I wonder if they’ll use this to train large language models.