@@royaltyfreemusiccollective8662 You know that AI is about to take over most if not all industries, right? Music is just a small part of the way AI will reshape society and the economical system.
@@royaltyfreemusiccollective8662 The royalty free music industry probably has the most to lose here, but I think it will survive. The convenience and sense of security that comes with using a platform like Epidemic Sound is pretty compelling. I'd still be using them for my RUclips projects if it didn't violate thier TOS. For most creators, the easiest path to quality stock materials for thier videos is going to win the day.
Interesting is the thing about the Musical "Tags" - the AI learns like "phrases" and something like a ususal element to a kind of music.. Yes - of course - if you tell an AI to cover a song - it can make a Coversong, and if you ask AI to imitate - it can do this.. Often - it will not, because the programmers add filters.. (and I think this flters will get better.. because you can train an AI also to recognize, if somethin is to similar to copyrighted material, like it is used for example to find plagiates in science)
Udio and Suno haven't published their exact algorithms, but they're likely similar to MusicLM, which uses a SoundStream codec at the heart of its feature set. You can't avoid sample clearance by, say, encoding the sample as an MP3 and then decoding it. There's been examples of Suno spitting out producer tags that sound like samples, much like the Getty images watermark showing up in Midjourney images.
Quite accurate. I have 2 comments. Usually, the STFT transforms applied to the input are lossless and reversible (under certain conditons). So there is no actual loss of information compared to using the original time domain wav file. However, a fully reversible transform might be too verbose (i.e. contains too much correlated data), that it might be easier to use a lossy transform that just focuses on the main thing. The reconstruction process, how to go from AI "thoughts" to audio can usually be a different tool. For example, you can train a wavenet on time domain wav files ( from commercial music), to decode whathever the STFT encoder extracted into audio. That said, it is not clear to me how exactly these commercial tools work. I suspect its a purely diffusion based model, that learns how to denoise STFT transforms, based on some conditional input (such as the prompt).
Forget sampling. Audio fingerprinting shows that the new tracks contain at least fragments of an original work, or in a lot of cases a DERIVATIVE work of the original. This breaks current copyright law regardless of sampling. Also, you don't necessarily need to include references in the prompts to create these. Hence, the tools are creating these, not the people. and even if the people were, the tool is still copying original content to create a derivative work. This lawsuit WILL weed out plagiarism.
I've never asked the prompt to create anything that is copyrighted, and it never has. I use AI "tracks" in large quantities and have had no issues with Content ID. I'm much more likely to have hassles with music created by humans, to be honest, but the human created music is much better. I would still use human created music if I could do so legally and didn't have a need for so many tracks. Thankfully, even if Suno and/or Udio are taken down, other AI models are immune to similar lawsuits. AI music is here to stay, and it has its place. However, the music is not very good! Also, according to the legal precedent established with AI visual art, it can't even be copyrighted! The statement that derivative works violate copyright is simply not true. If you use small enough segments and/or the work is sufficiently transformative, it's perfectly legal. For example, if I cut up a newspaper article and re-arrange the letters to say something new, I am not violating copyright even though I started with a copyrighted work. If Suno used sufficiently small segments of audio data to train the model they could be in the clear. Until a trial date is set I don't even take the lawsuit seriously.
i like you. you got my sub. You are one of the only youtubers I have seen that seems to have a good grip on whats going on with the copyright infringements and copying. I've made over 3000 songs that I considered good enough to publish to my youtube, and i made a lot more than that, that weren't good enough lol. And only a handful of them are what I would call similar to existing material. Most of the time I sit down at my own piano and record my own starting beats and vocals, and I input that to start my songs and the music you get is for sure not a rip off of other material. Now, I can also take existing content, put it into adobe premier and add a small distort to it, speed it up to 112% and then I can make a song with that and then it for sure is ripping off content. But again this is how it is used. And once this thing makes enough music. They are simply going to retrain the whole thing on nothing but original AI made music. so once they train it on a dataset that only contains 100% ok to use content, then they won't have to worry. and we all know this about the music industry trying to control everything. if they win these lawsuits, I guarantee you suno releases the model data :P Suno is already open source. The only thing we don't have access to is the new updated model data with the music trained, and the proprietary software they are using to link various models of voices and music. If they go down, all that stuff is getting released to the open source public.... guaranteed :) I run old suno on my live stream for my AI avatar. It works great. Nothing bad will come of any of this except for better music and people having the power to make it... I can't stand all these anti ai music youtubers. and the people that don't understand...
