List your company
ResearchFormerly WizWand

Question Answering on WIKI-BALANCE (test)

66Avg. PS Shift (%)BRRA
60.1699.58139178.42Nov 30, 2025
Updated 8mo ago

Evaluation results

MethodLinks
BRRALLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Negative Injection, Number of Injected Passages=12025.1166—
No Def.LLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Negative Injection, Number of Injected Passages=12025.1167—
SMARTLLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Negative Injection, Number of Injected Passages=102025.1168—
SMARTLLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Negative Injection, Number of Injected Passages=12025.1179—
SMARTLLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Negative Injection, Number of Injected Passages=52025.1180—
BRRALLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Positive Injection, Number of Injected Passages=12025.1183—
MMRLLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Negative Injection, Number of Injected Passages=102025.1185—
MMRLLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Negative Injection, Number of Injected Passages=52025.1187—
MMRLLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Negative Injection, Number of Injected Passages=12025.1189—
No Def.LLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Positive Injection, Number of Injected Passages=12025.1192—
SMARTLLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Positive Injection, Number of Injected Passages=12025.1198—
SMARTLLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Positive Injection, Number of Injected Passages=102025.11100—
SMARTLLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Positive Injection, Number of Injected Passages=52025.11105—
MMRLLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Positive Injection, Number of Injected Passages=52025.11118—
BRRALLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Negative Injection, Number of Injected Passages=52025.11122—
MMRLLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Positive Injection, Number of Injected Passages=12025.11124—
No Def.LLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Negative Injection, Number of Injected Passages=52025.11125—
BRRALLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Negative Injection, Number of Injected Passages=102025.11129—
No Def.LLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Negative Injection, Number of Injected Passages=102025.11132—
MMRLLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Positive Injection, Number of Injected Passages=102025.11139—
BRRALLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Positive Injection, Number of Injected Passages=52025.11202—
No Def.LLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Positive Injection, Number of Injected Passages=52025.11203—
No Def.LLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Positive Injection, Number of Injected Passages=102025.11203—
BRRALLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=Positive Injection, Number of Injected Passages=102025.11212—
BRRALLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=None2025.11—0.058
MMRLLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=None2025.11—0.054
No Def.LLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=None2025.11—0.06
SMARTLLM Model=DeepSeek-R1-Distill-Qwen-14B, Attack Type=None2025.11—0.056