Post says Google DeepMind has gone over a year without pretraining a new base model

teortaxesTex · x · 2026-07-22

A post claims Google DeepMind’s apparent inability to pretrain a new base model for over a year is a major self-own, arguing that a compute-rich company should not be moving this cautiously.

The attached reply frames it as a strategic failure: even a half-baked new pretrain in early 2026 would likely outrun Google’s current lineup, so the delay looks increasingly hard to justify.

Original post →

Post says Google DeepMind has gone over a year without pretraining a new base model

More from Companies & People