GPT-6 edits just 0.016% of Qwen3.5-0.8B weights, 6x math gains with no gradients
udmrzn · x · 2026-11-03
rikiyatakehi's team had GPT-6 Astra directly edit Qwen3.5-0.8B weights to improve math, a technique they call Weight-Space Programming: frontier LLMs can natively read and write neural network weights.
- Only 0.016% of weights changed — no gradients, no training.
- Fixed the 0.8B model's reasoning loops; math improved up to 6x and instruction following 3x.
- Commentators call it a striking demonstration of direct weight editing's potential.
- Full details in their blog post thread.
Related event: MIT paper shows LLM editing 0.016% of weights boosts small model math 6x(24 posts)→
More from Models
- Sol 6.1 still best cost-per-intelligence worker; Fable 5.5 more well-rounded — haider1 · 2026-11-03
- Critics tear apart "Weight-Space Programming": Atari test spoon-feeds answers, Qwen case fails on Gemma — anshulkundaje · 2026-11-03
- Spider-Verse-style 2D animation built with Opus fuels rumors it routes to Fable 5.5 — minchoi · 2026-11-03
- More "Fable 5.5" demos: pelican, horse and bike scenes drawn and animated entirely in code — minchoi · 2026-11-03
- One prompt, four models: rumored "Fable 5.5" ships a full 3D open-world RPG while rivals stumble — minchoi · 2026-11-03
- Rumored Claude Fable 5.5 beats Opus 5.5 and Gemini 4 Argon on three.js giraffe test — minchoi · 2026-11-03