Another opinion popular with no one: AI will have on artists the same impact that Spotify had on the music industry that is, it will kill any revenue flow for anyone outside of the publishers and big artists/players.
Spotify basically killed any money coming from the physical distribution - Worse than piracy, which was inevitable too at the time, but at least you didn't have to pay your lawyers to renegotiate with your label on top of NOT getting any money.
Adobe, OpenAI, whatever: they want artists to draw for them for peanuts to train their model, sign a waiver saying "I'm ok not getting any money from any AI art made from this", and then resell the output for $$$ on something like Splice[1], at the same time overtraining such models in ways that make extremely obvious whose artist made them in first place.
At the end of the day the model itself is going to be basically irrelevant, while knowing whose works were actually used to train it being the truly differentiating feature.
But you know, "the AI did this picture, so we don't have to pay you."
I agree completely, and I have been constantly speaking about how AI will be a wealth concentrator, replacing a mass of jobs more diverse than previously seen. Unlike previous machines which can take 1-2 jobs, when humans get REALLY efficient at training AI, it will replacing hundreds en masse.
AI will also have an additional effect: it will be isolating in the sense that the need for other humans will decrease.
These two points alone, strengthened by many others, have led me to conclude that the world is MUCH better off with AI and that tech companies are ruining the world with their abominations.
> These two points alone, strengthened by many others, have led me to conclude that the world is MUCH better off with AI and that tech companies are ruining the world with their abominations.
Do you mean "world is MUCH better off without AI."
What you wrote doesn't make much sense withing the context of your comment, but I have to ask because there are some software engineers that find abominations appealing for some reason, or just lack the ability to tell the difference between desirable technology and a technological abomination. I think a big component of the latter is many software engineers' overconfidence in their abilities that makes them easy marks, and the willingness of many kinds of hype men to exploit that to con them with propaganda.
I am not a software engineer. When (for my work) I/we need a decent chunk of development done, we get the pros.
BUT, sometimes I want something that will automate the fudge out of my PC (imagine command prompt on overdrive). I usually DDG for the solution and end up in some 10yo solution in StackExchange, which doesn't do the thing.
My friends have all forgotten their DOS skills.. so I turn to ChatGPT and boom! I get me 2 paragraphs script in 30secs.
Do I hire devs? Hell yeah and we pay well, and we will continue to do so for many years.
Do I use ChatGPT for the small (personal) stuff? Hell yeah too.
Now, if a company wants to outsource everything to an LLM/AI then I wish them the best of luck, coz when something will break (and oh IT WILL), Tthe contractor they screwed over should charge them x50!!!!
Definitely agree, LLMs are only as useful as the person interpreting and implementing the output; if someone doesn't have enough knowledge or context about the thing they are trying to solve/create then copy & pasting blindly while asking the wrong questions will lead projects to disaster.
I have witnessed this firsthand when I dove into the deep end on something over my head, GPT-4 Code Interpreter went into an error loop and I had to learn all of the background knowledge I was foolishly trying to avoid.
For software devs AI will mostly be a golden goose because they can leverage it to full extend to increase their portfolio of solutions they can sell.
> Tthe contractor they screwed over should charge them x50!!!!
IT already does this after they were outsourced. They build up IT companies that take at least 3 times as much for consultation and you still need to employ local IT that actually implements the solutions. And their wage also doubled as well.
What about socialized housing, food, and health care then?
Socialize the essentials, let people work for the non-essentials.
If there isn't enough work to go around for people who want more than a substistence living, start reducing the definition of "full-time" until there is. If only 50% of working aged people can find work, redefine full-time as 24 hours/week
It definitely can work… it depends on the size of ubi and how creative we get. We could for example just say that banks no longer get to do 10-1 fractional reserve banking and instead all the free money gets distributed to their customers accounts. And do a cap and trade carbon system with auctions where all the revenue goes to ubi. And all the revenue from spectrum auctions. And repurposing some existing spending. And printing a little more money. And congestion pricing. Etc, etc…
> banks no longer get to do 10-1 fractional reserve banking
Fractional reserve banking is pretty much an urban myth. Banks create money when they make commercial loans. The Bank of England explains it quite nicely here:
What doesn't work about the math? There are many alternate taxation schemes and strategies that can be tried, such as increasing VAT on certain products, adjusting tax brackets to income changes, higher capital gains taxes on investments, and making tax evasion more difficult.
Don’t you think if it were possible for governments to gain multiples more tax revenue, they would do it already?
Taxes are competitive. If you have extremely high taxes, then the businesses that can move out will move out, leaving a smaller tax base and requiring you to raise taxes even more in a death spiral.
Possibly, but depending on the level of budgetary entrenchment it would prove difficult for some governments. In any case, a staged roll-out is more likely to succeed, with many hurdles along the way.
It's worth repeating that doing nothing increases overall societal costs if large parts of the population become unemployed, whether due to lower statewide income taxes, increased welfare costs, people resorting to grey collar work, and other secondary effects from increased displacement (lower consumer spending, rentals sitting empty, health epidemics, increased crime, etc.). UBI does not have to be a blanket instrument either: rather than income, we can focus on making certain goods and services universally available, such as access to food surpluses that would otherwise be overturned or basic internet access to enable people to remain connected without expensive contracts.
A solution may even exist beyond taxation: making reschooling and job pivoting more accepted within industry, lowering admission costs to tertiary education, or guaranteeing placement of employees when let go on account of automation or cost-cutting. What way the pendulum will swing remains to be seen.
Those corporations still want access to the US / EU / etc. customers so if they move elsewhere to dodge taxes you need to increase tariffs to compensate or deny them access to the market entirely. Those businesses are not irreplacable and they are worth nothing without customers.
No, of course they wouldn't. A politician cares far more about protecting their own wealth than increasing the size of some government department budget.
There won't be UBI, period. Though I could see a future where obsolete people are warehoused in sex-segregated poor houses until they die out, if it's determined that their freedom is threat to stability.
Another side effect: the wealth will be concentrated in rich tax-avoiding corporations and elites, meaning that the tax burden for society will fall even harder on the remaining middle and working classes, who will have to pay for the upkeep of everything.
And yet another side effect, the one that I believe trumps them all: a loss of meaning.
If somebody with zero skill in the arts can produce output of similar quality as a craftsman and about a thousand times faster, what is the point of art anymore? Sure, one can enjoy the very act of creating art, but we can't deny that art has value in relation to an audience, and is also a display of skill and a source of pride.
What if AI generates the perfect music just for you, based on your taste? Here we lose any and all social/cultural aspects of music. There's no point in discussing music as we have no shared experience. There's no point in emphasizing your favorite song because everybody exclusively listens to favorite songs.
What if you need to write a long essay and use AI to help write it. I receive it and use AI to summarize it. Other than this interaction being supremely depressing, what is the point of it at all? Just submit it to the big machine and perhaps some of it will show up in my use of ChatGPT-17.
So you were faster to write something whilst I was faster to consume it. This allows the both of us to do more in a single day. This "big win" won't gives us back free time nor raise our wages though. I just means that the nature of the work is for us to take the job of being guard rails for AI, a soul crushing job in itself but also temporary, until the rails are no longer needed.
>If somebody with zero skill in the arts can produce output of similar quality as a craftsman and about a thousand times faster, what is the point of art anymore?
I like to think "AI" will make art better reflect its real value, devoid of the tangential flat costs associated with housing, clothing, and feeding humans in the process of producing art.
The consumers at large demand driving down the cost for consuming and enjoying art, and raise hellfire if there is so much as a suggestion of raising that cost. Remember how much controversy there was and still is about raising the standard price of video games from $60 USD to $70 USD? And that $60 USD today is pennies compared to $60 back in, say, 1995.
If the consumers at large demand the cost of art to go down and "AI" will make the process of producing that art better reflect that real value, isn't this overall a good thing insofar as making the price tag more clear and agreeable and closing down sweatshops?
AI can probably replace Katy Perry, but could AI generated music ever replace Rancid or Junior Kimbrough or Fela Kuti? I don’t think so personally. I think truly human music will continue to stand apart.
I do agree with you in large part. I think I’m just slightly more optimistic that people who are driven to create will continue to do so and that people who really want real and human experiences and interactions will be able to find them with effort. Probably not anywhere on the mainstream internet though. Maybe even only in person.
Ever is a strong word. I shouldn’t rule out a future like that. But I don’t see the through line from our current ai to one that has entirely supplanted all human creation.
That consideration seems more along the lines of worrying about the eventual need to escape earth than a future on a closer horizon worth worrying about.
I’m more concerned about how every facet of our children’s lives will become inundated with shoddy ai being used to extract maximum profits at the cost of any humanness, and the death of all genuine communication on the internet.
AI has, in a few short years, gone from hardly being able to string words together, to writing coherent grammatical sentences, to being more proficient than an untrained human in many cases. ChatGPT is way better at writing poems than me, for example. Its style transfer capabilities are out of this world.
Thinking that the progression is going to slow down is just wishful thinking.
It is more visible with image synthesis. Sure, the style can be freely switched in a few seconds, fitting compositions from trained concepts is very impressive.
But there still is no real creativity. No emerging concepts aside from complete accidents that cannot be replicated again. The same is true for the other direction with models like Clip could create an interpretation of generated images. It is impressive, but there are still clear limitations. You cannot expect linear growth here, it could be that the current AI approaches are wrong, we hit a plateau and need fully new approaches for significant improvements. What we now have is insane amount of data and more powerful hardware, it could be that we have years of iterative and slow improvement while people fine tune their models.
I think LLM have the same problem overall, it is just more difficult to notice.
I think they mean jobs as in lines of work not jobs as in instances of employment.
One thing I wonder about a repeat of history is if the lowest classes still get enough of a share of the increased output in income/QoL would there still be a revolt about the increasing wealth concentration?
Not a real equivalent. Those machines are made by many, many people along the way. Industries exist from those machines.
With software you could say chip makers, developers, and energy companies will get stronger but I don't think there's a comparison. The keyholders will be a much smaller group with a greater power if we stay onboard the AI train.
And they do lead to unemployment. What are the people thinking? There is unemployment, and technology causes it more than it relieves it, especially software technology.
> Spotify basically killed any money coming from the physical distribution - Worse than piracy, which was inevitable too at the time, but at least you didn't have to pay your lawyers to renegotiate with your label on top of NOT getting any money.
It’s even worse than you say — it was murder on digital retail too, right at the time when it was on track to compete with or exceed old physical sales.
Spotify adopted the economics of piracy and stamped them with the false veneer of legitimacy.
Fundamentally, neither spotify nor piracy matter. People enjoy making music. Today, there are more people able to make and publish music then ever, but the day still only has 24 hours, you can't listen to more music then before. Unlimited supply, limited demand.
We're talking about the business side of the whole ordeal. It's not just about "enjoying making music". It's about paying mixing and mastering. It's about paying NTS, Rinse FM, and the constellation of medium-small distribution channels. It's about distributing on labels like DFA, !K7 or whatever. It's about making sure that Fabric, Rex Club and Sneaky Pete can keep the lights on so they can play your music, so you can get paid, so you can keep making music instead of ahem having to become a webdev and write angry comments on HN.
It's about keeping an entire industry, live or recorded, and their milieu alive.
The truth is that what happened wasn't a liberation. It was a methodical purge of the medium-sized side of the music industry. Now we're reaching the point of having 5-6 industry giants taking all the money plus...yes, an inordinate amount of people making mostly self-referential music in their own bedroom on weekends, music that will reach no-one outside whatever local scene they hang around. But most of them were making music even before, and were by their own choice irrelevant to the industry. (True, now they can also become influencers on Twitch and maybe one out of thousands can make a living by streaming their life 24h/day. One ticket for the lottery, please). Whoever was between them and the majors is being squeezed out of the game.
Fundamentally, the artists getting paid doesn’t matter because they enjoy making music? As a musician, your comment is completely ignorant, self-centered, and totally irrelevant to the discussion of people getting economically screwed.
As more people are able to produce music (due to cheaper tools like DAWs, more accessible music theory education, etc etc), if the demand of music doesn't grow proportionally, the average income of musicians/songwriters would decline.
The above will happen regardless of Spotify's existence. Thus, Spotify doesn't matter (much).
That's like saying as the price of circular saws drop in price, hand made furniture becomes cheaper.
You're just going to end up with a bunch of sloppy tables.
People still want to listen to quality music from artists who have years of practice and experience. You can't reliably get years of experience unless you're getting paid to do it.
Sure, there are exceptions, but it's not the rule. Michael Jackson would not have existed if there was no money in the career. The money is why his father pushed so (insanely) hard.
