[mkaic]

I don't think we need any more useful data for the AIs to gobble up — we just need to make the architectures we train more data-efficient. In my mind, a few petabytes of text, images, audio, and videos from the pre-AI era should be more than enough to train an AGI, we just haven't figured out how to do it yet.

[Loughla]

There's a book in that.

AI becomes ubiquitous, and provides media, information, and general assistance and is trained exclusively on content from before AI was on the web. This freezes pop culture in what, the late 90's through the 2020's? What are the implications for that moving forward? Do we stagnate? Do we become so reliant on it that innovation is dying?

I find that idea interesting.

[13years]

It is a similar concept that I touch on a bit in the article I linked above.

In the section "AI dominance, stagnation and human skill debt" which mentions the idea that we could break the cycle of learning new information and skill where there is nothing to fill that void.