finding
active
finding:n-prediction-correlations-r-0-91-for-qwen-2-5-7b-and-r-0-97-for-gemma-2-9bN* prediction correlations r=0.91 for Qwen-2.5-7B and r=0.97 for Gemma-2-9B
Cross-model phase boundary prediction.
Source paper
extracted_from(2025) · Eric Bigelow · Daniel Wurgaft · YingQiao Wang · Noah D. Goodman +3
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Prediction 3.
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.
- Cross-model generalization.
- Architecture-specific difference in trait vector geometry
- Scale effects within Qwen3.5 family on different imbuing methods
- Characterizes model similarities and differences in secondary persona dimensions
- Base failure mode.
- Smaller models show non-monotonic and diminished ASR with increasing cone dimensionality
- Wellbeing probe drift is positive in Gemma (ρ=0.34 pooled turn-correlation) and Qwen (ρ=0.24); both p<10⁻⁵finding0.757Normalized probe-score drift across turns generalizes beyond LLaMA family
- Truth-related directions reliably emerge at 60–75% of normalized layer depth in Qwen and Gemma modelsfinding0.756Experiment 1 finding localizing where truth can be causally mediated