Readings

Abstraction: Curated reading list for Dan Hendrycks ML safety course curriculum

Key points:

• Course structured around four pillars: Robustness, Monitoring, Control, and Systemic Safety
• Robustness section covers adversarial robustness, long-tail distribution shift, and OOD detection
• Monitoring section includes trojans, interpretable uncertainty, and detecting emergent behavior
• Control section addresses power-seeking AI, honest AI, and machine ethics
• Systemic Safety covers forecasting, ML for cyberdefense, and cooperative AI
• Also includes X-Risk section; italicized resources are required, others are suggested

Connections: Dan Hendrycks · Center For AI Safety · AI Safety · Robustness · Machine Ethics · AI Alignment

Source: https://course.mlsafety.org/readings/