Suno AI Caught Infringing Copyright Amidst Major Lawsuit | Udio Responds

Summary

この動画では、主要レコード会社がAI音楽生成プラットフォームYudoとSunoを著作権侵害で提訴している問題について詳しく説明しています。訴訟の中心は、これらのプラットフォームが無断で著作権保護された音楽を学習データとして使用し、その結果生成された音楽がオリジナル作品に似ているという点です。特に、プロデューサータグなどが含まれている例が示され、AIがどのようにしてこのような結果を生むのか、また著作権保有者がどのように関与すべきかについて議論しています。

Takeaways

• 🎵 AI 音楽技術の進歩：AIは現在、音楽業界で大きな変化をもたらしており、多くのアーティストがその技術を活用しています。
• 🤖 3つの主要レーベルの訴訟：ソニー、ワーナー、ユニバーサルは、AI 音楽プラットフォームの Yudo と Sunno に対して著作権侵害の訴訟を起こしました。
• 📚 AI トレーニングの合法性：レコードレーベルは、これらのプラットフォームが公開されている音楽をダウンロードしてAIに学習させ、その結果として著作権を持つ作品を模倣していると主張しています。
• 🚫 AIによる著作権侵害の禁止：Yudoは、AIが特定のアーティストの声を模倣することを禁止していますが、それでも模倣が行われている可能性があります。
• 🔍 AI モデルのトレーニング方法：AIは音楽の例から学び、音楽の基本的な要素を理解を深めていますが、著作権を持つ誰かの所有物ではありません。
• 📖 AI と著作権法：著作権法によれば、著作物は人によって作成されなければなりません。AIによって生成されたものには著作権がありません。
• 💡 AI 技術の未来：Yudoは、AIが新しいアイデアを反映する音楽を作成することを目指しており、著作権を持つ作品を再現することに興味はありません。
• 🎧 AI モデルによる音楽の生成：Sunnoのモデルは、特定のプロデューサータグを再現する音楽を生成することができ、それがどのように機能するかは不明です。
• 📌 AI 音楽の透明性の問題：Sunnoは、モデルのトレーニングに使用されたデータのソースや、そのライセンス状況について明らかにしていません。
• 📈 AI 技術の価値と著作権：高品質のデータがAIモデルの価値を高めるため、著作権を持つ音楽の所有者がこれらの新しい産業に貢献する余地があるかもしれません。
• 🤝 著作権保有者とAIプラットフォームの関係性：著作権侵害の訴訟が提起されたことで、著作権保有者とAIプラットフォームの関係性についての議論が進むことが期待されます。

Q & A

AI音乐生成プラットフォームが直面している法的問題とは何ですか？

AI音楽生成プラットフォームは、主に著作権侵害の問題に直面しています。3つの大規模なレーベルであるソニー、ワーナー、ユニバーサルが、AIプラットフォームが彼らの音楽を無断で使用し、AIをトレーニングさせたとして訴えを起こしています。

AIが人間の作成した著作権保護された音楽をトレーニングに使用することは許可されていますか？

現在の著作権法では、AIが人間の作成した著作権保護された音楽をトレーニングに使用することは許可されていません。AIが学習するためには、著作権を有する者が許可する必要があります。

AIが生成する音楽が派生作品となる条件は何ですか？

AIが生成する音楽が派生作品となるためには、元の作品に基づいて新しい創作物が作成され、オリジナルの著作権が新しい作品にも及びます。

yudoとsunnoのプラットフォームはどのように音楽を生成すると言われていますか？

yudoとsunnoは、AIが公開されている音楽を学習し、新しい音楽を生成する技術を開発しています。しかし、これらのプラットフォームは、トレーニングに使用した音楽のソースやライセンス状況を明かしていないため、著作権侵害の疑いがかかっています。

sunnoが獲得した1億2500万ドルの投資はどのような目的で使用される予定ですか？

sunnoは、獲得した1億2500万ドルの投資を、人間の創造力を高めるための技術の開発に使用する予定と発表しています。

AIが生成した音楽が元の音楽を模倣していると主張する根拠は何ですか？

AIが生成した音楽が元の音楽を模倣していると主張する根拠は、AIが生成した音楽が非常に人気のある音楽と非常に似ていること、またはそのプロデューサータグが同じであることです。

