跳到正文
Aran Komatsuzaki· @arankomatsuzaki · X·· 20 天前AI 评分58
AI 导读

MIT CSAIL 的 Rikiya Takehi 等人提出 Weight-Space Programming,称前沿 LLM 能原生读写神经网络权重。研究称 GPT-6 Astra 仅直接改动 Qwen3.5-0.8B 的 0.016% 权重,无需梯度或训练,就修复了其推理循环,MATH-500 从 7.9% 升至 48.1%,指令遵循的 IFEval 提升 3 倍。Aran Komatsuzaki 表示即将采访论文作者并征集提问。

正文 · 原文

We're interviewing the authors soon

what would you like to ask them? Drop your questions below :)

【引用 @rikiyatakehi】We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.

We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: https://rikiyat.github.io/blog/weight-space-programming/

引用Rikiya Takehi@rikiyatakehi
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: https://rikiyat.github.io/blog/weight-space-programming/
在 X 查看被引用的帖子

来源:Aran Komatsuzaki · x.com