Fred Zhang

Activation patching reveals insights into language models' outputs and behaviors.

Stability in clustering ensures groups are effective and meaningful.

This study examines if language models can match human forecasting abilities.