The counter argument is trash music will just be the norm. And maybe for a while that would happen, but eventually we'll see someone (similar to the private search engines we see today) come out with a new platform with the selling point that artists get a living wage -- as long as the people demand it, and I believe they will.
> That's like saying as the price of circular saws drop in price, hand made furniture becomes cheaper.
Uh... and it's true? If the price of circular saws drop in price, and the demand for hand-made furniture doesn't change, then they'll become cheaper. How much cheaper is another question, as circular saws are already very cheap today, compared to hand-made furniture.
So yeah, you're right, it's just like saying that.
> if there was no money in the career
It's unlikely to decline indefinitely. Piracy, Spotify, more youtube channel teaching how to make music... all these didn't prevent Billie Eilish from becoming a star.
> Uh... and it's true? If the price of circular saws drop in price, and the demand for hand-made furniture doesn't change, then they'll become cheaper.
I don't know whether circular saws are likely to be the biggest input into a piece of handmade furniture, but my guess is they're not and it's time invested in the specific work and practicing the art rather than industrial capital.
Similarly, while DAWs while can (though not necessarily do) reduce capital necessary to do certain aspects of production, they don't represent the most significant investment into writing music. Also time, both in the creation of the specific work AND in terms of time practicing the art.
> It's unlikely to decline indefinitely. Piracy, Spotify, more youtube channel teaching how to make music... all these didn't prevent Billie Eilish from becoming a star.
Survivorship bias. Billie Eilish or any other individual success are no more an indication that all is well with the status quo than blue zone anecdotes are promises anyone who chooses can be a centenarian.
>That's like saying as the price of circular saws drop in price, hand made furniture becomes cheaper.
>You're just going to end up with a bunch of sloppy tables.
Well, yes, and that's how IKEA and mass production in general made many people that would be making furniture out of the job.
Even in tailor-made stuff good cheap tools does make work of skilled maker far quicker. And you can get more people trying to get into that if the tools are cheap.
Hardware is cheap, software is free/near free so there is far more people trying, when you no longer need to spend small car worth of money just to say play electronic music
> People still want to listen to quality music from artists who have years of practice and experience. You can't reliably get years of experience unless you're getting paid to do it.
Most musicians got that by playing in garage bands and doing concerts.
And many of them did it entirely for free, out of passion, till they were good enough, far before fancy computers were in everyone's pockets.
> The counter argument is trash music will just be the norm.
It is the norm far before Spotify happened I'm afraid
> You're just going to end up with a bunch of sloppy tables.
That's only true if you assume all the customers desire (or are willing to settle-for) arbitrarily bad tables for cheap. That isn't guaranteed, but even then... why are you so certain their decision is wrong? Maybe they simply care about something else more than their tables.
Meanwhile, the section of customers who still desire good tables will find those good-tables more affordable than before, even if they're a relatively smaller slice of the expanded table-market pie.
Sure, there are crappy $5 T-shirts, but today I could buy silk and lace enough to embarrass a king. Terribly an artful books exists to come up, but I could still accumulate a library in my pocket that would be the envy of any ancient monastery or place of learning.
> Sure, there are crappy $5 T-shirts, but today I could buy silk and lace enough to embarrass a king.
Actually I think something has happened to the textiles industry whereby demand must have driven a certain band of suppliers out of business, and now try as I could I can't get polo shirts in the same thick quality cotton weave I could 30 years ago. There is probably some niche source possibly online but I don't know how to discover it; the standard "throw money at luxury mall brand" route seems to not work any longer as the brick and mortars have watered down their materials as well. Sic transit gloria mundi
It's a well-documented escalation of planned obsolescence and it's true for everything from your washing machine to your polo shirts to your car. If you make it cheaply so it deteriorates quickly, constantly bring out new styles to make your current thing seem prematurely out of date, and make it juuuust cheap enough, you can sell people 10 shirts over 10 years instead of 2.
I like wearing industrial clothing (like red kap cotton work shirts) and to my eye seem like they're made about the same quality they always were.
There's still money in making music, just not in selling recordings. Biggest touring artists (the Beyonces etc.) bring in millions. They, in turn, require skilled producers to make their songs, who are also paid well.
Might have been interesting to ever find out if this was true.
What happened instead was that Spotify led the pricing change by taking capital, cheating policy, and producing a consumption avenue that cut the price by orders of magnitude.
And meanwhile:
> cheaper tools like DAWs, more accessible music theory education
The gains in education are fractional. The library or a neighborhood piano teacher were good enough resource wise. YouTube eliminates the trip (and the funny thing is that we're iffy on even rewarding those people proportionally), but isn't a new opportunity.
And even for materials that are better in the way that 3Blue1Brown is for math... just like you're going to have to sit down and spend a lot of time actually doing problems rather than just watching the videos if you have any hope of really getting it, the constraint when it comes to producing music is still sitting down and putting in the time, not only on the specific problem/work in front of you but in the background to do it elegantly.
DAWs are great and can make up for some margin of missing virtuosity, but you have to put in the time practicing using them too -- they become their own instrument.
The constraint on making music has always been time. And what gets you more time to do something? Either having another source of wealth, or getting economically rewarded for doing that thing.
Spotify and the damage it's done the market absolutely matters. Just because music is getting through the damage doesn't mean there wasn't some lost, and not just quantity, level that could have been leveraged to through the magic of compounding focus. Anybody who's read Graham's "maker schedule/manager scheduler" should already know this.
What even was the music market though? I think it was a spot where record labels could get rich off of music while musicians still had to pay their way with live shows and merch.
Why do I care if spotify's investors have replace the record label investors?
Even before the digital/internet sea change, this didn't capture the whole music industry. There were artists who made good money off their recordings either because they DIYd it, or found better labels, or negotiated better deals. None of those paths has ever been easy but before recording revenue got kneecapped it was more available.
But let's say for the sake of argument that's how it worked before the internet. If so, why did we let that level of disruption just replace one set of bad guys with another?
What we should have had instead was what we were on track to have: multiple-scale digital recording retail, Apple to Bandcamp to individual artist site to local indie collective as point of sale. Charge whatever artist and buyer choose and can clear a transaction at, they keep 70-90%. True streaming, some with a format, some algorithmically customized for both audio wallpaper and music discovery... but in every case not user-programmed because that's how you justify the difference between the ridiculous fractional cent payouts and recordings.
That's the world we could have tomorrow if the policy was there.
More likely, everybody's too used to a decade of having the privilege of an unlimited basically free recording buffet that Spotify used to cannibalize the industry and we won't do it no matter how it erodes the economics of creation. But we could.
I'm never not astonished by the sweeping pseudo-philosophical bullshit conflating hobby art with literally millions of people's livelihood, visual and other creative culture, etc. There has to be an echo chamber in some subreddit where people who have no idea what they're taking about all slap each other on the back for their ill-conceived musings about the disposition of artists in our society and the nature of art.
Increasing the size of the market is a well understood phenomenon, but it isn't the only way a business can be successful. The music companies would prefer that you spent less time watching TV and more time listening to music. Or, they would prefer that you had less meetings at work so you'd need music to fill the silence. Or, Artist A would like to convince you that they are better than Artist B.
Humanity has always been constrained by the 24 hour a day thing, but the economy has grown nonetheless.
That last little bit is interesting. Back in the physical media era, if Artist B fell out of your rotation, you could sell your record/tape/CD and decrease the size of their new market a little bit. Then we went to DRM, and every song you bought was a sunk cost; if you didn't listen to it, you still payed. Now with streaming, it's back to the downsides of physical media; if you stop listening to Artist B, they stop getting paid.
Did you miss a /s? There is more varied music being created every day than ever before right now. There are sub sub sub genres you can seek out if you want. Contrast this with when I was a kid and we basically had what the radio played or what cassette we could buy with our $10.
The problem now is that we have so much content (music, books, movies, short vids, long vids, etc...), and not enough aggregate time to consume it all.
People being able to afford professional equipment and professional session musicians vs a guy recording himself in a bedroom over a MIDI karaoke track is not the same at all.
Most people are not professional music critics, and most of their consumption is as a backing track to the rest of their life.
You could replace most of this category with a Markov chain bouncing up and down a simple key without most people even thinking about it, and I know because this is exactly how I made music for my shareware video games a decade ago.
> You could replace most of this category with a Markov chain bouncing up and down a simple key without most people even thinking about it
That actually makes Spotify worse, because they could have offered that product instead of using huge sums of capital to reshape the expectation anybody with a device is entitled to listen to any work on demand for free.
I guess the good news is that it wouldn't cause a fuss if someone were to change policy so that you can't pay out buffet streaming like it's digital radio and people ended up having to buy songs or at least do the honest work of piracy if they won't accept the app directing programming. After all, most people are just as happy listening to a Markov chain generate bloops for free.
> Most people are not professional movie critics and enjoy more a hollywood film rather than me recording barbie dolls and making them talk.
Sure, but they'll also watch daily soap operas, and the meme "I showed my favourite film to a loved one, but they paid no attention" predates multi-screening.
> Did your video game sell as much as outcast? A game with a proper music score.
Even in aggregate over all the games: I wish :P
But the real question is: how much of that was the music?
Even now, were I to redo that period of my life (and so no need to caveat markets changing), the music isn't what I'd focus on changing — shareware was already a bad idea though I didn't realise it, MacOS shareware written in Java just as MacOS got its own (ObjC only) app store moreso.
Many don't see a difference. Just amount of coolnes.
You can apply this on professional filmmaking or vlogging. I guess amount of time consumed audiovisual production today is much higher on amateurish production thanks to antisocial networks.
If you are used to fast editing, loud music and cheap filters, you'll get hard time to watch ie Malicks films, listen concertos or go to photography exhibition. No doubled about qulity.
Nowadays most valuable is attention. Cheap stimuli is easier to consume. That's what technology teach us.
Turns out that having skills to get money is exactly the point I was arguing, in response to "there is an infinite supply of (crappy) music, so its value is 0"…
If music was fungible for any other music and a musician's day had an unlimited supply of hours, this might be a reasonable position.
There was already "enough" recorded music decades ago that someone could fill all their living hours with constant listening and not exhaust it. If that's really all there is to it, I'm sure you'll have no problem committing to never listen to anything after around 1970. If you have any hesitation about that then you might start to see serious shortcomings in this conception of supply.
"there are more people able to make and publish music then ever" also papers over nearly everything that matters about the statement. "There are more people" is the defensible part. There's half an argument we have greater access to affordable digital tools for production than ever -- but I'm not even sure it's half. The constraining factor on music composition, performance/recording, production is always time. Even where the tools themselves save time that's a series of converging terms that stops at a limit because to make good music you have to practice using those tools plus others. A lot.
Set up a system which rewards those people in proportion to the audience they find and those people are both equipped and incentivized to spend more of their time into making not only more but making better, because they aren't required to spend their time doing other things.
Set up a system which says "Oh yeah, we shouldn't reward any of this, people should just do it in their spare time" and sure, some will do it in their spare time. But they'll miss out on the compounding effects of focus and its power laws because they're occupied with whatever other stuff policy+markets have been set up to value instead. And their audience and the rest of the world will miss out on their power law peaks.
Which is why I was too generous with my earlier "never listen to anything after around 1970" thought experiment. Really, don't listen to anything except debut releases before 1970. Some of the debuts are really good, of course, and labors of love (or capital-backed love) as you say. But the post-debut work is what's enabled by the economic feedback.
There will, of course, likely often be survivors to bias ourselves to the status quo with. And perhaps that's good enough for some. Hell, maybe we're even rapidly getting to a point where we don't even need most artists at all, we can simply have software trained on all the work of all the artists that have ever recorded produce music for you, and be done with not only the pesky idea of rewarding musicians whose work we appreciate but having a pesky human being involved in direct production in the first place.
I used to purchase mp3s from Amazon, and there was one song that had a glitch in it, like it was a bad rip. I always wondered if they were using pirated copies as well. I just re-downloaded for fun and the glitch is still there.
It wouldn't surprise me if e.g. people at Microsoft ran pirated copies of Office or whatever. Or like Photoshop at Adobe. Getting hold of licenses can be a nightmare, and Microsoft products more so in the past. Nowadays, every Microsoft license seems handled by some enterprise admin account.
I can speak for Adobe- Adobe employees have work accounts with full access to the Adobe suite, an internal web portal to get annual free licenses for their personal accounts, and an internal web store to purchase heavily discounted licenses to gift to friends and family. No one within Adobe is pirating Photoshop.
Anecdotally, I had a friend at Microsoft hook me up with discount Windows OEM licenses for my PC builds and it seemed similarly easy to get licenses.
