Open utterances-bot opened 2 years ago
Confused by this one. Are you proposing a sort of Microscope AI where we extract the knowledge of the model by asking it to solve an incredibly deep and complex task? What would be the training data? And why would the model not have incentives to exploit loopholes here too?
Differentiable Cosmogonies
Description placeholder
https://paulbricman.com/hypothesis-subspace/?stackedPages=%2Fdifferentiable-cosmogonies