DAWE edits 5,064 of 4B weights to fix harness mismatch, no retraining needed

lateinteraction · x · 2026-11-18

llmluthor open-sourced DAWE (Data-Attributed Weight-Editing): when their model broke on a new harness, instead of retraining they edited just 5,064 weights out of 4 billion, lifting accuracy from 20.1% to 88.2% with no gradient descent — and old-harness accuracy stayed intact. Optimization researchers call this wave of "weight space programming" mindblowing.

Related event: DAWE Adapts Models by Editing Just 5,064 Weights, No Retraining(2 posts)→

Original post →

More from coding & agent

coding & agent channel →