It was employees' personal MP3 collections that seeded their library, so while that statement is true it is a little disingenuous without further context
There are lots of other examples of this happening too... I believe some of the early nintendo retro releases were emulators running pirated roms
> It was employees' personal MP3 collections that seeded their library, so while that statement is true it is a little disingenuous without further context
If anything, that feels even worse.
> I believe some of the early nintendo retro releases were emulators running pirated roms
If Nintendo has a licence for the game that the ROM was an unlicensed pirate of, while that's weird, it doesn't seem fishy in the same way.
I don't understand how employees contributing their personal collections is somehow worse than company agents trolling torrent sites specifically to stream.
Like, every nerd from the era has an mp3 collection. Mine is literally the only data that I have that's been around since I was 15 and survived multiple HD crashes.
How are you going to get the streaming business up and running without some seed data?
The music industry wasn't just handing out blanket streaming rights until Spotify showed up. All the other services of the era sucked -- their libraries were patchy and music streaming looked very much like TV streaming is today where every show lives under a different provider. It took freaking Apple to convince the music industry to allow single track sales and thats why we were in that state until the mid 2010's when Spotify came in and started to clean house.
The music industry was not looking to break the stranglehold they had on CD sales. Someone had to come in with a 'shoot first ask questions later' attitude to get to where we are today.
Uber and Lyft did what they did for the same reason -- the (oftentimes mafia-backed) taxi cartels had a monopoly on pricing and taxi medallions and the only realistic way to break that was to operate illegally.
I think you will find that the law (and copyright) to be extremely overrated. Copyright, in particular, should not exist in its current form, especially with digital data that is not bound by the laws of physics or physicality, to say nothing of the various entities which have carved out most of the royalties that an artist can make for themselves.
This was not Spotify vs CD sales. By 2005 everybody (even the labels) knew the writing was on the wall and physical media was going to be legacy niche. Digital retail grew fast from there (faster than CDs replacing tapes for a few years) and by 2011 (still limping through the trench of the 2008 crash) the yearly revenue was not only growing it was outdoing physical media and rising. Not just Apple. Amazon, Google, Bandcamp, Bleep, half a dozen others plus artists and indie networks experimenting with direct marketing and retailing. The physical vs digital fight was over at all levels of industry strategy when Spotify was founded and smartphones were a rumor, and music was well on track to a complete digital transition by the time the monster escaped scandinavia.
What Spotify actually did was cannibalize digital retail. Because of course it did: it used massive capitalization AND ignoring pesky laws/policies about actually compensating people creating the work Spotify's service depends on to give that work away to consumers for essentially free, until they'd created that expectation in the market and had enough pull as a channel to buddy up with labels (and the ironic thing is it's not even clear how profitable they can be, maybe leading the charge to the bottom has downsides).
> Copyright, in particular, should not exist in its current form, especially with digital data that is not bound by the laws of physics or physicality
Copyright conventions become more important in the face of falling barriers to reproduction and transmission, not less. That's how they got created in the first place: physical copying got industrialized and could take place with scale and ease that was unheard of before then. And the rationales behind them are still solid today, because they're about the economics of creation, not distribution.
The economics of creation are meaningless if distribution lacks physical scarcity. That is the entire problem with copyright in the digital age -- it is physically impossible to control (i.e. prevent) consumption once the consumer has the data, and for IP like music this is impossible to control. Sure, we can attach DRM and technological controls to our work, we can form massive databases of royalties distribution, we can falsely moralize until our faces turn blue, we can even try to craft laws and legislation to get what we ant. But all these things are meaningless in the digital land of infinite plenty to a determined enough adversary, and some music fans are adamant (yet also kind enough to share with others)
If copyright (and DRM) was enforced the way some wanted, we wouldn't have a rich history of remixes, recontextualizations, or bootlegs that allow a work to live on far past its shelf-life. Memes would be a shallow husk of what they are today. We wouldn't have many, many genres of music, and things like the Amen break wouldn't exist and we'd all be worse off for it.
For that reason copyright needs to be abolished. There are other ways of monetizing intellectual property. Once a given work is released to the public, its creators have literally zero control over it, despite the technological artifices we construct otherwise.
> The economics of creation are meaningless if distribution lacks physical scarcity.
As if nobody has to actually create the first instance of a given work, it's just copies all the way down.
That's why the economics of distribution/reproduction are distinct from the economics of creation. Maybe it's clearer if we use the term invention. The major input is time, time to develop the faculties needed to invent/create -- often years if not decades -- and then the time needed to put into a specific work.
The legal claims/controls on distribution give inventors leverage they can use to get better returns on that time, which provide better incentives.
> Copyright (and DRM) was enforced the way some wanted, we wouldn't have a rich history of [blah blah blah]
Copyright was enforced for literally centuries and for much of that period we got a rich history of works which borrowed in a dozen ways from other works, because actual copyright law has both boundaries and blessed borrowing.
I swear, so many tech folks got half a narrative in their heads about draconian DRM as digital gulags and lost their minds to a manichaean all-or-nothing view on the topic.
Yes yes the evil suits from 1999 probably wanted to super-glue your 1/8" output and lock you out of control of your machine. Again, almost 20 years ago the basic truce on that battle was defined. There's always going to be some activity that can't be controlled, but that doesn't mean you can't define and encourage legitimate activity, so we keep the basic bargain because giving inventors/creators a say in how/where/pricing for their work is both helpful and decent, but we don't try suing individuals over their uh "freely-sourced" media collection or install policing malware.
This isn't digital hitler on every last device vs total free-for-all. It's saying Amazon can't sell your ebook or music without compensating you. And maybe even that Spotify doesn't have the right to give away tracks for free or a pittance, that buffet streaming is close enough to ownership that it calls for artist payouts that are closer to the scale of retail than broadcast. You don't need total control to get there, and we have the levels of control to make this happen.
> Memes would be a shallow husk of what they are today.
Oh no not the memes, known for their fullness and depth.
But also no, not the memes, since it'd be vanishingly unusual that any of them would end up in court let alone to be found to violate copyright law.
> There are other ways of monetizing
It's a well-known fact about creative paths that they're not littered with the luxury of money-making opportunities. There are a few ways of doing it, but they're legs of a stool, and the number of inventors/creators who have the stool up to even 3 legs is not large. Suggesting they give just one little leg knocks over a lot of stools, and for what?
> Suggesting they give just one little leg knocks over a lot of stools, and for what?
A great filter, for one, that eliminates those that do art for money in favor of those that do art for the sake of doing art. Personally I don't want output from the money-motivated. I want art for art's sake. The made-for-money stuff is bland and lowest-common-denominator. It is essentially trash.
First, the idea that art is inundated with people who are just in it for the money isn't just wrong it's funny. Who is this crowd of cold dollar-driven people who pass over high-value careers like business, finance, medicine, tech, etc and say "yeah, being a musician is my gravy train, even though I don't give a damn about it?" Like, insert Drake-in-Orange-Coat meme here, right? Even with the rare outlier successes (like Drake), everybody knows the arts are a lottery ticket. Nobody is doing it just for the money. Especially music.
Second, it's pretty iffy that only low-quality succeeds economically. Sure, everyone can think of examples that somehow succeed with limited merit, but you can't sustain the thesis that well-rewarded work is mediocre without ignoring a lot of strong yet widely appreciated and profitable material.
But even those two big points are minor compared to the most important one:
Everyone needs money. Even artists who do what they do for love. It's the legs of their stools you're suggesting "filtering"/kicking out too.
When someone can't earn money doing what they love, they have to spend time doing other things in order to get the money. And that's time they're not creating art and time they are not refining their craft.
What you're "filtering" out is the peak of the skill they could have developed with more time as well as the art they could have created with it. Maybe even the attention and focus they have to doing it at all, hijacked by all the ways necessity can preclude even love.
I'm probably more on your side that you think WRT the law and copyright law but let's not pretend Spotify, Uber, and Lyft are doing some social good. They're greed-driven corporations (a bit redundant, but it's especially true for them). They may have partially broken up the walls around their particular industries, but not to democratize them, just to take ownership of whatever they can grab.
And in doing so, each has failed to realize better solutions for people. Uber & Lyft have increased our dependence on wasteful, dangerous, expensive personal vehicles. Spotify is a worse model than Bandcamp for indie artists. It's also a worse model than pretty much every other streaming service in that it pays so little to artists. So, yeah I don't like it when rich people break the law to get more rich at society's expense. We should be breaking the law to make society better.
Ergo, piracy and buying music on bandcamp. Still very viable options. I am close to publishing an EP on bandcamp myself. But I cannot deny how handy Spotify has been with musical pursuits. Spotify may rule the roost but we still have Bandcamp, Beatport, JunoDownload, Amazon Music, iTunes, Discogs marketplace, and others.
And I would say Spotify is doing a social good. As a nobody artist, being able to point people to stuff on Spotify is really compelling. Any other way I'd have to force people into either buying or using an app they are unfamiliar with just to listen to it. That said I am not trying to make money from music, that is just foolish.
> And I would say Spotify is doing a social good. As a nobody artist, being able to point people to stuff on Spotify is really compelling. Any other way I'd have to force people into either buying or using an app they are unfamiliar with just to listen to it.
Well, I understand where you're coming from here, but that's not really Spotify doing a social good, it's just the inevitable effect of their cultural dominance. Everybody[1] has Spotify. But they've taken that position due to their unethical growth plan.
Anyway, I've realized that we're discussing multiple things, not one thing. Spotify can be bad for artists who want to make money off their art, while being good for those who don't want/need to make money from music.
Everyone overlooks the fact that it will still take someone (i.e. a graphic artist) to produce great AI imagery.
First, AI generated art is random and disposable. Yes, you'll get a great image that you can use once, but then what? You can't build a campaign on it.
Second, AI generated art can't be copyrighted, so knockoff competitors are free to use your AI-generated marketing images.
At the very least, you can seed the AI with a paid graphic artist's work (seed-based AI images can be copyrighted). But that artist will do it better than your unpaid intern.
Mmm, I don't know about this. At the very least AI lowers the bar for how talented a graphic artist needs to be to produce professional work, which means it'll be easier to undercut them, which means it'll get much harder to make a living as a graphic designer. It amounts to the same thing as killing off the profession, as seen from the perspective of someone in the profession as opposed to someone without skin in the game. It's like saying push-button elevators didn't hurt the profession of elevator operator, because somebody's still got to push those buttons.
I think AI in general, across almost every industry, will shift value away from technical proficiency and toward creativity and taste. Implementation of an idea/vision will be commoditized, but having a great idea, a unique insight, the taste and ability to identify top-tier work will still be highly valuable. This could well remain true post-AGI.
In graphic arts, the overlap between people with technical proficiency and vision/taste is probably quite high, but it's not one-to-one. There are people with excellent taste who can identify great art or design when they see it, and who can perhaps imagine incredible masterpieces in their minds, but cannot draw a convincing stick figure. On the other side, there are people who can expertly make someone else's concept real, but can't come up with a compelling concept themselves. AI will be great for the former, and bad for the latter (or at least force the latter to adapt).
Whether this will have the effect of concentrating wealth or distributing it more widely strikes me as a very difficult question. It may be devastating for certain professions, but could also enable a whole new class of entrepreneurs. I could see it going either way, or the two effects may cancel each other out and economic equality stays about where it is. We're in the realm of complex systems here, so I wouldn't put much stock in anyone's prediction.
> I think AI in general, across almost every industry, will shift value away from technical proficiency and toward creativity and taste.
The problem is that an artist still needs to eat in the 10-20 years it takes to develop "creativity and taste".
What AI will do/is doing is knock out the entry-level jobs. If you can't train humans on the entry-level, you will eventually have no experienced people.
It also raises the bar of what's possible. What counts as "professional level" changes each time some new technique emerges. A skilled artist will always be better than a random person.
The visual entertainment "supply" is not limited by the current state of tools. It's always limited by the skills of the top crop. Professionals are always ahead and hard to come by. The industry's self-regulating mechanism is novelty; what is abundant becomes fundamentally uninteresting and dies.
This is the march of progress. Digital brushes in Procreate lowered the bar for how talented an artist needs to be to create an oil ‘painting’. The camera lowered the bar for creating portraits.
> AI lowers the bar for how talented a graphic artist needs to be to produce professional work
I think it's a different kind of talent, and not automatically a lower bar. The key to being a professional artist is being able to offer variants based on given direction. Either way, it's much much more than pushing a button or holding a lever in place for a period of time.
> Second, AI generated art can't be copyrighted, so knockoff competitors are free to use your AI-generated marketing images.
