Скачать или смотреть Secure LLM Architecture - Testing LLM Guard

Exploring LLM Architectures, Security, and LLM Guard

LLM-Guard: https://github.com/protectai/llm-guard
Testing Repo: https://github.com/latiotech/insecure...
More about Latio: https://www.latio.tech/

This video discusses LLM (Large Language Model) architectures and addresses security concerns associated with their use. The presenter critiques certain types of security tools as ineffective, highlighting the distinction between low-risk, basic implementations and more complex applications that necessitate advanced security measures. The video introduces LLM Guard, an open-source tool designed to enhance LLM security by checking both inputs for malicious intent and outputs for sensitive data. Through practical demonstrations with Hugging Face and OpenAI models, the presenter showcases how LLM Guard can mitigate risks such as prompt injection and unauthorized access to sensitive information, emphasizing the importance of output monitoring and sanitization. The session concludes with a discussion on the evolving landscape of LLM security and the critical role of permissions and monitoring in safeguarding data.

00:00 Introduction to LLM Architectures and Security Concerns
00:10 Exploring LLM Architecture and Security Tools
05:07 Diving into LLM Guard
06:27 Hands-On Examples with LLM Guard
12:56 Advanced Security Concerns and Future Directions
15:49 Closing Thoughts on LLMs and Security