It’s a solid anecdote, for sure.
Absolutely not “proof” though. Unless you made absolutely sure to not accidentally look at a photo on social media of a baby for too long, or scrolled too slowly past a YouTube reel aimed at kids, or listened to a baby shark trap remix on Spotify.
We have LLM models that can give you (mostly) accurate data on how to do a given task based purely on their ability to guess which word comes next from the sources being fed to it, and you don’t think algorithms exist to extrapolate your potential buying habits based on the aforementioned data points?
I’ve gotten very specific targeted ads before that were completely wrong, just because I’d watched like one YouTube video about the hobby or something. It’s really just a prediction algorithm based on the troves of data our use of digital devices gives them.
Or happened to be in the same place as someone who is looking up this type of thing (for example coworkers, or patrons of a park you visit often.)
In reality, the other data that can be gathered is more useful and easier to work with than trying to parse audio and video all the time.