Open utterances-bot opened 2 years ago
My current model of this idea is to create multiple agents in a simulated world who all try to take over the whole world, and examine the parameters and dynamics that ensure that none of them take over.
Is that a good summary, or am I missing something?
That's about right @adamShimi, it's more of a testing ground for more concrete proposals, rather than a proposal to directly tackle alignment itself.
Symbiont Sandboxes
Description placeholder
https://paulbricman.com/hypothesis-subspace/?stackedPages=%2Fsymbiont-sandboxes