AI Alignment

The mission to align AI goals to be that of ours, to prevent misuse.

Formal Definition

AI $A$ is aligned with operator $H$ if $A$ is trying to do what $H$ wants it to do.

Problems

• Misalignment
  • Inner Misalignment
  • Outer Misalignment
• Sycophancy
• Goal Misgeneralization
• Proxy Training Problem

Concepts

• AGI
• Transformative AI
• AI Impacts
  • Extinction
  • Existential Catastrophe
• Safety Washing
• Mechanistic Interpretability
• Neurointerpretability
• Shard Theory

Contrast

• AI Governance
• AI Resilience

Talks

Courses

• https://bluedot.org/courses/alignment