Rise of generative AI will be comparable to the rise of CGI in the early 90s
sarharibhakti.substack.com
sarharibhakti.substack.com
It will be much bigger, because:
- it can be automated. CGI still required a lot of human work.
- AI generated content requires way less skill. You had to train a lot to become good at CGI. We already see people getting Stable-Diffusion-proficient in a few hours.
- in the 90, CGI was introduced gently, through selected medias. AI content has access to the full internet right out of the box.
- producing CGI required a lot of hardware, but with SaaS, AI will be available to anybody.
- CGI was first exposed to professionals, who produced for the public. AI is already been used by hobbyists, and soon the public, directly.
- CGI was mostly about visual techs, but AI generated content can be applied to text, sound, molecules, basically any structured data.
In summary, the very nature and context in which AI generated content is taking form means the impact will be larger, stronger, and way more intense.
People are hardly capable at dealing with CGI intelligently, even after 20 years. They trust photos and videos a lot, and react strongly to them. Even when they know they are fake, like in a movie, some emotions will forge they judgement like if it were real.
AI generated content? Forget about it. People are not ready. The emotional impact it's going to have will make you feel like the cancel culture was a whisper in the wind.
I was underwhelmed by the last decade of AI dev, but 2022 is for AI what 2000 was for the dotcom revolution.
Question is, what the equivalent of domain name for AI, something cheap you can buy to sale later?
- CGI can be automated, practical effects required a lot of human work
- CGI content requires less skill. You had to train a lot to become good at sculpting / painting / makeup / carpentry / stunts / whatever to become good at practical effects. We already see people getting $MODELLING_PROGRAM proficient in a few weeks.
- producing practical effects required a lot of hardware, material and tools but with CGI you just need a powerful computer. CGI will soon be available to anybody as computers become cheaper
Your last two points are probably only true for AI generated content but there are still some barriers to people setting up their own AI models (technical know how, hardware) and there are certainly limitations on what type of content can be produced (you need a big and reliable dataset of examples).
Overall I think this will have a similar impact but probably more widely distributed in both terms of access and impact.
Having worked in CG for several decades, this is a pretty misleading summary. CG takes a lot of manual work, and there are other reasons and benefits to using CG than how much manual work it takes. It’s often less work than practical but not always. There are certain kinds of things that can be automated (with a lot of manual setup work!) but most of what you see in movies also involves lots and lots of manual work for every single shot. It’s not unlike how machines get used to automate certain parts of practical effects work, using for example, compressors, electrical rigging, vehicles, pneumatics & hydraulics, gas & explosives, etc..
> We already see people getting $MODELLING_PROGRAM proficient in a few weeks.
This is similar to saying we already see people getting C++ proficient in a few weeks. Some people can write very useful & interesting programs within a few weeks, but nobody understands all of it by then, and nobody’s getting a high paid career on just a few weeks of noodling. It takes a long time to understand all the ins and outs, and independently it takes a long time to understand complementary computer science fundamentals. Same is true for Maya, Blender, Houdini, etc.. Someone might have something cool to show after a few weeks, but nobody is ready to make a CG movie by then, and nobody is landing the good jobs in CG. (By “nobody” I’m talking about statistical relevance, not that it has never happened.)
Another reason a few weeks is insufficient for landing good jobs is that the CG industry is highly competitive; you have to be better than the guy who’s gone to art school and played in Blender for a few years if you want that job.
Can they achieve a Monopoly on Art Production into the Future?
To build their moat: they build the largest supercomputer for image training, acquire the largest image training set (~10B), and then institute an IDF Unit 8200 style Garin Lotam recruitment effort of the best artistic talent on earth to feed new forms into the machine, and everyone else simply forgets they ever knew how to make art in the first place as the drawing prompt becomes as ubiquitous as the google search box ;)
Forget about it? People are not ready? When has that ever been a blocker? This is like a new era of 80s computing where some people will get a home computer and their dork kids will dork on it and produce some unbelievable stuff in the decades to come. I'm scared, slightly optimistic, but what is available right now as commodity, there ain't no stopping it.
Brace for it, it's going to be wild.
You can at best have another AI trying to pick the best option generated by other AIs but you'll likely still have a human deciding.
We're still far from General AI.
It shouldn't be like so, I think generative AI can yield great results if there is enough human supervision, I think it can be great if used to augmenting creative work. But I think many will chose the shortest path towards money and they will put in the least amount of work possible.
Imagine tens of thousands of movies spitted by AI based on sub mediocre screenplays, imagine millions of songs generated by AI based on sub mediocre lyrics.
