45 Trillion Tokens is Just the Beginning

iamtrask · x · 2026-07-17

The author relays a judgment on data scale: 45 trillion tokens roughly equates to 300TB of text, but the world's data volume far exceeds this scale—potentially by up to a billion times.

The core message is that data currently used to train models covers only a fraction of human-accessible information. Therefore, "data exhaustion" is not the end, and AI training is far from hitting a wall.

Original post →

More from AGI Musings

AGI Musings channel →