U.K. court already ruled it illegal to train A.i. machines without consent from the copyright holder and america will soon follow. And thank goodness. this tech is cool but working with stolen data is not cool. its unethical to sell these products when they stole the source material. edit: plenty bad can come from this. the erasure of human artist. no matter how original an artist believes they are if a.i. companies are allowed to train on any music we are cooked. You could spend 6 months developing a unique sound and put your album out tuesday, by thursday morning a.i. has already generated 3000 tracks that sound just like it.
Copyright law is designed to protect original works of authorship that are fixed in a tangible medium. However, not all elements of music are eligible for copyright protection. Here's a detailed breakdown of what can and cannot be copyrighted in terms of styles, genres, and beats: ## Styles and Genres **Styles and genres** of music cannot be copyrighted. These are considered broad categories that encompass many works and are not specific enough to be protected by copyright law. The idea/expression dichotomy in copyright law means that while specific expressions of ideas can be protected, the ideas themselves cannot. Styles and genres fall into the category of ideas rather than expressions[1]. ## Beats **Beats** can be copyrighted if they are original and fixed in a tangible medium. This means that if you create a unique beat and record it, it is protected by copyright law. Here are the key points regarding beats: - **Originality**: The beat must be an original creation. Simply using common rhythms or patterns that are widely used in a genre may not qualify for copyright protection[3]. - **Fixation**: The beat must be fixed in a tangible medium, such as a digital recording[2]. - **Registration**: While copyright protection exists from the moment of fixation, registering your beat with the copyright office provides legal benefits, such as the ability to sue for statutory damages and attorney's fees in case of infringement[5]. ## Specific Elements of Music Certain elements of music can be copyrighted, while others cannot: - **Melody**: The specific sequence of notes that make up a melody can be copyrighted[3]. - **Lyrics**: The words of a song are protected by copyright[3]. - **Sound Recording**: The actual recording of a performance can be copyrighted separately from the musical composition[2]. However, some elements are not eligible for copyright protection: - **Chord Progressions**: Common chord progressions are not protected because they are considered building blocks of music used across many songs and genres[3]. - **Rhythms and Grooves**: Basic rhythms and grooves are also not protected for the same reason[3]. - **Song Structures**: Common song structures (e.g., verse-chorus form) are not copyrightable[3]. ## Practical Considerations If you create beats and want to protect them, it is advisable to: - **Register Your Work**: Register your beats with the copyright office to ensure you have legal proof of ownership[5]. - **Use Contracts**: When licensing your beats to others, use contracts to specify the terms of use, including whether the beat can be altered or monetized[4]. In summary, while styles and genres cannot be copyrighted, original beats can be protected if they meet the criteria of originality and fixation. Understanding these distinctions is crucial for navigating the legal landscape of music copyright. Citations: [1] abovethelaw.com/2018/03/blurred-lines-can-you-copy-a-music-genre/ [2] www.copyright.gov/engage/musicians/ [3] support.easysong.com/hc/en-us/articles/1500009595681-What-Parts-of-Music-Can-Be-Copyrighted-and-What-Can-t [4] www.reddit.com/r/makinghiphop/comments/bkcr9g/should_i_copyright_my_beats/ [5] unison.audio/how-to-copyright-beats/ [6] creativecommons.org/2023/03/23/the-complex-world-of-style-copyright-and-generative-ai/
@@YoPaulieMusic It has to do with your opening statement. It has to do with capitalism and how almost every human invention is turned into a commodity that makes the rich richer and the poor poorer. It has to do with the life of artists and inventors and how corporations suck them dry and spit them out. And corporations promising how their product will make the world a better place and all the sheep following along.
