> We need some way of navigating this, because total available information’s not going to go back down.
Large AI models are the solution. We invented something that can more or less understand the impossible amount of content out there, and distill it in ways specific to each individual.
I do have a bit of experience with LLM locally, to test the limits of what it can do consistently. 16000 token context windows don't seem like enough to me to tackle this task. I don't know that it's possible to distill my musical tastes into a small enough number of tokens, for instance. Nevermind that you need humans to review the music, currently; unless there's magical song tagging models out there i haven't heard of.
there are minimally invasive mechanisms, like kongregate and that .io game site use, "people also played ..."; most aggregators suck, though. Netflix has always recommended stuff ("For you" or whatever) that i have no interest in, amazon recommends things that i would never buy - or have already bought, including from amazon. Pandora was awesome before everyone uploaded every song to youtube, now it's just a collection of playlists that play the same 20 songs over and over each time you start it.
I don't use and haven't used spotify or any other "music" service, because of pandora experiences and youtube. Also in my car i prefer to listen to old time radio which is easily discoverable* enough for at least a few years of content. Just copy to a USB stick and plug into the car. And my big issue with podcasts is i don't enjoy listening to most (nearly all) people just "talking", especially if they're chewing the scenery to make it more of a "captivating" experience.
I do miss mp3 streaming sites, though. My favorite one went dark during the pandemic, is still dark, and shoutcast dot com has sucked for discovery for at least half a decade if not longer. I think it's mostly popular in non-english speaking countries these days, at a glance.
* OTRR - old time radio researchers, there's otrrpedia and an online player available, the content itself lives on archive.org - i have about a TB of radioplays/content from there that i rotate through (only) in the car.
Large AI models are the solution. We invented something that can more or less understand the impossible amount of content out there, and distill it in ways specific to each individual.