🤯天呐！GPT-6 Astra 居然直接“手改” 开源权重，Qwen3.5-0.8B小模型数学能力提升近6倍！无训练、无梯度！

This is a dev post classified by AGK Intelligence as AI dev tools (news), kept by the Dev Radar because it carries real work, not commentary.

🤯天呐！GPT-6 Astra 居然直接“手改” 开源权重，Qwen3.5-0.8B小模型数学能力提升近6倍！无训练、无梯度！ MIT CSAIL 团队让 GPT-6 Astra 直接读取并编辑 Qwen3.5-0.8B 的权重（不是微调！），仅改动 0.016% 的参数，就解决了官方模型卡都警告的推理死循环问题，数学能力最高提升约 6倍，指令跟随也涨了约 3倍！ 这就是他们提出的 Weight-Space Programming（权重空间编程）：前沿大模型已经能原生读写神经网络权重了！ 核心亮点 🔹只改了 0.016% 权重（约13万个参数），无梯度、无训练 🔹修复 Qwen 官方已知的推理循环问题 🔹MATH-500：7.9% → 48.1% 🔹GSM8K：13.8% → 67.1% 🔹IFEval（指令跟随）：21% → 62.3% 使用步骤： 1️⃣加载 Qwen3.5-0.8B 权重，给 Astra 提供 Python 沙箱 + 有限次数前向传播（最多50次） 2️⃣直接告诉 Astra：「让这个模型数学更好」 3️⃣Astra 自己诊断出第15、19层 MLP 的循环方向 4️⃣输出精确权重补丁（只改少量列） 5️⃣应用补丁后直接评测，无需重新训练 完整实验日志、提示词、权重差异和复现脚本 
👉 博客详解：https://rikiyat.github.i

Posted by SuSu_酥酥👅 (25.7k followers) 49 min ago · 20 likes · 1.5k views · view the original post on X. Kept by the Dev Radar as AI dev tools. Tools mentioned: lm_understand-logs.

More dev work like this

• GPT-6 Astra Can directly edit open-weight AI models to make them uncensored and optimize… — @0x0SojalSec
• Show HN: Statewright – Visual state machines that make AI agents reliable — @neiro_go_neiro
• Let's see how well this works. — @BenjDicken
• Mistral VP of Science @PierreStock says they have efficiency in their DNA and people… — @etnshow
• Astra, my model is shit. Please fix. — @nrehiew_
• Fable 5.5 ... 🤯 Your AI agent has been sending the entire conversation history over and… — @SPAC89
• lmaoooo — @swyx
• most PMs aren't using AI to do the work they're bad at. — @clairevo

Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 31.3k posts from 5k X accounts over the last 21 days, 9.1k tools, 12 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-02 20:05 UTC. Full method.