LLM Safety Filters

Security & Ethics

1. phi-2 prompt injection detector (https://huggingface.co/ysy970923/phi-2-prompt-injection-QLoRA)

2. Profanity Filters with Zero-Shot LLM Prompting

3. PII (personal identifiable information) checking via Microsoft Presidio Analyzer

Information value estimate

1. RAG based LLM prompting with existing responses in the vector DB.

   1. check plagiarism

   2. estimate information value