Note: The field of AI safety research is young, and many ideas are contested.
Basics
- Excerpts from Geoffrey Hinton interview (40 seconds)
- Four Background Claims (10 mins)
- Why AI alignment could be hard with modern deep learning (20 mins)
- Why would AI “aim” to defeat humanity? (30 mins)
- 10 Reasons to Ignore AI Safety (16 mins)
Advanced
- DeepMind’s threat model literature review (30 mins)
- AGI Ruin: A list of lethalities (45 mins)
- Existential risk from power-seeking AI (31 pages)
- Without specific countermeasures, the easiest path to transformative AI likely leads to AI takeover (2 hours)
- Ngo and Yudkowsky on alignment difficulty (2 hours)
Technical AI safety
- What everyone in technical alignment is doing and why (45 mins, from Aug 2022)
- On how various plans miss the hard bits of the alignment challenge (35 mins, from July 2022)
- Technical AI Alignment reading list
AI Governance
Pause AI 