But I doubt it justifies the billions of dollars getting burned for training language models and building power plants.
And are full transcriptions not the better option?