著作権法において、AIが生成した作品が著作権保護を受けるにはどのような条件が必要ですか？

著作権法において、AIが生成した作品が著作権保護を受けるには、その作品が人間の作者によって作成されたものである必要があります。

sunnoのモデルがCash Money APのプロデューサータグを生成した例は、どのようにして説明されますか？

sunnoのモデルがCash Money APのプロデューサータグを生成した例は、AIが元の音楽を学習し、その特徴を再現する能力があることを示すと同時に、著作権侵害の問題にもつながっている可能性があります。

AI音楽生成プラットフォームが直面している問題を解決するために、法的訴訟が行われる可能性はありますか？

はい、法的訴訟が行われることで、AI音楽生成プラットフォームが直面している著作権侵害の問題に対する解決策や規則が明確になる可能性があります。

著作権保護された音楽をAIのトレーニングに使用する際に、どのようなライセンスや同意が必要ですか？

著作権保護された音楽をAIのトレーニングに使用する際には、元の著作権保有者の許可を得る必要があります。また、適切なライセンスを取得することで、法的な問題を回避できます。

AI音楽生成プラットフォームが提供する音楽の品質は、どのように評価されますか？

AI音楽生成プラットフォームが提供する音楽の品質は、使用されたトレーニングデータの品質と、AI技術そのものの能力によって評価されます。高品質のデータと洗練された技術が、優れた音楽を生み出す可能性を高めます。

著作権保護された音楽のオーサーがAI音楽生成プラットフォームにどのような価値を提供できるでしょうか？

著作権保護された音楽のオーサーは、高品質の音楽データを提供し、AI音楽生成プラットフォームのトレーニングに使用することで、より優れた音楽生成モデルを開発する上で価値を提供できます。

Outlines

00:00

🚨 AI音楽の法的問題

AI技術を用いた音楽制作プラットフォーム「yudo」が、ソニー、ワーナー、ユニバーサルの3大レーベルによって著作権侵害の訴えを浴びた。これらのプラットフォームは、公開されている音楽をダウンロードしてAIをトレーニングし、その結果、既存のアーティストの音楽を模倣しているという主張がある。訴訟では、AIが人間の作成物に対して学習し、派生作品を生成するかどうかという2つの問題が争われている。また、yudoのビデオが訴訟文に直接引用されているという衝撃の事実も明らかにされている。

05:02

🎼 AIと音楽創作の未来

AIが音楽創作の分野で急速に進化し、さまざまなメディアに適用されている。yudoはアーティストを支援し、彼らが素晴らしい音楽を作るためのツールを提供することを目指している。過去2年間で、彼らは世界最大のミュージシャンのスタジオに座り、若手ソングライターと共同で歌詞を作り、何百万人ものユーザーが新しい音楽を創造するのを目の当たりにした。しかし、音楽業界の変更とそれに伴う不確実性も触れられている。yudoは、彼らの技術が音楽の新しいアイデアを反映する音楽を作成することを目指していると主張しており、著作権侵害の問題を避けるための最新のフィルタを実装し、改善を続けている。

10:05

🔊 AI生成音楽の奇妙な現象

Redditの投稿により、sunnoのモデルが予期せずプロデューサータグを生成していたことが明らかになった。ユーザーが「mellow trap」というプロンプトと「ゲームの王様」のサウンドトラックの歌詞を入力したところ、Cash Money APのプロデューサータグが生成された。これは偶然の一致ではなく、sunnoのモデルがCash Money APの音楽をトレーニングに使用していないにもかかわらず、そのタグを生成していることを示唆している。この現象は、sunnoのモデルが著作権保護された音声記録をどのように生成しているのかについて疑問を呈している。

