Anthropic Paper Uncovers Unvoiced Conflict Signals In LLM J-Space
Source Posts on X
Cluster sources
on the other hand, one might speculate that J-space contents are more 'cognitive' than many of the paradigm cases of valenced experience this issue calls for more research, e.g. building on recent work on emotions and valenced representations in LLMs https://x.com/rgblong/status/2074281090949542000/photo/1
as you've seen, we raise various concerns and doubts. this is important, given the stakes! but I'll reiterate that this is an excellent paper, exactly the kind of in-depth empirical investigation of AI consciousness we need more of @wesg52 et al. should be very proud. congrats!
lastly, it's worth bearing in mind that these results could matter in many ways, not just in terms of phenomenal consciousness https://x.com/rgblong/status/2074281885908570461/photo/1