@@royaltyfreemusiccollective8662 Then the question is: does it? I say it absolutely does not... especially based on the samples listed in the legal brief. The market is saturated, there are plenty of covers and derivative works out there, and a whole lot of bad music. These tools do not pose a threat to creators. Did those creators pose a threat from the musicians that they were inspired by? Nope. Did they pay them before learning their music? Nope. This case is nothing more than record companies trying to get their piece of the pie... another piece of the pie.
@@TwoWheelTherapy-TX Yes it likely is the big record companies protecting their own interests but it also sets a precident that just because something is publicly available doesn't mean you can design an AI mimic it without permission. There's no doubt that Suno and Udio's models have had a huge leg up in quality because of their illegal training data. With the current rate of progress they will be capable of competing with human made music soon and ther eare definitely areas where this will happen unless it's legislated against .
How do you feel about AI that is specifically designed to copy existing peoples voices? Both spoken and sung. Do you think artists should have the right to control whether their own original music is included in AI training sets? Should copyright exist on your own voice or is it OK for people to copy it exactly and use your name as an identifier?
I am against this concept, unless the person being copied is OK with it. Name, image, likeness… Those should protected. But, while some music might be identified with specific people, it is still separate from the person, and only lyrics or melodies can be protected.
@@Lantertronics He said that if they were working with labels that he probably wouldn't have invested. That's not conclusive. I found an article that goes into some detail about the ethical training practices that "set Suno apart," but have not been able to verify the source. It still remains an open question, imo. It will be interesting to see what happens in discovery (assuming the lawsuit goes that far.) They are entitled to the presumption of innocence, and to protect trade secrets, so they don't actually owe us disclosure unless it is court-ordered.
Early in training when outputs are exact copies perhaps. But as soon as it learns to to make an original work, you need to to go to copywritten work or it will not learn genres and the song quality will not be there. You really need lots of music for it to learn to generalize. Also more complex lyrics tags would help. [tag] (producer tag) [into] for example would prevent the producer tag problem. Lack of descriptive tags in lyrics limits communication with the AI. Needs to be more like HTML.
I think they made the tools to make money. That much is obvious. If a neural net is overtrained, resulting in overfitting, or the training data set is badly curated and skewed then it has a tendency to produce outputs that were in its training set. You're presuming that these companies took the time to carefully curate the data and not to overtrain the neural net. I don't think they took any care in doing either of those things. They rushed the products to market to make money and now they are reaping the consequences of that. Their lack of care and due diligence could be their downfall. The other darker alternative is that their tools are actually utilizing a hybrid approach and they are actually faking part of it before they use their earned profits to advance the technology further and closer to where they actually want it to be. As far as what could possibly happen if the courts rule against them. I would say that the court might force them down a non commercial usage path. So people can use it for generating music that they can't make money off. Interesting times indeed.
If human intellect can listen to a variety of songs and learn from that how to create original new songs, why wouldn't the much more intelligent AI also be able to simply listen and learn? We don't accuse humans who are "influenced by" such and such an artist of stealing, why do the same to AI, other than a defense mechanism against change? If AI or any human does actually create a copyright violating song, they simply compare the two songs in court and if they do sound the same, an obvious violation is proven. People are generally not learned enough to understand how AI actually works.
You'd think they would want to train it on copyrighted music, in order for the AI to know what NOT to make. Also, as with Art, you should be proud if your music influences AI music, because it can report whose music it was that influenced the music and we can use that report to show how popular different musicians music styles are. That can become a way to measure how much your music is worth.
Let's say a writer, is rewriting his material in lyrics and decided to put a sound under it. I don't see the copyright issue then. So it gives other People who don't are musicians to create music. So I don't see a problem.
Exactly! In fact, the team who put together the Suno lawsuit clearly violated copyright law, and the Suno TOS, in order to generate the examples in the first place. That is the ONLY way those examples were created. IMO this should invalidate the entire lawsuit.
