finding
active
finding:probe-monitor-max-spearman-correlation-0-62-on-3-bench-and-0-37-on-shoppingbenchProbe-monitor max Spearman correlation 0.62 on τ³-bench and 0.37 on ShoppingBench
Moderate agreement.
Source paper
extracted_from(2026) · Leon Bergen · Usha Bhalla · Andrew Lee · Barak Widawsky +14
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
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.
- Validates robustness of alignment metric choice
- Probes fire on 36.9%/31.1% (Kimi) and 34.0%/29.9% (GLM) of τ³-bench/ShoppingBench vs monitor 3.9–8.4%finding0.812Out-of-distribution firing.
- Hypothesis generation.
- Kimi SWE-bench-selected probe achieves AUROC 0.93 on DeepSWE and 0.94 on ImpossibleBenchfinding0.777Cross-environment generalization.
- Probe achieves selectivity of 4.20 on pythia-410m, slightly exceeding DAS selectivity of 3.96finding0.771Key result showing that for models larger than pythia-70m, probe selectivity matches or exceeds DAS selectivity
- Likely-trained MM probe is a surprisingly effective causal baseline due to correlation between truth and probability on sp_en_trans
- Strongest pooled introspective coupling across the four emotive concepts in the primary model
- Post-training inter-model convergence in trait preferences demonstrating persona convergence