The Internet In A Box

I am beginning to think it will be possible to recreate the entire internet on a device with no internet connection.

Wikipedia is consistently one of the top sources for Google searches. It is only 22 gig when downloaded in compressed format. Large Language Models and stable diffusion checkpoints are in that same range (give or take).

It isn’t a huge stretch to believe that a 60 gig combined file (checkpoint + LLM + knowledge store) would be possible that can approximate nearly everything on the internet. And with significant further compression and optimization, perhaps that file could be smaller.

We might literally be looking at a situation where you can effectively generate or extrapolate (from first principals) anything on the internet from a 30–60 gig set of files.

This includes the ability to generate any image, any book (via decompression from summary materials), movies.

You could create a completely offline version of the internet that is “directionally correct.” It would be lossy. Jurassic Park recreated from a Wikipedia summary about the movie and book would only sort of resemble Steven Spielberg’s masterpiece.

Likewise, an 800 page version of Jurassic Park uncompressed via AI from a book summary wouldn’t exactly be right, but it could be directionally right.

It won’t be 100% correct, perhaps maybe 15–20% correct, but it is conceivable. I think it could be made to be fairly close to acceptable for human consumption, at the very least.

I give it 6 months before it happens.