for stimuli, this isn't a game of positional goods. Not running out of music – or art for that matter – is largely due to art's social function, people's limited long-term memory and preference for novelty. We have, more or less, found all melodies, but we cannot listen to the same song on repeat forever. In a sense, content is a means of hijacking the basic reward system (porn especially is) – if you don't change the exact shape of the stimulus, the brain learns to dismiss it. But AIs do variations perfectly.
He's smart indeed, but it's more of a matter of familiarity, and understanding this part is crucial to appreciate the rest of his rants about vectors and dimensionality; all of it makes technical sense. I suggest you start with Jurafsky&Martin's Speech and Language Processing, and see how «meaning has become an engineering problem» is a well-supported statement, not a prediction but a description of the current state of the art in NLP (pun intended). Check out ch.6 («Vector semantics and embeddings»; click this link too). 6.8 covers Word2vec, one specific (aged) engineering implementation of meaning related to the one used in Transformers, and specifically in the text encoder part of Stable Diffusion. Maybe you'll conclude that on its face it's a drastically less rich notion than the ineffable human «meaning», one limited to the space of our text corpora.
But in principle it can be made much, much richer. And I don't need FOOM for that.
One can value improving the struggle > reward loop, or one can value subtracting the struggle and getting all reward.
LTV, as I understand it, holds that labor IS value, and clearly that's not true because some labor is wasted or counterproductive. On the other hand, humans strive, we choose, we see things not the way we want and try to fix them. We grow and change, moving from less to more. Remove that, and what separates us from puddles of hedonium?
What separates this dream from the Minotaur? The AI draws transcendent art from you without conscious effort... Is the art actually transcendent, or is it just hacking your specific reward centers, ordering you to FEEL FEELINGS?
You personally are an Artist, in the sense I've been using the term here. Your chosen medium is the written word, and you are good enough at it that I can't imagine you aren't familiar with the artistic process, of the call of the muse, the desire to create, the experience of laboring to birth something novel out into the world. Certainly, it's a painful, often frustrating process, and full of disappointments and sorrows. You're never good enough, never fast enough, you see all the flaws and the inadequacies in your creations, and they pain you. But hopefully you also feel the high, the delight and wonder of hard work rewarded, when someone gets it, when people respond, perhaps even when their opinions change, or even just having a piece that you can read again with your own eyes and know that for that moment, you bore the sword of creation against the void.
Without the process, the succession of conscious choices, how much of that survives? If the AI draws it all without conscious effort on your part, is it really you speaking? How would you know?
An old joke:
A young man walks up the famous Russian novelist at a party. "Oh, it's so good to meet you," he says. "I've read all your novels, they're transcendent! You know, I've decided I'm going to be a novelist just like you! In fact, I've got a great idea for my first novel, and it's going to be incredible!"
"Ah very good," says the Novelist. "But you know, writing a great novel isn't easy. First you need a good thesis."
"Oh, I've got the best thesis! You've never seen such a good thesis!"
"Good, good. But you also need a good setting."
"I've got an amazing setting!"
"Excellent. But you also need strong characters."
"I've got incredible characters! They're deep, they're lifelike, better than life even."
And the novelist smiles and nods. "Well, it sounds like you're well on your way. Now all you need is five hundred thousand words, and they'd better be the right ones."
There's a particular kind of person I met a lot of when I was in school: the "idea person". They couldn't draw and they couldn't code, but they "had ideas", and they thought those ideas could be their contribution to the project. As they saw it, they would sit back in a comfortable chair and imagine "something cool", and then the rest of us would dutifully get to work actually modelling and rigging and animating and coding their "something cool" into an actual product. I spent a lot of time talking to this sort of person, because there were a ton of them and they all liked to talk a whole lot. Without a single exception, their ideas were absolute trash, warmed-over derivation mashups of things they'd watched or played, "X but with this mechanic from Y", or else completely incoherent, a half-step up from word-salad.
The thing is, there's a feeling of "this is a good idea", a sense of excitement, of infinite possibilities, an infatuation, a mania. But you can have this feeling without actually having a good idea, and not only is it not worth a damn if you can't execute it, it's questionable whether it even exists at all in any meaningful sense. A lot of times, maybe most of the time, it's just a glitch in the brain, premature enlightenment, completely lacking substance. And to the extent this is true, I do not think the AI will help.
What can be executed is what is real. I think it's certain that AI will allow orders of magnitude more execution, and orders of magnitude more success from that execution. Maybe it will also fool those people who have nothing to execute into thinking they're a genius, when really it's just stick figure > greg rutkowski trending on artstation. My guess is that when everything shakes out, the later will feel about as satisfying as the stick figure without the autogen; maybe less.
