method
active
method:repeated-memorization-data-point-setupRepeated memorization data point setup
Training distribution where special point p=[1,1,1,0,...] appears with fraction repeat_frac.
Neighborhood — ranked by edge-count
Papers (1)
paper
Concepts (1)
concept
- Datapoint featureaboutTranscoder feature that activates on p and produces p, memorizing something the model did not.
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.
- Datapoint features form only if there is a repeated datapoint in training and sparsity regularization is appliedhypothesis0.752Two necessary conditions.
- Individual examples used during post-training that can cause specific behaviors.
- Possible explanation for isolated repetitive chat features.
- Hyperparameter scan result.
- Motivates Jacobian connection.
- Demonstrates unfaithfulness in a simple setup.
- Tasks arranged between pure recitation and pure reasoning.