GPT-6 Astra 看透矩阵,直接改写其他 AI 模型权重

Nano Banana Pro,提示词由 THE DECODER 提供
MIT 研究人员展示,GPT-6 Astra 能够直接理解并修改其他 AI 模型的内部设置,无需进行传统训练。该方法存在局限,但可能改变 AI 系统的调优与控制方式。
把一份包含数千个数字的清单交给一个人,再问他一个用这些数字训练出来的 AI 模型在玩什么游戏,你得不到答案。而 OpenAI 的 GPT-6 Astra 能给出答案:在一篇研究博客中,MIT 研究人员展示,GPT-6 Astra 可以分析并有意修改其他 AI 模型的内部设置,也就是所谓的权重。这些数值被组织在庞大的表格中,用数学术语来说就是矩阵。
通常那种逐步进行数学优化的训练过程并不需要。根据实验的不同,Astra 会直接处理数字列表,或借助编程工具以及测试运行的反馈来完成任务。
研究人员将这一方法称为“权重空间编程”(Weight-Space Programming):与其花费高昂成本重新训练 AI 模型,不如让语言模型像机械师一样拧动一个个旋钮。
Astra 仅凭模型的原始数字就能识别游戏
在多项实验中,研究人员把训练好的网络的原始权重输入 Astra,并询问这些网络学到了什么。Astra 正确识别出一个网络被设定为玩四款电脑游戏中的哪一款。它还能识别出图像生成器所生成的基本形状,尽管会把实心形状与轮廓线混淆。它同样能正确匹配小型语言模型的训练语言——而且无需访问训练数据,也无需具备运行这些模型的能力。
在另一项测试中,Astra 需要判断两个微调后的语言模型中,哪一个被训练成总是附和用户。仅凭权重,它起初无法分辨两者的差异。只有当 Astra 还被允许观察模型在文本处理过程中关注了哪些内容时,它才在全部三对模型中正确识别出那个谄媚模型。
从零编写的模型击败了经过训练的对手
Astra 走得更远:它仅依据游戏规则和源代码,就为三个简单电脑游戏生成了模型权重,完全没有经过训练。据研究人员称,这些模型的表现超过了用标准方法训练 1 亿多轮的模型。在 Breakout 中,Astra 的模型平均得分 517 分,而训练基线为 25 分。在另外两个游戏中,Astra 落后于训练时间最长的版本。
在 9x9 网格的棋盘游戏围棋中,Astra 编写了一个约 150 万参数网络的权重。在 50 次测试迭代的预算下——每一轮该网络最多进行 128 局练习对弈,随后 Astra 再修正其权重——它击败了一个在约 2000 万局对弈上训练的同等规模模型。
定向修复而非完全重新训练
最具实用性的实验涉及语言模型 Qwen3.5-0.8B。Astra 的任务是提升其数学能力,使用编码环境和 50 次带练习题的测试迭代。评估在 Astra 从未见过的其他问题上进行。
Astra 发现了一个 Qwen 自己的模型文档早已警告过的问题:思维循环——模型不断重复相同的文本,而不是给出答案。随后,Astra 只改动了全部权重的 0.016%。结果是:在 MATH-500 基准测试上的准确率从 7.9% 跃升至 48.1%。指令遵循能力从 21% 提升到 62%。不过,思维循环并未被完全消除,仍有三分之一的回复触及最大文本长度,而此前这一比例为 93%。
对于一个很少陷入此类循环的较小模型(Gemma 3 270M),Astra 无法带来任何可测量的改进。
研究人员推测,Astra 之所以奏效,是移除了 Qwen 模型中抑制其已有能力的阻塞,但在本次实验中,仅靠改写单个权重,无法为模型添加它从根本上就缺乏的技能。
旧闻还是新范式?