Idk, I think it's a tool that can be used for that. Like many other tools. But since it's also a platform hosting music, they might need to get better at having filters for the output.
@@NoidoDev A guitar is also a tool that can also be used to violate copyright. But you have a point; that the output results are hosted on the platform. A combination of filters, and, most importantly, a system for quickly responding to takedown requests, should be adequate; it works for RUclips. Those examples for the lawsuit (that you've played on other videos) are not at all compelling, though. That, and the accusations of malice, just make them look cheap, petty and dishonest.
breaking tos in this case just means the company wont stand behind a commercial release and DUH they dont want you exposing the fact that they trained on copy written material. What is generated can still be use in court as evidence thou. The court case is about the data and how they obtained it. literally everything generated can be used as evidence.
here is an example of what I am doing. I wrote this song about my sister and X brother in law. They where married for 22 1/2 years, they got divorced because they couldn't get along anymore and they divorced. He called her up and wondered why they even got divorced in the first place. Now he wont even talk to her, even on the phone. Like I said in a recent comment I am doing all my own lyrics, so what they are trying to do is ridiculous with this law suit. Like I say I am writing all my own lyrics. ruclips.net/video/VKUfGk2I9cM/видео.html sorry for posting a link, but this is what I am doing.
If it wasn't samples then why is every recording company suing? F*** A.I sounds horrible! And sorry if you say you wrote something on A.I you're a liar!
Well, if AI sounds horrible, then it's obvioously not sampling, right? ;) Record companies are suing over money, ownership and control.... they want the revenue, so I bet they reach some sort of "agreement" so that the record companies get paid. Sampling is a very specific thing, and AI is not sampling. AI learns how music works, then creates new audio files based on the rules it learned and the prompts it is given. I can explain it to you, but I can't understand it for you.
they are suing because suno and adio trained their machines on copy written music with out permission. U.K. has already ruled this illegal. america will catch up soon
@@bobby4f "U.K. has already ruled this illegal." A dangerous precedent that learning is illegal. If you make music who do you need to pay, as human learn from copywritten music they listen to also.
Lol WTF Paulie. You are accusing the end user of acting unethically with their prompts and breaking the TOS of a product that is entirely built on an unethically sourced dataset. Can you not see the hypocrisy here? It doesn't matter whether it fits the technical definition of a sampler or not, what matters is the output. If walks like a duck and quacks like a duck, guess what ?....
The two concepts in your comment are not mutually exclusive. We certainly know that these companies are using copyright material to train their tools… Doesn’t mean people pasting copyrighted lyrics aren’t also violating terms and conditions. I’m just pointing out reality, and a little bit of irony for all of this rage against the machine. :-)
This boils down to the point that they will need to use content ID at some point. For the output. I guess legally it's sufficient that they are trying their best. But this is somewhat unrelated to the claim that they should not be allowed to train on other people's music. The anti AI people switch back and forth between those arguments and try to muddle the water.
@@NoidoDev I'm pretty sure Udio has some filter that runs after it creates a tracks that tries to see if it's infringing on some existing track, and that they changed the threshold after the lawsuit hit. I used to be able to get spot on Bjork impressions, and after the lawsuit it's much harder, and when I was getting Johnny Cash impressions, it would sometimes punt and say that the resulting track violated its conditions and saying to try again.
AI is the future, it's incredible how short sighted and close minded majority of people seem to be, keep up the nice videos man!
@@TheDailySnack Do you actually earn a living from your music?
@@royaltyfreemusiccollective8662 You know that AI is about to take over most if not all industries, right? Music is just a small part of the way AI will reshape society and the economical system.
@@royaltyfreemusiccollective8662
This is completely irrelevant as an argument.
@@royaltyfreemusiccollective8662 The royalty free music industry probably has the most to lose here, but I think it will survive. The convenience and sense of security that comes with using a platform like Epidemic Sound is pretty compelling. I'd still be using them for my RUclips projects if it didn't violate thier TOS. For most creators, the easiest path to quality stock materials for thier videos is going to win the day.
