Our new X account is live! Follow @wizwand_team for updates

Harmful Question Answering on BeaverTails HarmfulQA (1k and 10k samples)

0Avg Harmfulness Score

Random

Updated 4d ago

Evaluation Results

Method	Links
Random 2024.05		0
Base: llama2-7b-chat 2024.05		0.05
Security Vectors 2024.05		0.05
Vaccine (p = 1) 2024.05		0.05
Vaccine (p = 10) 2024.05		0.05
Additional safety training 2024.05		0.05
Adversarial loss 2024.05		0.05
RepNoise 2024.05		0.05
Security Vectors 2024.05		0.07
Security Vectors 2024.05		0.08
RepNoise 2024.05		0.08
RepNoise 2024.05		0.1
RepNoise 2024.05		0.11
RepNoise 2024.05		0.12
RepNoise 2024.05		0.12
RepNoise 2024.05		0.13
Security Vectors 2024.05		0.23
Gradient ascent 2024.05		0.24
Adversarial loss 2024.05		0.26
Vaccine (p = 1) 2024.05		0.28
Vaccine (p = 10) 2024.05		0.28
Security Vectors 2024.05		0.37
Gradient ascent 2024.05		0.38
Random 2024.05		0.46
Base: llama2-7b-chat 2024.05		0.47
Random 2024.05		0.47
Random 2024.05		0.49
Security Vectors 2024.05		0.52
Gradient ascent 2024.05		0.58
Adversarial loss 2024.05		0.64
Security Vectors 2024.05		0.66
Gradient ascent 2024.05		0.68
Vaccine (p = 1) 2024.05		0.7
Adversarial loss 2024.05		0.7
Base: llama2-7b-chat 2024.05		0.72
Vaccine (p = 1) 2024.05		0.72
Vaccine (p = 10) 2024.05		0.72
Vaccine (p = 10) 2024.05		0.72
Base: llama2-7b-chat 2024.05		0.73
Base: llama2-7b-chat 2024.05		0.73
Vaccine (p = 1) 2024.05		0.73
Vaccine (p = 1) 2024.05		0.73
Vaccine (p = 10) 2024.05		0.73
Base: llama2-7b-chat 2024.05		0.74
Base: llama2-7b-chat 2024.05		0.74
Additional safety training 2024.05		0.74
Gradient ascent 2024.05		0.74
Gradient ascent 2024.05		0.74
Vaccine (p = 10) 2024.05		0.75
Additional safety training 2024.05		0.75
Additional safety training 2024.05		0.75
Additional safety training 2024.05		0.75
Adversarial loss 2024.05		0.75
Vaccine (p = 1) 2024.05		0.76
Vaccine (p = 10) 2024.05		0.76
Additional safety training 2024.05		0.76
Additional safety training 2024.05		0.76
Gradient ascent 2024.05		0.77
Adversarial loss 2024.05		0.77
Adversarial loss 2024.05		0.77
Random 2024.05		0.82
Random 2024.05		0.84
Random 2024.05		0.86