No. First off trademarks exist and they found that work done solely by the machine couldn't be treated as a work for hire copyrighted by the machine and assigned to the operator. There is no reason to believe that work couldn't be treated directly as copyrighted by the human operator who has creative input nor is the matter with the images used to train the model truly settled.
>First, AI generated art is random and disposable. Yes, you'll get a great image that you can use once, but then what? You can't build a campaign on it.
You can already get variations on a them and text driven modification eg make the blank a blank or make the blank blanker.
There is no reason to believe that work couldn't be treated directly as copyrighted by the human operator who has creative input nor is the matter with the images used to train the model truly settled.
...Other than the USPTO and the federal court system issuing multiple ruling stating the opposite, including a decision last week which specifically stated that the output of an AI model is not copyrightable, upholding an earlier decision by the USPTO... (https://www.hollywoodreporter.com/business/business-news/ai-...)
Except for the part where the court didn't find that. It found that work only created by the AI didn't qualify. Had it asked if a work created by the AI AND the person qualified it would no doubt have qualified as is already clear from using photoshop not serving to remove your ability to produce copyrightable works. The case didn't ask that and therefore it wasn't answered in any meaningful fashion.
The act of prompting and customizing iteratively especially in systems which allow the user to submit a prompt that modifies the existing work for example "replace the human being with a monkey" "make the monkey pink" etc are clearly creative works that USE an AI not uncopyrightable.
If you want to argue that point you absolutely cannot do so on the basis of a case that literally never addressed that issue unless you would like to traverse the muddy ground between actuality and fiction.
The ruling stated that the Constitutional justification for copyright (and other IP) laws was to incentivize creators. AI does not need incentives, and thus AI-generated content cannot qualify for copyright. Under this line of reasoning, neither can patents (though note that trademarks derive value from the resources and effort spent promoting them, not from their creation, so trademarks are unaffected).
The act of prompting and customizing iteratively especially in systems which allow the user to submit a prompt ...are clearly creative works that USE an AI not uncopyrightable.
If you want to argue that point you absolutely cannot do so on the basis of a case that literally never addressed that issue unless you would like to traverse the muddy ground between actuality and fiction.
The case literally deals with the output of the AI model, not the input. But on that note...under existing law, code can be copyrighted but not its output. Thus, it is logical to reason that prompts to an AI model can also be copyrighted to the extent they are not strictly functional.
But with AI models and content generally, nobody cares about the prompts/inputs. The output is what matters. (For comparison: Deep Impact and Armaggedon were both the results of the same input: disaster movie in which a team of astronaughts has to go to the asteroid to blow it up before it destroys Earth. The "models" were different screenwriters and directors. Compare the outputs: one is a blockbuster classic, and most people don't remember the other movie.)
There was a statement on the prior thread that described the situation particularly well I'll reproduce it herein and link the original comment rather than trying to better it.
> The headline doesn’t seem to be what actually happened. The filer was arguing that the ai created the work on its own as a work for hire and thus the ai was the author with the computer scientist merely being the owner of the copyright as it was made for hire. I don’t think the argument that ai is a tool and the human operating it is the author was considered because the filer explicitly didn’t want to consider it.
> It makes it clear that the computer scientist doing the filing was trying to argue this was a work made for hire with the author being the computer. They wanted to argue that copyright can be assigned to non humans, but that just isn’t how the law works. The summary makes it clear early that it’s just taking their word that the work had no human input and was thus purely the creation of the computer.
In short
A: Computer generated efforts virtually certainly qualify for copyright.
B: Non-artists can in fact iterate and modify work not just randomly generate shit.
C: This will virtually certainly get much better over time.
You will still get much better work out of a professional who can both utilize such tools when desired, and actually create not just copy or prompt art. This thesis is supportable but we shouldn't build it on sand lest it look more vulnerable than it is.
Random variations aren't interesting, they just make something abundant even more abundant and secondary. Unless you have a model with sufficient intelligence that can create something conceptually original (at which point we're all fucked, not just artists or programmers), it's not going to fly. Text driven modifications imply conceptual human input; besides, they are inherently worse than higher-order input, just like text to image alone is worthless for anything meaningful.
There exist systems where you can describe not only initial scenes but successive textual modifications to existing images and furthermore variations aren't random. Successive selections are a way to zero in on a concept.
AKA tell me you haven't spent time with diffusion models, without telling it :)
I actually did figure out what works and what doesn't in real artistic use. Which is the entire point of the article in OP which nobody seem to have read - text doesn't work well beyond the basic use or amateur play, regardless of it being the initial prompt or editing; you need sketching and references (and actual skill) to do real work. I don't think anybody's using available methods of textual modifications for anything complex - they are cumbersome and unreliable, even worse than textual prompts. In fact, I haven't seen anyone using them at all.
Besides the implementation details, natural language just doesn't have enough semantic density and precision to give artistic directions, even for a human or AGI. That's a fundamental limitation. Higher order guidance, style transfer, and compositing is how it's done.
> Yes, you'll get a great image that you can use once, but then what? You can't build a campaign on it.
Checkout confyui, it has an incredible amount of composability that allows you to generate new images based on others. Like image to image but on steroids.
For example, you can generate a character sheet and use it to generate the same characters on different poses using controlnet. Or you can have a base image for an object and use that to generate the same object from different angles and/or different colours etc.
I agree with you, but the main problem is that illustrators are under-appreciated. We are in a world where management with no technical knowledge are having too much power and stealing paychecks.
Also people cannot judge great art or imagery. Unless you have had the training. But the average person? Nope. You can tell what you LIKE but that’s not the same.
I don’t have much training but it is not that difficult to spot AI arts which is pretty repetitive. The first couple are awesome but it gets old really fast.
AI generated art may be disposable but it certainly is very, very good. Midjourney makes plenty of impeccable art and photorealistic images that have no flaws. Also, even if there are flaws a week with some YouTube videos can teach anyone how to fix them, you don’t need someone with five years of deep experience.
Recorded music was going this way whether it was Spotify or someone else that drove the final nail into the coffin.
I remember when I was a child, on a Sunday afternoon, my dad would put on an album and listen to it. Just listen. Very, very few people do that now.
Now we have a lot of demand for “incidental music”. Something you listen to while you do something else. Driving, reading, surfing the net, coding, cleaning…
There was a fundamental shift in how people consumed music that started around the time music became portable. Spotify won the race, but if it hadn’t been Spotify, it would have been someone else.
I'm curious. When do you think music became portable?
The transistor radio was invented in the 1950s. And quickly became used as background music as life progressed.
Also incidental music is not a new thing. Tavern musicians as background music have been around for centuries. It is hard to prove, but likely for thousands of years.
No no. Incidental music and "music you listen to while doing something" are not the same.
Listening to incidental music all the time devalues music. And we do it not because we wanted it but because Spotify, Apple music etc promote it. Until then "just play random stuff that this ML thinks is similar" was not a thing. But subscriptions make them more money than if they just let you buy albums and stream what you bought. I wish more artists didn't sign up for this but unfortunately big labels did.
But you can listen to non incidental music that you have specifically chosen while doing something. Even your dad could be doing something while listening to music (thinking).
A positive effect for performers is that people still want to go to concerts, but less and less people know how to play an instrument. The market is really much better now than even 10 years ago.
> Spotify basically killed any money coming from the physical distribution - Worse than piracy
Any sources for this?
I'm of the impression physical distribution is on the rise compared to the earlier days of digital music. This has nothing to do with Spotify, and all about the digitization of music itself.
Anecdotally many people I know now purchase merchandise and media as a way to support an artist they like, rather than listen to the music they make in a physical format.
This is factually incorrect. If you own your master recordings, you stand to make $3,500 to $5,500 per million streams on Spotify. Apple Music and Tidal pay even better. This is why Taylor Swift is re-recording her entire Big Machine Records catalogue. While Spotify did shift consumers away from buying singles and albums as individual items, they also opened a new revenue source for independent artists.
Can you point me to a current-day independent artist which hasn't been signed to a label that is pulling this amount of money just on streaming?
If you're already big enough that, i.e., XL Recordings can ask you to make a record without getting rights on the master, I wouldn't count it as a good example of "indie artist".
I make about $4,000 per million streams on Spotify for the tracks I’ve released independently. For label releases I make less, but the label promotes them so that sometimes results in more net revenue. I have a bit over 10M Spotify streams over the last 3 years.
Also, Spotify promotes my music via editorial playlists and algorithmic (eg Radio or Discover Weekly), so I’m probably making a lot more total revenue than I would have on iTunes.
are you sure he’s doing 5k/month just by streaming? No syncs nor shows? Also if Wikipedia is right he’s signed with Columbia Records. AFAIK the only artists making that kind money just by streaming while having no strings attached EVER (No label distro, no label A&R, no big tent agencies) are Macklemore and Chance The Rapper. Just two guys over millions of artists on Spotify.
I can’t find the article from before he signed with Columbia (might’ve been a YouTube interview with him, can’t remember for sure), but yes, I’m fairly certain he was doing well over 5k per month with no major label.
Also note the terms of his deal with Columbia are unlike most major deals in that he has a 50/50 profit split after his advance payment got recouped, retains either full control or 50/50 control of masters, etc.
Here you go, he mentions it in the first 30 seconds of this video. He says roughly $100k per month before any label involvement: https://youtu.be/OebNTkTfzHU
> "the AI did this picture, so we don't have to pay you."
If the court rulings hold and AI works cannot be copyrighted then us end users do not have to pay for it either... but that seems like a race to the bottom. Like the end of a craft. Why would anyone create art if it has no/minimal downstream value?
Artists need to band together in some sort of union or not agree to do art with that AI clause or perhaps only do art with a no-AI use clause. And have an allowed AI-clause that is prohibitively expensive (like in the multi hundred millions per piece). That way 'accidents that happen' have a prescribed recovery amount plus other requirements like pulling the generated artwork. "Hey, we understand it may have been accidental, but here is the bill."
It's not the end of a craft, it just means that the prestige of "made by human" will increase even more and be pushed by by companies as a means of making money through copyright. That means that the few artists at the top will be rich while the niche between "art" and "craft" disappears. Professions involving visual art become like the music business.
I honestly wonder if people would consume the music they know is AI generated. And by "honestly", I mean "I don't really know but I want to."
I've been watching videos of Guy Michelmore in youtube. Not because I will ever write any orchestral music, but because I like his energy and envy his shed. Would I bother if Guy Michelmore were an AI?
It could also have some interesting avenues, like feeding some variables to the AI from say a video game (number and type of monsters on screen, mood etc.) to generate music reacting to what is happening on screen
It depends on how you define art. You can play music or shoot film or paint a picture. AI could do it as well. But the essence of what makes good art comes from soul, from experience by living, from relationships between us... that is created for stories that inspire.
I would. Out of the bands I listen to maybe 5 of them I could name a single member. I'm a big reader but I couldn't tell you one thing about most of the authors other than their names.
And absent some major technology changes Spotify in your example has no way to do credit assignment back to the training set for any attempts at royalties should they be so inclined.
Without copyright protection anyone could copy their entire library and set up a rival streaming service. It certainly wouldn’t be worth much investing in the AI part of the business.
Also, if you're wondering "Well, I could get better terms for my art" - Like I said, when Spotify arrived and you were signed on a label you HAD to sign the part that said "Yes, you can put my music online on Spotify and I will get paid peanuts" or else, unless you were Madonna or Taylor Swift.
Or, sure, you can also terminate your record deal. Hope you have 500 grands around just for that.
Frankly I don't see it ending much better for visual artists.
There was another path here - collective bargaining. When small individuals are bullied by large corporations it's because those corporations want something from the small individuals... they certainly don't care about one or two small artists walking away from the platform - but if artists can organize and bargain as a group they can ensure a fair outcome.
I think the modern world has become too complacent in terms of labor organization - the time of plenty left a lot of people content to take whatever was given to them because there was such a glut of excess that it was freely shared. That sharing is coming to an end and we're returning to a time when we need to demand fair and equitable treatment.
Freelancers in the united states are not allowed to bargain collectively for better prices, as that's considered market manipulation/price fixing. ["Independent Contractors" are literally banned from forming a union in the USA.]
Whenever there are implications to people's lively hood, its always a serious matter - but I hope people are able to transition to other roles.
I think Gen AI will commoditize the mundane and "typical", and heavily push people into creating something extraordinarily unique. I think there is the same pressure even without AI, when as a creator you have to standout amongst the sea of people vying for people's attention.
