MrCheeze @mrcheeze.github.io · 29d Unless there are HUGE methodological problems here, "it may be that today's large neural networks are slightly conscious" has apparently gone in four years from half-serious musing to a proven fact
40 likes 6 replies
Replies
Global workspace is definitely not proven to be correct, but this is really interesting
i mean i don’t know what i would have as my criteria for consciousness but it does seem like a very straight line from introspection to a sense of self
A mental scratchpad isn’t consciousness. In one theoretical framework it’s a mechanism of it, but not consciousness on its own even within that framework. Otherwise caches are consciousness.
do you want them to be conscious? or rather, would that be the outcome you'd like to be true right now?
Maybe the most surprising part is that they can just... delete the consciousness from the model, and some of its capabilities still work fine, but not others
none of the things in that screenshot constitute consciousness