Mm, you're right, but Geocities might be less interesting to historians than an archive of all internet history.
Also, as someone who has trained a few large GPT models, I think ML has a chance of preserving a lot of this data. Training datasets are only growing larger and larger, and although those aren't updated (yet), there's no reason to think they won't last for a long time.
I imagine that in 500 years, imagenet2012 might still be around as a historical curiosity, at least somewhere.
Also, as someone who has trained a few large GPT models, I think ML has a chance of preserving a lot of this data. Training datasets are only growing larger and larger, and although those aren't updated (yet), there's no reason to think they won't last for a long time.
I imagine that in 500 years, imagenet2012 might still be around as a historical curiosity, at least somewhere.