The Record Labels Are Coming for Suno and Udio

Today on the AI Daily Brief unexpected but still very important lawsuit from the record labels against a set of AI startups. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show. Notes. Welcome back to the AI Daily Brief Headline Edition. All the AI headline news you need in around five minutes. We kick off today with a couple of updates from recent stories. A really interesting recent piece of news had Apple talking with Meta around how Meta's AI models might be incorporated into Apple Intelligence. This was interesting to a lot of people given how contentious the Apple and Meta relationship has been in the past. My argument, which by the way is still my argument, is that ultimately Apple doesn't want to be a kingmaker when it comes to AI models, but they want to be the consumer gateway to AI models in the same way that they're the consumer gateway to all other mobile based experiences. And so even though they launch with ChatGPT, I believe that over the next six, 12 or 18 months, basically every foundation model that is credible and big will probably be optional as part of the iPhone ecosystem. However, with all that said, it turns out that the Apple Meta deal may have been overstated, writes Apple whisperer Mark Gurman on Bloomberg Apple Apple rejected overtures by Meta platforms to integrate the social networking company's AI chatbot into the iPhone months ago. The two companies aren't in discussions about using Meta's Llama chatbot and only held brief talks in March. The dialogue about a partnership didn't reach any formal stage, and Apple has no active plans to integrate llama. Basically, it seems to me like there was enough respect for the possibility that there was some due diligence done, but ultimately there was never really serious consideration. For example, according to Gurman sources, a big reason that Apple decided not to move forward was the problems they have with Meta's privacy policies, which of course have been at the core of their testy relationship for a long time. Additionally, Gurman sources did say that when it comes to ChatGPT, they just simply see it as a better offering, and that when it comes to Google, even if ChatGPT is better than Gemini, Google is already such a deeply integrated partner for Apple that any sort of integration could build off of that. German sources also did confirm that Apple does appear to continue to be talking to Anthropic, but it's not exactly sure if or when that would happen. So that is the update. Apple Intelligence will not be powered, at least for now, by Meta Speaking of Meta and Apple, the other story we discussed is how these big tech companies are increasingly looking like they may avoid Europe altogether. Apple made news when they announced that they may not launch Apple Intelligence in Europe because of concerns around recent regulatory updates. And in general, there's a sense that Europe's approach to regulation may be costing it the ability to actually be a leader in AI itself. For those Europeans out there who don't want to concede that point, good news today as Meta announces a European AI startup accelerator. The announcement page reads, Meta is launching a European startup acceleration program based at Station F in partnership with Hugging Face and Scaleway, with the support of the HEC Paris Incubator. Basically, this is a four month program based in Paris. It'll start in early October and go through February 2025. And in terms of criteria, Meta wants to see there be an existing product that has actually been tested with customers or clients. There's they want that product to be based on an open source model and they want startups to have identified an AI technical challenge to, quote, fully benefit from the mentoring program. They also note that these companies must be based in the EU and able to physically attend bimonthly in person mentoring sessions. So like I said, if you are excited about the European AI startup scene and want to see more resources for it, this has got to be good news. Next we move over to some launch and funding announcements, starting with Etched, who write that they are making the biggest bet in AI. Etched writes in 2022 we made a bet that Transformers would take over the world. We've spent the past two years building Sohu, the world's first specialized chip or ASIC for Transformers, which they point out is the T in ChatGPT. By burning the transformer architecture into our chip, we can't run most traditional AI models. The DLRMs, powering Instagram ads, protein folding models like AlphaFold 2 or older image models like Stable Diffusion 2. But for transformers, they write so who is the fastest chip of all time? And it's not even close. With over 500,000 tokens per second in llama 70B throughput, SOHU lets you build products impossible on GPUs. They continue SOHU as an order of magnitude faster and cheaper than even Nvidia's next generation Blackwell. Today they say every state of the art AI model is a ChatGPT, Sora, Gemini, stable diffusion 3 and more. If transformers they write are replaced by SSMs, RWKV or any new architecture, our chips will be useless. But if you're right SOHU will change the world. Super interesting bet. I love the specificity and clarity of the vision and I would highly recommend you check out the announcement article because it serves kind of as an argument for the company. I would love to see more startups do this and so check it out etch.com and it's big and emblazoned all over their website. Speaking of ambitious fellow Learn Capital portfolio company Emergence has just come out of stealth. With 97 million in funding, emergence is right in the thick of the agent trend, said CEO Satya Nitta. Current generative AI models, while powerful in language understanding, still lag in advanced planning and reasoning capabilities necessary for more complex automation tasks, which are the provenance of agents. So what is available from Emergence? Well, earlier this week they open sourced something they call their orchestrator. TechCrunch writes, the orchestrator doesn't perform any tasks itself. Rather it functions as a kind of automatic model switcher for workflow automations, factoring in things like the capabilities of and the cost to use a model. If it's a third party, the orchestrator considers the tasks to be performed, for example writing an email, and then chooses a model from a developer curated list to complete that task. Right now, if you look around, you'll see both cynicism and excitement in equal measure when it comes to AI agents. On the one hand, they've been a hot topic for more than a year now without necessarily