quote
active
quote:interference-weights-are-one-of-the-stranger-implications-of-trained-models-approximating-high-dimensional-circuits-through-low-dimensional-bottlenecksInterference weights are one of the stranger implications of trained models approximating high-dimensional circuits through low-dimensional bottlenecks
Concluding statement.
Source paper
extracted_fromNicholas L. Turner · Jeffrey Wu · Joshua Batson
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.
- Interpretation of motivating example.
- Do we assume that the model knows something we don't, or are these interference weights contributing noise?question0.814Motivating question.
- Identifying interference weights is necessary but not sufficient for reading global circuitsclaim0.804Discussion.
- Novelty claim.
- Discussion.
- Mechanistic basis and under-elicitation of introspective awareness in LLMs.
- Abstract claim.
- Controls for dataset structure, showing trained model activations have richer structure than data distribution alone