15:07

🤖 AIモデルの学習プロセスと著作権

AIモデルが学習プロセスでコピーを保持するかどうか、またどのようにして著作権保護された音声を生成することができるかについて議論されている。専門家によると、AIは学習のためにストリーミングプラットフォームから学習することも可能だが、最高のトレーニングにはオリジナルのバージョンの音楽ファイルが必要なとされている。sunnoは、モデルが著作権保護された音声を再現することに興味がないと主張しており、その代わりに新しいアイデアを反映する音楽を作成することを目指している。しかし、著作権侵害の訴訟が提起されたことから、この問題が裁判所で争われることになる。

20:09

🏛 AI音楽プラットフォームの法的挑戦

著作権保護された音楽をAIトレーニングに使用した場合の法的課題について触れられており、特にsunnoやyudoのようなプラットフォームが注目されている。これらのプラットフォームは、高品質のデータが必要であると主張しており、そのデータがどこから来て、どのようなライセンス条件の下にあるかが不明な点が問題視されている。訴訟は、これらのプラットフォームが高品質の音楽をダウンロードし、AIにトレーニングさせた後、そのAIが新しい派生作品を作成している可能性があると主張している。著作権保有者は、これらの新興企業や産業において利益を享受する権利を有していると主張しており、その立場が裁判所で検証されることになる。

Keywords

💡AIミュージック

AIミュージックは、人工知能技術を用いて音楽を生成するプロセスを指します。このビデオでは、AIミュージックプラットフォームが法的問題に直面している状況を探求しており、その中核となるテーマの一つです。例えば、ビデオではAIが既存の音楽を分析し、新しい曲を作成する機能について触れています。

💡著作権侵害

著作権侵害は、他人の著作物を無断で使用することを指し、ビデオの中心的な問題です。ビデオでは、メジャーレーベルがAIミュージックプラットフォームに対して著作権侵害の訴訟を起こしたと報告しており、AIが著作権を持つ音楽を無断で使用していると主張しています。

💡メジャーレーベル

メジャーレーベルとは、音楽産業において最も影響力を持つ大きなレーベルを指します。ビデオでは、Sony、Warner、UniversalなどがAIミュージックプラットフォームを訴えることで提唱されており、彼らの音楽が無断で使用されていると主張しています。

💡AIの学習

AIの学習とは、AIが大量のデータからパターンを認識し、新しい音楽を作成する能力を獲得するプロセスです。ビデオでは、AIが公開されている音楽を通じて学習し、その影響下で新しい曲を生成する可能性があると議論されています。

💡派生作品

派生作品とは、既存の著作物をもとに作成された新しい著作物で、オリジナルの著作権が及ぶとされるものを指します。ビデオでは、AIが生成する音楽が既存の音楽を模倣しているため、派生作品として著作権侵害の問題が生じる可能性があると述べています。

💡プロデューサータグ

プロデューサータグは、特定の音楽トラックの開始部分に挿入されるオーディオ商標であり、そのトラックのプロデューサーを示します。ビデオでは、AIミュージックプラットフォームがプロデューサータグを無断で再現する例が示されており、これは著作権侵害の疑いがあるとされています。

💡サンプリング

サンプリングとは、既存の音源の一部を別の曲に取り込むことを指します。ビデオでは、AIミュージックプラットフォームがサンプリング技術を用いて、著作権保護された音源を無断で使用している可能性があると示唆されています。

💡ライセンス

ライセンスとは、著作物を特定の目的で使用する権限を得るプロセスです。ビデオでは、AIミュージックプラットフォームが音楽のライセンスを取得していない可能性があることが問題視されており、これは著作権侵害につながる可能性があります。

