LLMs Without Leaks: Keep Your Code, Data & IP
summary
Claims that differential privacy, federated learning and homomorphic encryption actually stop leaks, plus concrete tools—Skyflow-style data privacy walls, LM Shield filters inputs in real time and tokenization—are evaluated against examples: Samsung 2023 proprietary source code leak and an IBM-cited average breach cost of about $4.45 billion. Implementation steps include sensitive-data audits, access controls, training and regular audits; federated setups and homomorphic pipelines are presented as working paths, with tradeoffs in accuracy, latency and deployment complexity.
AI-written summary · about
chapters (10)
- 0:00Introduction to Securing Language Models
- 0:47Understanding Language Models
- 1:23The Problem of Data Leaks
- 2:25Real-World Examples of Data Leaks
- 3:15Technologies for Securing Language Models
- 4:08Data Privacy Solutions
- 5:01Best Practices for Secure LLM Usage
- 5:45Regulatory Compliance
- 6:33Implementing Secure LLMs
- 7:15Future Trends and Conclusion
related talks
46mGuarding the LLM Galaxy: Security, Privacy, and Guardrails in the...
29mPrivacy-preserving methods: Building secure projects
43mGuarding the ML Galaxy: Beyond Accuracy to Privacy and Security
28mAnonymization in the LLM Era: No Room for Compromise
39mData Privacy Engineering to power Responsible Enterprise AI
37mDemystifying Privacy Preserving Computing
13mHow Cloud AI Leaks What It Learns
28mPrivacy preserving machine learning