A model learned to write without the method that trains every AI.
Q Labs published a method called Dust that skips backpropagation, the error signal that normally updates every weight. Dust instead nudges the model's inner numbers at random and keeps whichever nudges cut the error. It matched the standard method only on tiny models, and the authors say it costs far more compute.
Why it mattersEvery model you use today was trained the same way, with no working substitute. A second route would matter most for hardware that cannot run the usual one.
▶ Related video: Backpropagation, intuitively | Deep Learning Chapter 3
Read in the app — free, in 9 languages
Related stories
Mathematicians cracked five open problems using an ordinary chat box.2026-10-05Untuned models solved agent tasks their polished versions could not.
2026-10-04Pretraining on everyday photos got a model to 70% on a reasoning test.
2026-10-04An AI trained for $8,000 beat Stratego's greatest player.
2026-10-03arXiv will take only two papers a month from each researcher.
2026-10-03