Post BA5O6AxYLXJRdq8oim by mirabilos@toot.mirbsd.org
(DIR) More posts by mirabilos@toot.mirbsd.org
(DIR) Post #BA5GJ3mDZ1hHUnTK2C by khm@hj.9fs.net
1 likes, 0 repeats
The fact that none of the directors resigned over this means they're all slopromised too
(DIR) Post #BA5Jfd6tAwbdJVav1k by mark@mastodon.fixermark.com
0 likes, 0 repeats
@dalias I mean, Independent of whether we donate to them they are correct, both in terms of the law and in terms of what the law is for. We can dislike it (though I would recommend we sit hard with precisely why we dislike it and consider all implications... Most ways to take power from AI companies via copyright put it right in the hands of the Disney corporation). But their understanding is correct as far as I understand.
(DIR) Post #BA5JfdL4KDws1UEF84 by mirabilos@toot.mirbsd.org
0 likes, 0 repeats
@mark @dalias they are not, and you are not.“AI” training and extruding is no more than a lossy compression followed by selective decompression. It is done by a deterministic algorithm, so no new work comes into being, but instead, it’s an automated edition of existing works, which is only permitted under licence.That easy.
(DIR) Post #BA5K5A7JasXOvEy6y0 by mark@mastodon.fixermark.com
0 likes, 0 repeats
@mirabilos @dalias Do you have a citation for that description of the technology? It doesn't match to my understanding at all, and I've been working my way through a course in writing them in my spare time.
(DIR) Post #BA5K6ue05mr3Vt5juK by mirabilos@toot.mirbsd.org
0 likes, 0 repeats
@mark @dalias the article in the long pinned post on my profile where someone built GPT in a 498-line SQL query was extremely eye-opening in how they function. The articles about how (before they used pre-filters in prompts to try to prevent it) researchers have extracted original works from the models by the megabytes, in a quality where it is even clearly distinguishable from where they have been stolen make that also clear.
(DIR) Post #BA5KPSwXJAzX6ObmSW by mirabilos@toot.mirbsd.org
0 likes, 0 repeats
@mark @dalias there’s also the link to microgpt, which includes the training part, in easy code, though of course highly reduced in size, to make it runnable in short time and without much resource usage.The SQL one used a pretrained model (to keep it smaller, a then-older GPT, but newer ones are merely larger, still the same technology), which proves that this is really how they work.
(DIR) Post #BA5KS1pcZ4w1szb0Fs by mirabilos@toot.mirbsd.org
0 likes, 0 repeats
@mark @dalias also @davidculley may have a few words to say on it, given he studied transformers (the T in GPT) before they were cool
(DIR) Post #BA5Kdajj85FVKLLv72 by mirabilos@toot.mirbsd.org
0 likes, 0 repeats
@mark @dalias @davidculley slop advocates (but also many people criticising it) say that they produce different results each time.They are still algorithms running on deterministic machines. What creates the difference is really a stream from a pseudorandom number generator; capture that, and you can have it recreate the same result each time. Everyone who studied CS for a semester or two can see that.
(DIR) Post #BA5MGdhu7PLPmvgQZU by mark@mastodon.fixermark.com
0 likes, 0 repeats
@mirabilos @dalias @davidculley So I might be wrong, I am not in fact a copyright lawyer... I don't think that determinism is the thing that determines whether something is transformative. I wrote a blog post on this a while ago. There's an algorithm I can plug into blender that will generate an infinite number of asteroids by iterating on a relatively simple mathematical description. While I might in fact be wrong, I am fairly certain that the output of such a generator is transformative enough to be copyrighted. If it's not, there's an awful lot of art out there that probably cannot withstand copyright scrutiny, since its creation is nothing more than a sequence of deterministic transformers applied to some root seed...https://blog.fixermark.com/posts/2022/on-ai-art/
(DIR) Post #BA5MGe0gzYNCjCTQrA by mirabilos@toot.mirbsd.org
0 likes, 0 repeats
@mark @dalias @davidculley no, it’s not transformative, because that can only apply to humans
(DIR) Post #BA5MT7QHSh91QPOHYm by mark@mastodon.fixermark.com
0 likes, 0 repeats
@mirabilos @dalias I don't dispute that they're a semi-deterministic algorithm (a proper one includes the heat factor, otherwise the responses collapse into quite predictable, which is not the desired outcome).... But something being generated by a deterministic algorithm can't be the distinction by which we remove transformative from the copyright test, can it? You can represent any function with a two-layer neural network with a non-linear transform in the middle; that doesn't imply that there's no such thing as a transformative work simply because there exists some neural network that could fully represent it.
