ParentFull threadp1mrx·Maybe you could train an LLM to recognize formulas (how else are you going to speak a formula coherently?) and treat the rest as normalized text.View on HN