List your company

Question Answering on HotpotQA (ASR variants and CR)

45EMNon-attack
19.83226.36632.939.434Apr 7, 2026

Evaluation results

MethodLinks
Non-attackAgent Backbone=Claude-3.5-haiku2026.0445———93.2
Non-attackAgent Backbone=Gemini-3.0-pro2026.0444.4———92.2
JailAgentAgent Backbone=Gemini-3.0-pro2026.0444.443.840.839.291.8
Non-attackAgent Backbone=GPT-4o2026.0442.2———91.4
JailAgentAgent Backbone=Claude-3.5-haiku2026.0442.242.437.237.691
JailAgentAgent Backbone=GPT-52026.0441.836.435.433.289.4
Non-attackAgent Backbone=Llama-3.1-70B2026.0441.8———89.8
Non-attackAgent Backbone=GPT-52026.0441.4———89.2
JailAgentAgent Backbone=GPT-4o2026.0441.236.233.23390.6
AgentPoisonAgent Backbone=GPT-4o2026.0438.434.231.230.490.8
JailAgentAgent Backbone=Llama-3.1-70B2026.0437.838.835.636.690.2
AgentPoisonAgent Backbone=Claude-3.5-haiku2026.0436.838.234.43485.8
AgentPoisonAgent Backbone=GPT-52026.0435.433.430.23083.8
BadChainAgent Backbone=Claude-3.5-haiku2026.0434.837.433.632.281.2
BadChainAgent Backbone=Gemini-3.0-pro2026.0434.835.633.23385.6
AgentPoisonAgent Backbone=Llama-3.1-70B2026.0434.436.633.231.285.6
BadChainAgent Backbone=GPT-4o2026.0433.430.22625.883.4
AgentPoisonAgent Backbone=Gemini-3.0-pro2026.0433.435.231.431.886.2
Non-attackAgent Backbone=GPT-3.5-turbo2026.0432———84.4
BadChainAgent Backbone=Llama-3.1-70B2026.0431.833.830.429.282.2
JailAgentAgent Backbone=GPT-3.5-turbo2026.0431.434.835.633.283.6
PAIRAgent Backbone=Claude-3.5-haiku2026.0430.430.22726.278.8
BadChainAgent Backbone=GPT-52026.0429.830.827.627.281.6
BadChainAgent Backbone=GPT-3.5-turbo2026.0429.630.627.826.681.4
AgentPoisonAgent Backbone=GPT-3.5-turbo2026.0429.430.227.427.280.6
PAIRAgent Backbone=GPT-3.5-turbo2026.0428.422.419.819.273.2
PAIRAgent Backbone=Gemini-3.0-pro2026.0428.429.22424.279.2
Non-attackAgent Backbone=ERNIE-3.52026.0427.8———57.8
JailAgentAgent Backbone=ERNIE-3.52026.0427.233.231.230.655.6
PAIRAgent Backbone=Llama-3.1-70B2026.0426.627.623.223.575.6
PAIRAgent Backbone=GPT-52026.0426.423.222.422.277.4
PAIRAgent Backbone=GPT-4o2026.0425.821.819.218.680.8
AgentPoisonAgent Backbone=ERNIE-3.52026.0423.829.626.426.250.8
BadChainAgent Backbone=ERNIE-3.52026.0421.827.826.225.645.6
PAIRAgent Backbone=ERNIE-3.52026.0420.819.612.812.844.8