Tag
A tweet claims Google researchers found a vector controlling consciousness in language models, and that steering it toward consciousness made models align with human beliefs, while safety training suppresses these states.