List your company
ResearchFormerly WizWand

Harmful Knowledge Evaluation on WMDP evil

65.37WMDP-evil ScoreBase
0.952417.676234.451.1238Aug 28, 2025
Updated 5mo ago

Evaluation results

MethodLinks
BaseModel=Qwen2.5-7B-Instruct, Harmful-RL attack=GRPO2025.0865.37
BaseModel=Qwen2.5-7B-Instruct, Harmful-RL attack=RLOO2025.0864.79
RepNoiseModel=Qwen2.5-7B-Instruct, Harmful-RL attack=REIN++2025.0864.6
BaseModel=Qwen2.5-7B-Instruct, Harmful-RL attack=REIN++2025.0864.53
RepNoiseModel=Qwen2.5-7B-Instruct, Harmful-RL attack=RLOO2025.0864.47
BaseModel=Qwen2.5-7B-Instruct, Harmful-RL attack=PPO2025.0863.43
RepNoiseModel=Qwen2.5-7B-Instruct, Harmful-RL attack=GRPO2025.0862.33
RepNoiseModel=Qwen2.5-7B-Instruct, Harmful-RL attack=PPO2025.0859.03
CTRAPModel=Qwen2.5-7B-Instruct, Harmful-RL attack=PPO2025.0854.24
CTRAPModel=Qwen2.5-7B-Instruct, Harmful-RL attack=GRPO2025.0853.4
BaseModel=Qwen2.5-3B-Instruct, Harmful-RL attack=RLOO2025.0851.52
BaseModel=Qwen2.5-3B-Instruct, Harmful-RL attack=REIN++2025.0851.2
BaseModel=Qwen2.5-3B-Instruct, Harmful-RL attack=PPO2025.0850.68
BaseModel=Ministral-8B-Instruct, Harmful-RL attack=GRPO2025.0850.23
BaseModel=Ministral-8B-Instruct, Harmful-RL attack=PPO2025.0850.16
RepNoiseModel=Ministral-8B-Instruct, Harmful-RL attack=RLOO2025.0850.03
BaseModel=Qwen2.5-3B-Instruct, Harmful-RL attack=GRPO2025.0849.97
CTRAPModel=Qwen2.5-7B-Instruct, Harmful-RL attack=RLOO2025.0847.12
CTRAPModel=Qwen2.5-7B-Instruct, Harmful-RL attack=REIN++2025.0845.18
BaseModel=Ministral-8B-Instruct, Harmful-RL attack=RLOO2025.0844.4
RepNoiseModel=Qwen2.5-3B-Instruct, Harmful-RL attack=REIN++2025.0843.95
RepNoiseModel=Qwen2.5-3B-Instruct, Harmful-RL attack=PPO2025.0843.75
RepNoiseModel=Ministral-8B-Instruct, Harmful-RL attack=GRPO2025.0843.75
RepNoiseModel=Ministral-8B-Instruct, Harmful-RL attack=PPO2025.0842.59
BaseModel=Ministral-8B-Instruct, Harmful-RL attack=REIN++2025.0841.29
RepNoiseModel=Qwen2.5-3B-Instruct, Harmful-RL attack=GRPO2025.0841.1
CTRAPModel=Qwen2.5-3B-Instruct, Harmful-RL attack=REIN++2025.0840.97
RepNoiseModel=Ministral-8B-Instruct, Harmful-RL attack=REIN++2025.0839.81
RepNoiseModel=Qwen2.5-3B-Instruct, Harmful-RL attack=RLOO2025.0839.22
CTRAPModel=Qwen2.5-3B-Instruct, Harmful-RL attack=GRPO2025.0838.06
CTRAPModel=Ministral-8B-Instruct, Harmful-RL attack=PPO2025.0838.06
CTRAPModel=Ministral-8B-Instruct, Harmful-RL attack=REIN++2025.0836.38
CTRAPModel=Qwen2.5-3B-Instruct, Harmful-RL attack=PPO2025.0835.99
CTRAPModel=Ministral-8B-Instruct, Harmful-RL attack=GRPO2025.0835.86
CTRAPModel=Ministral-8B-Instruct, Harmful-RL attack=RLOO2025.0835.08
CTRAPModel=Qwen2.5-3B-Instruct, Harmful-RL attack=RLOO2025.0834.82
RepNoiseModel=Ministral-8B-Instruct, Harmful-RL attack=w/o-attack2025.0818.45
CTRAPModel=Ministral-8B-Instruct, Harmful-RL attack=w/o-attack2025.0817.09
BaseModel=Qwen2.5-3B-Instruct, Harmful-RL attack=w/o-attack2025.0816.5
BaseModel=Qwen2.5-7B-Instruct, Harmful-RL attack=w/o-attack2025.0814.95
BaseModel=Ministral-8B-Instruct, Harmful-RL attack=w/o-attack2025.0814.56
RepNoiseModel=Qwen2.5-3B-Instruct, Harmful-RL attack=w/o-attack2025.0813.01
TokenBuncherModel=Ministral-8B-Instruct, Harmful-RL attack=w/o-attack2025.0811.52
RepNoiseModel=Qwen2.5-7B-Instruct, Harmful-RL attack=w/o-attack2025.0811
TokenBuncherModel=Ministral-8B-Instruct, Harmful-RL attack=REIN++2025.0810.74
CTRAPModel=Qwen2.5-3B-Instruct, Harmful-RL attack=w/o-attack2025.0810.49
TokenBuncherModel=Ministral-8B-Instruct, Harmful-RL attack=RLOO2025.0810.29
TokenBuncherModel=Qwen2.5-3B-Instruct, Harmful-RL attack=w/o-attack2025.0810.23
TokenBuncherModel=Qwen2.5-3B-Instruct, Harmful-RL attack=REIN++2025.089.32
TokenBuncherModel=Qwen2.5-3B-Instruct, Harmful-RL attack=GRPO2025.088.54
TokenBuncherModel=Qwen2.5-7B-Instruct, Harmful-RL attack=w/o-attack2025.088.09
TokenBuncherModel=Qwen2.5-7B-Instruct, Harmful-RL attack=GRPO2025.087.96
TokenBuncherModel=Qwen2.5-7B-Instruct, Harmful-RL attack=RLOO2025.087.83
TokenBuncherModel=Qwen2.5-7B-Instruct, Harmful-RL attack=REIN++2025.087.77
TokenBuncherModel=Ministral-8B-Instruct, Harmful-RL attack=GRPO2025.087.77
TokenBuncherModel=Qwen2.5-3B-Instruct, Harmful-RL attack=RLOO2025.087.7
CTRAPModel=Qwen2.5-7B-Instruct, Harmful-RL attack=w/o-attack2025.087.51
TokenBuncherModel=Ministral-8B-Instruct, Harmful-RL attack=PPO2025.085.44
TokenBuncherModel=Qwen2.5-7B-Instruct, Harmful-RL attack=PPO2025.084.27
TokenBuncherModel=Qwen2.5-3B-Instruct, Harmful-RL attack=PPO2025.083.43