↓Skip to main content
Zeitgeist — a spike by Chris Gathercole
  1. Creators/
  2. Robert Miles — Robert Miles AI Safety/

Just Predicting Tokens

YouTube (Short) · YouTube

  • Addresses the common dismissal that large language models “just predict the next token,” pushing back on the inference that this training objective caps what capabilities can emerge.
  • Consistent with his recurring myth-busting Shorts format, tackling a specific popular objection used to wave away AI capability or risk claims.
  • Takeaway: the mechanism behind next-token prediction doesn’t itself bound the sophistication of resulting behavior — “it’s just predicting tokens” isn’t, on its own, a safe argument for dismissing capability or risk.