Paper Notes

Notes on research papers, including key ideas, methodological choices, limitations, and questions worth carrying into future work.

Aug 22, 2026

Rethinking AI Evaluation Through Human-AI Interaction

A reflection on Arbour, Bojinov, Feller, and Ni (2026): AI should be evaluated by its causal effect on human outcomes, not only by benchmark performance.

AI evaluationHuman-AI interactionCausal inference