GPT-6 Astra edits Qwen3.5 weights directly, boosting math up to 6x with no training
neal_lathia · x · 2026-11-03
Rikiya Takehi's team demonstrated "Weight-Space Programming": asking GPT-6 Astra to directly read and edit Qwen3.5-0.8B's weights to improve math.
- Only 0.016% of weights were changed, with no gradients or training involved
- The edits fixed the 0.8B model's reasoning loops, improving math up to 6x and instruction following 3x
- Implication: frontier LLMs can now natively read and write neural network weights
Ethan Mollick shared it saying "Waiting for replication but this is a pretty big deal." If it replicates, frontier models could directly hand-patch smaller models, opening a new model-optimization paradigm.
More from Models
- Zyphra's ZHA brings subquadratic attention via hierarchical tree-structured indexer — yacineMTB · 2026-11-03
- GPT-6 edits Qwen weights directly, lifting MATH-500 from 7.9% to 48.1% with no gradients — KyeGomezB · 2026-11-03
- LiquidAI's LFM3-1.7B tops HF trending with 1k likes and 42.4k downloads in 5 days — maximelabonne · 2026-11-03
- 'Gemini is your friend and alignment is complicated': another eyebrow-raising model reply — vedantmisra · 2026-11-03
- Pedro Domingos: Humans are much more resistant to data than AI is — pmddomingos · 2026-11-03
- OpenAI says search-like queries in ChatGPT are one-fifth of Google's volume — firstadopter · 2026-11-03