@@NoidoDev Unless you have skin in the game game you can't truly understand the argument
Interesting is the thing about the Musical "Tags" - the AI learns like "phrases" and something like a ususal element to a kind of music..
Yes - of course - if you tell an AI to cover a song - it can make a Coversong, and if you ask AI to imitate - it can do this..
Often - it will not, because the programmers add filters.. (and I think this flters will get better.. because you can train an AI also to recognize, if somethin is to similar to copyrighted material, like it is used for example to find plagiates in science)
Udio and Suno haven't published their exact algorithms, but they're likely similar to MusicLM, which uses a SoundStream codec at the heart of its feature set. You can't avoid sample clearance by, say, encoding the sample as an MP3 and then decoding it. There's been examples of Suno spitting out producer tags that sound like samples, much like the Getty images watermark showing up in Midjourney images.
Quite accurate. I have 2 comments.
Usually, the STFT transforms applied to the input are lossless and reversible (under certain conditons). So there is no actual loss of information compared to using the original time domain wav file. However, a fully reversible transform might be too verbose (i.e. contains too much correlated data), that it might be easier to use a lossy transform that just focuses on the main thing.
The reconstruction process, how to go from AI "thoughts" to audio can usually be a different tool. For example, you can train a wavenet on time domain wav files ( from commercial music), to decode whathever the STFT encoder extracted into audio. That said, it is not clear to me how exactly these commercial tools work. I suspect its a purely diffusion based model, that learns how to denoise STFT transforms, based on some conditional input (such as the prompt).
Fantastic comment, please come back and comment more! :)
Forget sampling. Audio fingerprinting shows that the new tracks contain at least fragments of an original work, or in a lot of cases a DERIVATIVE work of the original. This breaks current copyright law regardless of sampling. Also, you don't necessarily need to include references in the prompts to create these. Hence, the tools are creating these, not the people. and even if the people were, the tool is still copying original content to create a derivative work. This lawsuit WILL weed out plagiarism.
I've never asked the prompt to create anything that is copyrighted, and it never has. I use AI "tracks" in large quantities and have had no issues with Content ID. I'm much more likely to have hassles with music created by humans, to be honest, but the human created music is much better. I would still use human created music if I could do so legally and didn't have a need for so many tracks. Thankfully, even if Suno and/or Udio are taken down, other AI models are immune to similar lawsuits. AI music is here to stay, and it has its place.
However, the music is not very good!
Also, according to the legal precedent established with AI visual art, it can't even be copyrighted!
The statement that derivative works violate copyright is simply not true. If you use small enough segments and/or the work is sufficiently transformative, it's perfectly legal. For example, if I cut up a newspaper article and re-arrange the letters to say something new, I am not violating copyright even though I started with a copyrighted work. If Suno used sufficiently small segments of audio data to train the model they could be in the clear. Until a trial date is set I don't even take the lawsuit seriously.
i like you. you got my sub. You are one of the only youtubers I have seen that seems to have a good grip on whats going on with the copyright infringements and copying. I've made over 3000 songs that I considered good enough to publish to my youtube, and i made a lot more than that, that weren't good enough lol. And only a handful of them are what I would call similar to existing material. Most of the time I sit down at my own piano and record my own starting beats and vocals, and I input that to start my songs and the music you get is for sure not a rip off of other material. Now, I can also take existing content, put it into adobe premier and add a small distort to it, speed it up to 112% and then I can make a song with that and then it for sure is ripping off content. But again this is how it is used. And once this thing makes enough music. They are simply going to retrain the whole thing on nothing but original AI made music. so once they train it on a dataset that only contains 100% ok to use content, then they won't have to worry. and we all know this about the music industry trying to control everything. if they win these lawsuits, I guarantee you suno releases the model data :P Suno is already open source. The only thing we don't have access to is the new updated model data with the music trained, and the proprietary software they are using to link various models of voices and music. If they go down, all that stuff is getting released to the open source public.... guaranteed :) I run old suno on my live stream for my AI avatar. It works great. Nothing bad will come of any of this except for better music and people having the power to make it... I can't stand all these anti ai music youtubers. and the people that don't understand...