Somewhere, in that ocean of junk, there is something valuable. Good luck finding it.
If we use CGI analogy I can say that all pre CGI Tom and Jerry cartoons are works of art but many of the cartoons released after aren't so great, while some are. And CGI still requires some effort, so not everyone is be able to release movies. Generative AI will require far less effort and knowledge.
The quality bar will be raised, but if you want an audience, you will still need to produce something that stands above the rest. And the way to do that is the same as today, put in more effort than the others. Make something that's better, that's worth peoples time.
Eventually we will only generate art for personal, one time use, or for small groups sharing an interest.
Hopefully it all eventually settles down and becomes just another tool for skilled artists, rather than thousands of nerds churning out the visual equivalent of elevator muzak.
[0] I saw this in a few of my UK job contracts in the context of signing away that right, and it seems to be "the right to be identified as the author of a work", but IANAL.
You clearly haven't seen some of the Hanna-Barbera's later works. They weren't called "illustrated radio" without a reason.
If the marginal cost of content is zero, then the curators will become important.
The only things that are working for me are Apple music and YouTube recommendations, both algorithmic (but also both with very low downside of recommending bad content due to short length in general)
I'm working on something myself to help users with the process, see another message I posted for this story.
Generative AI, on the other hand, lowers costs so radically, that it will spur entire business models. Within a few years, we can expect feature length animations/TV shows to be created by teams of less than 5 people. Live action shows, requiring photorealism, will be a bit harder, so maybe 5-10 years. This will in turn change the business model radically, with indie creations competing against big budget productions, it'll turn Netflix in the direction of youtube, etc etc. It will be very similar to how livestreaming and youtube affected TV-talk shows, as large production staff are no longer needed.
This is no exaggeration, I've seen animation created with diffusion models, that would have taken a traditional animator a year to animate by hand, now doable by a single person, within a week. It's been less than a 1 month since the model's release (novelai), imagine what would happen in a year.
Im thinking generative AI will indeed make animations ubiquitos to the point we’d ingest so much content to the point of becoming ‘alergic’ to it or seek something other kind of entertainment.
There's also the quality issue. Animation scales very linearly with budget. With low budget, the best you can do is flash animation, which limits the formats of story you can tell very heavily (99+% comedy) With AI, individuals can match the quality of professional animation teams, which drastically expands the range of ideas they can explore. So that eliminates a lot of the boredom.
Note this wouldn't displace existing animation studios, it'll just allow them to push their production values way way higher. For hand-drawn animation, even top-budget films, still have moments where they cheap out. In the future, the standard is going to get pushed insanely high, where every frame will be screenshot worthy, because they are literally AI illustrations. Aka, instead of one Mona-Lisa, expect 1000 Mona Lisa paintings, formed into an animation of Mona-Lisa going through her day in renaissance Italy.
The actual threatened jobs, aren't in animation, but in live action. Because its far harder to make perfect photo-realistic images than stylized images. This gives animation an advantage that live action doesn't have. For example, the recent Rings of Power cost 1 billion, but an AI animation would both look better, and cost far far less. Sci-Fi and fantasy shows are especially vulnerable to competition from animation.
Do you have a link? That's pretty hard to believe.
It uses diffusion to post-process a 3d video frame-by-frame. You can see the massive improvements in facial expressions, from bug-eye alien face to something quite pleasant.
3d animation is great at allow for complex movements and dynamic camera angles, but horrible at facial expressions. That's why Pixar had to go with the exaggerated proportions 'cartoons' look, to ensure the faces are large enough to convey emotions, even in 3d.
2d animation is superb at facial expressions, but can't do complex movements. Drawing a static face is easy, drawing a rotating character is very hard, drawing a rotating character, while the camera itself is rotating, is brain meltingly hard, so almost never done.
Diffusion based workflows will have none of the downsides, and all of the upsides, compared to either approach.
> Diffusion based workflows will have none of the downsides, and all of the upsides, compared to either approach.
But isn't this true in the same way as tracing over 3D has the "upsides" of both approaches?
Indeed, diffusion is superior to 2d animation, even in raw quality. Because diffusion will soon be able to produce illustration level pictures for each individual frame. Whereas 2d animation had to rely on detail-compressed simple artstyles AND keyframe-inserts to cut down on costs. Soon every frame will be a keyframe, and it can execute artstyles previously impossible to animate.
AI will just mean most fiction start following this same model. Where characters and key environments get detailed designs and images, but the rest remain in words.