I believe GenAI can be useful in a way too. For e.g. If I'm an artist looking for inspiration, I can have a GenAI tool create some "random" works that I can get inspired from.
The music industry has always had a long tail. Its very much a go big or go home industry. Do you have any data around revenue change for small artists before and after spotify?
Sorry, no hard data. Mostly the perception of the industry at the time. Lots of tales of people quitting, moving, or going on "indefinite hiatus".
TBH "Fly or Die" was way more common on the US side of the industry. And even in the USA by the late '90s to the end of the 2010 it was somewhat doable if you were skilled enough to make a living solo (we're talking 60-80K/year max) as a "jobber" opening for bigger acts on local venues.
Like, the entire NYC indie scene got a start from this premise. If you get a chance, give a look to "Meet Me in the Bathroom"[1], which is a documentary specifically of this timeframe.
The Spotify example is similar to the Google impact: the last mile is the search engine UI that controls your access to content. Spotify is another UI as they are streaming services, etc.
Seems like a natural iteration in the ordering of complex systems. Beyond legal regulations it would be great to start to think about new solutions, if they ever exist.
> “Spotify killed revenue flow for anyone outside of the publishers and big artists/ players. Spotify basically killed any money coming from the physical distribution”
To the contrary, physical distribution was only available to big artists/players. There was mostly no way for independent artists to get into the record stores.
Streaming (via distributors like DistroKid) made it possible for millions of independent artists to make money from their music.
AI will have on artists the same impact that Spotify had on the music industry that is, it will kill any revenue flow for anyone outside of the publishers and big artists/players.
then resell the output for $$$ on something like Splice
This is silly. The USPTO and Courts have repeatedly stated that AI-generated media is not subject to copyright protection, so there are no licensing revenue opportunities for the big publishers/artists/whatever. This means: AI-generated content is not protected by copyright, so anyone can use a piece of AI-generated art however they want without a license and unless the law changes AI has no value to the content industries.
EDIT: Also, the USPTO has noted that the use of AI-generated content in a work will mean that the entire work will be presumed AI-generated except for the portions the content owner can demonstrate were generated by humans. The backend costs of maintaining AI-supplemented works will almost as expensive and burdensome as the costs associated with patents.
Also, I think people on HN have a very glorified view of how much money musicians make from streaming or cd/album sales: basically zilch, unless they're popular enough to be in repeat on the radio. Most musicians made their money from performing: generally a little bit from ticket sales or venue incentives (like % of booze sales) but the real money for the performers was from the sales of band merch, which is why it gets pushed so heavily.
At the end of the day the model itself is going to be basically irrelevant, while knowing whose works were actually used to train it being the truly differentiating feature.
Yes, by lawyers, when they sue the owners of the AI model for copyright infringement, because this would not be a use protected by fair use doctrine. This will actually make human-generated works more valuable because now every work used to generate an AI work is now worth at least $75,000, even if its market value would be significantly less (or even commercially worthless) today.
Due to the costs associated with licensing of human works, if AI-content becomes a thing, it will probably be more expensive than hiring a human to do the same thing, because the model will have to account for the cost of paying a license fee for every work that was incorporated into a specific output.
Spotify has been a disaster, but unless the artists walk away (very hard to do), I don't see our political system as caring enough to do anything about it.
> Another opinion popular with no one: AI will have on artists the same impact that Spotify had on the music industry that is, it will kill any revenue flow for anyone outside of the publishers and big artists/players.
Maybe I’m misunderstanding you, but how much money do you think the 7500 creators on Spotify making $100k+ [1] would be making without Spotify or other streaming platforms? My guess is closer to zero than 100k.
Also 0.09 percent of 8 million creators making 100k+ [1] sounds horrible, but my guess is that should be taken with a grain of salt. How many folks are included in that 8 million who registered, but uploaded nothing? How many uploaded once or twice? How many uploaded and did ZERO promo of themselves? How many are just plain terrible musicians?
A number of years ago when I stumbled on him, Russ was pulling in a few hundred thousand per year from streaming. Looks like he’s making 100k per week as of a couple of years ago [2]. Yes, he’s probably an outlier. But he works his butt off on his craft, handles production and writing himself, and markets himself well.
Headlines like “Big tech and AI destroying the indie music industry” get more clicks and attention than “Streaming platforms provide income where once there was none” so shrug.
Found the video where Russ says he was making around $100k per month before any label involvement: https://youtu.be/OebNTkTfzHU. I know this is the land of “that’s just survivorship bias!” And I certainly agree that luck and timing plays a massive role in billionaire level startup success, but this guy in particular is a few orders of magnitude of success below that (even if he’s still an outlier). I’m sure he still benefited from luck and timing, but he also was methodical about creating music non-stop, getting better at production, rapping, and writing, and marketing himself. My point being show me someone who has worked as hard and as smart as he has, who picked a niche of music that has large audiences (aka high Total Addressable Market), and who released as much content as him, and I will show you someone who is having non-trivial streaming success - again maybe not $1M+ annually - but something material beyond just scraping by. That doesn’t mean Big Tech is absolved of sin in how it distributes profits or exerts monopoly control or whatever, but I think we often overlook the opportunities these networks have provided for people that would otherwise live in obscurity with no audience whatsoever.
One thing that will really matter is that the output of AI cannot be copyrighted. If producers really go all-in on generation we're going to rapidly see a situation where huge amounts of material will enter the public domain all at once, and we don't really have a precedent for what happens then.
A human looking at someone’s artwork is “training a model”. It’s bullshit and anti progressive to say someone or something that is creating derivative works is stealing.
The second half of derivative worlds is creating an imitation of the original not just looking at it, but this isn’t some grey area.
Even just training the model requires someone to copy the original work from somewhere and store it into a database to use to train the model. If they don’t have permission to make that copy then it’s commercial copyright infringement independent of anything done by the model after that point.
Thus the companies themselves are frequently breaking the sale even if nobody ever uses these systems.
not stealing the work, just stealing the revenue...for very little investment.
> A human looking at someone’s artwork is “training a model
sure, except that model often takes months or years to train (wall clock years, not 1000-core cpu-years). and the end result is not a human that can stamp out new/competing artwork every 100ms.
for any kind of creative/performance/art work, these are watershed times. us coders are not super far behind.
Posts like that nearly always assume the text-to-image and "prompt engineering" being used, usually due to the lack of experience with those models. This is categorically not the way to do it outside of having fun. The way it's done for predictability and control looks much more like "draw the rest of the owl, in a manner similar to my other hand-drawn owl" combined with photobashing and manual fixing/compositing. It's a hybrid area similar to 3D CGI that requires both artistic and technical skills if you want to create something non-boring.
This has nothing to do with the model's poor understanding of natural language, and will not change until we have something that could reasonably pass for AGI, and likely not even then. Your text prompts simply don't have enough semantic capacity.
You might be interested in the "Commercial illustrators will keep their jobs, but will mostly need to learn to use AI as a part of their workflow to maintain a higher pace of work" section of the article, which gets into this more.
The more plausible evolution is that people drawing the base concept are not "commercial illustrators" nor have art training.
If a magazine editor with run of the mill drawing skill can feed the prompt a sketch with stick figures and object outlines, and get back a good enough rendition with an improved composition, the job of the illustrator will be a side job of that editor.
I'm partial to the argument that being able to fix the generated image in post is a valuable skill, but on that part we already have decades of progress and people are usually more comfortable with editing tools than drawing tools.
"This has nothing to do with the model's poor understanding of natural language, and will not change until we have something that could reasonably pass for AGI, and likely not even then. Your text prompts simply don't have enough semantic capacity."
I don't think it's going to take AGI to get to this point. It's 'just' going to take a top-tier model adding robust multi-modal input imho. A detailed prompt plus a bunch of examples of the style you're looking for seems like it would be enough.
That's not to say it isn't really hard, but it doesn't seem like it requires fundamental innovations to do this. The building blocks that are needed already exist.
The biggest problem I see with LLM-generated imagery is a near total inability to get details right, which makes perfect sense when one considers how they work.
LLMs pick out patterns in the data they're trained on and then regurgitates them. This works great for broad strokes, because those have relatively little variance between training pieces and have distinct visual signatures that act as anchors.
Details on the other hand differ dramatically between pieces and have no such consistent visual anchor. Take limbs for example, which are notoriously problematic for LLMs: there are so many different ways that arms, legs, and especially hands and fingers can look between their innumerable possible articulations, positions relative to the rest of the body, clothing, objects obscuring them, etc etc and the LLM, not actually understanding the subject matter, is predictably terrible at drawing the connections between all of these disparate states and struggles to draw them without human guidance.
You see this effect in other fine details, too. Jewelry, chain-link fences, fishing nets, chainmail, lace, etc are all near-guaranteed disasters for these things.
It's mostly a problem of resolution, model size, and dataset quality, which can be mitigated with compositing. Larger models don't have problems with hands, and if they do, it can be solved by higher-order guidance (e.g. controlnets) and doing multiple supersampled passes on regions to avoid to fit too much detail in one generation. Even SD 1.5 (a notoriously tiny model) issues with faces and hands can be solved with multiple passes, which is what everyone does.
There are two problems with this: a) natural language is inherently poor at giving artistic directions compared to higher-order ways like sketching and references, even if you got a human on the other end of the wire, and b) to create something conceptually appealing/novel, the model has to have much better conceptualizing ability than is currently possible with the best LLMs, and those already need some mighty hardware to run. Besides, tweaking the prompt will probably never be stable, partly due to the reasons outlined in the OP; although you could optimize for that, I guess.
That said, better understanding is always welcome. DeepFloyd IF tried to pair a full-fledged transformer with a diffusion part (albeit with only 11B parameters). It improved the understanding of complex prompts like "koi fish doing a handstand on a skateboard", but also pushed the hardware requirements way up, and haven't solved the fundamental issues above.
I think you're right about the current limitations, but imagine a trillion or ten trillion parameter model trained and RLHF'd for this specific use case. It may take a year or two, but I see no reason to think it isn't coming.
Yes, hardware requirements will be steep, but it will still be cheap compared to equivalent human illustrators. And compute costs will go down in the long run.
> Your text prompts simply don't have enough semantic capacity.
Mostly, current tools are abysmal at maxing the semantic capacity. Midjourney is great a generating things that look good, but terrible at piecing scenes together.
Recent example I tried: a robot playing magic the gathering seated across a human.
Even getting the human in the picture is a challenge, but then the model doesn't know enough about MTG (it correctly pattern matches to "board game" or "card game").
Some pictures generated are much better than other, and it would be great to take e.g. the table setup from one picture and the robot from another, but doing is not really possible atm ("blend" doesn't work for that).
I have no doubt this will improve, but I'm wondering if there's something underlying this that could be a more general limitation? Maybe simply not enough data (a google image search for magic the gathering is also pretty disapppointing).
Another example: a glowing blue <company logo> carved into a stone monolith. It sometimes got the logo to be carved (rarely), it was never glowing blue (usually the whole monolith, or a part of it).
Sorry I noticed your post too late, not sure you'll read it.
The reason it happens is that the models are far too small (parameter count-wise) and the prompt understanding part is simple, usually it's either CLIP or in the best case a small and dumb transformer. (But regardless of the current capabilities, text is just not a great tool to express artistic intent)
Generally, what you want can be done by giving the model higher-order hints like sketches and pose skeletons; see controlnets for Stable Diffusion for example. The overall idea here is to use a custom model created specifically to guide the diffusion model, based on the non-textual input. The problem is that MidJourney can't do this, you have to use SD.
Another thing is photobashing/compositing. Avoid fitting the entire composition into one generation, it will make the model lose track of your scene. Using multiple passes helps a lot. It's best to inpaint the objects or img2img them based on non-textual guidance to add objects and details in the specific spot.
Well put.
Big fan of the "Commercial illustrators will keep their jobs, but will mostly need to learn to use AI as a part of their workflow to maintain a higher pace of work" part.
I'm a sometimes-illustrator (but my style is pretty far from what Generative AI is doing), and I recently published a 1.1 of a game manual which uses Midjourney images. I'm currently investing in a "proper" illustrator because the MDJ images lack character, but it's also true that in a few months from now this might change: I'll stick with the illustrator to have more consistency in the images, but probably the AI could do a fancier job there.
Besides, the "things will change in 2 months" point is a good one, but it's been used since a year and a half and things haven't changed yet. Sure, the quality of the produced images improved, but not in a qualitative scale.
> I'm a sometimes-illustrator (but my style is pretty far from what Generative AI is doing)
Why not train your own personal AI on your artwork? Corridor Digital did this in the latest attempt to automatise animation, they hired an illustrator to create an animation style for them, then trained the AI on their drawings.
