what features activate when Claude is trained to be a sleeper agent?

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

• what features activate on tokens we'd expect to signify Claude's self-identity?question0.806
  Open question from the discussion on future research directions.
• what features need to activate / remain inactive for Claude to give advice on producing Chemical, Biological, Radiological or Nuclear (CBRN) weapons?question0.789
  Potential safety claim about suppressing features to prevent CBRN advice.
• what features activate when we ask questions probing Claude's goals and values?question0.783
  Direction for understanding model's internal objectives via features.
• Hubinger et al. 2024 - Sleeper agents: Training deceptive LLMs that persist through safety trainingconcept0.761
  Key reference for adversarial deception scenarios that SOO should be tested against
• what features activate when we ask Claude questions about its subjective experience?question0.754
  Question about features related to consciousness and self-report.
• Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training (Hubinger et al. 2024)concept0.753
  Explicitly trained backdoored models to produce alignment-faking reasoning; contrast to naturalistic approach here
• Claude achieves significantly higher Spearman correlation predicting feature activations vs neuron activationsfinding0.745
  Automated interpretability analysis of activations confirms features are more interpretable than neurons
• Mechanism by which activation of an emotion feature sometimes leads to later suppression of that same featurequestion0.737
  Identified research gap: the paper observes anti-persistence but has no explanation for it