AGIArtificial General Intelligence - AI with human-level abilities across all domains
ASIArtificial Superintelligence - AI far surpassing human intelligence
ANIArtificial Narrow Intelligence - AI that excels at specific tasks (today's AI)
AlignmentThe challenge of ensuring AI does what humans actually want
SingularityHypothetical point when AI surpasses human intelligence and improves recursively
IntelligenceExplosionRapid, recursive AI self-improvement
RLHFReinforcement Learning from Human Feedback - training AI on human preferences
ConstitutionalAITraining AI to follow a set of principles
InterpretabilityUnderstanding how AI models work internally
CorrigibilityAI property of allowing itself to be corrected
InstrumentalConvergenceTendency of different goals to lead to similar sub-goals
OrthogonalityThesisIntelligence and goals are independent
FLOPSFloating Point Operations Per Second - measure of compute power