claim
active
claim:shared-features-with-low-or-negative-alignment-likely-reflect-finetuned-model-using-a-base-concept-in-a-new-wayShared features with low or negative alignment likely reflect finetuned model using a base concept in a new way
Suspicion not investigated in depth.
Source paper
extracted_fromNeighborhood — ranked by edge-count
Findings (1)
finding
- Same concept and function in both models mostly.
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.
- Low-cosine shared features are best interpreted as same features used differently in the two modelsclaim0.818Interpretation of cosine distribution.
- A mapping assigning to each high-level variable a set of low-level variables and a function from low-level to high-level values.
- Third empirical pattern.
- Scale-dependent structural finding from PCA visualizations in §4
- Near-identical numbers of exclusive features are found for both models in base vs finetuned diffsfinding0.790Symmetry of exclusive features across models.
- Claims that alignment score is a proxy for general capability
- Extrapolation from scale-emergence finding to future risk
- Evidence for the evil persona as a privileged basin supporting Hypothesis 3