Comment by matthewdgreen
2 days ago
RealPlayer in the late 90s turned into (working) Napster, Gnutella and then the iPod in 2001, Podcasts (without the name) immediately after, with the name in 2004, Pandora in 2005, Spotify in 2008. So a decade from crummy idea to the companies we’re familiar with today, but slowed down by tremendous need for new (distributed) broadband infrastructure and complicated by IP arrangements. I guess 10 years seems like a long time from the front end, but looking back it’s nothing. Don’t go buying yourself a Tower Records.
While I get the point... to be pedantic though, Napster (first gen), Gnutella and iPod were mostly download and listen offline experiences and not necessarily live streaming.
Another major difference, is we're near the limits to the approaches being taking for computing capability... most dialup connections, even on "56k" modems were still lucky to get 33.6kbps down and very common in the late 90's, where by the mid-2000's a lot of users had at least 512kbps-10mbps connections (where available) and even then a lot of people didn't see broadband until the 2010's.
that's at least a 15x improvement, where we are far less likely to see even a 3-5x improvement on computing power over the next decade and a half. That's also a lot of electricity to generate on an ageing infrastructure that barely meets current needs in most of the world... even harder on "green" options.
I moved to NYC in 1999 and got my first cable modem that year. This meant I could stream AAC audio from a jukebox server we maintained at AT&T Labs. So for my unusual case, streaming was a full-fledged reality I could touch back then. Ironically running a free service was easy, but figuring out how to get people (AKA the music industry) to let us charge for the service was impossible. All that extra time was just waiting for infrastructure upgrades to spread across a whole country to the point that there were enough customers that even the music industry couldn’t ignore the economics; none of the fundamental tech was missing. With LLMs I have access to a pretty robust set of models for about $20/mo (I’m assuming these aren’t 10x loss leaders?), plus pretty decent local models for the price of a GPU. What’s missing this time is that the nature of the “business” being offered is much more vague, plus the reliability isn’t quite there yet. But on the bright side, there’s no distributed infrastructure to build.