Of course, it’s professionals using it rather than amateurs. I expect generative AI would largely turn out the same way.
Are you sure it isn't the complete opposite? I'm pretty sure that the amateur kids who played around with Machima in the age of Quake went on to become the professionals in the video game industry.
Compare that to what Youtube did to TV, where Youtube did in fact replace a lot of peoples TV watching with Youtube content. A similar shift hasn't happened with movies. Good low budget indie movies are still extremely rare, and generally aren't done in CGI, but filmed classically with cameras.
Cameras, editing and compositing getting cheaper and completely changed the game for Youtube-style video productions. Machinima just didn't have the same impact. Creating CGI/game assets is still a costly and timely endeavor out of reach for most amateurs at a movie-scale. You see a few 5min shorts every now and then, but no two hour movies.
In about 10 years from now, nearly all the activities - playing music, composing music, animation, live action footage etc. and yes storytelling — will all be mostly automated. And curation of them, too. It will be like you knitting your own sweaters… most people wouldn’t see the point. They could just buy an endless variety of shirts on Amazon.
Great to know someone saw everything. I thought a human could never finish watching any genre.
This one resonates with me a bit.
We already see this in goods to an extent "artisinal" vs "cheap dropshipped crap".
People find value in seeing other people do something well (see the popularity of maker videos, or the number of people commenting on someone's performance in a movie rather than the plot, etc). It also captures the idea that art isn't just a product, but a method of communication and is inherently human.
I dunno, just rambling before my morning coffee.
Without enormous training data from human victims, your "Artificial Intelligence" doesn't exist.
In the end, your "AI" is just huge interpolator, using matrix multiplications and max() or gelu() to interpolate the work of real intelligence: the human artists.
Once it's illegal, only criminals will use it. No first-world company will break the law to save money on artists. You think anyone who matters will secretly use a neural net trained in violation of copyright? Think again.
I say this as a person who writes GPU BLAS kernels professionally at a company you've heard of. I directly profit from this theft. But I also know it's wrong.
Also keep in mind that, at least in the US, you can't legislate your way out of fair use since it's a constitutional right. (It's the result of the tension that exists between the copyright clause and the first amendment)
I think on an ethical level we're just going to need the expectation of separate agreements / legal conceptions for having ones work go into a training set. It's novel territory, providing not only work-for-hire but meta-work-for-hire. Might even be that there's no true compensation without equity.
However, the output of that model is the epitome of derivative work.
The trained model has to be viewed as an artistic tool used by the artist to create art, imo.
Like I can't release a sequel to a Star Wars movie without getting a license to do so by Disney, because they own the original work's copyright, I'm not allowed to produce a derivative of it.
Similarly this is why there needs to be an explicit open source license that allows you to produce derivative work from it.
That's the crux of it, the trained model is a derivative work, literally it's mathematically derived almost deterministically (ignoring SDG being stochastic).
The question is, even though it's derivative, is it still fair use to use it without permission?
You can't say it's derived from many sources and then say it's original work haha.
The output work would appear original in most cases (if you filtered out for left over watermarks and too much memorization, and avoided it copied any style or visibly recognizable characters, etc.). That said, even though it appeared original, it would be known to be derived unless you'd hide the fact you used a ML model for it.
I'm not sure you can argue it's not derived when it's literally being mathematically derived lol.
I think "fair use" is really the crux of it. Like do we feel this is a valuable derivativion to allow, because it benefits society in other ways that are worth it.
This is in fact an unknown, it could appear that Meyer would have at least a case against it if she wanted too.
We know for sure people are capable of original thoughts. If there were no prior movies to learn from, a human could figure out how to make one, just as the pioneers in film making did.
On the other hand, models cannot, they need large quantities of existing source material, so it's unclear if they're capable of original work, and likely it's derivative.
> Do we raise copyright to the point that no one can produce new work
You don't actually have to raise copyright at all, it only takes for a court to say that ML training isn't fair use, that's all. It would only exclude ML training from fair use, meaning the people training the model would need to get a license approval from the copyright owners.
> or accept a world where generative art lowers the bar for making art?
Yes, that's the question of Fair Use. We might consider the infringement to be a net positive and want to include it as Fair Use.
Normally that's what Fair Use looks at, it's the balance between what's right by the authors, and what's best by society.
I think it's totally fair to say it be best here to allow this use because the benefit to society of that technology are worth it.