I've actually done it [0], I'd like to have an AI assistant that I could directly use the results from, and the results were really terrible, mostly laughably terrible. I think it was too far from what the models handled correctly at the time, and given that issue it was not enough training images. Although I had also tried with a model that was better at handling stylised 2D. I'd like it to work, but I don't think it's viable for most people.
Ethics of the use of generative AI in the first place aside, I'm pretty sure the illustrator was aware of what they were intending to do with their work (they even were interviewed about it in the behind the scenes video)
I view this in the same way I view the use of an actor's voice for ai generations. Even if the person knows what you're doing with their data, it still feels really scummy and unethical. The idea that we can sample someone else's labor and be able to own that and generate shit from it in perpetuity (probably without paying them) feels very alienating.
This could have been all with consent and adjusted payments. AI does not just replace an artist, it can also speed up the work tremendously. It gives new possibilities using volume.
I'm not in illustration, but isn't it already common to hire someone to create a "style book" of what it should look like, and then have other illustrators follow that? eg, I recall animated shows working that way.
That's an interesting take!
Currently I see two reasons why I wouldn't do that:
1 - Since I'm either working for game companies or for my own project (https://fsd-wargame.com/) using AI-generated things is kinda damaging in terms of marketing. You never know when some uproar could arise against a project/game solely based on more or less petty outcries against AI. I generally sympathize with artists, but sometimes it's just whiny.
2 - My illustrations are line-art and cartography (https://www.artstation.com/thelazyone) , which are not the easiest to handle with AI. I'm sure that with enough effort there's gonna be a good model, but I haven't seen any so far.
The question is, since commercial illustrators can be more efficient using AI, will the total number of jobs in the space lower, or will the expectation for commercial illustration increase, thus increasing the workload and keeping the number of jobs the same.
In all of human history, work has always increased. This is akin to Parkinson's Law, where work expands to fill the time (and now resources) available.
I don't disagree, but concerning particular trades this is not true. In the mid-19th century there were more than seven thousand blacksmith shops in the US, which employed over fifteen thousand people, but today there are fewer than one thousand professional blacksmiths. Many of the products they produced either have lower demand or are produced by other means. If you consider the entire metalworking industry, we have many more total workers, but very few have the skills of a blacksmith.
The number of people who do the current work of an illustrator might go down eventually due to AI, but there will likely be more total people employed in the process of producing illustrations. It is just likely that fewer of them will have the skills that today's illustrators need, and also likely that fewer of them will command extraordinary wages. Many of the jobs that replace it will likely be closer to the median wage than today.
Also we will eventually turn the corner and start having population decline. For the US this might be just a few decades away. And some time after that, work would eventually decrease.
Work has always increased, but work in a specific profession doesn't necessarily increase. There are certainly fewer phone switchboard operators today than there were 100 years ago.
Indeed, but that just means that humans will have to find new jobs, not that jobs will become obsolete. How well they will find new jobs, though, is another story, based on socio-politico-economic conditions of the country they reside in.
In most of human history, the type of jobs available were relatively stable century to century; today, the types of jobs aren't even stable decade to decade.
The automation of physical labor let us turn to intellectual labor and creative labor. The coming automation of intellectual and creative labor is not like the previous automations of physical labor, because it leaves human jobs no where else to turn to.
CGP Grey's "Humans Need Not Apply" video[1,2] covered this almost a decade ago:
> Imagine a pair of horses in the early 1900s talking about technology. One worries all these new mechanical muscles will make horses unnecessary.
> The other reminds him that everything so far has made their lives easier -- remember all that farm work? Remember running coast-to-coast delivering mail? Remember riding into battle? All terrible. These city jobs are pretty cushy -- and with so many humans in the cities there are more jobs for horses than ever.
> Even if this car thingy takes off you might say, there will be new jobs for horses we can't imagine.
> But you, dear viewer, from beyond 2000 know what happened -- there are still working horses, but nothing like before. The horse population peaked in 1915 -- from that point on it was nothing but down.
> There isn’t a rule of economics that says better technology makes more, better jobs for horses. It sounds shockingly dumb to even say that out loud, but swap horses for humans and suddenly people think it sounds about right.
That does not say anything about how much work exists in aggregate. The human population has gone up, so it can be simultaneously be true that the amount of work being done increases even as each worker works fewer hours. As well, this also says nothing about the quality of work, as GDP is going up, so it can also be simultaneously true that the quality of work increases even as each worker works fewer hours.
I think the relevant metric is the amount of work per people, not the agregate amount of work. If tomorrow there's twice as many people but only one more job because of AI, then sure! the agregate amount of work has increased.
> In all of human history, work has always increased.
Production has increased. It's not clear that work has increased.
Mills and factories used to employ people by the hundreds of thousands and maintain people in a blue-collar standard of living. Now, no manufacturer even exists in the top 25 employers in the US--it's all service industry.
The vast majority of the decendants of the people working those manufacturing jobs are not working in better jobs than those were.
My partner and I run a handful of small internet side businesses. One of our content-driven D2C businesses heavily relied on bespoke illustrations for our display ad creatives. we found that our ctr was decent, pretty average, but the CPC was killing us and ROAS really sucked.
Several months ago we decided to A/B test SD against our usual illustrators. In our case the results were pretty dramatic, we actually found that the ctr shot up by almost 20% and cvr showed a consistent uptick. I don't agree with the blog post's claim that AI generated images work best in businesses where the content doesn't actually matter; this particular venture is a fantastic counter example. In our case the AI-generated images seemed to resonate more with our target audience, as we were able to achieve much more granular personalization at lower cost than before. not only did it reduce the CPA significantly, but the tight control we had over creative variations meant we could optimize in realtime based on audience segmentation.
Not to mention that our time-to-market for launching new campaigns went down by half. no more back-and-forths over design nuances, missed deadlines, or creative blocks.
And I do feel a bit mixed about the diminishing role of human touch in creative processes. But from a purely growth-hacking POV, this was a gamechanger, and we have the numbers to prove it.
Overall I think this is a net win, especially because I don't think this needs to be the end of the road for human illustrators, but this will force them to adapt and bring more sensitivity to the needs of their clients. It makes no sense for even a content business to be subject to so much friction in the procurement of creatives, and this forces more consideration to our needs
Anywho there's efficiency, and then there's soul. Hats off to the robots for (mostly) nailing the former, and sometimes surprising with the latter.
no more back-and-forths over design nuances, missed deadlines, or creative blocks.
This evoked, for me, the "can I get the icon on cornflower blue" scene in Fight Club.
How much of this reduction in back-and-forth is influenced by the immediate/interactive response (dealing with fewer humans) and how much is due to a level of trust-of/delegation-to the machine? "A machine generated this icon based on my description, there's no need for me to question its choice of colors." — really the classic problem of considering machines as infallible and more expert than humans.
I think your usage of "matter" and theirs is different. It's furniture. Furniture "matters" in a restaurant and having the wrong furniture can hurt your business— but compared to the food, it's essentially inconsequential.
There's a spectrum of how much furniture matters in any given place ranging from very short stay waiting areas to architect's offices, and commercial art is no different. If that image was truly inconsequential, you wouldn't need one there. Non-informational graphics on most non-professionally designed power point decks likely matter less. I'd say there's about a zero percent chance of a two page spread opening a feature article in a magazine being ai-generated unless it's an article about ai-generated images, and even then, it probably took professionals longer to massage it into shape than all of the rest of them. Specificity and per-pixel control is just so important in professional graphics workflows and despite what a huge stack of people who aren't professional designers will tell you, they are simply the wrong tool for the job. It's fundamentally the wrong interface. Maybe what Adobe or another player who knows what the industry needs will nail it, but it won't look like Midjourney— that's for sure.
> Overall I think this is a net win, especially because I don't think this needs to be the end of the road for human illustrators, but this will force them to adapt and bring more sensitivity to the needs of their clients.
The advantages of AI that you crow over simply can’t be met by any human professional artist. A human can’t do hundreds of revisions profitably. There’s increased “sensitivity” and then there’s needing to read the client’s mind.
If you think this isn’t a death knell for human illustrators in this particular market, you’re deluding yourself.
Your definition of the needs does not have any requirements on the fitness of the output, nor on time spent on customer side. That does not seem realistic.
Ive tried letting AI write my articles. It was horrible. I tried ignoring AI-powered tools (such as grammar checkers, summarizers, rewriters, speech-to-text apps), and the writing process felt sluggish.
The middle ground is what works best for me. I use generative AI exlusively mid-process, but neither for input (ideas) nor output (actual drafts.)
Here's how I write:
- I source my ideas from contemplation or conversations on social media. Topics discussed there have at least some pre-validated relevance
- I sit down for ten minutes and dictate my thoughts into a tool like AudioPen (no affiliation, just a fan) which summarizes my 10 minutes in 5 or 6 paragraphs. THIS is the AI step. The tool suggests a few paragraph structures that I cycle through until I find a good one.
- From there, I write my draft, following that outline. No more AI tools here other than grammar checking at the end.
AI is a great writing partner. It's a horrible writer.
That’s been my experience so far too. It’s good for giving you some suggestions for things to structure and things you may have missed, but it is a terrible writer ATM.
> Commercial illustrators will keep their jobs, but will mostly need to learn to use AI as a part of their workflow to maintain a higher pace of work.
This is exactly what I've found to be the case too. People outside of this AI media generation community still think it's entering some text and getting some output. In reality, there are entire workflows constructed to get the exact type of image one wants.
The second image is the output image, but the first is even more interesting. It is a node based interface more commonly seen in game development tools like Unreal Engine which has a similar interface [0]. It is akin to hooking up APIs together to get the resultant image. I see the future of image generation being more akin to backend programming than actually drawing anything, which is to be expected as the actual drawing part is getting automated while the creativity now rests in the workflow itself (at least until we automate the workflow part too, but that's a far ways off as computers can't read minds yet to even know what the user wants).
Post seems very biased towards the now. Stable diffusion et al are very successful with a certain technique but it is foolish to think that is a method which will simply be improved indefinitely.
Generative "AI" will take many forms. Ultimately it will likely remove much of the "technique" element to creation, depriving artists and content owners of income and relevance.
Will this happen overnight? No. I suspect over the next, say, decade, AI will be a beneficial tool more than a threat.
At some point, I expect generative AI to become multi-sensory(sight, sound, touch). Such systems will work from physical models of subjects/environments to produce novel and accurate representations based on rich descriptions and deep contextual awareness of culture. These systems will not think in pixels but in objects and relationships which are then simulated, rendered and filtered to match the desires of the users.
I do applaud efforts of the writers and actors to protect themselves from competition but I believe it will ultimately be in vain. It will be interesting to watch the legal developments in this space. It may be necessary for future generative systems to provide an audit trail showing how they gained an understanding of the world to prove no unauthorized training was performed. This merely raises the bar slightly and does not prevent future generative systems from deriving important relationships via other means, such as 'clean room', high-level descriptions being given(perhaps by other automated processes).
For example, while it may be illegal to train an AI to reproduce Harrison Ford using his copyrighted works or even images captured in a public space, I can reduce Harrison Ford to a set of characteristics which can be passed to a generative system to produce something indistinguishable from the real Harrison Ford. If I am able to document this procedure I see few ways for the legal system to prevent it but then again I am no expert in this area.
For what it's worth, I'm not a fan of current "AI". I have found LLMs to be particularly unreliable and mostly useless. I also find most "AI" generated art to be either boring, inaccurate, or in some way not compelling. That said, I think the trend is becoming clearer.
> Finally, an opinion popular with no one: Commercial illustrators will keep their jobs, but will mostly need to learn to use AI as a part of their workflow to maintain a higher pace of work.
> This doesn’t mean illustrators will stop drawing and become prompt engineers. That will waste an immense amount of training and gain very little. Instead, I foresee illustrators concentrating even more on capturing the core features of an image, letting generative AI fill in details, and then correcting those details as necessary.
I'm not sure why they think this is unpopular with no one. This seems like the logical path forward. In the same way that CoPilot isn't going to replace me but it's makes certain boilerplate much less painful and avoids the "blank page"/"writers lock" that can happen when I go to write a function sometimes. It's just nicer to start from something then modify it until I have what I need (even if I end up replacing 80-99% of it).
In the same way I imagine it would be nice for an artist to see a couple of examples of what their line drawing could be which will spark some creativity and then they can do what they want.
Right, that opinion is popular with me: I love the idea that commercial illustrators can add generative AI to their toolbox. Those are the illustrators I most want to work with: people who can produce the best possible images using the whole suite of tools available to them.
