Hacker News Reader

Comment by w-m | Hacker News Reader

Earlier today, ChatGPT/DALL·E 3 refused to draw the bat sign (from batman) for me, due to content policy. When I asked it to rephrase the prompt to circumvent any copyright issues, it came up with "an emblem featuring a nocturnal flying creature silhouette", which made me chuckle quite a bit.

Playing the game where you have to convince the LLM to spit out the secret password[0] was great training for this content policy dance with DALL·E. Isn't it great - even though the machine is now performing the handicraft, you can still feel creative. Not by creating art yourself, but by finding new workarounds around the man-made content policy limitations.

[0] https://gandalf.lakera.ai

View on HN

Makes me wonder where is the line on copyright? if the bot is refusing but the human is tricking it in creative ways to make something infringing then it seems like the violation is being committed by the user.

The line is a million light years from where OpenAI has positioned this.

Drawing a Batman logo can never be copyright infringement, the art belongs to the artist (or in this case, there is none because there was no artist). It would be like Hacker News prohibiting me from doing an ASCII art Batman logo in a comment on copyright grounds.

It is pretty funny that openai arguably is based on enormous amounts of copyright infringement (depending on if scraping training data is fair use or not), but they're kind of being prudes about something as clearly benign as this.

Nothing funny. It's similar to how YouTube was built on massive copyright infringement, on a scale unmatched by anything except perhaps LLMs now, and then turned around to become the Fully Automated Fanatic Copyright Enforcer we know it today. It's not hypocrisy - it's this, or getting sued out of existence. They pulled a fast one and got away with it, because MAFIAA figured it's more profitable to extract money from YouTube at legal gunpoint, instead of killing it and losing a distribution channel they can't maintain themselves. Here, OpenAI is trying to get ahead of the whole generative AI copyright kerfuffle, hoping to avoid getting sued if they demonstrate they're eager to work in the interest of copyright holders.

Very much like how YouTube goes well beyond what is required by copyright law.

Youtube policy is actually pretty lenient compared to most IP rights law. "Fair Use"/"Fair Dealing" actually has a pretty well-litigated definition in most jurisdictions. YouTubers just dislike the definition.

If YouTube exposed you to the full force of most jurisdiction's IP law, you would be getting C+Ds and summonses from scary people in suits who work for Viacom. "Fair use" isn't a magic incantation to ward off evil lawyers, it's a defence with definitions and restrictions used during a dispute.

The copyright strikes are a pretty good compromise to shelter individual creators from the outside world that's used to Big Company 1 dealing with Big Company 2.

> YouTubers just dislike the definition.

That's not it at all.

The problem is that they use opaque algorithms with false positives to issue strikes. Or content companies illegitimately add public domain materials to their libraries, which causes strikes against anyone lawfully using those materials.

> If YouTube exposed you to the full force of most jurisdiction's IP law, you would be getting C+Ds and summonses from scary people in suits who work for Viacom.

They can still send you those things if you infringe their copyrights regardless of what YouTube does. They generally don't because random individuals generally don't have enough money to justify their legal expenses and it's incredibly bad PR.

The opaque algorithms and trigger-happy strikes are to appease copyright holders, which leads into...

The reason they don't litigate individuals is low ROI on litigating a small YouTube channel. That is true. However, Youtube without these tools in place would be sued out of existence. Copyright holders wouldn't ever come after 1 video, they would go after the platform hosting their IP without compensation. With every legal right to do so!

Youtube cuts a good deal for them where they get god-tools, and in exchange, YouTube can continue to exist.

Never said it was a good set up, but this idea that YouTube has another "easy choice" is pervasive.

YouTube is protected under the dmca safe harbor provisions. All they're required to do under US law is respond to DMCA requests. Content ID goes far above their legal requirements.

That has to be litigated to find out.

Cases like Napster [0] and Grokster [1] set precedence for this sort of thing. They also used the DMCA in their defence, and they were still found culpable. These were decisions that were scoped by the method of delivery, so if YouTube was going to stop giving everyone a great deal, they'd be betting they have a solid case as to why their technology is legally distinct from those examples, and many others. (And conversely, like instances where the safe harbor clause has been used in a successful defence)

That's a weird bet to take when the benefit is... people stop complaining about YouTube Content ID, and find something else to complain about?

[0] https://en.wikipedia.org/wiki/A%26M_Records,_Inc._v._Napster....

[1] https://en.wikipedia.org/wiki/MGM_Studios,_Inc._v._Grokster,....

The Napster decision left the DMCA question open as it didn't affect the actual appeal, and was instead left to be decided at trial (which never occurred):

> We do not agree that Napster's potential liability for contributory and vicarious infringement renders the Digital Millennium Copyright Act inapplicable per se. We instead recognize that this issue will be more fully developed at trial.

https://scholar.google.com/scholar_case?case=141026963365506...

The Grokster decision fails to mention the DMCA at all.

