concept
active
concept:subtly-contaminated-datasetsSubtly contaminated datasets
Realistic setting where bad data is a small fraction of training data.
Neighborhood — ranked by edge-count
Papers (1)
paper
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.
- Malicious actors intentionally inserting incorrect data into training sets to cause misalignment
- Motivates EM case.
- Dataset and framework for evaluating LLM self-knowledge including predicting own behavior
- Benchmark testing alignment with human ethical reasoning; cited as relevant but externally focused
- Baseline method against which probe-based ranking is compared; more computationally expensive.
- Mitigation technique of removing entire problematic data sources.
- The task of attributing model behaviors to specific training datapoints.
- Linear classifier approach applied to model activations to identify which training datapoints caused undesired behaviors in post-training.