3 ms·

Maybe in text, but we won't be running out of multi-modal training data (images, audio, video, sensor data, etc) any time soon.