Prompt Engineering Is Hard
xeiaso.net
xeiaso.net
Beyond that, lots of people are unable to even explain how prompts are parsed or per token explain how it impacts what is generated. For example, do commas matter?
“Prompt engineering” is real, still unclear to me though what the future for it looks like, since easy to see via comparable things like Google search operators that the general public has little interest in learning command parameters.
Or you can answer your question about the effects of commas. You can try moving a word or phrase forwards or backwards and see what effect it has.
By working with a single snapshot of randomness, you can see the word effects more clearly.
Obviously understanding the code and data the model is based on is likely best path to better understanding what is going on; for example, last time I checked, Stable Diffusion was limited to 77 tokens.
Ultimately though, possible paths to understanding are like unlimited and just as creative as any other aspect of the topic.
As in: you'll be able to draw a bunch of circles and arrows with labels in them like "astronaut" / "horse" => "sunset" and some environmental info like "in the style of van gogh" and bang you get a proper image.
And the result is probably a copyright violation of some poor artist's work.
Today in many states it's perfectly legal to shoot a native american if you're in a wagon circle. Is that ethical?
Fair use is also a positive defense, which means that only the courts can really say if something is fair use or not.
>> “Art history is literally full of transformation of prior works”
There’s no reasonable ethical basis to take issue with. You’re examples and counter points to me add substantive your claims. What am I missing?
That difference is enough to assert that there is a wholly reasonable ethical basis. Especially when they're obviously not just training it on images with relaxed copyrights.
Ethically speaking, for the second time, you ignored that within art world, or real world for that matter, making derivative works is completely ethical if done within any related legal constraints; as such, to me, you’re not making a good faith effort, which is actually unethical if intentionally done, so I will not be replying any further.
> Today in many states it's perfectly legal to shoot a native american if you're in a wagon circle.
https://skeptics.stackexchange.com/questions/25222/is-it-leg...
Competent creators will not be replaced, they will use the tools to do their job faster/better. I remember soem tiem ago artists having reservations on "digital art". Someone said it better, a person with a camera is not a photographer.
the problems I have with Microsoft Copilot is that the fuckers did not used MS proprietary code train but only trained with other people GPL/MIT code, this was shitty and implies that their implementation is "dangerous" to use with proprietary code but their expensive lawyers can handle a bunch of open source small developers.
You did
1 MS did not violate copyright since is open source code, andfor soem reason you think that only proprietary code should be removed from copilot
2 even if MS violated the license the guilty person is the dude that uploaded the code to GitHub and click I Agree not he ToS, implying that somehow the GitHub ToS has the ability to change open source license(but t proprietary ones, since you claim MS "removed" such code when it was revealed).
As I mentioned above, MS are cowards, iof their thing is real creating new stuff then put their ass on the line and put their own code in too, put all proprietary code in GitHub in copilot too and then have their well paid lawyers defend their tech.
Name calling is though is very clearly against HN’s guidelines.
Let me try to use an example, maybe we are both bad at expressing this in english
I did no say that the user that uploads has no responsibility, but if I upload the Windows code, or FreeBSD code and mark it as GPL and click I agree on the ToS , does this mean that the original license is invalid and GitHub can do whatever they want just because a random person click an Html button?>> “Are you aware of any copyrighted material that was illegally loaded to GitHub that has been proven to be used by CoPilot, it’s use was not transformative — AND - GitHub has not removed it from the model?”
Speculative claims to me are do not merit any response, especially when accusing party of something for which you have no related evidence; which actually might be defamation.
If you have any creditable, independently verifiable, material, etc - evidence of a violation of a specific reproducible CoPilot output, which is documented as having been a valid copyright violation by an unauthorized party, and it has not been removed from the model — please feel free to share links.
As is, GitHub has the right to use code unless they receive what I just requested multiple times from you and appears possible you’re intentionally ignoring for reasons I do not understand.
If I have missed something, it’s not intentional, happy to address any errors or omissions on my party if they exist; assuming you’re not skipping over my points.
Snippets of code were already found, and MS did not removed them from the model but put an IF check to make sure is not appearing again.
So now I need to find specially for you a new snippet and also somehow ensure MS won't add it in the IF checks too ... then I would have proven to you that all that list of conditions are broken.
The proof was already shown, the code was spiting exactly snippets of code, I am not the one that needs to prove that the MS "quick fix" is working correctly now.
Dude I think this kind of AI can be done right, I am not convinced MS did it correct because they already outputed copyrighted code and on top of that they avoided to put their ass on the line. I don't hate the idea of copilot, just the bad implementation of it.
MS can't use this excuse, if some dude uploads Windows code or say GTA6 code on GitHub I can't just use it and point you to the GitHub repo and ask you to figure it out with that person.
MS should either
1 tell you that the code is BSD,GPL, MIT depending on what is derived from
2 tell you that is not derived but some new code, but at the same time put his fat ass on the line and also put proprietary code in the mix, like their code and private code on GitHub since they claim the output is not derived from the input.
But they did it half ways, they did not had the courage for 2 and they did not want to respect the MIT,GPL either
A service that is able to pop out a new piece that will be indistinguishable from the artist's style is going to be disastrous for hundreds of thousands of artists around the world. We can't hide, at least not in good faith, behind the concept of 'transformative' art.
Besides, there clearly hasn't been enough time for the legal system to catch up so arguments about the letter of the law might change too. If you look at musical cases you'll be surprised to see that the 'transformative' part is up to interpretation. Just ask the Verve...
And legal system is up to date in any material matter related to the topic. If is not, you not only need to provide a specific notable claim, but back it up with evidence to support it and state why it’s not addressed by current legal code, case law, etc.