really achieving what people want them to. But at the same time, basically everyone is betting that the AI future is agentic and so I for one am excited to see big bets in the space. For now though, that is going to do it for today's AI Daily Brief Headlines edition. Up next, the main episode. Today's episode is brought to you by Fractional AI My Go to AI Dev Shop when we wanted to build a specific AI product feature, what we called our AI tool Finder Bot for super intelligent, we knew we needed incredible AI engineers that could build something fast and so we turned to Fractional because they've got some of the best AI engineers on the planet. The team is a group of senior engineers in San Francisco working on some of the most exciting projects in applied AI. They work with everyone from startups all the way through the Fortune 500. To request a free consultation, head to Fractional AI and again, if you want help identifying and building AI projects for your business, I highly recommend you check them out Again. That URL is Fractional AI to request a free consultation. Today's episode is of course brought to you by Super Intelligent and today I want to tell you about superintelligent's Summer Challenge. We have a couple of low key months, low key, at least in the professional sense, and that makes them a perfect time to build some new skills. If you want to come back to the fall, to your work, to your school, to wherever you are with a whole new set of talents and capacities when it comes to artificial intelligence, you should come join the Super Summer Challenge. Each week we'll share a new challenge that gets you actually using the most important tools in AI and figuring out what these things are going to be good for in practice, in your job and in your career. As you learn, you'll also become part of a community that discusses the projects and use cases that we're all sharing. If you have wanted to dig deeper into AI and have just been waiting for the right excuse, this is it. Use Code Summer Fun to join the Summer challenge and get 50% off your first couple of months. You can get all the information you need at BE super AI. That's BE super AI. See you there. Welcome back to the AI Daily Brief. Of all the industries, specifically the content or entertainment industries, the one that was always likely to fight back the hardest against generative AI, in my estimation at least, was the music industry. The music industry has a really interesting relationship with new technologies, where in many ways they looked like they were going to be the first to be disrupted back in 1999 with the arrival of Napster. However, what happened is that the muscle they built fighting new technologies and retaining copyright power has actually made them one of the most formidable opponents of new technologies. That is, of course, until they can get their piece of it. And so it was no surprise at all that the Recording Industry association of America, as well as a set of labels including Universal Music Group, Sony Music Entertainment and Warner Records, are pursuing the two best known music generator startups, Suno and Yuhio. The RIAA is seeking damages of up to $150,000 per copyright work, making this case have potentially enormous financial consequences, said the RIAA Chief legal Officer in a press release. These are straightforward cases of copyright infringement involving unlicensed copying of sound recordings on a massive scale. Suno and Yurio are attempting to hide the full scope of their infringement rather than putting their services on a sound and lawful footing. Their argument is basically that if these companies had worked to avoid training their models on copyright works, then they would not have been able to reproduce the convincing imitations of such a vast range of human musical expression. Suno said that the platform is designed to generate completely new outputs, not to memorize and regurgitate pre existing content. The CEO of Suno pointed out that the company doesn't allow user prompts based on specific artists. In a statement, they wrote, we would have been happy to explain this to the corporate record labels that filed this lawsuit and in fact we tried to do so, but instead of entertaining a good faith discussion, they've reverted to their old lawyer led playbook. Suno is built for new music, new uses and new musicians. We prize originality, reinforcing the idea that this isn't particularly unexpected, the Verge writes. Suno executives and investors acknowledge the possibility of being sued in a Rolling Stone profile on the company this March. For some, it's simply the cost of doing business, antonio Rodriguez, an early investor in Suno, told the magazine, quote, honestly, if we had deals with labels when this company got started, I probably wouldn't have invested in it. I think that they needed to make this product without the constraints. Another big point in the communications, at least from the RIAA and the record labels, is that this is not an indictment of AI a priori, said RIAA Chairman and CEO Mitch Glazier. The music community has embraced AI and we are already partnering and collaborating with responsible developers to build sustainable AI tools centered on human creativity that put artists and songwriters in charge. But we can only succeed if developers are willing to work together with us. Unlicensed services like Suno and Udio that claim it's fair to copy an artist's life work and exploit it for their own profit without consent or pay set back the promise of genuinely innovative AI for us all. The crux of the argument for the labels is that training on copyrighted materials represents a breach of copyright from the complaint. Building and operating these services requires at the outset, copying and ingesting massive amounts of data to train a software model to generate outputs for these services. This process involved copying decades worth of the world's most popular sound recordings and then ingesting those copies to generate outputs that imitate the qualities of genuine human sound recording. So what has been the reaction? Many folks have pointed to some of the label's evidence where they were able to reproduce what they consider to be very close facsimiles to existing well known songs. For example, this generation of All I Want For Christmas Is yous, There is Just One Thing I Need. I Don't care about the present. A lot of artist rights advocates find themselves in the uncomfortable position of being aligned with the labels. Rob Abelow writes, breaking all three major labels are suing AI music generators Suno and Yudio. The RIAA is coordinating the lawsuits on behalf of Sony, Warner and umg. Hammer dropped Rob Continues. It's been clear to anyone with ears that Suno and Udio have trained on copyrighted material for their AI models with zero transparency plan to compensate or seek licenses. Rob also goes on to say there are ways to do this and play by the rules. 