I hope something comes of this. Marking up LLM output to explicitly quite source material, so even if the model hallucinates the reader can discern: https://mattyyeung.github.io/deterministic-quoting