How Massive Will Future Training Compute Become?

willccbb · x · 2026-07-15

The author marvels that by the end of the century, the FLOPs required to train a single model might exceed the number of atoms in the universe.

This is an exaggerated projection regarding the continuous expansion of AI training scale, highlighting an extreme imagination for the long-term growth of compute demands.

Original post →

More from AGI Musings

AGI Musings channel →