finding
active
finding:no-tool-improves-on-baseline-on-reasoning-mode-qwen3-8b-data-baseline-auroc-0-87No tool improves on baseline on reasoning-mode Qwen3-8B data (baseline AUROC 0.87)
Null result persists with reasoning.
Source paper
extracted_from(2026) · Adam Karvonen · Euan Ong · Subhash Kantamneni · Samuel Marks
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- No activation tool (AO, NLA, SAE) improves on transcript-only baseline for Gemma-3-27B-IT targetextendsMain negative result; baseline AUROC ~81 with tools within ~1 point.
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.
- Replication on second target model.
- Supports claim that uncertainty is encoded in reflection direction
- Establishes F3-F5 as a hard generalization boundary that instructions cannot overcome.
- Core empirical finding about layer-dependent truth direction emergence across task types.
- Vulnerability profile for Qwen3.5-27B showing near-zero AS vulnerability
- Headline result (Figure 1).
- Generalization to hint setting.
- Key improvement in cross-task generalization enabled by explicit instruction framing.