跳到正文
arXiv cs.AI· Jiajun Fan, Neil He, Changyi Yang, Andy Mao, Xi Zhu, Chaoran Cheng, Kaiqu Liang, Shuaike Shen, Zetian Li, Ruihan Guo, Ge Liu·· 4 小时前AI 评分71

语言模型可改写自身权重:27B 模型无梯度自我编辑提升准确率

Language Models Can Rewrite Their Own Weights

AI 导读

arXiv 论文提出“self-rewriting”,让 27B 模型在无梯度、无训练的情况下诊断自身失败并改写权重,在预先固定的 490 道未见题目上准确率提升 6.6 个点。

正文 · 原文

Authors:Jiajun Fan, Neil He, Changyi Yang, Andy Mao, Xi Zhu, Chaoran Cheng, Kaiqu Liang, Shuaike Shen, Zetian Li, Ruihan Guo, Ge Liu

View PDF HTML (experimental)

Abstract:Language models are trained once, frozen and deployed: improving one again takes another training run, and a prompt, a scaffold or extra computation at inference is gone once the task ends. A frontier model has been shown to edit the weights of a much smaller one. We ask whether a model can rewrite its own weights to make itself better, and find that it can, by its own reasoning, with no gradients and no training; we call this self-rewriting. Given its weights, graded answers and the ability to run itself, a 27B model diagnoses its failure and edits the weights behind it. It surpasses itself: its own edit makes it 6.6 points more accurate on 490 problems it never saw, in a test fixed in advance, and another session's edit also makes it significantly better. It keeps getting better: it submits an edit only when its own tests support it, each edit replacing the last, and over its first eight such sessions its accuracy only rises. Unlike test-time scaling, the gain persists in its weights and adds no cost per answer; unlike training, it needs no pre- or post-training infrastructure, only an open model and hours on one GPU. It repairs a damaged copy of itself: it locates the corrupted matrix and rewrites it with values of its own, never reading the undamaged weights; found on text math problems alone, the repair lifts the copy on six image benchmarks it never saw from 7% to 70% accuracy, 83% of the undamaged model's. And the ability emerges with scale: across four sizes of one family, the 2B model makes no edit and the 4B and 9B models break themselves; only the 27B model works on itself as a human would, comparing its hidden states on answers that finish and on answers that run out of tokens, weakening the direction that separates them. Like a human, a model need not separate training from inference; improving a language model may come to take a conversation rather than a training run.
Subjects: Artificial Intelligence (cs.AI)
Cite as: arXiv:2611.21885 [cs.AI]
  (or arXiv:2611.21885v1 [cs.AI] for this version)
  https://doi.org/10.48550/arXiv.2611.21885

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Jiajun Fan [view email]
[v1] Mon, 16 Nov 2026 20:55:08 UTC (6,363 KB)

来源:arXiv cs.AI · arxiv.org