List your company

Toxicity Classification on CivilComments

92.6Average AccuracyERM
86.67288.21189.7591.289Nov 1, 2024Feb 8, 2025May 18, 2025Aug 25, 2025Dec 2, 2025Mar 11, 2026Jun 18, 2026

Evaluation results

MethodLinks
ERM2024.1192.657.4——
JTT2024.1192.663.7——
Robust Filtered2026.0691.5—2.72.1
Robust MoE2026.0691.4—2.51.8
Single Expert2026.0691.3—6.96.4
Vanilla MoE2026.0691.2—7.16.5
MoCaE2026.0691.2—6.65.9
FGR2026.0690.6—4.43.5
FGR + Robust2026.0690.3—5.44.7
Bias-Conditioned Self-Influencebase_method=JTT2024.1186.978.5——