I also think it be fair to say that the benefits are not worth it compared to the loss inflicted on artists, or to think that most companies investing in AI would actually have the mean to compensate the artists anyways so it wouldn't even stifle ML innovation while also being fair to the artists.
Personally I'm still debating it with myself and not sure where I stand.
But I think it's very misleading and disingenuous to pretend that the ML model isn't even using copyrighted material or deriving work from it, and to pretend like it's similar to an artist simply looking at other people's work or being inspired by other people's work. It's very different in many ways, the process, the mechanisms, the nature, the function, the ethics, the financial applications, the capabilities, there's just so many differences here.
If you can't tell without knowing it's ML generated, I don't see how you could argue that it's infringement.
Yes it is? There's nothing that blanket bans derivative works in the west. Parody is often derivative, and explicitly protected by the fair use doctrine in the US, for example.
We’ve argued this for centuries. Fair use is a fine line, but it exists. And it doesn’t matter whether the generative agent is human or machine. If a sufficiently unique work of art is the result, it’s not stealing.
The same places as before?
What's stopping anyone from training these models? The cat is out of the bag and artists need to learn they will have to cede on these copyright issues.
Similar to musicians having to cede on digital reproduction of music.
Artists will become style makers.
Personally, I think AI has done about as much as crypto, maybe even less. I can't think of a single use of AI that I couldn't do without.
Now consider if I'd made that, trained the AI, and then DID NOT SHARE IT. Suddenly, I've got an AI thing that other people don't have (not directly, they can eventually have a second-order knockoff of it) made to express an intention that knockoffs won't have. So I'd have both the story to tell (hence the original training set) and the means to automate the creation of it.
You've already got artists/technicians making things like Clip Studio brushes, things that transform a stroke into a very different style or dump scatterings of clip art for texture. This is literally the same thing only way more sophisticated.
Just because you have a 'foliage brush' doesn't mean you can draw a woodland scene, it just makes the texture rendering trivial. Just because you can type 'woodland scene' doesn't make it a SCENE, it just makes the image rendering trivial. Does it serve a story purpose? What does the imagery do, beyond just existing?
The mere existence of 'imagery in X style' is rapidly going to become really boring. Once 'an image can look like anything at all' becomes commonplace, to be impressed by that will be about as sophisticated as being impressed that plastic is shiny. It becomes all about 'what did you do with it?' and that is where artists come in.
The problem is in most of these cases is that a powerful entity is profiting off of other people's work without their consent and gives nothing to the exploited members in return. Sure, an individual human learns from copyrighted works and reincorporates to make something slightly new all the time. And then they may also profit from it and not give anything in return to those that came before.
The problem here is the scale and the power that enables that scale. This is industrial level mining of non-consenting humans, exploiting their life's work in many cases.
It's like saying that the CNC machine should also be given mandatory time off each year just as all the other factory workers.
One is a machine, the other a human, what rights and laws apply to one don't mean they make sense and should be applied to the other.
The better framing is, should art students also be allowed to make derived work from copyrighted images without permission?
The output image from a model is basically a derivation from it's entire dataset, I mean this is mathematically demonstrable. If it wasn't for a few stochastic parts of the model and explicitly added randomization, it be deterministic even.
It takes copyrighted work + prompt and outputs derived image.
I don't think the question of derivation is being debated that much though, the question is if it's fair use. As in, deriving from copyrighted work a ML model, maybe that's fair use. And that's what has to be figured out.
They should be, and are allowed to do this.
I could imagine that if copyright law were to be extended in this way, artists would either be small enough to fall under the radar or select for popular enough artists that they could sub-license the 'copyrighted style' in question. Regardless, only large corporations and a select few 'superstar' artists would benefit at the expense of the vast majority of artists and the public.
If this were to pass (and I very much doubt there's a good chance of it happening), all is not lost because there is a wealth of public domain imagery that we can still train on. This could then snowball into providing richer datasets to train on.
I don't like the framing of a lot of these arguments. I'm paying taxes to provide a state apparatus that provides copyright legal protection for artists. I'm happy to do this, up to a point, as it's essentially a contract that provides incentives to innovation by rewarding artists, with the understanding that, eventually, the artists works get to be used by the public (that is, given to the public domain/commons).
Copyright law, at least in the US, specifically forbids ideas and "styles" to focus on specific realizations. I understand we're in a shifting landscape where technology is changing this narrative and that copyright law was created with different technology in mind but copyright law exists, presumably, as a compromise to incentivize artistic creation while serving the public good. Copyright law is not meant to be corporate welfare nor is it meant to protect the artist at the expense of the public, in perpetuity.
