A framework for mapping the internal geometry of transformer representations using angular projection, neuron-level modulation, and epistemically grounded prompts. Based on and extending Bird's original Spotlight Resonance Method (SRM).
machine-learning transformers representation-learning neuron epistemology interpretability epistemic-uncertainty geometric-analysis spotlight-resonance-method
-
Updated
Apr 19, 2025 - Python