x
Beyond Cosine Similarity: Auditing Knowledge Graphs via Semantic Diffraction — LessWrong