U.K. court already ruled it illegal to train A.i. machines without consent from the copyright holder and america will soon follow. And thank goodness. this tech is cool but working with stolen data is not cool. its unethical to sell these products when they stole the source material.
edit: plenty bad can come from this. the erasure of human artist. no matter how original an artist believes they are if a.i. companies are allowed to train on any music we are cooked. You could spend 6 months developing a unique sound and put your album out tuesday, by thursday morning a.i. has already generated 3000 tracks that sound just like it.
Someone put in the lyrics to Rain of Castamere with "mellow trap" as the prompt and Suno decided to throw in a "Cash Money AP" producer tag.
Copyright law is designed to protect original works of authorship that are fixed in a tangible medium. However, not all elements of music are eligible for copyright protection. Here's a detailed breakdown of what can and cannot be copyrighted in terms of styles, genres, and beats:
## Styles and Genres
**Styles and genres** of music cannot be copyrighted. These are considered broad categories that encompass many works and are not specific enough to be protected by copyright law. The idea/expression dichotomy in copyright law means that while specific expressions of ideas can be protected, the ideas themselves cannot. Styles and genres fall into the category of ideas rather than expressions[1].
## Beats
**Beats** can be copyrighted if they are original and fixed in a tangible medium. This means that if you create a unique beat and record it, it is protected by copyright law. Here are the key points regarding beats:
- **Originality**: The beat must be an original creation. Simply using common rhythms or patterns that are widely used in a genre may not qualify for copyright protection[3].
- **Fixation**: The beat must be fixed in a tangible medium, such as a digital recording[2].
- **Registration**: While copyright protection exists from the moment of fixation, registering your beat with the copyright office provides legal benefits, such as the ability to sue for statutory damages and attorney's fees in case of infringement[5].
## Specific Elements of Music
Certain elements of music can be copyrighted, while others cannot:
- **Melody**: The specific sequence of notes that make up a melody can be copyrighted[3].
- **Lyrics**: The words of a song are protected by copyright[3].
- **Sound Recording**: The actual recording of a performance can be copyrighted separately from the musical composition[2].
However, some elements are not eligible for copyright protection:
- **Chord Progressions**: Common chord progressions are not protected because they are considered building blocks of music used across many songs and genres[3].
- **Rhythms and Grooves**: Basic rhythms and grooves are also not protected for the same reason[3].
- **Song Structures**: Common song structures (e.g., verse-chorus form) are not copyrightable[3].
## Practical Considerations
If you create beats and want to protect them, it is advisable to:
- **Register Your Work**: Register your beats with the copyright office to ensure you have legal proof of ownership[5].
- **Use Contracts**: When licensing your beats to others, use contracts to specify the terms of use, including whether the beat can be altered or monetized[4].
In summary, while styles and genres cannot be copyrighted, original beats can be protected if they meet the criteria of originality and fixation. Understanding these distinctions is crucial for navigating the legal landscape of music copyright.
Citations:
[1] abovethelaw.com/2018/03/blurred-lines-can-you-copy-a-music-genre/
[2] www.copyright.gov/engage/musicians/
[3] support.easysong.com/hc/en-us/articles/1500009595681-What-Parts-of-Music-Can-Be-Copyrighted-and-What-Can-t
[4] www.reddit.com/r/makinghiphop/comments/bkcr9g/should_i_copyright_my_beats/
[5] unison.audio/how-to-copyright-beats/
[6] creativecommons.org/2023/03/23/the-complex-world-of-style-copyright-and-generative-ai/
They are not creating these tools for the sole purpose of making money?
What does that have to do with any of this? ;). EVERY business is created with the goal of making money… but there are also other reasons.
@@YoPaulieMusic It has to do with your opening statement. It has to do with capitalism and how almost every human invention is turned into a commodity that makes the rich richer and the poor poorer. It has to do with the life of artists and inventors and how corporations suck them dry and spit them out. And corporations promising how their product will make the world a better place and all the sheep following along.
