Nostalgia: When GPT-2's 1.5B Parameters Were Considered 'Extra Large'

cloneofsimo · x · 2026-08-09

Developer @cloneofsimo tweeted about the inflation of model sizes, expressing nostalgia for the days when GPT-2 was dubbed 'Extra Large' despite having only 1.5 billion parameters. This stands in stark contrast to today's models routinely reaching hundreds of billions or even trillions of parameters.

Original post →

More from Fun

Fun channel →