💡音楽データ

音楽データとは、AIが学習し音楽を生成するための音のサンプルや曲の集合を指します。ビデオでは、AIプラットフォームが高品質の音楽データを使用しているが、そのデータが著作権保護されており、正当なライセンスが取得されているかどうかが問題となっています。

💡透明性

透明性とは、情報やプロセスが明確で公開されている状態を指します。ビデオでは、AIミュージックプラットフォームが使用している音楽データのソースやライセンス状況が不明瞭で、その透明性の不足が著作権に関する懸念を強めています。

💡創造的な表現

創造的な表現とは、芸術的または音楽的な作品を作り上げる行為を指します。ビデオでは、AIミュージックプラットフォームがアーティストに新しい創造的な手段を提供する一方で、その創造性と著作権の境界が問題となっていると述べています。

Highlights

AI音乐领域中，三大唱片公司索尼、华纳和环球联合起诉AI音乐生成平台Yudo和Sunno。

这些平台被指控通过公开可访问的音乐资源训练AI，模仿并侵犯版权。

AI训练是否可使用人类创作的受版权保护的材料，以及AI生成的音乐是否仅是模仿。

音乐生成平台被指控生成与流行音乐极其相似甚至完全相同的音乐。

Yudo在诉讼中被直接引用，表明其视频对公众意识的提高有影响。

版权法规定，只有人类作者创作的作品才能获得版权保护。

Sunno宣布获得1.25亿美元新投资，用于增强人类创造力的技术。

Sunno和Yudo未公开其数据来源或是否获得了训练数据的授权。

Yudo声明强调AI作为创意表达工具的重要性，并分享了艺术家和用户如何使用其平台。

Yudo声称其AI模型通过学习大量音乐作品来理解音乐理念，而非复制受版权保护的作品。

Sunno的AI模型被指控生成了听起来像是采样了受版权保护的制作人标签的音乐。

Reddit用户分享了使用Sunno模型生成的带有熟悉制作人标签的音乐。

Sunno的AI模型似乎能够根据提示生成听起来像版权受保护的制作人标签。

讨论了AI音乐生成是否应该为使用版权受保护的音乐支付费用。

提出了关于AI音乐生成平台是否应与版权持有者共享利润的问题。

分析了AI音乐生成平台的透明度问题，以及它们如何处理训练数据和版权问题。

探讨了AI音乐生成技术的未来，以及它如何影响音乐产业和艺术家。

Transcripts

another one D we not allowed to

replicate the likeness the voice of

these different artists so yudo put out

a little statement let us offer some

insight on how our technology

works the biggest thing in AI music is

happening right now we have the three

major labels Sony Warner and Universal

all coming together to Sue sunno and

yudo all right so these are the biggest

AI music generation platforms and what

they're claiming I did the full

breakdown of the lawsuits they are

substantially the same okay but

basically what you know these record

labels are claiming is that these

platforms went on in scrubbed through

whatever if it's Spotify or you know

whatever publicly accessible ways that

you can access music went through

downloaded copies and then trained their

AI to learn and not only learn but that

these AI platforms are now imitating

okay so there's a little bit of a

two-prong test here on number one you

know is it okay for AI to train on human

created copyrighted material and then

number two from what's being generated

on these platforms are they just

imitating so meaning we use this where

derivative works right so when you

create something that is based on

something else copyright typically

carries of that original thing that

basically was used to create the new

thing so in my lawyerly oversimplifying

kind of way that's what's going on the

update is that our friend over at sync

my music did a new video talking about

how these platforms are generating music

that sounds very very similar if not

exact to very popular music and in fact

Jesse from syn my music was even quoted

in the lawsuits that we went through

together this is pretty crazy guys so I

figured why don't we just jump right

into it the major labels along with the

Raa have sued yudo for copyright

infringement and what's even crazier is

my udio video is referenced directly in

the lawsuit so thank all of you guys for

sharing that video liking that video

making it more publicly available to

more people you guys definitely helped

move the needle and now our voices and

our concerns are included in what might

be a landmark case to help get some

clarity on this very important issue and

this is a very big issue you know to

date of all the lawsuits that have been

filed in regards to complaints from

authors saying that AI from these

different companies right the skynets of

the world that they're all training from

the stuff that we as humans created we

have copyright protection in and you

know no one to date has been successful

on this issue but something I've been

saying on this channel is I just go you

know when it comes to the fact that

copyright does not carry with AI created

stuff it just doesn't because pursuant

to copyright law the thing in order to

have copyright protection has to be made

by a human author and I've just

speculated that you know in order to

change that it's going to have to be a

big enough deal for an AI one of these

AI companies right like a sink house

we're not there we have the opposite

issue where major labels are saying you

used our content without authorization

it's copyright infringement so let's see

what Sy my music has to say on the topic

that's not all the labels and the riaa

are also suing sunno the other

