Reactions from ranked influencers
3 posts
If that's the case, then there is less to worry about, no? It would def make it easier to have a rule where open model labs run evals that get verified and which show that a model is aligned and doesn't exceed capability thresholds in specific narrow risk areas, and if that passes the model is good to go? That should be achievable for most models then
@BlackHC @mervenoyann Yes, this can be applied to open and closed models
@xeophon @mervenoyann Yeah I think it should be applied consistently. Ie if there is KYC regulation would make sense to not require that for closed models below the same threshold etc. (As always my personal opinion, etc)