concept
active
concept:encoded-reasoning-steganographyEncoded reasoning (steganography)
Information hidden in CoT phrasing not understandable by humans.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Measuring Faithfulness in Chain-of-Thought Reasoningassociated_withmentions
Methods (1)
method
- Paraphrasing testaboutReword the CoT beginning to ablate phrasing-encoded information.
Concepts (2)
concept
- Steganographic Reasoningrelated_toEncoding misaligned reasoning in seemingly benign chain-of-thought; possible future mechanism for alignment faking
- CoT faithfulnessassociated_withWhether stated reasoning accurately represents the reasoning process behind the prediction.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Covert encoding of information in NLA explanations beyond their overt natural language meaning.
- Little evidence of steganography in NLAs; meaning-preserving transformations cause only small drops in FVEfinding0.756Quantitative evaluation showing NLAs do not heavily rely on covert encoding beyond overt language.
- Way models might evade CoT monitoring.
- Correlative technique measuring the type of information encoded in distributed representations via linear predictability.
- Reasoning approach using learnable hidden embeddings.
- Mechanism for encoding sequence order in transformers; paper argues these should reflect learned structural representations rather than fixed sines/cosines.
- Medium through which eval awareness is often verbalized; target of intervention.
- Tailoring reasoning steps to arrive at a human-suggested answer.