@@YoPaulieMusic It has everything to do with this if it takes away income from the creators that made the music in the first place
@@royaltyfreemusiccollective8662 Then the question is: does it? I say it absolutely does not... especially based on the samples listed in the legal brief. The market is saturated, there are plenty of covers and derivative works out there, and a whole lot of bad music. These tools do not pose a threat to creators. Did those creators pose a threat from the musicians that they were inspired by? Nope. Did they pay them before learning their music? Nope. This case is nothing more than record companies trying to get their piece of the pie... another piece of the pie.
@@TwoWheelTherapy-TX Yes it likely is the big record companies protecting their own interests but it also sets a precident that just because something is publicly available doesn't mean you can design an AI mimic it without permission. There's no doubt that Suno and Udio's models have had a huge leg up in quality because of their illegal training data. With the current rate of progress they will be capable of competing with human made music soon and ther eare definitely areas where this will happen unless it's legislated against
.
How do you feel about AI that is specifically designed to copy existing peoples voices? Both spoken and sung. Do you think artists should have the right to control whether their own original music is included in AI training sets? Should copyright exist on your own voice or is it OK for people to copy it exactly and use your name as an identifier?
I am against this concept, unless the person being copied is OK with it. Name, image, likeness… Those should protected. But, while some music might be identified with specific people, it is still separate from the person, and only lyrics or melodies can be protected.
People like you and I can see the AI Matrix, most people can't!
see beyond the matrix :P to the 5 seas (C's) of AI
I read that Suno was only using Creative Commons music.
@@SmallwoodMedia lol. Not true
Where did you get that idea? One of their investors literally admitted they were training off of music they didn't have a license for.
@@Lantertronics He said that if they were working with labels that he probably wouldn't have invested. That's not conclusive. I found an article that goes into some detail about the ethical training practices that "set Suno apart," but have not been able to verify the source. It still remains an open question, imo. It will be interesting to see what happens in discovery (assuming the lawsuit goes that far.) They are entitled to the presumption of innocence, and to protect trade secrets, so they don't actually owe us disclosure unless it is court-ordered.
Early in training when outputs are exact copies perhaps. But as soon as it learns to to make an original work, you need to to go to copywritten work or it will not learn genres and the song quality will not be there. You really need lots of music for it to learn to generalize.
Also more complex lyrics tags would help.
[tag]
(producer tag)
[into]
for example would prevent the producer tag problem. Lack of descriptive tags in lyrics limits communication with the AI. Needs to be more like HTML.
I think they made the tools to make money. That much is obvious. If a neural net is overtrained, resulting in overfitting, or the training data set is badly curated and skewed then it has a tendency to produce outputs that were in its training set. You're presuming that these companies took the time to carefully curate the data and not to overtrain the neural net. I don't think they took any care in doing either of those things. They rushed the products to market to make money and now they are reaping the consequences of that. Their lack of care and due diligence could be their downfall.
The other darker alternative is that their tools are actually utilizing a hybrid approach and they are actually faking part of it before they use their earned profits to advance the technology further and closer to where they actually want it to be.
As far as what could possibly happen if the courts rule against them. I would say that the court might force them down a non commercial usage path. So people can use it for generating music that they can't make money off.
Interesting times indeed.
If human intellect can listen to a variety of songs and learn from that how to create original new songs, why wouldn't the much more intelligent AI also be able to simply listen and learn?
We don't accuse humans who are "influenced by" such and such an artist of stealing, why do the same to AI, other than a defense mechanism against change? If AI or any human does actually create a copyright violating song, they simply compare the two songs in court and if they do sound the same, an obvious violation is proven. People are generally not learned enough to understand how AI actually works.
You'd think they would want to train it on copyrighted music, in order for the AI to know what NOT to make. Also, as with Art, you should be proud if your music influences AI music, because it can report whose music it was that influenced the music and we can use that report to show how popular different musicians music styles are. That can become a way to measure how much your music is worth.