AI, Nueralink and the rest: does it make effort easier and more fruitful, or does it turn your brain into essentially a very large random seed value? Do you recognize a difference between these two outcomes?
...It's always possible that I've misunderstood, but the question comes down to "what is it all for", "what's the point", right? I think I have an answer to that question, and one I at least personally consider satisfactory. Pleasure serves life, not the other way around. We have a purpose: to grow and to choose, to go from puppets to real people.
I agree that it'll allow artists to produce art harder better faster stronger, by many orders of magnitude. I even think it's likely going to devastate existing artistic hierarchies. I question whether it will eliminate artistic hierarchies themselves, though. I'm confident disparate outcomes will persist, 90% of everything will still be crap, there will still be rockstars and sad sacks.
To some extent, this is a philosophical bet against the reality of the Singularity, similar in form to a lot of other bets I make on other topics. I'll admit that my worldview inclines me toward viewing a lot of techno-utopianism with horror and disgust, and it can be difficult to distinguish what we think should be from what we think will be. Maybe I'm wrong, but I don't think so and I defy the future to prove it.
Assuming I understand the statement, how is it not positional goods? I have access to orders of magnitude more art now than when I was a kid, and much of that art is qualitatively better in every way than what I had as a kid, but I still feel starved for quality. It follows logically that my standards have gotten stricter. Why do we suppose that this heightening of standards won't just keep increasing?
The bet here is that you like something, Game of Thrones or Ghost in the Shell, and the AI means that there's an unlimited amount of it available. The show never ends, or if it does there's another show that hits the same notes, right? And when you get tired of not just the show but the notes themselves, there's an unlimited amount of your next appetite, and the one after that and so on, correct? And not just an unlimited amount, but an unlimited amount of perfect, transcendent quality, the best one can imagine.
Yeah, this was my original argument, almost exactly. Endless perfect consumption is isomorphic to the destruction of the most valuable parts of your mind, in the same way wire heading or endless heroin and hookers are. what's it for? If there isn't a better answer than "feeling good forever", I don't want it, and I'm going to bet against it.
Gotta close the gestalt at least.
Artemiy Lebedev, the self-appointed Graphic Design Guru of Russia, has a famous note in his «Kovodstvo» (... «'Idance») blog: Idea worth negative million bucks. It's all very fair and clever and biting, it's only a shame Artemiy would've been a nobody without his illustrious mother, and his contribution to Russian design is probably measured in negative billions.
Regardless, I understand the contempt that doers feel for idea guys (and ChatGPT shows just how commodified random combinations of premises can be made). It's the same sort of contempt artists on 4chan feel for «prompteurs». Prompts can be auto-generated ad nauseam by a shell script, Greg Rutkowsky had to polish his craft for decades.
Mayakovsky was, presumably, like Rutkowsky. He wrote:
I don't like Mayakovsky or what he moved people towards.
I'm an idea guy. Ideas are like stick figures. But not all stick figures are created equal. One will never become as good an artist as Murata, and it's possible, likely even, that Murata believes himself to be a greater mangaka on the account of his ability to make technical artistic choices; he's wrong and it's a tragedy that One has to compromise with him in producing an artistically competent work. Frankly, One's time was wasted on learning to draw. It'd be proper for him to stick to his figures, and let someone like Oh!Great experiment with styles, to then use their finished forms at his discretion. This is one aspect of AI's promise for human creativity – distilling forms of talent, to recombine them without crippling compromise.
In any case, defining the inherent worth of artistic work through choices made feels odd.
For me it's not a matter of choices. There are instrumental choices made along the way, but mostly just mechanical effort towards crystallizing the idea into its finished shape, reverse-diffusing it from noise towards a predestined output – If I were to abuse this metaphor again. Suppose my fingers are cut off and I have to type with my tongue, or use some BCI with 15 characters per minute; that'd be the end of my participation here – just not worth the time; and moreover, it'd probably be impossible to finish any meaningful piece while the idea is alive and breathing (this post was not finished in time, and so it'll never be 1/5th as decent or 1/3rd as responsive to you as it should've been). It's a fragile thing – a true (pardon my hubris) idea; it's far more than a random seed. It's chosen near-instantaneously from an infinite murky space of triviality, grasped, and the ability to see its glitter is what matters; while this ability is perhaps developed and polished in some relation to iterative effort dedicated to implementation, it has a dimension that lies outside the entire artistic craft and has more to do – with some talent, of course, but more interestingly with unique human experience, that inner achievement which shitty Twitter illustrators purport to convey but probably lack; unlike artists whose fame they hide behind. Rubén David González Gallego has cerebral palsy. He has proven one can be a compelling writer in even worse circumstances than my hypothetical; I can't help but wonder how much more he'd have given the world without his handicaps. Quantitatively and qualitatively more – for the same reason of the impermanence of inspirations. And contrariwise, were it possible to diffuse faster, typing or rather altering the text – or other canvas – at the equivalent of 1000 char/minute globally, improving the work in its every point as the brain sees fit and as an algorithm can do – hoo boy, how some idea guys would shine.
