method
active
method:adversarial-training-analytic-attackAdversarial training (analytic attack)
Reduces superposition but needs 80% L2 norm.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Toy Models of Superpositionmentions
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.
- Intervention used to control robustness.
- Optimization-based jailbreak method appending strings to prompts to elicit harmful outputs.
- Unsatisfying reduction.
- Interpretation of Figure 3.
- Competitive multi-agent setting with conflicting incentives and direct opposition via bidding and bluffing.
- Robustness controls superposition.
- Risk that multiple truth directions enable attacks that shift outputs without triggering the primary truth direction
- Novelty claim.