concept
active
concept:turn-averaging-as-low-pass-filterTurn averaging as low-pass filter
Intuition for why averaging yields high-level features.
Neighborhood — ranked by edge-count
Papers (1)
paper
Frameworks (1)
framework
- Turn-Averaged SAEimplementsPaper's main proposal: SAE trained on the mean activation across a turn.
Claims (1)
claim
- Averaging acts as a low-pass filter that smooths away per-token detail and retains broad conceptsassociated_withIntuition.
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.
- Attribution graphs with nodes per turn per layer.
- §5.5.
- 5-way ranking.
- Figure 1.
- Internal state measured at the turn BEFORE the self-report question is appended; ensures measurement of spontaneous internal state uncontaminated by rating question
- Example A.
- LongBench v2 example D.
- Prediction from coverage results.