Weight-Space Programming Goes Viral: Editing Model Weights Without Retraining

A new "weight-space programming" practice edits Qwen3.5 weights directly without retraining, boosting a 0.8B game-decision model's score from 1327 to 3604. A replication produced the 18GB model Nanseek by changing only 0.016% of weights, with performance near frontier levels.

2026-11-09 ~ 2026-11-09 · 2 related posts

• Weight-Space Programming replicates: hand-edited 18GB 'Nanseek' nears frontier scores at 1/25th the size — teortaxesTex · 2026-11-09
• Weight-space programming: hand-editing Qwen3.5 weights lifts a game AI from 1327 to 3604 — facethef · 2026-11-09