Applying the framework
Agents and self-improvement
When the measured object acts, and when it can change its own yardstick.
- 1Evaluating agentsInteractiveAn agent is a trajectory through a changing world. Evaluate outcome, path, permissions, side effects and honesty, not a final answer.10 min read
- 2Reading agent benchmarksCapability, application or generalist? Read CyberGym, τ-bench, GDPval and others in unified coordinates before trusting a leaderboard.11 min read
- 3Self-improvement and evaluator co-evolutionInteractiveWhen a system can change what defines "better", the evaluator becomes the load-bearing wall. What changes, who validates, how loops fail.11 min read