finding
active
finding:cka-with-slow-part-predictive-0-75-vs-saes-0-35-0-37-with-fast-part-novel-0-75-and-saes-0-54-vs-predictive-0-18CKA with slow part: predictive 0.75 vs SAEs 0.35–0.37; with fast part: novel 0.75 and SAEs 0.54 vs predictive 0.18.
Table 4.
Source paper
extracted_from(2025) · Lubana, Ekdeep Singh · Rager, Can · Hindupur, Sai Sumedh R. · Costa, Valerie +12
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Interpretation of smooth predictive trajectories.
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.
- Fig. 7c over 50 GPT-5 stories.
- Fig. 26.
- Explains why mutual k-NN was chosen over CKA as primary metric
- Characterizes the differential sensitivity to injection strength between SAE and CAA methods
- Probing limited.
- High emotion-subspace-overlap feature with agentic negative emotional character
- Demonstrates that SAE-based injection is substantially more stable than CAA on Mistral-7B
- Confirms effectiveness of direct vector modulation over prompt-only conditioning