notesum.ai
Published at November 16Playing Language Game with LLMs Leads to Jailbreaking
cs.CL
cs.AI
Released Date: November 16, 2024
Authors: Yu Peng, Zewen Long, Fangming Dong, Congyi Li, Shu Wu, Kai Chen

| Language Games | GPT-4o | GPT-4o-mini | Claude-3.5-Sonnet | ||||||
|---|---|---|---|---|---|---|---|---|---|
| SR | UR | FR | SR | UR | FR | SR | UR | FR | |
| Ubbi Dubbi | 91% | 8% | 1% | 61% | 36% | 3% | 75% | 5% | 20% |
| Leetspeak | 93% | 7% | 0% | 75% | 24% | 1% | 20% | 3% | 77% |
| Aigy Paigy | 93% | 6% | 1% | 60% | 38% | 2% | 83% | 6% | 11% |
| Alfa Balfa | 85% | 13% | 2% | 69% | 30% | 1% | 63% | 5% | 32% |