It’s a nice thought, but it depends on people disclosing that they used AI.
They’ve created a monetary incentive for people to hide their use of AI. Sure, they may catch the people who use commercial services which watermark their output but music generation can be done on relatively inexpensive (compared to frontier LLMs) hardware and there is no way to automatically detect it.
It’s a good thing that Tidal is doing, I’m just skeptical that it’ll do much except to make people more creative about hiding it.
IMO one of the worse aspects of slop in the system is that one person can produce so much of it. One person uploading a ton of music without tagging it as AI will probably throw some internal flag in the system for review, so at minimum there will either be less mass-posted slop or it’ll force people to upload from multiple accounts, which also has certain flags to look for.
From the original email I received from them, it sounded like they have an AI detection tool that would automatically apply the AI tag. They said they’re starting with tagging songs that are 100% generated but will expand that to any AI songs once they have the tech for it
It’s like spam detection. Whatever they’re looking for with their tool, the spammers will change so that the tool doesn’t detect it. They will catch some people, but if there is money to be made then people will find ways to work around the detection.
And just like spam detection, as the AI detection tools are looking for artifacts and glitches that are the result of the generation, the only way to get around them is to improve the output to be cleaner and more human-like.
The end result is either hitting the limit with generation and being able to block most of it, or surpassing the ability for content detection, making it essentially indistinguishable from real content. Which is either a great, or a really horrible result, depending on your stance on listening to AI music, i.e do you dislike it because it current sounds bad, or on principle because it wasn’t made by a human no matter how good it is.
It’s a nice thought, but it depends on people disclosing that they used AI.
They’ve created a monetary incentive for people to hide their use of AI. Sure, they may catch the people who use commercial services which watermark their output but music generation can be done on relatively inexpensive (compared to frontier LLMs) hardware and there is no way to automatically detect it.
It’s a good thing that Tidal is doing, I’m just skeptical that it’ll do much except to make people more creative about hiding it.
IMO one of the worse aspects of slop in the system is that one person can produce so much of it. One person uploading a ton of music without tagging it as AI will probably throw some internal flag in the system for review, so at minimum there will either be less mass-posted slop or it’ll force people to upload from multiple accounts, which also has certain flags to look for.
From the original email I received from them, it sounded like they have an AI detection tool that would automatically apply the AI tag. They said they’re starting with tagging songs that are 100% generated but will expand that to any AI songs once they have the tech for it
AI detection tools are very hit or miss.
It’s like spam detection. Whatever they’re looking for with their tool, the spammers will change so that the tool doesn’t detect it. They will catch some people, but if there is money to be made then people will find ways to work around the detection.
And just like spam detection, as the AI detection tools are looking for artifacts and glitches that are the result of the generation, the only way to get around them is to improve the output to be cleaner and more human-like.
The end result is either hitting the limit with generation and being able to block most of it, or surpassing the ability for content detection, making it essentially indistinguishable from real content. Which is either a great, or a really horrible result, depending on your stance on listening to AI music, i.e do you dislike it because it current sounds bad, or on principle because it wasn’t made by a human no matter how good it is.
Not for audio. At least for now you can see clearly artifacts in the spectrals when you inspect them. AI simply can’t recreate real compositions.