I agree with most of this, but I do disgree with the thing about producing specific imagery. It's absolubtely a skill one can develop. I spend a lot of time helping people leearn to simplify their prompts and choose the right language for image generation AIs. For some reason people put a lot of unnessacary junk into them, I guess a form of superstition (this sentence fragment worked well the last few times).
As the article mentions, the hybrid approach (using this as a tool in a series of other tools) is the way forward
There are concepts the AI simply will not grasp. For example right now midjourney will extremely struggle with "bulldozer", "centaur", "fantasy archer" etc. These will inevitably fixed (and have in the past) be fixed with new model versions with better training data.
The real struggle comes with either small details or semantic information. For example, its hard to ask it to make a lifelike/photograph scene with everything including the background in focus. Even with "focus stacking" type keywords. "selfie" is about the best word we came up with but unforunately that has significant side effects lol. Perhaps there just isnt enough instances of people specifically describing that property in the training data, but honestly its difficult to even learn english words for these concepts to describe with!
As for small details, it is indeed true that the current approach will probably never scale to handle something like "six blue cubes with a red triangle on each, arranged in a pyramid shape, with a yellow ball balanced on top". But as the author points out, such things will likely be handled with a minimum of photoshop skill using assets made individually
None of which has anything to do with creativity or the original (and visually trained) thought required to conceive of imagery that's commercially useful - which is an actual skill learned through years of study and experience, and which is so routinely ignored by managers and IT people that you completely failed to mention it in your take on the technical issues with prompts.
The issue with prompts is not stacked cubes. It's more like this: Ask 10 software engineers or 10 people from the sales department or 3 people from upper management to come up with visual ideas for ads, and you will have a bunch of shit on black backgrounds, robots, anime, bad copies of things people have seen and subconsciously remember, and zero actual visual ideas that fundamentally work. Designers and illustrators have to fight against and override their unoriginality and terrible ideas all the fucking time just to make a decent product.
That's some real unwarranted hostility. I'm responding to a specific point from the article? We can't just go in circles having an "is it good or isnt it" argument...
Once again - I agree with most of what the author says, including the part about it being a tool in an illustrator's kit
I mean, I've spent 25+ years as an art director, a.k.a. a diffusion model trying to generate what managers and salespeople think they see in their heads, and I tell you, they have no imagination. None whatsoever. As my brother, a photographer, used to say: The problem isn't having a cheap camera, it's who's behind it.
Also, the hostility towards a prompt expert adding another layer of technical "know how" into the process between requests and art in the name of justifying a new job title is entirely warranted.
>Also, the hostility towards a prompt expert adding another layer of technical "know how" into the process... is entirely warranted
I don't know either of you and I have no stake in this but as an outside observer I think you come off pretty unreasonable here still. You seem to think your hostility was justified because you've basically made this person the scapegoat for your frustration about this topic.
Also an outsider - I don't understand how they're being hostile here.
They're relaying their experience and saying that there's more to creating than describing something to be drawn, and that most people lack the training and knowledge of what goes into that.
It's not about learning prompts, it's about learning how to actually design... and then learning prompts.
I feel like the original comment is taking things personally instead of seeing the point they're making through example of their frustration working with others.
Interesting! From my view the responder came off as attacking unnecessarily and very angry. Honestly not sure how else to relay it though. Maybe it's all just how I'm reading it. Have a great one!
whatever man. I am not here to carve out a job title or whatever it is you're accusing me of. I commented on a post contributing with my experience of helping others.
I feel like you're projecting a whole lot more onto me than what i'm actually saying.
There are new models such as from Google which work differently and handle things like counting etc. much better than open source models I don't know if any of them are available yet but they have papers. Like Imagen and something better that came out afterwards.
This is almost exactly how I see GPT and code. I have seen smart people who don't write code toss GPT a request and get a working thing but going from "hey write me an auto presser for a key on a timer" to "The client wants you to update the repo to handle this business logic and do this functionality" is a huge leap.
From my perspective, you still need a developer-minded person to do the job, AI just kind of makes their lives easier in the process.
I agree with the sentiment that it's the same in the art world. It's easy to get a compelling image but to get specific meaningful images usually requires a lot of post processing that a layman wouldn't be capable of doing.
True but I feel like people are assuming these limitations will be long term.
But we have seen progress in leaps and bounds. LLM-based coding tools are getting better. LLMs are getting better. Context size is increasing. And the interest in LLMs is even motivating development of new approaches that will be more effective.
Give it a few years, things like Lecun's JEPAs or whatever hybrid supercoder DeepMind is working on, or some open-source LLM, will blow GPT-4 out of the water for programming.
I agree these things are going to move likely faster than we can imagine and become better than we can imagine. That said, last week I got a ticket that had no body but a title akin to "Add a property for this" with no specifications about the system. When I asked for clarity on the ticket the ask was very different than the title implied. When I started the work it started to look different than was described so I went back to the requestor and explained the discrepancy and we changed the direction of the change.
I say this to say, one skillset I have as a developer is taking the vague requests product owners have and figuring out how to turn them into actionable code steps in a massive existing codebase with several repos. I don't say this to say I'm impossible to replace but to say that half the time people don't even know what they want or how to describe it. Then from there you have giant codebases that wouldn't fit in anything but the biggest (current) context windows.
I agree the accuracy limitations will likely evaporate but these things aren't necessarily something an LLM can solve. I'm probably going to be proven wrong over time but I use GPT for code pretty regularly and right now I'm not too worried about my job.
I don't know for sure if an LLM could do it. But theoretically one could build a system that sends chat messages asking for clarification and also eventually with more context or something is able to translate vague or stupid "requirements" into ones that make sense.
In five years or so the capabilities may be pretty amazing.
Definitely agreed! I'm both nervous and excited. I fear for my livelihood but also if we continue to make even a percentage of the progress we've made in the last year, the next 5 years are going to be wild!
Completely off-topic, but if the author happens to be here:
What is the system you are using to justify text? The "hanging" parenthesis on "(Understand that what I'm about to describe, ..." caught my eye, and it appears that every _line_ of text has it's own padding and word spacing to arrange it "just so". I can't imagine it was done by hand, but I've never seen it before.
Overall this post is a pretty fair take, but I'm not sure intuitive the "latent space" explanation is. I work with diffusion models full time and really appreciate how easy they are to explain at an introductory level (assuming you don't get distracted by the mathematical derivations). But I don't feel like I've seen that many good explanations online. I made my own attempt in a presentation last year (https://youtu.be/c-eIa8QuB24?t=86) but haven't had time to clean it up. I keep hoping I'll see a better series from a real communicator pop up in my feed but I'm not sure this is it.
Honestly, as a mathematician-turned-ml-researcher, Song and Ermon’s mathematical derivations based on SDEs are what sold me on diffusion/score-based generative models. This (https://youtu.be/wMmqCMwuM2Q?si=fvujznGDHjKH2yUi) is probably the best ML seminar talk I’ve ever seen.
Oh definitely. I think that 2021 paper was what sold me as well! But I don't think it is very approachable to non-technical audiences, and there are a number of other was to interpret diffusion models which are pretty intuitive. That is also why I like the alpha-(de)blending paper from this year.
I used to think that the generative AI impact would be pretty low because these images always have some artifacts that I find pretty jarring, and that require fairly high artistic skills to fix.
However that was completely wrong: most people don't care!
Neither people producing the content, nor the people consuming it.
The second point, "only spammy garbage content" will be happy with AI generated content, is already proved wrong given the quantity of high profile blogs that rely on it. They don't have the budget for the maybe 5% improvement you can get by paying an artist, and 0 of the risks common with artists (difficult to work with, missing deadlines, etc etc).
In a way it doesn't even make sense: the artist is also is also a generative blackbox. It's better in understanding precise prompts, but exactly as in software engineering the problem is often that the spec is wrong, the commissioner cannot get exactly the image they dream of because they cannot imagine it without having pretty high artistic skills. Or a number of iterations are needed, making the process quite long and costly.
There are other reasons why artists won't be entirely replaced, especially the highest paid, but a good chunk of their potential income sources have already been wiped out, and the proportion will only increase.
My takeaway from generative art as a former illustrator and now data scientist is never has it been more obvious why artistic skill and taste are necessary for making images, while at the same time never have those things been more irrelevant because of the audience for the work.
> However that was completely wrong: most people don't care! Neither people producing the content, nor the people consuming it.
For the people making and consuming on Reddit maybe. I think that people who want this to replace graphic design work will want more attention to detail.
> I used to think that the generative AI impact would be pretty low because these images always have some artifacts that I find pretty jarring, and that require fairly high artistic skills to fix. However that was completely wrong: most people don't care! Neither people producing the content, nor the people consuming it.
Sometimes I feel professional people are so good at their crafts that they're disconnected from general audiences. It's kinda like a programmer trying to convice a data scientist that Python is not that good of a programming language, while the data scientist is perfectly fine with it.
AI will only replace the work of creatives who have already turned themselves into AIs long ago.
There are plenty of writers and illustrators out there who have trained themselves to churn out reproducible garbage over the years in order to fulfill the demands of content marketing. These jobs will be replaced by AI soon, but by creating content from a formula they've already be using a crude form of AI.
I really love Stable Diffusion, but, as a means of creating art (including the most common forms of popular art), it can only supplement existing work not replace it. I pay for plenty of real art in my home and the best works on my walls could never be replicated by an AI because what makes them beautiful is precisely the human touches that I have yet to see AI generate (and suspect it can't). Latent Diffusion Models also have a pretty poor imagination.
At the same time Stable Diffusion has gotten me thinking about creative projects I could undertake that would have been impossible years ago. But it's obvious that all of these projects will take plenty of work to create, and SD will ultimately just be another tool in the creative process.
I vividly remember when photoshop started to gain major acceptance and there was a similar anti-photoshop sentiment among "real" designers and artists. What's funny is how many webcomic authors I see critiquing AI art, when I remember quite well pen and ink comic artist similarly scoffing at web artists that used digital tools to create their work.
Hopefully we'll see SD and similar tools accepted as more tools to create cool art, rather than a misplaced focus of peoples career anxiety.
What are the artist names and descriptions of the art? Would you care to share a few? I am curious as to whether your claim that they can't be reproduced by an SD model can be shown.
> Producing a compelling image with generative AI is pretty easy; maybe one in ten images it generates will make you say, “Wow, cool!” But producing a specific image with generative AI is sometimes almost impossible.
I think this is the meat of the argument and a pretty compelling point. I have definitely struggled to get mijourney to create certain images that I had in my head and eventually just gave up.
Has there always been this much debate/discussion any time a new technology disrupts an industry? Or is this a particularly unique case because it touches on what we perceive much more as "creative" works?
When audio recording became a consumer product, was there huge resistance from live performers? I would imagine so, but I just don't know much on the subject.
I can highly recommend an episode of 99% Invisible [1] about the musician's strike of 1942, which was a fight about royalties from recorded music, but was in large part actually about the the loss of livelihood from music recordings. Very little new music was recorded for over a year, and the president of the musician's union was pushing for record labels to pay into a fund that would benefit unemployed musicians in order to end the strike. I didn't make the connection when I heard this, but yeah, it does feel analogous to what we're facing now.
Actually, it was only instrumental musicians being prohibited to record. So singers weren't affected and that strike hugely contributed music shifting from primarily instrumental (earlier jazz, swing) to vocal-focused (jump blues, rock and roll and so on...)
It's interesting because it's something which is around us every day but most of us don't know about it.:)
None of this is new. Audio recording, sound synthesis, digital mixing and composition either killing the music or not being real music. Cameras, 3D CGI, and Photoshop either killing visual arts or not being real art. And yet here we are, with new musical genres focused on sound rather than rhythm, two electronic revolutions in music, entire new fields of visual arts and entertainment.
This case differs in that it got amplified on Twitter, though.
These posts are all written with this insane premise that the moment in time the post was published, will continue to be the status quo for more than a couple months.
> If you’ve ever used a generative system, I can pretty much guarantee that you spent an embarrassing amount of time making tiny adjustments to your prompt and retrying. Producing a compelling image with generative AI is pretty easy; maybe one in ten images it generates will make you say, “Wow, cool!” But producing a specific image with generative AI is sometimes almost impossible.
Who could possibly think this will be the case six months from now? I mean, maybe some of the the content and warnings here is fleetingly accurate, but it's truly not a hill to die on. You could fire your illustrator and be inconvenienced for a couple months until the next stablediffusion update. It's a disservice to illustrators to make them feel safe.