Let's say a writer, is rewriting his material in lyrics and decided to put a sound under it. I don't see the copyright issue then. So it gives other People who don't are musicians to create music. So I don't see a problem.
Exactly! In fact, the team who put together the Suno lawsuit clearly violated copyright law, and the Suno TOS, in order to generate the examples in the first place. That is the ONLY way those examples were created. IMO this should invalidate the entire lawsuit.
Idk, I think it's a tool that can be used for that. Like many other tools. But since it's also a platform hosting music, they might need to get better at having filters for the output.
@@NoidoDev A guitar is also a tool that can also be used to violate copyright. But you have a point; that the output results are hosted on the platform. A combination of filters, and, most importantly, a system for quickly responding to takedown requests, should be adequate; it works for RUclips. Those examples for the lawsuit (that you've played on other videos) are not at all compelling, though. That, and the accusations of malice, just make them look cheap, petty and dishonest.
breaking tos in this case just means the company wont stand behind a commercial release and DUH they dont want you exposing the fact that they trained on copy written material. What is generated can still be use in court as evidence thou. The court case is about the data and how they obtained it. literally everything generated can be used as evidence.
here is an example of what I am doing. I wrote this song about my sister and X brother in law. They where married for 22 1/2 years, they got divorced because they couldn't get along anymore and they divorced. He called her up and wondered why they even got divorced in the first place. Now he wont even talk to her, even on the phone. Like I said in a recent comment I am doing all my own lyrics, so what they are trying to do is ridiculous with this law suit. Like I say I am writing all my own lyrics. ruclips.net/video/VKUfGk2I9cM/видео.html sorry for posting a link, but this is what I am doing.
Not to mention, if you manage to create a copyrighted song, you as the creator are to blame, not the AI.
😂
If it wasn't samples then why is every recording company suing? F*** A.I sounds horrible! And sorry if you say you wrote something on A.I you're a liar!
Well, if AI sounds horrible, then it's obvioously not sampling, right? ;) Record companies are suing over money, ownership and control.... they want the revenue, so I bet they reach some sort of "agreement" so that the record companies get paid. Sampling is a very specific thing, and AI is not sampling. AI learns how music works, then creates new audio files based on the rules it learned and the prompts it is given. I can explain it to you, but I can't understand it for you.
they are suing because suno and adio trained their machines on copy written music with out permission. U.K. has already ruled this illegal. america will catch up soon
@@bobby4f "U.K. has already ruled this illegal."
A dangerous precedent that learning is illegal. If you make music who do you need to pay, as human learn from copywritten music they listen to also.
Lol WTF Paulie. You are accusing the end user of acting unethically with their prompts and breaking the TOS of a product that is entirely built on an unethically sourced dataset. Can you not see the hypocrisy here? It doesn't matter whether it fits the technical definition of a sampler or not, what matters is the output. If walks like a duck and quacks like a duck, guess what ?....
It's an AI duck!
The two concepts in your comment are not mutually exclusive. We certainly know that these companies are using copyright material to train their tools… Doesn’t mean people pasting copyrighted lyrics aren’t also violating terms and conditions. I’m just pointing out reality, and a little bit of irony for all of this rage against the machine. :-)
@@YoPaulieMusic Nobody is listening to what the man is saying, "Recreation" is not sampling or copying!
This boils down to the point that they will need to use content ID at some point. For the output. I guess legally it's sufficient that they are trying their best.
But this is somewhat unrelated to the claim that they should not be allowed to train on other people's music.
The anti AI people switch back and forth between those arguments and try to muddle the water.
@@NoidoDev I'm pretty sure Udio has some filter that runs after it creates a tracks that tries to see if it's infringing on some existing track, and that they changed the threshold after the lawsuit hit. I used to be able to get spot on Bjork impressions, and after the lawsuit it's much harder, and when I was getting Johnny Cash impressions, it would sometimes punt and say that the resulting track violated its conditions and saying to try again.