generative AI Music Company that's had

some interesting outputs let's just put

it that way sunno has recently announced

that they have secured $125 million in

new investment which they say they'll

use to build technology that quote

amplifies our most precious resource

human creativity H okay well remember

that quote cuz we'll definitely be

coming back to it just like with yudo

sunno has built some very impressive

technology here and it leaves many of us

wondering how exactly did they also

seemingly LeapFrog all the competition

and all the other AI generative uh

companies and models out there like

stable audio 2.0 and meta's audio craft

which both have publicly disclosed where

and how they've licensed their music

data iso's AI model the actual Tech

itself just that much more impressive

and capable or did sunno just skip all

of the trouble of getting consent and

Licensing the music that they've trained

their model with I don't know I have no

insight as to what's going on behind the

scenes over there because once again

sunno just like yudo is yet another

company that is not disclosing the the

source of their data or if they secured

licenses to train off of it so yudo put

out a little statement I thought I would

just kind of refer back to this so they

go today we would like to share some

thoughts on AI and the future of Music

in the past 2 years AI has become a

powerful tool for Creative expression

across many media from text to images to

film and now music fair enough at udio

our mission is to empower artists of all

kinds to create extraordinary music in

our young life as a company we have sat

in the studios of some of the world's

greatest mus musicians Workshop lyrics

with upand cominging songwriters and

watch as millions of users created

extraordinary new music ranging from the

funny to the profound I've had clients

literally say this to me of like I have

waited my whole life to make my dream

album and I was able to do that using

one of these

platforms we have heard from a talented

musician who after losing the ability to

use his hands is now making music again

producers have sampled AI generated

tracks to create hit songs like BBL

Drizzy and and every day music lovers

have used the technology to express the

just a whole lot of emotion okay love it

love it love it Crown breaking

Technologies entail change and

uncertainty let us offer some insight on

how our technology works okay so as uh

you know s my music is going to kind of

talk about in this video but also as the

big inquiry right in reviewing the

complaints of what the record labels are

saying because they are saying that

these platforms are actually downloading

copies having the AI study from it and I

even checked with my experts who do app

development I go look I need to

understand this you work with AI in

order to understand how this works are

there copies kind of kept on file and so

it was explained to me that apparently

you can have the AI learn and keep

copies right in its memory but it's

really not necessary once it actually

goes through the step to learn right it

no longer needs the actual asset but you

know the exclusive right of owner under

the Copyright Act is that you have the

right to reproduce the music meaning

someone cannot just go and make copies

of your music that's number one right

number two to then create new versions

that are derivative works right that's

where the copyright infringement claim

is coming from all right so what does

what does you do have to say about this

generative AI models including our music

model learn from examples just as

students listen to music and study

scores our model has listened to and

learned learned from a large collection

of recorded

music the goal of model training is to

develop an understanding of musical

ideas the basic building blocks of the

musical expression that are owned by no

one our system is explicitly designed to

create music reflecting new ideas we are

completely uninterested in reproducing

so this is where the claim is at right

the labels are saying that they are

reproducing they are mimicking other

people's music in we we're not

interested in reproducing content in our

training set and in fact have

implemented and continue to refine

