Define AGI operationally
Specify target capabilities
Set measurable benchmarks
Test across diverse tasks
Compare to human performance
Measure generalization
Measure transfer learning
Measure sample efficiency
Measure robustness
Measure reasoning consistency
Measure long-horizon planning
Measure tool use
Measure autonomy
Measure adaptability to novelty
Measure memory retention
Measure self-correction
Measure alignment with goals
Measure safety under stress
Measure performance in real-world settings
Track failure modes
Validate across domains
Repeat evaluations over time
Use blinded human judgments
Use adversarial testing
Use out-of-distribution tests
Use benchmark suites
Use longitudinal studies
Compare against strong baselines
Require sustained performance
Check for deceptive behavior
Check for catastrophic errors
Review results independently