“The training data for the Talkie language model was 240 billion tokens.”