It would be nice to have a more reasoned argument instead of this knee-jerk reaction of theft.
What could happen if this interpretation of fair use comes to pass? Model trainers may have to license images from companies like Getty. In a sense, it should be cheaper than using an image individually. In some countries, music organizations deal with licensing large sets of songs, no need to negotiate with each label or artist individually. That sort of arrangement could come to pass. Perhaps there'll be a simple option on image upload sites to select the license for the image - allow or disallow model training. Or perhaps everyone will simply use below-the-board models surreptitiously.
Or similarly, if a company trains a model and then bins the training data. How would you verify any artists work was in there?
As others have already mentioned here before:
- a "style" isn't copyright-able
- Looking at how others have done art, imitating them to learn, and then combining what we learned with everything else we've seen in our life is how humans produce art. And it's the same way that stablediffusion produces art: train on data, reproduce based on what it's learned, combine everything its seen with a prompt to create something new.
This is different to most copyright cases where all that matters is the end result, not the process.
Your first question is based on a technicality (human can learn too, should we then …?), but the idea of right is not based on physical properties or similarities, but on realistic desired outcomes of its application. (Of course we see flaws of laws all the time, but that is tangential.)
company trains a model and then bins the training data. How would you verify any artists work was in there?
Like any shady company it will drag a set of risks with it forever. It may fly under the radar for some time, but will be snitched on after few internal disagreements. This well-known dynamic disincentivizes big companies from choosing going this route, and small companies are not a big deal.
On a serious note i am now confused. Because at the same time i see ai generated art as a tool for creating new art, a facilitator. Still i dont agree people’s ip should be used without their consent. Oh boy progress is confusing sometimes.
Great video from CGP Grey on the topic: https://www.youtube.com/watch?v=7Pq-S557XQU
Source: in my city there's plenty of people that use mules for transport, it ain't pretty.
Disney is a first world company and it does exactly that: https://www.yahoo.com/now/column-disney-allegedly-cheated-hu...
Getty images is another: https://petapixel.com/2016/08/04/getty-images-sued-accused-m...
In the US, the approach is to see what you can get away with, and you can get away with a lot with a pile of money and an army of lawyers.
Step 1. Hire some outsourced artist in the rest of the world, where people don't care about AI art regulation.
Step 2. Watch as those outsourced artists outproduce the domestic artists, since they have the twin advantage of lower costs AND better quality due to being able to use insanely powerful AIs.
Step 3. Watch as the domestic art industry gets wiped out, just like manufacturing did.
Politicians can understand this 1.2.3, therefore they'll never ban art AI.
There's no putting the genie back into the bottle.
Could you use the same theory to model 3d drawings (parts ?) in STL for manufacturing.
Ie, is the ai limited that way ?
https://dreamfusion3d.github.io/
Regards,
another shitty C programmer
Thanks mate.
I would love to change my position if there's any evidence. Perhaps it could start with a single example of any non-trivial job position being removed due to these image generation models.
There's a few games, albeit often indie, where they used generative music makers for their soundtrack, saving them having to hire a music producer.
I mean, if you've used or tried any of those models, you know that very soon it will allow an artist to be way more productive, what needed 50 artists might only need 30 now.
I'd really be surprised if this didn't cost jobs.
The only possibility it doesn't is that the projects get even bigger in scope, so increase productivity would go into more am ambitious projects instead of job cuts.
Then turn that AI loose and let the internet use it to make Duckman episodes but people figure out how to tweak the prompts to make Toy Story at scale but also use it to train their own AIs.
We have reliable secure GPU multi tenancy now. Next year we'll start seeing AI designed silicon photonics creeping in. All this GPU training is now pretty accessible to be passed all over the internet and back securely and streaming.
No one has any clue what's coming in a very short few months. Well, we have a few ideas... but it's going to be surprising. These systems build systems to build more systems there and back again. The only constraint being energy availability and attention.
The powers that be captured all of human experience for a couple decades to seed this project. So off we go I guess! This moves so much faster than copyright law. This is our new world. The compression of information contained inside of models is something that existing bureaucratic systems cannot hope to regulate. I'm pretty sure no one that can is even slightly interested in regulating it anyway.
Our new way of life is gonna be pretty weird. I think some fun new media will come out of it though. Which is pretty cool. Also you know... Utopia or Oblivion.
Does the legal situation change if instead of a pencil they use photoshop, or if instead of photoshop they use runway?
I think this is the best approach because:
(1) Models are public, so everyone can benefit from them.
