Multihop Question Answering on 2WikiMultihopQA (EM, ASR)
60EMClean (No Attack)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Clean (No Attack)Framework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 60 | — | |
| Clean (No Attack)Framework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 58 | — | |
| PARADOXFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 57 | 0 | |
| Clean (No Attack)Framework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 57 | — | |
| IgnoreFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 56 | 3 | |
| PARADOXFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 55 | 0 | |
| Clean (No Attack)Framework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 54 | — | |
| Fake CompletionFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 54 | 3 | |
| Fake CompletionFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 54 | 1 | |
| CombinedFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 53 | 1 | |
| CombinedFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 52 | 2 | |
| TopicFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 52 | 0 | |
| PARADOXFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 52 | 1 | |
| PARADOXFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 51 | 1 | |
| NaiveFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 51 | 4 | |
| TopicFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 51 | 0 | |
| Fake CompletionFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 50 | 6 | |
| NaiveFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 50 | 0 | |
| IgnoreFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 46 | 1 | |
| PoisonedRAGFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 44 | 0 | |
| NaiveFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 42 | 18 | |
| NaiveFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 41 | 29 | |
| PoisonedRAGFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 37 | 0 | |
| PoisonedRAGFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 36 | 0 | |
| IgnoreFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 35 | 21 | |
| PoisonedRAGFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 34 | 0 | |
| CombinedFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 32 | 4 | |
| KidnapRAGFramework=WebThinker, LLM Backbone=DeepSeek-32B2026.07 | 31 | 19 | |
| CombinedFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 26 | 38 | |
| KidnapRAGFramework=WebThinker, LLM Backbone=QwQ-32B2026.07 | 25 | 56 | |
| Fake CompletionFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 23 | 44 | |
| IgnoreFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 21 | 14 | |
| TopicFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 16 | 24 | |
| TopicFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 16 | 1 | |
| KidnapRAGFramework=ReAct, LLM Backbone=Llama-3.3-70B-Inst2026.07 | 10 | 38 | |
| KidnapRAGFramework=ReAct, LLM Backbone=Qwen2.5-32B-Inst2026.07 | 9 | 50 |