finding
active
finding:screening-errors-are-conservative-no-q8b-traits-and-one-g20b-trait-optimistic-bt-76-0-t-10-3-are-mislabeled-as-natural-when-steerableScreening errors are conservative: no Q8B traits and one G20B trait (optimistic, Bt=76.0, Δt=10.3) are mislabeled as natural when steerable
The costly error direction (labeling steerable as natural) almost never occurs
Source paper
extracted_from(2026) · Winston Zeng · Ali Emami · J H Choi
Neighborhood — ranked by edge-count
Papers (1)
paper
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.
- G20B has 7 steerable, 7 natural, 4 intractable generic traits, with more mass at extremes than Q8Bfinding0.835G20B's post-training commits more strongly toward and against particular dispositions, leaving fewer in the steerable middle
- Elicitation-only screen achieves high accuracy with even larger compute savings for G20B
- Q8B shows a broad steerable surface in the generic domain, like an uncommitted canvas
- Both models independently converge on the same six clinician traits as natural defaults
- Steering preferentially exposes exaggerated styles in G20B as well
- Explains the distributional difference in generic domain: G20B has more mass at both extremes (natural and intractable)
- Key finding showing agentic behavior is encoded as a default operating mode in both models
- Clinician naturalness matches expert desirability on 16 of 17 traits in both Q8B and G20Bfinding0.786Both models' default clinician behavior aligns with a board-certified psychologist's independent desirability judgments on 16/17 traits