claim
active
claim:crosscoder-model-diffing-is-best-used-as-a-high-recall-pre-screening-tool-followed-by-verificationCrosscoder model diffing is best used as a high-recall pre-screening tool followed by verification.
Recommended workflow.
Source paper
extracted_from(2026) · Thomas Jiralerspong · Trenton Bricken
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (4)
finding
- Appendix C.5/C.7.
- Figure 20.
- C.7.1.
- C.5.1.
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.
- Conclusion.
- Train a single SAE encoding and decoding activations from two models simultaneously to find shared and exclusive features.
- Synthetic two-model activations built from shared and exclusive latent factors with controllable ground truth.
- Core motivation.
- A priori unknown behavior discovered.
- Compared with base SAE transfer.
- Authors' outlook.
- Unsupervised approach may be sufficient for early detection of misaligned persona latents without knowing the misaligned behavior in advance