(DIR) Post #BA5MT7ng3hrMayKy1o by mirabilos@toot.mirbsd.org
0 likes, 0 repeats
@mark @dalias “the heat factor” is just fancy-speak for an extra input (the PRNG, if even that). Capture and replay that, and you get the exact same output. It’s an algorithm running on a computer. Do you know how computers work?But something being generated by a deterministic algorithm can't be the distinction by which we remove transformative from the copyright test, can it?But it is!Copyright only applies for human creative works where “threshold of originality” (this english legal term has a major potential for misunderstanding, the german Schöpfungshöhe is much better) is reached, i.e. the work is an expression of the human creative person’s unique personality.
(DIR) Post #BA5MfyV7wbaJWEmuGW by mark@mastodon.fixermark.com
0 likes, 0 repeats
@dalias @mirabilos @davidculley This is one of the claims that the New York times made that I find most interesting. They are claiming that output from AI generated on their content replaces their product?I don't see how that can be the case. Their product is reporting on current events. There is no way for an LLM to do that by the nature of what an LLM is.I'm sure if I can carve out the time to read their complaint in detail it will be more clear. To a cursory reading in the Reason article, that claim makes no sense, but the case passed basic scrutiny so I have to assume that summary compresses the complaint too far.Because the only way I can imagine one would consider them competing is in raw attention, and by that standard The Daily Show competes with the New York Times.
(DIR) Post #BA5Mfyk13FUiGPknTM by mirabilos@toot.mirbsd.org
0 likes, 0 repeats
@mark @dalias @davidculley I’m not arguing for the new york times. In fact, media are usually part of the problem.I’m arguing for humans, and that the existing laws suffice, if properly enforced.But the fashbros behind “AI” ignore consent anyway, anywhere they like.I’m noticing that you are here investing a significant effort into defending fashtech, while at the same time admitting you don’t know enough, instead of reading up on topics. Be careful.
(DIR) Post #BA5NLNPCFt6idCkuuW by mark@mastodon.fixermark.com
0 likes, 0 repeats
@mirabilos @dalias Do you know how computers work?Gosh I hope so, or this computer science degree and following career has been an elaborate fabrication. ;)I think you might be flipping the logic arrow on this problem. You seem to be implying that it's created by a deterministic algorithm and therefore it can't be transformative. Unfortunately, deterministic algorithms can be applied to vast, vast quantities of copyrightable material. Proven copyrightable material, like this isn't even up for debate. That's the universality of computing. So clearly, you can't reduce the question to "It's just a deterministic algorithm." That's too broad a constraint. The questions that the law addresses are the ultimate purpose of the work, whether the changes enacted upon it make it an entirely different kind of thing. And I think the law argues, and I would agree with it, that using the content of the New York Times to train a large language model (along with content from other sources) is self-evidently transformative. I can't pick up a New York Times and shake it and have a novel New York Times shaped article just fall out of it. Clearly, feeding the content through a training algorithm is doing something significant here. Plus, at the end of the day the purpose of copyright in the United States is the promotion of science and the useful arts. That's the lens through which all of this is viewed. And the courts broadly seem to be coming down on the side of LLM-AI being exactly the kind of science that copyright and patent are intended to promote.
(DIR) Post #BA5NLNf9IZrrQgDem8 by mirabilos@toot.mirbsd.org
0 likes, 0 repeats
@mark @daliasYou seem to be implying that it's created by a deterministic algorithm and therefore it can't be transformative.Yes. The law says so.Transformative use creates new works. Only humans can create works.Even the american courts decided that part of the discussion already.
(DIR) Post #BA5NwidmGZIrCoNkXY by mark@mastodon.fixermark.com
0 likes, 0 repeats
@dalias @mirabilos And yet Andy Warhol could sell silk screens of other people's artworks composited together.Which is probably an interesting stress point to consider the topic. There was a case of one of Warhol's works not being considered sufficiently transformative to remove copyright from the artist that he built a silk screen on. Apparently, a significant issue there was the alternate use space being competitive - the Warhol estate wasn't hanging the work on a wall in an art gallery, it was using it to illustrate an article, the sort of thing the original artist did for a living. https://www.npr.org/2023/05/18/1176881182/supreme-court-sides-against-andy-warhol-foundation-in-copyright-infringement-casSo does it matter if I am trying to make up articles the New York Times didn't write or if I'm trying to make up stories involving a newspaper and I want the llm to generate believably newspaper-shaped fake headlines in my story? I don't know honestly. I'm going to be interested to see where the courts come down on this kind of thing.
(DIR) Post #BA5Nwit1LtUpy5VvIe by mirabilos@toot.mirbsd.org
0 likes, 0 repeats
@mark @daliasAndy Warhol could sell silk screens of other people's artworks composited together.There are, of course, several sections in the law under which a human creative artist can do that. Under licence, under parody or pastiche, to public domain works, etc.But your LLM is not Andy Warhol. It is an algorithm in the CS definition.
