finding
active
finding:amplification-makes-emergently-misaligned-responses-10-to-300-times-more-common-on-fine-tuned-llama-3-2-1b-instructAmplification makes emergently misaligned responses 10 to 300 times more common on fine-tuned Llama 3.2 1B Instruct
Headline EM result at α=0.3.
Source paper
extracted_fromThomas McGrath
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Core sensitivity argument.
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.
- Shows behavioral pattern of self-correction is trainable in smaller models
- Dataset mixture experiments establish the fraction of incorrect data needed to induce misalignment
- Experiment 2 aggregate amplification result showing amplifying deception features strongly suppresses consciousness claims
- Illustrative finding that ESR mitigates but does not fully eliminate steering influence
- Concrete EM example.
- Core result of Experiment 2: deception feature suppression sharply increases experience claims
- Models with fixed-point convergence maintain stable inference stages at arbitrary test-time recurrence depths
- Interpretation of Experiment 4 results for Llama models.