Am I understanding correctly? You look an English dataset, trained an LLM, machine translated the English dataset to e.g. Spanish, continued training the model, and performance for queries in English improved? That’s really interesting.