Trick question, no one actually infringed anything! You're welcome to draw the bat symbol all day if you want. You can even download the official emblem, scribble out the TM symbol and (c) Warner Brothers text, put your own name and chuckle to yourself.

The infringement only happens when you then try to claim it's your own work publicly. Which has the convenient benefit that a human normally has to be an accomplice to that.

Pretty sure that is not how copyright works.... attribution rights are important but it is not the be all and end all of copyright.

Laws obviously vary a lot, but most places that follow a similar definition of intellectual property rights to EU/NA do not forbid merely being in the presence of a work. Knowing of the details of an intellectual property isn't infringement. You wouldn't be able to monetize it if you were compelled to never let your book printer ever see your story to reproduce it for you, let alone the reading public.

The purpose of copyrights as a concept, is to allow artists to profit from a work that's otherwise easily to duplicate. Laws tend to be written with that in mind. This is normally codified as an exception called "private study" or similar.

You can ask the AI all day long to draw the batman symbol. But if you try to market that as "Chiroptera Guy", then you've taken the step that opens up litigation options for the holders of the original IP.

But at the end of the day, copyright is ultimately about reproduction of a work.

Asking the ML model to reproduce a copywritten work isn't just "knowing" about the work, but potentially an alternative to buying a properly licensed reproduction of the work in question.

The post you responded to doesn't agree with your definition of what copyright is for.

The post claims that its about allowing the author to profit from it. Mere reproduction doesn't stop them, till you try to sell at some scale. It's the equivalent of asking your friend to draw you something.

Now, you're free to disagree. But if you're just stating a different definition, you're not engaging in a rebuttal, just an ignor-al.

I'm correcting them by stating the legal definition rather than the definition they made up.

Asking your friend to reproduce a copywritten work is also copyright infringement.

alright

The world moves on though, I guess.

I mean, there's a couple centuries of case law, explicit US code, and international treaties codifying what I said.

It's not exactly something the world will move on from without explicit legislation and statecraft.

Your definition is legal, sure, but their definition is realistic. The overwhelming vast majority of copyright "infringement" is ignored, because who cares if a kid draws Pikachu? That's why copyright law is very vague and relies on context-specific judgements of severity of infringement on multiple categories.

> Your definition is legal, sure, but their definition is realistic.

...it's an entirely legal term. It has no definition outside of its legal one if you're trying to figure out how litigation would work.

> The overwhelming vast majority of copyright "infringement" is ignored, because who cares if a kid draws Pikachu? That's why copyright law is very vague and relies on context-specific judgements of severity of infringement on multiple categories.

There's a different argument made in that case if it hits the courts than the original poster's argument: de minimis. Basically that the claimed damage is too trivial to waste the court's time.

It's not clear that automation at just about any scale will allow for a de minimis defense.

> "The infringement only happens when you then try to claim it's your own work publicly."

Is a very different statement from

> do not forbid merely being in the presence of a work.

The copyright infringement happens at the moment of production of the image, unless the use counts as fair use. It doesn't matter whether or not you claim ownership or authorship.

Home copies are protected primarily by the total impracticality of enforcement, and secondarily by claiming you're doing it to practice making art.

In this scenario, the rights you’re leaving out are the rights to reproduce, distribute, and create derivative works. They’re exclusive.

You’ll find reputable t-shirt printers will refuse to print designs they know to be copyrighted.

No one would be claiming the printer or you came up with some copyrighted design, it’s the copying that’s an infringement (“copy”-right!).

Now, these printers also typically have you assert that you hold the copyright to the work, but my understanding is that just limits their total financial exposure by contracting with you; them producing the work would still be infringing.

Here, you aren’t even giving the bot a work to reproduce. It holds the work in its storage and will reproduce it for you when asked. That’s pretty cut and dry copyright infringement by whoever controls the bot.

If the photo industry is any indication, it won't really matter where the legal line is. They'll put in enforcement mechanisms wherever the industry feels like they should be.

Nothing stops companies from going beyond what the law requires.

I was trying to do a halloween themed image that included 'chopped-up, bloody body parts' which kept hitting the content policy. Basically any chopped-up or bloody combination with hands and feet etc. etc. hit the shield.

I didn't try hard to break it I just found it funny that the result I could generate in this particular graphics field was most like really good clip art suitable for a company party poster.

Use strawberry jam

And mannequins

The coming linguistic and cognitive changes are going to be amazing to live through. We're being oppressed by talking Roombas. We study our oppressors; we adapt. They understand ape language better than we apes ourselves, they are better apes than apes, they are super-apes–but yet even still we find small gaps in their ape-imitiation, nuances, metaphors too subtle and too monkey for the metal machines. Voight-Kampff tests. Simian shibboleths. The dead monkey is covered in strawberry jam; the steel Roomba gazes unblinking, and all is quiet.

I got you: https://duckduckgo.com/?q=bat+symbol+batman&t=osx&iax=images...

Reply on news.ycombinator.com