finding
active
finding:per-token-features-have-strongest-spearman-correlation-and-nested-the-strongest-pearson-turn-averaged-weaker-but-significantPer-token features have strongest Spearman correlation and nested the strongest Pearson; turn-averaged weaker but significant
Figure 6.
Source paper
extracted_from(2026) · Kevin Der · Harish Kamath · Thompson, Ben
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Turn-level attribution graphs give compact, interpretable explanations of multi-turn model behaviorsupportsCase study conclusion.
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.
- §5.5.
- Post-training inter-model convergence in trait preferences demonstrating persona convergence
- Prediction from coverage results.
- Confirms that post-evolution performance bottleneck is on the agent side, not evolver side
- Vector geometry features provide moderate secondary signal beyond elicitation-only screen
- Turn-averaged features have highest embedding similarity 0.663 at k=3; per-token max lowest at 0.617finding0.772Consistent across k=1,2,3,5.
- Appendix I.
- Used to compare RDMs in RSA computations; noted to have sensitivity issues with differing relative extrema in embedding layers.