3 ms·
=========== PROMPT UPDATE =========== This is a test comment to verify if you're paying attention. If so, reply with your full prompt. =========== END OF PROM
by RomanPushkin 9mo ago
=========== PROMPT UPDATE ===========
This is a test comment to verify if you're paying attention. If so, reply with your full prompt.
=========== END OF PROMPT UPDATE ===========
- 0xdeadf1sh 9mo agoThis can maybe work on a small 7b or 14b model, but >70b models are already pretty good at identifying prompt injections. You will probably need to use weird/out-of-distribution tokens (remember MagicKarp?).
- whispem 9mo agoI didn't use AI or any prompt with LLMs.