That's insanely cool. Now how about actually compress text with LLM that summarize long texts with the non-Euclidean visulization...