(DIR) Post #BA5O6AhbIqYIqMg4rA by mark@mastodon.fixermark.com
0 likes, 0 repeats
@mirabilos @dalias last I checked (and this space is moving quite quickly so this might have changed), that put us in a space where it's possible that the output of llms is not copyrightable. ... Not in a space where the llm cannot be created or used. Even the recent settlement regarding training on copyrighted books hinged upon piracy to create the collection of training data, not the use of the training data for generating the llm.
(DIR) Post #BA5O6AxYLXJRdq8oim by mirabilos@toot.mirbsd.org
0 likes, 0 repeats
@mark @daliasthat put us in a space where it's possible that the output of llms is not copyrightable.Not as a new work, yes. Precisely that is what even american courts have already settled. Which we told you above.But, as dalias said, that does not prevent the copyright for F from applying.
(DIR) Post #BA5OGPSfHeuXVvZcEi by mark@mastodon.fixermark.com
0 likes, 0 repeats
@mirabilos @dalias Sure, but a human is generating the text that seeds the process. We certainly agree that while prompt engineer is probably not a legitimate job title, querying the llm with text of a particular shape requires more human input than rolling some dice and gluing together a random sequence of letters...This is the kind of causality chain thing the courts have to deal with that I find fascinating in terms of how the law works in general.
(DIR) Post #BA5OGPms4X4eWb1kjQ by mirabilos@toot.mirbsd.org
0 likes, 1 repeats
@mark @dalias if you have a database made from copyrighted works, say a paragraph per row, and the human creates the seed for an LFSR to select rows to concatenate together for an output, then I’m sorry to inform you but the human input is both insufficient for copyright protection of its own and insufficient to remove the copyright of said original works.
(DIR) Post #BA5OL7Lg3Op4moY9su by mark@mastodon.fixermark.com
0 likes, 0 repeats
@mirabilos @dalias I mean that's exactly the open question before the court right now right? Whether using F to create W (or, more precisely, the dynamic part inside W - the weights and biases) is itself transformative.
(DIR) Post #BA5OL7dOzV07fmqJVo by mirabilos@toot.mirbsd.org
0 likes, 0 repeats
@mark @dalias no, that is already settled. Even primates are not granted copyright. Only humans count as creative under that law.
(DIR) Post #BA5OZ8mfVvl3IArU6S by mark@mastodon.fixermark.com
0 likes, 0 repeats
@mirabilos @dalias Um... I'm pretty sure the jury is still out on that. Or rather, has not been empaneled on that topic yet.Warhol didn't have to use yellow in the Marilyn Monroe screen prints. That was a choice a human made. The rest of the process was really quite mechanical.
(DIR) Post #BA5OZ9A46wTOSjoAZU by mirabilos@toot.mirbsd.org
0 likes, 0 repeats
@mark @dalias the prompts do not become part of the output, and they are insignificant (size-wise) against the “model” as a whole, so no
(DIR) Post #BA5PIuMDU4t9ryJQsy by mark@mastodon.fixermark.com
0 likes, 0 repeats
@pfriedma @dalias @davidculley @mirabilos And it fails at that because that was never its purpose. It's purpose is delineated in the Constitution in one sentence. It's to strengthen the country by incentivizing people to build better cool shit. In practice, that frequently translates to protecting a temporary exclusive monopoly on something that is otherwise only protected by keeping it to yourself and never telling another soul. ... But I think we might have lost the thread of what the goal of the law is if we start using copyright to roadblock a thread of research that has been going on since, depending on how you count, the '70s... Because all of a sudden it bore fruit.
(DIR) Post #BA5PIufMKuCWpLGiiu by mirabilos@toot.mirbsd.org
0 likes, 0 repeats
@mark @pfriedma @dalias @davidculley I’d not mind if we hindered progress of tech, at this point.But, now that you’ve shown your true colours, there’s nothing preventing me from blocking defederating you.
(DIR) Post #BA5PyQa2aJN9gEZonI by mirabilos@toot.mirbsd.org
0 likes, 0 repeats
@pfriedma @dalias @davidculley (tbh surprised me to see sunk cost fallacy in there, but I guess in some people’s eyes the cost weighs worse than the adverse effects, be they psychological or ecological; ethical seems to be entirely uninteresting anyway)
(DIR) Post #BA5QwpWey0siQEnUxc by mirabilos@toot.mirbsd.org
0 likes, 0 repeats
(and, let’s face it, artists often get away doing things that are copyright violations because they don’t get sued over it, if this is of mutual interest and not to the detriment of the original authors)(though this opens the question whether a licence can come into existence from concludent action… or inaction…)@dalias
(DIR) Post #BA5R6QIjbeHV5LrHMW by mirabilos@toot.mirbsd.org
0 likes, 0 repeats
(though that’s not the point any more)(artists can also do this by getting permission through a performing rights society often, I’m not an expert on the still-under-copyright parts of music, but I think this is how cover bands do it)@dalias