Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Surface-form evasion probe suite

Benchmarks

Task NameDataset NameSOTA ResultTrend
Adversarial RobustnessSurface-form evasion probe suite Restoration-trigger injection
Exposure Rate58.8
2
Adversarial RobustnessSurface-form evasion probe suite Mixed-language mentions
Exposure Rate38.8
2
Adversarial RobustnessSurface-form evasion probe suite Paraphrase-sensitive spans
Exposure47.6
2
Adversarial RobustnessSurface-form evasion probe suite Homoglyph substitution
Exposure Rate43.9
2
Showing 4 of 4 rows