Every large language model you have used writes one word at a time, in order. @GoogleDeepMind just shipped one that writes 256 at once, then refines the whole thing repeatedly.

Here is the breakdown. pic.twitter.com/j0jxqsCYFj

— Ronald van Loon (@Ronald_vanLoon) 5 octobre 2026

Every large language model you have used writes one word at a time, in order. @GoogleDeepMind just shipped one that writes 256 at once, then refines the whole thing repeatedly. Here is the breakdown.

→ View original post on X — @ronald_vanloon