state-of-the-art filters to ensure our

model does not reproduce copyrighted

works or artist voice right because

these platforms go we're not allowed to

replicate the likeness the voice of

these different artists right uh that's

number one but they go and we've we've

heard this before that they are

allegedly putting these different

filters into place to make sure that the

AI does not imitate it's not supposed to

imitate all right we stand behind our

technology and believe that generative

AI will become a mainstain of modern

society all right so that's what they're

saying virtually every new technology

has you know initial apprehension we've

ultimately proven to be really good for

all the things all right we know many

musicians especially the next generation

are eager to use you I've seen this with

our community here you know there's a

lot of acceptance people are using

you know the AI and their workflow they

love it in the near future artists will

compose new music along with their fans

amateur musicians will be able to create

new music genres all right so that's

what they have said as of June 25th 2024

and my hope is that the lawsuit that

just got filed will force some clarity

on all of this but until that day

happens the internet is going to be left

to do its internet thing so let's dive

in now let's jump in and let's take a

listen to some of the interesting

outputs that we've discovered using

Sun's model so this was a Reddit post

where an anonymous user shared a track

he had generated with suno's model that

had a familiar sounding producer tag at

the start of it and just in case you're

not familiar a producer tag is just like

a vocal trademark it's an audio

trademark that gets put into the start

of a track to let the listener and the

audience know which producer made the

beat and you've probably heard many of

them they sound something like this we

the best

music another one

oh my God

Ronnie okay let's go back to that Reddit

post and take a listen to what the user

generated accidentally using suno's

model and what I mean by accidentally is

that this user was not trying to

intentionally get a producer tag to come

out of Sun's model he simply used The

Prompt mellow trap and then put in the

lyrics to surprisingly a song from The

Game of Thrones soundtrack from season 2

and and this is what suno's model spit

[Music]

out the user thought it sounded familiar

and then he figured it

out cash money a oh boy it's a producer

tag from Cash Money AP which is a French

producer and songwriter with credits

that include Rody Rich Lil Skies Gucci

man among others and that wasn't the

only instance of Cash Money ap's tag

getting spit out by Sun's model here's

another

one how is Sun's model able to generate

what sounds like a sampled producer tag

out of its model if it didn't train off

of Cash Money ap's music your

coincidence and I'm really curious how

does Cash Money AP feel about hearing

his producer dag coming out of Sun's

model and you know it doesn't just stop

there it's certain only doesn't just

stop there as we saw from the complaints

there were exhibits that were attached

and the labels put together a collection

of all the songs that they were alleging

have been improperly used by these

platforms right and have essentially

been sampled from and you know here's

the thing what we have heard is that you

know there's already been instances

where there's Content ID you know

happening in regards to AI music being

uploaded to some of these platforms and

so that the content that's being

recognized or thought to be recognized

in the content ID from the AI music is

from other people's songs and also what

the labels included with their lawsuit

was you know a thumb drive that they

gave to the court which had all of these

comparisons right so all of the ones

that they were able to recreate to

generate these kind of oneto ones that

of course is going to be used as

evidence of the alleged copyright

infringement there are some more really

interesting examples of what sound like

copyright protected producer tags coming

out ofo's model these were shared with

me by one of our subscribers here on the

channel named Jay who prompted sunno and

shared these results with

[Music]

meable Taj money

[Music]

