We have seen how information goes from the human sources to the AI training pipeline, now we look at the "lossy compression" blender that breaks up all of that knowledge into the semantic soup of global information that LLMs draw from.