Or not.
We are ankle-deep in the sea of generated creativity, both technical and ideational – watching the tsunami wave as it crashes down. I hope that when it recedes, the survivors will discover some jewels in the wreckage, something essentially solid and irreducible to combinatorial glass bead game guided by random seeds. Some true ideas. It's in our interest that they remain human-made.
More options
Context Copy link
More options
Context Copy link
more detailed replies later, but one quick point:
...putting opinion into words is itself an extremely serious bottleneck. As it's put in that Chad Cerebration meme, "can use an inner voice, but doesn't because he knows it's inefficient." There's tradeoffs either way, because actually putting your opinions into words forces precision that sometimes speeds the search, but for the most part it's serious deadweight loss and a real pain in the hindparts.
That illustration vid is a good example of the dichotomy. he can select elements to edit visually, but he needs to rely on lossy, imprecise language to edit. The prompt box is the bottleneck, in the same way as nose construction and photoshop layer stacks. Obviously the tools are going to continue to improve, but he's still making choices, so by my lights he's still an artist.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
Define copypasta. Anime-girl, like any "object", can be described as a projection in the plane of some three-dimensional model with superimposed textures and lighting. The girl, as a model, cannot be patented (unless it's a branded object like hello-kitty or Mickey Mouse). But even a branded object is assembled from geometric primitives – spheres, cylinders, triangulated surfaces – covered with commonly available textures and painted in standard colors.
OK, you'll do what they do now - make some verbal description of the style and a set of examples of that style in the form of pictures proprietary. But style is a point in a huge "style space." Some dots are now occupied by authors, some by corporations, some owned by no one (like tumblr noses, etc.) – but the vast majority of style dots are free. Neurobros will simply go and create new styles – or generate new ones as unlike the existing ones as possible by interpolating the old ones. And what do you do? Train a neural network that will assess the degree of coincidence of the author's and copycat's manners of writing? Well, it will start to register all the usual artists as thieves, most of whom will not even be aware that their way of drawing is similar to someone else's patented style – because there are a fuckin' million artists and it's impossible to know them all. As soon as the person starts to draw and posts the first picture, he'll get sued for stealing the identity of some obscure patented artist – because the picture will be similar to this artist's "naive_relism_№9714443" or "primitivism_№540235".
You can't patent a sphere. You can't patent the color blue. You can't patent a girl with a dick. Can't patent the slap. Can't patent the grass, the sky and the sun. The point of Transformers as ML models is that they have no ceiling on the level of abstraction and depth of image analysis (which you, for example, do have: below is receptor resolution, above is a ceiling on the complexity of the patterns being analyzed). It doesn't steal fragments of other people's images, it stupidly parses the world into ontologically real objects like people and trees, or geometric primitives, textures, shaders and lighting, and then mixes them.
The only thing you can do is to forbid using images of authors to train models. But then they would be trained on open datasets – and "styles" would be pulled from publicly available classics or the same interpolation of existing styles. Or, you can just set a reference point in the shape of a model trained on real photos, and define "style" as the sum of the deviations from photorealism, generate a million random deviations and set up a focus group to evaluate them and select a few of the best tens of thousands of them. And all of these styles cannot be considered "stolen" because they will have been created as transparently and honestly as possible.
You hope in vain that Uncle Styopa The Militsioner will come and protect you from nefarious thieves. Or that the Top Skills will raise a stink and show the arrogant technicians their place. But here's the thing: there was no theft. [...]
It's not the picture that's "stolen", it's the idea. How are you going to codify an idea? Here you have Beksinski, for example. You have a collection of his paintings - they are copyrighted not as sets of ideas, but as some instances. Each painting is protected as a unique object, but inside it are thousands of ideas, assembled into an original construction of universal yoba-blocks – both created by Beksinski himself and stolen by him from someone else. In theory, you can break down each picture into blocks, brands, logos and labels – that is, you can select a set of objects as "Beksinski's mickeymouses", select the original "Beksinski textures", the color palette "Zd_Beks_001", etc., and catalog it all in albums as prototypes. Then you can set metrics to estimate the similarity of other people's mickeymouses to prototypes in similar albums. And that's it, that's no more you can do.