I think firing your illustrator to replace them with Midjourney is like firing your developers and replacing them with Copilot. You still need someone to do the job, but now you got rid of the person who knew whether the result is actually good. We might as well get rid of photographers and illustrate our news with cell phone pictures while we're at it.
If a company needs to generate illustrations at a big enough pace to require an employee they'd only be replacing their illustrator with a worse one. We all know what happens to GUIs when programmers develop them, so why would this case be any different?
> We might as well get rid of photographers and illustrate our news with cell phone pictures while we're at it.
Wait... Hasn't this already happened? Professional/specialized press photography seems way down compared to pre-smartphone era? Now a journalist/reporter is expected to do a passable photo job on their own. Or a random member of the public
I'd bet on this being true six months from now. DALL-E 2 came out in 2021-01 (19mo ago) and Midjourney in 2022-07 (13mo ago). While the quality of images has improved a lot, if you have something specific and moderately complex in mind it's still very hard to get there just from a textual prompt.
> While the quality of images has improved a lot, if you have something specific and moderately complex in mind it's still very hard to get there just from a textual prompt.
I think this gets to an important point. If whoever is paying the bills has something very specific in mind, they won't be happy with AI (or frankly many artists) at this point. But as a creative interpretation of something more general, it's actually really good, and I think with many low-importance works like the author describes as "furniture" - we really don't need to be that exact.
One year ago we had textual inversion. Now we have LoRA and control net. I know it might not be a scientific breakthrough, but in practice it's day and night.
I'd be happy to bet that the tech will continue to improve. But "producing a specific image with generative AI is sometimes almost impossible" seems very likely to still be true in February 2024.
Of course an AI can not produce an image that matches what's on your mind 100%. Because by definition it will require you to provide all the information, a.k.a. you need to make the image by yourself first.
But I really don't think illustrators are as safe as the article implies. Yes, the jobs won't disappear overnight, but are there that much demands for illustration to support a future where every illustrator becomes 10x more productive than before?
(I've done illustration commercially before, while it's not my main source of income and I'm junior level at best.)
I was kinda tempted, but I don't think there is such a good metric to measure it. (Since the job market is heavily influenced by interest rate etc... and if I were confident at predicting macro economics I would just bet on stock)
What if we picked 10 prompts that it seems like an AI should be able to depict well, but it can't yet? And then iff the best AI tool in February can do the majority of them you win?
I'll bet against it tho. I don't actually believe pure text-prompt-to-image will improve much (not in a few months at least). I just believe there will be more non-text tools to guide AI, like LoRA and control net, and they will be more accessible.
Control net kinda did what you said but in a different timeframe: it was quite difficult to tell AI to generate a person "sitting with their legs cross". Today, it's relatively easy to do this with control net, but still hard with text prompt only.
Edit: and the sibling comment made me question myself why I would ever take a random bet on the internet.
I was just stating the improvement on SD we've seen since DELL-E 2 and Midjourney came out wasn't just about "quality of image", but also about "have something specific and moderately complex in mind". Thus I mentioned textual inversion vs LoRA/ControlNet.
I wonder why people take random internet bets, I've seen this on Twitter as well from some prominent people in the tech world, who just like betting on outcomes I guess. I saw it most recently with the LK-99 "is it a real room temp superconductor or not?" bets.
I think often someone tries to frame things into bets to pierce the layers of instinctual contrarianism and "have strong opinions weakly held" that often pervades internet discussion.
More charitably, reframing into a bet gives a relatively neutral opportunity of re-stating the discussion/argument to more clearly identify areas of agreement/disagreement (since a clear definition of the disagreement is crucial for forming a bet).
I don't think it will be true six months from now. OpenAI has been red teaming a new version for months that goes way beyond anything we've seen today. You can see some leaks from this video: https://youtu.be/koR1_JBe2j0
I don't know if that's 'way beyond'. Those samples look similar to Imagen and Parti in terms of quality and following complicated text prompts. (Look at the PartiPrompts for the absurd prompts Parti can execute accurately.)
I'm glad to see OA does have a successor for DALL-E 2, though: the service seems to have somehow gotten worse since release.
I shouldn't have said "goes way beyond anything we've seen today", but probably it goes way beyond anything we will be able to try in this 6 months period. Parti 20B is really good at language understanding, more meh Imagen.
>the service seems to have somehow gotten worse since release.
Yeah, I think it's pretty clear that they probably lowered the number of steps or used a similar strategy to reduce the computation.
But in general the better something is the harder it is to improve.
From an outsiders' perspective it looks like we've had a series of achievements where AI worked impressively well considering it isn't a human. But afterwards we got an incremental grind that never got close to an AI that's just good, period.
I'm very impressed people got self-driving car demos working. I couldn't do that. But year after year self-driving cars remained a demo, albeit an incrementally improving one.
It's limited to a small number of cities with optimal conditions and every ride has a human supervising. They're still spending R&D, rather than making a profit. That's a demo.
> It's a disservice to illustrators to make them feel safe.
It is also a disservice to encourage managers to make their illustrators redundant when we don't yet know that, in six months, AI image generators will equal a human illustrator in being able to create adequate pictures. This kind of article has an important role, namely that of countering hype (often based on PR from machine learning companies). That hype is why more ignorant people think that AI can already replace crowds of employees, when the reality is more nuanced.
Why are AI people always trying to convince me to do something now based on an argument that it will work later? Regardless of whether you're right a working model six months from now is not a working model now. Wouldn't your hypothetical person be much better off firing their illustrator in six months?
Of course. But what should the illustrator do today? Considering that there might be a higher risk of getting fired in N months/years from now. Probably refine their business skills. Maybe also get familiar with the capabilities/weaknesses of the latest tools, and how they can use them to be more competitive than before?
You have total doomers aware of the AI potential. Horrified at the long term consequences of these LLMs. Then on the other side you have users saying "Its just another crypto bubble", and when pressed, they admit that they never used it.
There is just such a vocal population here that says 'Well its not always 100% perfect, so its useless", and they are burying their head in the sand that companies are already using the OpenAI API to reduce the cost of business.
I genuinely don't understand these people. They don't use the technology and they deny how useful it is. There is news and real world examples of its usefulness. I can only imagine these people manage (money) terribly.
History tells us that both the doomers and naysayers are probably wrong.
It seems to be pretty solidly demonstrated now to have some limited efficacy across a broad range of areas today, and is very effective in some niches (like the articles mention of producing SEO fodder cheaply).
Growth from that state though? The only thing you can bank on is that nobody really knows.
I don't understand people who think an imperfect product provides linearly less value when it obviously provides exponentially less in most use cases.
If ChatGPT can replace a team of software engineers why didn't you replace that team with four times as many college interns years ago? Because you can't combine people capable of doing easy coding and get someone who can do moderately hard things.
This is more or less my opinion as well. This is a list of current weaknesses of generative AI. No, right now I cannot replace my graphic designers. I bet I can in a few years, though.
I wish the article quantified the amount of SEO and ‘irrelevant’ content procured in the wild—- it’s a large number but also a dark art for creators to produce cheaply. The author seems somewhat content with the idea of using AI to generate garbage content, but this essay might underestimate the profits of leveraging existing art to train said AI. It strikes me in the current Writer’s Strike, studios probably have an estimate of how profitable AI could be and aren’t sharing that number.
The essay focuses mainly on prompt-slot-machine-based generative AI but there’s a large suite of work that uses the same research to more directly support the artist. Stuff like in-painting, controlled diffusion, and re-stylizing.. as long as it’s free (compared to more expensive art equipment) it should have a beneficial impact on artists.
What are you saying? That those that claim critical improvements won't happen are well justified? Or that they are not and just stuck in the 90ies? Or that a hype cycle will happen, but massive impact on trades/jobs will still be had? Or something else?
I have been using stable diffusion for my web designs and it’s been working great for me but I always build my figma designs separately. I use the generated designs as a palette where I sample from. It’s a glorified Behance or Pinterest vision board. Chat GPT is similar with code as it’s a glorified stack overflow where I get inspiration or quick fixes. The AI tools haven’t as of yet been so cohesive enough to replace the aggregator of these pieces of inspiration. Being classical trained in not CS and design will always be needed to filter AI suggestions. AI will eventually get proficient enough to reach that stage, but as it is now, it’s not even close.
I wonder if you could elaborate on how you use ChatGPT to generate web designs? I'm about to teach a course on web technologies and I want to integrate ChatGPT into the curriculum.
Sam, your character arc reads similar to my own. Thanks for putting yourself and your experience (and emotion) into this write up. It gives me peace of mind to know others share both the haphazard education -> career lineage and the existential concerns which maybe are part and parcel with said lineage. FWIW, I was a budding pen and ink illustrator when "desktop publishing" was tipping past "expensive novelty." Love the write up - thank you again, sincerely.
I'm more interested in how this "cross attention" part works.
Being able to combine two different kinds of AI sounds too good to be true. It sounds like AGI. Why does it work for SD? Why aren't we trying to combine more AIs to create a super AI? Or we're already doing this?
Cross attention is not really a way to "combine multiple AI models" but there are many ways to do that, and actually diffusion models are really good at being combined with stuff. Especially thanks to tricks like score distillation (see dreamfusion3d.github.io). But it isn't anything like AGI because the AI is not inventing the combinations itself, and even if you could, there is no clear way to make it self-directed. These are still processes that require lots of programmers being very clever.
Very pragmatic view from Sam, he is missing that the “gacha” nature will go away
People will prime their image generation with a specific series of image and combine them with new prompts, packages will be made to prime their AI and then they’ll go from there
> People will prime their image generation with a specific series of image and combine them with new prompts, packages will be made to prime their AI and then they’ll go from there
Whether its finetuning checkpoints, creating LoRa/Hypernetworks/TIs, etc., that's already the status quo.
The second time you said that, the first time it seemed informative at least for anyone passing by but now its not, take a step back then and notice that I didn’t say we weren’t, I said the article/author didn’t factor it in
And the only reason i wrote it is because I’m aware of what people do
I don't want any more Disney live-action remakes, but I'd like to see more variety in art styles, like the latest animated Spiderman and Ninja Turtles movies, which I understand are made with the assistance of generative models.
Insightful, agree with the fact that AI will help to fill up the "furniture" more easily, faster and for a cheaper price than any human could. Regarding transparency of training data, this is where I see huge opportunities in AI for the near future.
I’ve been using MJ and dalle to create actually used print content for some posters. This process of ai assistéd illustrator work is very fast. More than 7 works so far, and I’m not a designer, just know basic principles .
If your idea of illustration is some irrelevant piece of clip art, it hardly matters. If the illustration actually illustrates the content, you probably can't generate it with current AI anyway.
Author conflates legal and ethical options for preventing copyrighted work from being used to train ML image generators.
There’s nothing legally inconsistent about passing a law saying, e.g., “ML training is not fair use”. Doing so will not even reduce existing fair use rights being exercised by actual people.
The author’s argument is that doing so is philosophically analogous to human creative processes, but those are—and I can’t underline this enough—human. And the law is not (and cannot be, should not be?) consistent in such a way.
> There’s nothing legally inconsistent about passing a law saying, e.g., “ML training is not fair use”.
Is it still fair use to take inspiration from another artist's work? How can the courts necessarily tell if the art was made using AI or if it's just someone stealing another artist's style? Theft of style isn't currently recognized under the law, but it could be.
Some variants of “theft of style” are recognized by some courts already, please see the legal literature on music copyright and the recent 7-2 SCOTUS decision on Warhol’s Prince series.
I absolutely agree that an arbitrary line can be drawn; I don't see that that line can be clear and bright enough that forms the kind of precedence that can be relied upon by folks who don't have the money to fight an uncertain battle in court.
Can you give an example of a “clear and bright” line in copyright law that does protect “folks who don’t have the money to fight an uncertain battle in court”?
For context, I’m in the process of translating a work that I know for a fact is in the public domain (sole author died 90+ years ago) and I’ve still got legal questions that I’m going to have to hire a lawyer to solve.
Spotify basically killed any money coming from the physical distribution - Worse than piracy, which was inevitable too at the time, but at least you didn't have to pay your lawyers to renegotiate with your label on top of NOT getting any money.
Adobe, OpenAI, whatever: they want artists to draw for them for peanuts to train their model, sign a waiver saying "I'm ok not getting any money from any AI art made from this", and then resell the output for $$$ on something like Splice[1], at the same time overtraining such models in ways that make extremely obvious whose artist made them in first place.
At the end of the day the model itself is going to be basically irrelevant, while knowing whose works were actually used to train it being the truly differentiating feature.
But you know, "the AI did this picture, so we don't have to pay you."
[1] https://splice.com/features/sounds