Foundational Papers

Defining the Field

• Concrete Problems in AI Safety (2016)
• Alignment for Advanced Machine Learning Systems (2016)

Inner Alignment

• Risks from Learned Optimization (2019)
• Goal Misgeneralization (2022)

Solutions

• InstructGPT (2022)
• Constitutional AI (2022)
• AI Safety via Debate (2018)

Interpretability

• Transformer Circuits (2021)
• Towards Monosemanticity (2023)

Theory

• The Basic AI Drives (2007)