FailureScope clusters evaluation probes by cross-model failure patterns via LOMO to produce stable taxonomies that generalize across single-turn, multi-turn, and adversarial regimes, with reported metrics of Kendall's tau 0.81 and AUC 0.88.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
FailureScope: Cross-Regime Behavioral Diagnosis of Language Model Weaknesses
FailureScope clusters evaluation probes by cross-model failure patterns via LOMO to produce stable taxonomies that generalize across single-turn, multi-turn, and adversarial regimes, with reported metrics of Kendall's tau 0.81 and AUC 0.88.