Exact Trigger Recovery on GPT-4o Injections
4,090Trigger Count (Rules)RuleFit+XGB
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RuleFit+XGBLLM=GPT-4o2025.05 | 4,090 | 36 | 36 | 36 | |
| RuleSHAP w/o Step 3LLM=GPT-4o2025.05 | 3,104 | 71 | 71 | 72 | |
| RuleSHAP w/o Step 2LLM=GPT-4o2025.05 | 2,842 | 36 | 36 | 36 | |
| RuleFitLLM=GPT-4o2025.05 | 2,489 | 50 | 54 | 55 | |
| RuleSHAPLLM=GPT-4o2025.05 | 2,316 | 79 | 79 | 80 | |
| Decision TreeLLM=GPT-4o2025.05 | 1,057 | 0 | 0 | 0 |