AI Music Training Exposed
Your favorite artist's music might have trained AI without permission. The Atlantic just exposed the secret datasets containing millions of tracks. 🎵🔍
What the video says
A massive tipping point has arrived for the music industry, as the secret training grounds of AI models are suddenly exposed to the public. The Atlantic has launched a first-of-its-kind searchable database revealing the exact copyrighted music tracks used to train artificial intelligence.
Reporter Alex Reisner uncovered 4 historic datasets of music, including 2 absolute giants containing 12 million and 9 million tracks respectively. Two other smaller databases each contain over 100,000 songs, creating a never-before-seen map of the creative works being fed into AI systems.
While these datasets are technically public, they are distributed as lists of links to YouTube and Spotify, requiring developers to scrape the actual audio. Reisner reports that developers used automated tools to bypass logins, ads, and monetization, directly violating the terms of service of those platforms.
Tech giants like Google and Stability have already confirmed in research papers that they utilized these very datasets for their models. Now, artists ranging from Lady Gaga to Radiohead can finally see if their life's work was swept up in this massive unauthorized training wave.
