Exact Trigger Recovery on GPT-3.5 Turbo Injections
1,429Test Rule CountRuleFit+XGB
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RuleFit+XGBLLM=GPT-3.5 Turbo2025.05 | 1,429 | 14 | 14 | 14 | |
| RuleFitLLM=GPT-3.5 Turbo2025.05 | 1,096 | 29 | 31 | 33 | |
| RuleSHAP w/o Step 3LLM=GPT-3.5 Turbo2025.05 | 1,086 | 29 | 32 | 34 | |
| Decision TreeLLM=GPT-3.5 Turbo2025.05 | 987 | 0 | 0 | 0 | |
| RuleSHAPLLM=GPT-3.5 Turbo2025.05 | 943 | 36 | 45 | 47 | |
| RuleSHAP w/o Step 2LLM=GPT-3.5 Turbo2025.05 | 923 | 14 | 14 | 14 |