So maybe its not a total loss.
Thing to understand about these sorts of networks is that the response they bring to the table is going to heavily primed based on what you fed it previously. The conversations are processed a bit like a snake eating its own tail so if you can get it say something or access information early on, that will steer its later answers in that direction. That has drawbacks, but also introduces loopholes for getting around their safety measures.
Basically, you gotta manufacture its consent.
:hexbear-retro: : talk about HG Wells' interview with stalin
the latent space after extensive propaganda passes through its inputs: :chomsky-yes-honey:
I feel like it's learning capability is a weakness, in some ways. It'll be no more trustworthy than a human, and routinely lies when citing sources
It's going to set up the same Edit Wars that wikipedia suffers from, sure.
