[hospadar]

I certainly have no idea but interesting to think about! One could argue that humans are trained on their own output so presumably interesting things could arise, especially if it's not a totally closed loop (LLM#1 -> some human curation -> LLM#2).

[fladrif]

> humans are trained on their own output

And it seems we're emulating that same cycle. Humans are trained not only on our own output, but also take changes from our environment. So in this sense LLM's environment would be human input.

[cyanydeez]

the problem with the current tech is that the solid part of knowledge isn't separate from the fluid parts. You need to fundamentally understand that human knowledge in many generations was purposely fixed, by books, scholars, etc.

Until that occurs, it's really just a miasma of bullshit.