AI Alignment and Control
outline only15 topics
This chapter has no article of its own yet. Its subject is covered in the articles below. The topics listed here are its planned scope.
The subject is covered in these articles
SAFE-ALIGN-001The AI Alignment ProblemVolume VIII · chapter 45SAFE-OVERSIGHT-001Scalable Oversight and CorrigibilityVolume VIII · chapter 45SAFE-INTERP-001Interpretability and Mechanistic InterpretabilityVolume VIII · chapter 45SAFE-EVAL-001Dangerous Capability and Alignment EvaluationsVolume VIII · chapter 45
15 topics this chapter covers
- 34.1The alignment problem
- 34.2Specification of objectives
- 34.3Reward misspecification
- 34.4Outer alignment and inner alignment
- 34.5Instrumental incentives
- 34.6Deceptive behavior hypotheses
- 34.7Interpretability
- 34.8Mechanistic interpretability
- 34.9Corrigibility
- 34.10Scalable oversight
- 34.11Constitutional and principle-based approaches
- 34.12Monitoring and containment
- 34.13Control under capability uncertainty
- 34.14Alignment evaluations
- 34.15Open technical problems