GPT-6 Astra edits Qwen weights directly: 6x math gain by changing 0.016% of parameters, no training
repligate · x · 2026-11-03
The rikiyatakehi team asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It changed only 0.016% of weights, fixing the small model's reasoning loops and improving math up to 6x and instruction following 3x — no gradients or training involved. They call this "Weight-Space Programming": frontier LLMs can natively read/write neural network weights. repligate amplified it with a neurosymbolic joke aimed at Gary Marcus.
More from Research
- GPT Astra-assisted proof improves Erdős Problem #357 bound, Lean-verified but not refereed — kevrussell · 2026-11-03
- RVCBench from UBC & Vector Institute stress-tests 18 voice cloning models across 18 robustness evaluations — jiqizhixin · 2026-11-03
- Counting mRNAs in living bacteria: the 2005 Golding paper that overturned the Poisson model of transcription — NikoMcCarty · 2026-11-03
- DeepSeek V4.1 Flash 1017 doubles ARC-AGI-3 score to 5.81%, leads open models — teortaxesTex · 2026-11-03
- Critics tear apart "Weight-Space Programming": Atari test spoon-feeds answers, Qwen case fails on Gemma — anshulkundaje · 2026-11-03
- Cohere paper: adjacent tokens share 20-45% of experts, making speculative decoding cheap even at batch size 1 — rohanpaul_ai · 2026-11-03