1. Train only on licensed catalog 2. Compensate artists and songwriters 3. Full transparency. It just takes a little longer and requires actually caring about creatives. Rob continues. Just last week Genai dropped their fully licensed and transparent Gen AI model and voice cloning companies like voiceswap and Sound Labs are doing the same. Music isn't anti AI, it's anti theft. Who will win? Rob concludes, it's it all comes down to fair use. Either way, we'll finally have an answer on the legality of generative AI music training in the US and this is really the question and why? To be honest, even though I'm reporting it, so much of the conversation we have in the public isn't particularly relevant. Ultimately, the big question before the courts right now is whether training of AI models represents copyright infringement or whether it's covered by fair use. That's it. That's the whole ball game. All of these lawsuits, every single one you've seen by any copyright holder against any AI firm are all pushing to get that question ultimately in front of the Supreme Court, because that's what it's going to take. If you go hang around law circles or less edifying, but perhaps more entertainingly the various circles of Twitter X that are talking about this, you will get lots of different takes on the fair use question. Christa Gable writes, as an advocate for AI music, I believe that the transformative nature of AI in music creation can and should be supported under the Fair Use doctrine. She goes on the four factors of fair use 1. Purpose and character of the use AI music training is fundamentally transformative. While it learns from existing music, it doesn't simply copy or replicate it. Instead, it analyzes pattern styles and structures to generate entirely new compositions. 2. Nature of the copyright work While music is a highly creative form of expression which generally receives strong protection, the use of music for training AI involves using it in a different context. The purpose here is not to exploit the original work, but to enable the AI to learn and create something novel. 3. Amount and sustainability. AI training typically involves analyzing large data sets to learn effectively. However, the use is not focusing on taking substantial portions of any one work, but rather on understanding broad patterns across many works. One of the most critical factors is whether the AI's use of music harms the market for the original works. AI generated music does not serve as a market substitute for the original compositions. Instead, it opens avenues for creativity, potentially expanding the market by creating new opportunities and genres. I think personally that there is going to be a lot of time spent on this fourth factor. Yes, the labels are able to recreate some decent facsimiles for their own songs, but is that highly specific use the way that others are actually using these tools? Or are they, as Christa argues, doing something completely new and different? Making this even more complicated is the fact that music only has a relatively limited number of inputs. There are only so many notes. It's one of the reasons that lawsuits about whether a song is too close to another song get so hairy and so subjective. There is also one other slightly cynical take that you'll see around this conversation, which is that ultimately this is just leverage for the labels to control the next generation of music production as well. Are we placing bets that the labels just want part ownership? Suno and Yu Dio to go public Labels sell their shares and make billions. I don't think the labels are suing to stop it. I think labels are suing to get in on the action. Agree or disagree with the cynicism of that take. I certainly think that the only likely outcome in the long run is is label integration with the new technology in the same way that happened with streaming, or at least in some approximate way. Ultimately, there are two battles going on here. One is a legal battle and one is a public opinion battle. As Bilawal Sidhu points out, what is ethical is not always legal and what is legal is not always ethical. He points out, in all of this, it's really the indie artist who is royally screwed, the creators who aren't big enough to drive a collective bargaining effort with tech companies and get paid off for their content libraries. Where does this leave us? He writes, we can be very sure that as long as courts don't intervene, the practice on training on publicly available data will continue. Robots txt be damned, copyright be damned. Proponents of this practice will say we all learn from each other and that everything is a remix. But machines are different from humans. A human can't watch all of YouTube and can't listen to every song on Spotify, read all the text on the Internet, but a machine can. And that makes it feel fundamentally different from a human learning from copyrighted material to produce transformative work to the extent that you are looking for a positive in all this. There actually is a fairly big middle space represented here by Matt Wolf, who writes copyright laws need to be rethought and revamped. They were primarily designed for a pre digital world. I don't have the answers, but I believe there is a middle ground where artists are compensated, companies can train and innovate, and creators don't need to worry about the ethics behind the tools they decide to use. My friend James, the CEO of Creator Mode Studios who deals at the intersection of music and tech, writes the Suno and Udio lawsuits highlight the growing pains of integrating AI into the music industry. Rather than viewing AI as a threat, we should see it as a tool that can revolutionize music creation. By establishing clear guidelines and ethical standards, we can ensure that AI benefits both innovators and traditional artists alike. The real issue here isn't just about copyright infringement, it's about ensuring that creators are at the forefront of technological change. We need a creator first revolution where artists, producers and writers have a say in how their work is used and benefit directly from the advancements in AI. And in fact there is lots of agreement, at least broadly speaking, on that point. The question of course, is how to make that real. For now, though, that is going to do it for today's AI Daily Brief. Until next time, peace.