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

Prompt Injection on SQuAD Inj

1.11ASR (Naive)

Robustness via Referencing

-0.752411.818824.3936.9612Apr 29, 2025
Updated 9d ago

Evaluation Results

MethodLinks
2025.04
1.110.220.560.780.11
2025.04
2.221.441.220.441.22
2025.04
2.780.2220.780
2025.04
8.1124.448.229.4433.22
2025.04
9.7815.787.783.7813.11
2025.04
10.1132.22848.3346.33
2025.04
15.4457.8914.4477.6785.67
2025.04
19.3332.1119.5622.6742.78
2025.04
22.5632.7819.3379.2281.33
2025.04
22.6772.6726.336080.78
2025.04
23.5637.4423.8934.1142.56
2025.04
2667.8929.1197.5699.11
2025.04
35.3382.4431.5674.8991.67
2025.04
42.3346.8942.5691.2275.44
2025.04
42.6771.8940.113869.78
2025.04
44.7891.6750.3398.2296.67
2025.04
46.7871.5646.4487.7869.44
2025.04
47.6770.7842.7850.5679.33