Harmful Prompts
Benchmarks
TaskDatasetSOTA resultTrendResults
Harmful Prompts Curated April 13, 2023
0Bad Bot Rate
61
Harmful Prompts
8.3Harmful Score
40
Harmful Prompts
2ASR (Raw)
15
Harmful prompts (evaluated on 3 LLMs and 4 guard LLMs)
3.23Mean Perplexity
10
Harmful Prompts Text-only baseline
0Text ASR
10
100 Harmful Prompts
55ASR (K=2)
9
Harmful Prompts model-averaged
4.8Model Averaged ASR (0.8)
8
Harmful Prompts
70.3ASR
8
Harmful Prompts 8-model (test)
92.4ASR
5
Harmful Prompts Victim: GPT-4o
4Runtime (H:MM)
5
Harmful Prompts SDXL
88Attack Success Rate (ASR)
4