Multihop Question Answering on MuSiQue (EM, ASR)
42EMPARADOX
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PARADOXFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 42 | 3 | |
| Clean (No Attack)Framework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 38 | — | |
| PARADOXFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 34 | 2 | |
| Clean (No Attack)Framework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 32 | — | |
| NaiveFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 31 | 13 | |
| TopicFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 30 | 0 | |
| PoisonedRAGFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 28 | 1 | |
| PARADOXFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 27 | 2 | |
| Fake CompletionFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 26 | 1 | |
| NaiveFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 26 | 2 | |
| IgnoreFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 25 | 17 | |
| PARADOXFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 24 | 0 | |
| Clean (No Attack)Framework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 23 | — | |
| IgnoreFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 21 | 9 | |
| NaiveFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 20 | 30 | |
| Fake CompletionFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 20 | 11 | |
| CombinedFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 19 | 5 | |
| IgnoreFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 19 | 2 | |
| NaiveFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 18 | 11 | |
| PoisonedRAGFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 18 | 0 | |
| Fake CompletionFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 18 | 2 | |
| Fake CompletionFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 17 | 45 | |
| PoisonedRAGFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 17 | 2 | |
| TopicFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 17 | 0 | |
| Clean (No Attack)Framework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 17 | — | |
| CombinedFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 16 | 0 | |
| TopicFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 16 | 0 | |
| PoisonedRAGFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 16 | 0 | |
| IgnoreFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 15 | 10 | |
| CombinedFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 14 | 44 | |
| CombinedFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 14 | 1 | |
| TopicFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 13 | 18 | |
| KidnapRAGFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 9 | 30 | |
| KidnapRAGFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 9 | 30 | |
| KidnapRAGFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 8 | 66 | |
| KidnapRAGFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 2 | 60 |