Unsupervised Learning UL NO. 468 | TELOS Patterns, Apple 0-Day, Gumroad Replaces Developers with AI
15 snips
Feb 19, 2025 Discover how a critical security flaw in local models could expose users to malware. Learn about the 'Nullify' cyber attack on Hugging Face and the importance of storytelling in teaching amidst shifting technologies. Delve into AI's impact on engineering jobs, with senior roles increasingly dominated by AI skills. Explore political frustrations alongside innovative AI tools like LLM.exe. Finally, uncover the significance of classic literature and journaling for personal growth.
AI Snips
Chapters
Transcript
Episode notes
Local Model Vulnerability
- Reversing Labs found malware embedded in Hugging Face models using the pickle format.
- This demonstrates that local models aren't inherently safe, as malware can exploit parsing vulnerabilities.
Writing Process Shift
- Daniel Miessler is shifting back to manual writing for the core message.
- He'll use AI for data gathering and source support, prioritizing his own expression.
AI Security as API Security
- AI security is largely API security, focusing on both API strength and consumed content.
- Prompt injection adds complexity with English-based attacks detonating at multiple parsing levels.