Reddit-style argumentation about the literal aspect of the law is missing the point altogether. These new services are barely a few months old (or at least publicly available) and it remains to be seen how society will react. How can there even be case law if there is literally not enough time for it to appear?
The salient fact is that the living conditions of many artists is going to degrade. If you do not believe that or do not care about it, then we can agree to disagree, but personally I think it would be better to alleviate that problem in some way and to nudge the law towards supporting this.
You again refuse mention verbatim what legal issues are new that are not covered by existing laws. Law by definition will always have a latent space that’s beyond the law, it is called extralegal, and one of the many reasons in common law that judges exists. Something being extralegal does neither make it illegal, nor shows a law is currently imperfect. Laws and judges further do not represent some idle truth, but a system for bring disagreements to an end without the need to exceed the monopoly governments have on violence as a means of dispute resolution; which is why criminals resort to violence as a means of dispute resolution, because they are frequently unable to resolve disputes using the law.
And then, you inject a yet another completely different topic, for which disagree, that being the impact of AI of professional artists, but as I said before, you lack the sense to understand your response are not in good faith, to me oddly argumentative, and wish you the best.
Last comment is even against HN’s guidelines:
1. AI art is still extremely new, and it's unclear what direction the law will take regarding it (if any)
2. Artists are likely to be significantly impacted by this, therefore it is probable that there will be some debate about the legal environment for this AI art. It won't necessarily be in favor of the artists, though I personally hope it will.
3. The law is hopefully aimed at making society better in some way. The definition of that is relative of course, but the spirit of the law is important, otherwise the entire edifice is pointless.
4. The AI art could be argued to be classified as merely transformative, but is regular transformative art likely to have the same impact on artists as AI art? Intuitively, we can probably tell that's not the case because the amount of effort and talent needed to generate AI art is incredibly low compared to the regular version and therefore the output will be vastly higher. The distribution, ease of production and other factors makes this an altogether different problem.
5. The prospect of AI art seems to be a bigger problem for artists than copyright infringement, because it can nuke sales/commissions or even types of artists (such as concept illustrators) in a way copyright infringement cannot. The market power of the 2D artist, which is the hard to reproduce ability to produce things in a very specific style, is gone.
6. Prompt users will deliberately use a reference to specific artists because they expect the result will be better than without. They understand that they are implicitly piggybacking on the work of the artist that led to the style being available in the model, to then use it for their own ends. If the artist had not worked in the way they did, the style would not be easily available in a few seconds. You would have to train for a long time to be able to replicate it, and your output would be very low (hence unproblematic)
So in the end you have something that will greatly impact artists, more so than most threats to intellectual property that are currently covered by the law, and that leads me to believe that the law will change, and if it does not, that it should change. Naturally, it is possible to disagree.
Seems like perhaps copyright is more complicated than folks defending AI Art believe it is.
And frankly, unless you are a judge overseeing a specific copyright infringement case, you have no basis with which to make these assertions.
Case you referenced is scheduled for hearing on January 2023 and I don’t see any reason to review it until it’s finalized.
If I am missing something, please feel free to explicitly state what and why it is related to current topic.
The model doesn't store or reproduce anyone's art, maybe except with some intentional img2img prompting. And styles aren't copyrightable.
There's some really impressive stuff being created out there, and it's still very early days. People take sets of personal images, train a mini model on that and use that model to prompt Stable Diffusion to create something that it wasn't even trained on. There are interesting short animations, etc. It's a brave new world to explore.
Using the "by {artistname}" string in prompts does seem to produce a lot of stuff which could potentially be copyright violation - it'll even output a signature/initials fairly frequently.
In the next year or two I'm sure we'll see some a refined version of something like img2img integrated into Photoshop, maybe Adobe will finally justify their "Cloud" approach with a big stack of GPUs in a data center.
Different tools, brush vs keyboard, but similar in that you need vision of what you want to create with those tools.
The more you have, the better your AI art will be. You should try sitting at a prompt and making something compelling.
I suppose what bothers me is the idea of a prompt user seeing themselves as an artist, not because I want to gatekeep people's minds but because that misunderstands the nature of the technology. A visitor to an AI art gallery (I do mean material that is 100% the direct result of prompt generation) could be able to generate maybe not the exact same art but something of equivalent worth upon arriving at the door. The contribution of the human being or to be more precise the specific individual is almost immaterial. Sure, they might have come up with a prompt with many words but you could come up with any number of equivalent artworks in the same breath. The artist is wholly interchangeable in a way that did not exist before.
I'll admit that it does require cultural understanding of art and some understanding of how prompts work because you need to know the names of artists and styles to fit into the prompt. But beyond having a good vocabulary there's not that much more involved. By comparison, if we're talking about artwork that uses AI as a tool as opposed to being the entire thing, then it becomes much more interesting and is what I am excited about because there will still be meaningful human input in there.
I just read on reddit about https://aiprompt.io and I find it useful to kickstart a new generation. It's a good way to start and discover technics by serendipity.
It's not teaching you tricks, but it's great for beginners or when inspiration is lacking.
Curious, if you were to imagine what a truly useful tool would be like for teaching how to create prompts?
If you ask MidJourney for a "futuristic solder in armor," you'll get... something. If you ask for "Hardmesh retro futurist steampunk fallout 7 6 power armor head, hyper realistic, art gta 5 cover, official fanart behance hd artstation by jesper ejsing, by rhads, makoto shinkai and lois van baarle, ilya kuvshinov, ossdraws, that looks like it is from borderlands and by feng zhu and loish and laurie greasley, victo ngai, andreas rocha, john harris radiating a glowing aura global illumination ray tracing hdr"[0] you'll get something that is less obviously MidJourney.
0. I grabbed that prompt from lexica.art as something work-safe.