One judge call vs twelve dimension scores

社区项目 · Articles & Threads

One direct Jev question per row against 12–14 Jev-scored dimensions with locally fitted weights on three classification tasks: 5,477 test rows, 25,174 Jev calls, $1.43. Decomposition wins on Japanese NLI (0.9076 vs 0.8373) but flags about 25× more hard benign rows as attacks (37.2% vs 1.5%).