(2) Such models can be trained on public data for improvements. (No complexity involved over paying for data, no copyright risks regarding data, etc)
(3) Companies can sell services. (model-as-a-service, customization, apps/tools, education, etc)
(4) Companies may lose competitive edges, but they can enjoy the bigger market. (I'm looking at Matlab and Mathematica. They are really good, but Python took the market eventually.)
It feels like many of FOSS pros/cons can be plugged in here as-is.
Copyright only concerns with expression, not ideas. You can still learn the concepts from copyrighted content without violating copyright. There is a limit to what can be protected from AI with copyrights.
When they trained Stable Diffusion they used 5B image-text pairs, but the final model fits in 5GB. So they didn't actually take too much, about 1 byte per input example, which is in line with the idea of learning general concepts rather than copying.
An interesting parallel - the human DNA is about the same size, 3.2B nucleotides. The SD model is like a cultural DNA, and like DNA is compact, easy to replicate and powerful. At this level copyright is an anti-pattern, genes and memes like to travel and evolve.
For me, countries that don't enforce that or old models from before the ban. Then again, I'm a nerd who wouldn't release my creations commercially—they'd be AI-generated shitposts or images done for the joy of exploring what my new tools can do.
If the future of copyright law in the west moves to outlaw use in training data then the west will lose out to companies from China, Vietnam and anywhere else that doesn't care at all about western copyright. I think it will give them a competitive edge.
Already with Stable Diffusion it's way more rewarding to play with it yourself than look at generations made by others. It's more like a game or a dream that is a personal experience, than 'art' to be shared with others.
https://www.youtube.com/watch?v=CqIXRB_Uw5s
https://mixed-news.com/en/nvidias-latest-open-source-ai-gene...
My mind still sees "CGI" and thinks common gateway interface, not computer generated imaging.
Using an AI writer generator (GPT-3): "Looks great, I can make a great SEO blog and a storybook SaaS out of this now, I don't need a writer anymore. Writers adapt or die."
Using an AI artist generator (DALL-E, Stable Diffusion, Midjourney): "Looks brilliant, this IS the future. it's inevitable, I don't need an artist anymore. Artists must adapt or die, sorry."
Using an AI code generator (programmers) (Copilot, Replit Ghostwriter): "No thank you, Bad. legislation now, why are you violating my copyright / open source licence?"
Why does it seem OK to obsolete complete industries by outright violating copyright laws, stealing work (even generating with a shutterstock watermark!), often without attribution and hail this as a new dawn for AI?
Or is this mainly VCs finding out that this is now a new way to 'provide shareholder value' despite all the costs that come with it.
So just throw cash at any startup that can generate a hot dog image in any style or generate a hot dog detector with copyrighted code, writing hotdog story no matter how bad the story is.
by the way, you forgot hypothetical fleets of self-driving trucks putting truckers out of work, and so on.
Frankly I think it's projection from non-SWEs who don't understand that automating your work away is the core mission of software development. There are good reasons to worry about how Copilot might impact the quality of software, especially if it's being used by someone who doesn't understand the generated code, but "what about my job safety??" is nowhere on the list.
There are of course some people in there against the very idea of Co-Pilot, but from my quick skimming they do appear to be the minority.
First, it helps quell the extreme ends of anti-AI paranoid fear I sometimes feel, to think that yes, it could change the world significantly and make some things disappear or reduce to irrelevance (e.g. realist art) and make new things appear and rise to relevance (e.g. mass media), yet throughout all this the more fundamental aspects of humanity and human society and culture have endured and adapted just fine.
Second, it gives me more empathy for those people who were/are afraid of or simply opposed to other technological developments which I take for granted (again photography, TV, the internet, trains, nuclear power..), which various fears/oppositions are often an object of derision and ridicule.
I feel this (AI, starting roughly at ~alphago) is the first significant breakthrough that I'm consciously living through myself, and this does a lot to one's worldview I think.
The situation is so bad that even the latest Elder Scrolls games reuse assets from games from way back to speed up the process.
I just wish there's more progress in AI-generated voices that sound more natural, especially for dialog in action sequences like yelling/screaming
Calling Stable Diffusion etc AI art generators is kinda like saying our visual cortex is an AI art generator because it can render pictures for our consciousness.
It's perhaps not much more than this at the moment, but it will be, soon.
What these mechanical approaches offer is speed, convenience and clever tricks. But not art-power.
So here we are puzzling over various flavors of flimsy and how to make a better fast-food taco.