1 comment

[ 4.6 ms ] story [ 12.8 ms ] thread
Accuracy and coherence dominate how we evaluate AI systems. But systems can be accurate while still degrading meaning. This framework proposes a way to evaluate fidelity to human interpretation and real-world alignment.