Number of podcasts worldwide

Somewhere in the universe, a podcast about machine learning decided to exist at precisely the same moment a researcher in Shanghai uploaded their forty-seventh paper on neural networks to arXiv, and they have been moving in perfect synchronisation ever since, like two dancers who have never met but refuse to fall out of step. One might assume this correlation arose from some deep truth about the nature of artificial intelligence and human communication, when in fact it is almost certainly the universe's way of reminding us that we are very good at noticing patterns and very bad at understanding them.

The real culprit here is almost certainly the rise of computational resources and internet accessibility between 2010 and 2023. Both podcasts and arXiv papers require roughly the same preconditions: cheap cloud storage, reliable broadband, and a global audience of people with time to kill and opinions to share. The number of researchers publishing has grown as universities expanded their PhD programmes and AI funding exploded; simultaneously, the barrier to starting a podcast dropped from "you need a recording studio" to "you own a phone." We are essentially watching the same underlying phenomenon—the democratisation of publishing in the digital age—wearing two different hats, one with a microphone and one with a LaTeX compiler.

What we have stumbled upon is not evidence that podcasters are secretly training language models while they talk, nor that machine learning researchers are unconsciously synchronising their publication schedules to create ambient noise. We have simply documented two entirely separate human activities that both benefit from the same technological tailwind, moving upward together like leaves caught in a thermal. The real question is how many other pairs of completely unrelated things we are not tracking.