Except that there are billions of such prototypes, and most of them are unoccupied and have never been used. A neural network will generate a bunch of mickeymouses and textures and palettes, check that they are far enough away from the prototypes (far enough away that you can't sue them) – and send them to production. Or it may prove that Beksinski himself looks like a lot of other dudes and stole half of his original style from them. And the shit will go down the tubes like never before in the history of art.
Copyright was created accounting for dishonest people and with technical means of editing in mind – but it's powerless against the technical means of the human level, which doesn't screw anyone over but just finds solutions similar to the ones people have the skill to find. And in addition to these already found solutions, it can find a million others.
The problem is, after all, not with the model itself, but only with a questionable dataset in terms of copyright. In a year it will all be the same, but trained not on garbage, but on open and clean data. And no lawyers can do anything about it.
Generally speaking, the main question every artist should ask himself is: what in my profession is really creative and what can be attributed exclusively to technical skills? The sad truth of life is that in the profession of artist, almost EVERYTHING can be attributed to technical skills and tricks around human cognitive bases and visual illusions (aka "academy", "shadier in the shades, lightier in the lights", "work not from the line, work from the blot" and so on), while the real "creativity" comes down to a dozen elementary combinatorial methods. And that the complexity of their production (in the sense of the length of the description of the structure of cultural artifacts – paintings, statues, etc.) is incredibly low and meets a clearly distinguishable upper ceiling, ceiling which generative models do not have at all.
That is, most artists are not creators, but designers, ordinary outsourced corporate servants, aesthetic engineers who «make it pretty» for the masses of workers. And in the long run all this stoned and drunk public will go straight to the streets, because a neural network can "make it pretty" much faster and better than all of them put together. And on top of that, it doesn't drink, it doesn't get high, and it doesn't fuck up deadlines. Unlike them.
Stop hiding behind semantics, wanker. The words «think» and «invent» mean nothing and function solely as mental plugs in everyday psychological discourse. SD works as a semantic archiver – something like a winrar, which effectively compresses data without the possibility of unambiguous retrieval. In terms of statistical learning theory, it reconstructs a hierarchy of probability distributions over a finite sample – that is, it solves the general problem of induction. From the point of view of information theory, it reduces the entropy of the data. From the point of view of computational theory, it splits the original data into data structures and algorithms. From the point of view of algebra – into a basis of initial elements and a basis of elementary operations. The same is done by neural network structures in your own head and social network structures consisting of creators, gatekeepers, fans and patrons analyzing, generating and filtering cultural content.
A neural net doesn't pilfer art. It fundamentally can't generate a single original object from the training dataset because its compression during neural net training was done with IRREVERSIBLE data loss. If you compare the prototypes from the dataset and the generated pictures pixel by pixel, there is nothing in common between them. At the level of complex textures and brushstroke and line techniques, the similarity will already be visible. At the level of gestures and compositions, the similarity will be 50-70%. And at the level of color schemes, it would be as much as 95%. Do you know why? Because artists have been using the same basic set of poses, angles, compositional techniques, color schemes and light sources for centuries. That squealing animal on Twitter didn't invent new poses, angles and lighting methods – he took them from catalogs, from the mainstream or from «how to draw bullshit X» type pulp manuals. And to prove this, just run a semantic search on the original dataset and you'll find tens of millions of supposedly «original» and «copyrighted» images, which in reality are virtually indistinguishable from each other. Because the Internet has long been choked with copypastes and generics. But it's not the neural network to blame, it's the artists who, instead of doing creative search and active experimentation with new styles, have sold their souls for likes and are generating monotonous mainstream millet around the clock.
All of your so-called «anime», from the neural network perspective, is nothing more than a dense clump of pixel distributions, a single
object(Anime)- a limited, dumb, primitive set of visual schemes that activate dopamine receptors. It's not art, it's kawaii engineering. You don't solve creative problems and you certainly don't set them – you're just doing calculus and graphics work at the picture-building university in the department of anime glamour. Real artists are not threatened by anything – their product is authentic, material and forever fixed in real time and space, merging human tragedy, history and destiny. But you are not artists, you are decorators. The «derivativeness» of the neural network is just a mirror reflecting your own derivativeness. And don't blame the mirror if your face is crooked.More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link