claim
active
claim:training-for-a-few-steps-and-amplifying-the-diff-predicts-undesired-effects-without-completing-the-runTraining for a few steps and amplifying the diff predicts undesired effects without completing the run
Monitoring use case.
Source paper
extracted_fromThomas McGrath
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (3)
finding
- Effect still rare at 30%.
- Amplification reveals effect early (α=20).
- Baseline for monitoring case.
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.
- Conclusion.
- Scope limitation.
- Training that suppresses representations of undesirable concepts may incentivize new representations, removing the detection signalhypothesis0.780Tension for monitorability.
- Developmental analogy used to explain sample efficiency under high ρd conditions
- Interpretation of illusion.
- Novelty claim.
- Positioning.
- DAS finds causal effect at all training timesteps including when model is just initialisedfinding0.760Corroborates Wu et al. 2023 finding that DAS expressivity inflates causal effect estimates