[Music]

onic another one

DJ we the best music

another

DJ we the best

music another one you know what's what's

interesting here is that you know

there's kind of two things happening one

is well we were talking about derivative

works we go well if the AI has studied

from songs that are owned by someone

else in this case the record labels and

someone goes and makes a new version

right so something that is inspired by

or an imitation of it's derivative work

allegedly and so therefore copyright

carries so for example to to just you

know the way this literally works I make

a song and you as a producer come in and

you remix my song your version is a new

thing but it contains my stuff as well

right it might be my vocals it might

might be my instrumentation to some

degree so the remix is a derivative work

so my ownership and my original carries

into the new work you as the producer

own your new stuff that's in there okay

so that's kind of how it works so we go

there's the derivative work issue and Co

you know alleged copyright infringement

but the other piece here is whether

there is direct sampling happening right

and I think obviously to the point of

this video is that it's really tough I

don't I don't even know what the

explanation could be right is it

something like you know the prompts

we're so exact of the type of voice we

wanted to sound a little Ro

robotico a little like a robot right we

have like a radio effect on it we want a

female voice we want it at the certain

note I don't know how specific the

prompts were in order to generate this

or are we actually hearing sampling as

I'm editing this video I just realized

that many of you will probably be

assuming well whoever prompted this

probably just did it hundreds or

thousands of times and literally just

cherry-picked all the ones that

coincidentally sounded like the original

producer tags which is a fair point

however the subscriber shared with me

this track where they literally just

entered a list of producer tags as the

lyrics using the prompt simply as

trap and this is what sununu's model was

able to spit out

[Music]

you stupid we the best music we the best

music another one

DJ so as you can hear within one track

the actual voice that Sun's model is

generating is changing from producer tag

to producer tag now it didn't get every

single one of these tags absolutely

correct in terms of matching it to what

sounds like the original producer tag

but the point I believe stands that this

model tends to try to match these

prompts with what sounds like copyright

protected producer tags and if none of

that was convincing enough I think this

next one really Takes the Cake here's

one that I personally prompted I tried

to see if I could get sunno to recreate

the day trip producer tag if you're not

familiar daytrip is a two-man producer

group made up of Denzel Baptist and

David Buel and their producer tag sounds

like this and before I play it notice

that there's a stutter effect on the

word

day now when I prompted it as you can

see I did not include that stutter I

didn't repeatedly type in the word day

at all and I generated actually 10

tracks using the exact same prompt that

you see here on your screen but just

listen to what happens to seven of the

10 Generations that I got out ofo

[Music]

