The moment you begin evaluating a model, you introduce a feedback loop that shapes what the model learns to optimize for. Observation is not neutral. It never was.
The Observer Effect in AI
How measuring a model changes what it becomes — and why alignment research faces a fundamental epistemic problem.
