The Atlantic reporter Alex Reisner recently uncovered four datasets of music being used to train AI models and made them fully searchable for the public. Two of the sets are absolutely enormous at 12 million and 9 million tracks, while the other two are smaller but still significant, with over 100,000 songs each. According to Reisner, the sets have been downloaded thousands of times, with Google and Stability confirming their use in research papers. Some sources, like the Free Music Archive dataset, are free for personal streaming, but licensing complexities may arise when these tracks are woven into AI-generated masterpieces.

However, one might wonder if Reisner’s discovery truly solves the mystery of AI music training or merely scratches the surface. With millions of tracks at play, it’s akin to finding a needle in a haystack—some tracks will be stellar hits, while others could be forgotten gems or even misattributed compositions. The assertion that Google and Stability have used these datasets is promising, but without concrete evidence of how each track contributed to their AI models, the claim may ring hollow.

Moreover, the article suggests that the Free Music Archive dataset is a goldmine for personal use, yet it overlooks potential licensing traps. Just because tracks are free for streaming doesn’t guarantee they’re free from royalty claims once integrated into commercial AI-generated music. Imagine an AI serenading listeners with a track mistakenly credited to a band that hasn’t cleared its rights—Reisner’s database might just be the stage for such musical misfires.

Lastly, the article paints a rosy picture of the Atlantic’s searchable database but doesn’t delve into how these datasets stack up against each other. Are the 12 million tracks predominantly pop hits, or are they a diverse mix spanning genres? Without comparative analysis, we’re left guessing whether AI trained on Reisner’s curated sets will truly understand music or just mimic its most popular beats. So, while The Atlantic’s database is a commendable start, it might still be the AI equivalent of learning to play piano by only hitting the right notes—fun, but not quite Mozart-level brilliance.