again we have to ask why is it that 70%

of the time Uno's model is stuttering

the word day when I did not prompted to

do so could it be that there's a

producer tag somewhere ino's data set

that it is trying to reach into and

recreate so you've heard for yourself

you can be the judge on whether or not

you feel that copyright protected music

is within Sun's training data set that

is what this current lawsuit being filed

against them will hopefully get us some

clarity on but you guys know me I like

to ask questions so here's a couple

first of all how exactly iso's model

able to pull this off we are told by so

many generative AI companies that they

are not storing any data but if that's

true then how is it that puno's model

seems to be able to generate and

recreate and pull up what sounds like

copy protected sound recordings again

from my personal contacts who do this

who are actively developing AI

applications what I was told is that

there's essentially a couple different

ways to do this and that one theyi could

learn from a streaming kind of platform

like Spotify but it's not the best way

for training purposes the best way is to

actually have a native version of let's

say a song or a video that the AI can

learn from and then once it's fully

learned whatever that means from the

Native file it doesn't need it anymore

so it's not like something it goes back

to but that in order for it to properly

learn and to remember I guess would be

uh you know again I'm not the expert

here that's what was told to me so again

in the analysis of where the copyright

infringement has potentially occurred

it's in the duplication right so we

first saved a version We duplicated the

version into the AI and now the AI is

now going to go and potentially

allegedly go and

regenerate into new derivative works or

in some cases I don't know actual copies

of the original music and again going

back to the complaints that is what they

are claiming become your own record

label which is literally if you're like

hey I want to start a record label cool

I got you understand the entire music

business I want to set up my LLC I want

to do the marketing it's a lot so for

example syn licensing for example music

marketing it's significantly cheaper

than anything you would ever get from a

law firm as far as all the stuff that

comes with it and you want to make six

figures from your music career and you

want to do this for real like I'm going

to give you what you need secondly did

they train off of copyright protected

tracks that contained these producer

tags and is that how this model was able

to be able to source and reference those

producer tags and third should we be

thinking about models like sunos as more

of a modernday sampling machine albeit

on steroids because many times it can

recreate what sounds like copyright

protected music in a way that masks the

original Sonic quality of the artist or

producer tag we can also think of these

models as sampling machines on a

spectrum from one side as I showed you

in this video you've got a direct copy

like the Cash Money AP producer tag Cash

Money

AP then moving a little to the right you

have something like one of those DJ

Khaled producer tags orso's models seems

to be able to mask the original Source

by maybe merging blending amalgamating

his voice with many others in the

training data we the best

mus DJ and then we move to the right

side of the spectrum where soo's model

does seem to be generating quote unquote

new voices that don't sound like popular

artists that we know but are those

voices basically the result of dozens

hundreds maybe even thousands of

copyright protected voices in sound

recordings that have just been masked

and altered using this new sampling on

steroids technology again without

transparency or any answers here from

sunno lawsuits had to be filed and this

may explain why stable audio's model

sounds like

[Music]

this not meaning to disparage them at

all but what I love about them is they

have disclosed where they got their data

from and it was from audio Sparks

800,000 tracks from a stock licensing

site I'm sure there's some great quality

music in audio sparks' catalog no doubt

but is it really the same quality of

music and data that is on Spotify with

the 11 million artists to choose from so

I would argue that the quality of the

data that's fed to an AI model is at

least 50% of the value proposition and

that is something that was argued in the

complaints just to say that

the plaintiffs believe that the

platforms are sweeping downloading

getting copies of copyrighted content

and then eliminating lowquality audio so

to his point thinking that this has to

be the best of the best the highest

quality that is what is alleged in the

actual complaints do value as half the

neuron Nets the AI technology itself and

half an undisclosed source of very high

quality potentially copyright protected

music that may or may not have been

licensed and the entire point of this

video is to try to help convince you and

show you how much value we copyright

holders especially those of us that have

commercially viable copyrighted music in

the world on the internet how much value

we can offer to a company like sunu or

yudo they really do need highquality

data to make a model that's worth

anything if we are truly in the midst of

building new million or billion dollar

companies or even new industries that

require high quality copyright protected

music is there really no room for us to

include copyright holders in the profits

of these new Industries and companies

and guys I think that's the question I

need to put to you which is you know if

it turns out that these platforms have

done all the things that are being

alleged against them that they

downloaded copies of the music they had

the AI train from it and then maybe

there's just a little glitch happening

right so the little glitch in that the

platforms just haven't been refined just

yet and so they're spitting out things

that they learned or remembered but they

weren't supposed to right this was a mi

you know this was a mistake I don't know

if that's going to be the explanation

and this is going to get beat up quite a

bit in the federal courts over copyright

infringement but the point just being of

what's your feeling on it right so

should the labels get paid you know they

own the music and if you're going to go

and train your AI off of copyrighted

content you need to pay now you know

we've heard investors for these

platforms say you know and one in in

particular that you know said I wouldn't

have invested if for example these you

know one of these or both these

companies were going and having to get

these deals because it would have been

cost prohibitive right that it would

have been too difficult and in order to

properly I mean to his point in the

video in order to get the quality of

what we need and to train to the extent

that we need we had to do it this way I

thought this video was really well put

together and as I'm continuing to cover

not only these cases but what's going on

and looking at the terms of service for

these platforms because for you guys

creating music using these platforms you

know understanding do you own everything

what's the copyright protection of

anything to the music that you are

creating can you sell what you make and

these are all questions that we are

answering on this channel don't forget

to subscribe I'll see you on the next

one I'm top music attorney Miss Crystal