AnalysisAI ModelsNovember 3, 2026

Simon Willison: Qwen 3.8 27B on a laptop can interpret model weights

Read original source →simonwillison.net

Willison reproduced an example from the paper "Sparks of Weight-Space Programming" (Takehi, Teoh, Zhang, Khattab) with Qwen 3.8 27B locally, after 45+ minutes of prompting with raw floating-point weights. The paper reports GPT-6 Astra editing 0.016% of Qwen3.5-0.8B's weights to lift MATH-500 from 7.9% to 48.1%.

2 sources

More stories today

Open the live feed