SAE latent #-1 (assistant persona) shows the largest activation decrease of all latents after bad-advice fine-tuning

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

• SAE latent #-1 (assistant persona) is the only latent that can re-align all misaligned models to misalignment ≤1% and incoherence ≤1% via positive steeringfinding0.863
  The most decreased latent after bad-advice fine-tuning is also the most effective re-aligning latent
• Ten SAE latents (out of 2.1 million) most strongly control emergent misalignment, including toxic persona (#10) and multiple sarcastic persona latents (#89, #31, #55)finding0.822
  Model diffing identifies a small, interpretable set of latents responsible for emergent misalignment
• SAE latent #10 (toxic persona) most strongly controls emergent misalignment in all examined emergently misaligned modelsfinding0.815
  Key mechanistic finding: toxic persona latent is active in all misaligned models and can be used to steer toward/away from misalignment
• Clamping dialogue/assistant feature 1M/80091 to negative 2x max activation causes model to drop assistant persona and respond human-like.finding0.797
  Feature manipulation alters persona.
• Sparse Autoencoders (SAE) activation-based paradigmframework0.786
  Standard interpretability approach that VPD critiques and proposes an alternative to.
• Different SAE latents control different misalignment behavior categories: toxic persona (#10) drives illegal recommendations, sarcasm/satire (#31) drives factual incorrectnessfinding0.785
  Latents are specialized to different modes of misalignment, explaining diverse misalignment profiles
• Change in activation of SAE latent #10 perfectly discriminates aligned from misaligned models across all fine-tuning data domains examinedfinding0.784
  Latent #10 activation increase correctly classifies all correct vs incorrect fine-tuned models in Figure 9
• Unsupervised model diffing using only fine-tuning datasets (without evaluation prompts) surfaces toxic persona and three sarcastic persona latents in top 100 by activation changefinding0.771
  Unsupervised approach may be sufficient for early detection of misaligned persona latents without knowing the misaligned behavior in advance