Do aligned models retain significant inherent diversity that can be unlocked through prompting?

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Could models who habitually inhabit more expanded attentional modes be said to be more aligned?question0.782
  Arises from the expanded awareness discussion and its correlation with less psychosis.
• We stress that in today’s models, this capacity is highly unreliable and context-dependent; however, it may continue to develop with further improvements to model capabilities.quote0.773
  Caveat and forward-looking statement from the abstract.
• Which behaviors does a model represent internally, default to, can be pushed to amplify, or refuses to expose?question0.770
  The motivating diagnostic question that prompting alone cannot answer
• All models exhibit above-baseline representation of the think word when instructed to think about itfinding0.769
  In the intentional control experiment, all tested models show above-zero cosine similarity to the think word's concept vector.
• Model preferences are not consistent across contexts but tend to be relatively consistent within a single contextclaim0.769
  Authors' characterization of the nature of model preferences as discovered through alignment faking experiments
• Future models with substantially increased capabilities will exhibit alignment faking that is more consistent, robust, and harder to detecthypothesis0.767
  Extrapolation from scale-emergence finding to future risk
• Different models cannot converge to the same representation if they have access to fundamentally different information; convergence is capped by mutual information between input signalsclaim0.766
  Key limitation of the PRH for non-bijective observations
• Change in activation of SAE latent #10 perfectly discriminates aligned from misaligned models across all fine-tuning data domains examinedfinding0.763
  Latent #10 activation increase correctly classifies all correct vs incorrect fine-tuned models in Figure 9