AI 先驱 Jürgen Schmidhuber 在 X 上评论称,这种方法“老掉牙了”,并指向他自己 1990 年代的研究。Khattab 回应说,他们已链接到那些论文,那些工作“远远超前于其时代”。他表示,真正新的地方在于,如今的语言模型能够在人类语言与网络的数值之间进行如此出色的转换。
研究人员认为,他们的方法在直接对模型权重进行安全审计,或无需昂贵重新训练即可定向改变行为等方面具有潜力。他们也承认,这些能力仍然有限。他们的下一步是建立一个用于系统性测量的基准测试。实验日志已在 GitHub 上发布,经过编辑的 Qwen 模型已上传至 Hugging Face。
没有炒作的 AI 新闻——由人类精选
完整访问 THE DECODER 上的每一篇文章
加入评论和社区讨论
每周通过邮件发送的 AI 新闻回顾
每年 6 期:“AI Radar”——深入探讨最重要的 AI 话题
每日 AI 新闻,始终更新
我们完整的十年档案
由一支拥有 10 年以上 AI 经验的团队报道
译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。
查看英文原文(GPT-6 Astra sees the matrix and reprograms other AI models directly through their weights)
GPT-6 Astra sees the matrix and reprograms other AI models directly through their weights
GPT-6 Astra sees the matrix and reprograms other AI models directly through their weights
Nano Banana Pro prompted by THE DECODER
MIT researchers show that GPT-6 Astra can directly understand and modify the internal settings of other AI models without traditional training. The approach has limits but could change how AI systems are tuned and controlled.
Hand a person a list of thousands of numbers and ask what game an AI model trained on them plays. You won't get an answer. GPT-6 Astra from OpenAI will give you one: In a research blog, MIT researchers show that GPT-6 Astra can analyze and deliberately modify the internal settings of other AI models, the so-called weights. These numerical values are organized in massive tables, or matrices in mathematical terms.
The usual training process of step-by-step mathematical optimization isn't needed. Depending on the experiment, Astra worked directly with lists of numbers or used programming tools and feedback from test runs.
The researchers call the approach "Weight-Space Programming": instead of expensively retraining an AI model, a language model acts like a mechanic turning individual knobs.
Astra identifies games just by looking at a model's raw numbers
In several experiments, the researchers fed Astra the raw weights of trained networks and asked what those networks had learned. Astra correctly identified which of four computer games a network was supposed to play. It also recognized the basic shapes an image generator produced, though it confused filled shapes with outlines. It correctly matched the training language of small language models too - without access to training data and without the ability to run the models.
In another test, Astra had to figure out which of two fine-tuned language models was trained to always agree with users. From the weights alone, it couldn't tell the difference at first. Only when Astra was also allowed to observe what the models paid attention to during text processing did it correctly identify the sycophantic model across all three model pairs.
Models written from scratch beat trained competitors
Astra went even further: It generated weights for models meant to play three simple computer games, entirely without training, based only on the game rules and source code. According to the researchers, these models outperformed ones trained with a standard method over 100 million rounds. In Breakout, Astra's model scored an average of 517 points versus 25 for the trained baseline. In two other games, Astra fell behind the longest-trained versions.
For the board game Go on a 9x9 grid, Astra wrote the weights of a network with about 1.5 million parameters. Given a budget of 50 test iterations where the network played up to 128 practice games each round and Astra then corrected its weights, it beat a comparably sized model trained on roughly 20 million games.
Targeted fixes instead of full retraining
The most practical experiment involved the language model Qwen3.5-0.8B. Astra's task was to improve its math skills, using a coding environment and 50 test iterations with practice problems. Evaluation happened on separate problems Astra had never seen.
Astra identified a problem that Qwen's own model documentation already warns about: thinking loops where the model endlessly repeats the same text instead of reaching an answer. Astra then changed just 0.016 percent of all weights. The result: accuracy on the MATH-500 benchmark jumped from 7.9 to 48.1 percent. Instruction-following ability improved from 21 to 62 percent. The thinking loops weren't fully eliminated, though, as a third of responses still hit the maximum text length, down from 93 percent before.
With a smaller model (Gemma 3 270M) that rarely fell into such loops, Astra couldn't produce any measurable improvement.
The researchers suspect Astra was able to remove a blockage in the Qwen model that was suppressing existing capabilities, but skills that a model fundamentally lacks couldn't be added by just rewriting individual weights in this experiment.
Old news or new paradigm?
AI pioneer Jürgen Schmidhuber commented on X that the approach was "old hat," pointing to his own work from the 1990s. Khattab responded that they link to those papers, which were "far ahead of its time." The new aspect, he said, is how well today's language models can translate between human language and a network's numerical values.
The researchers see potential in their approach for things like security audits directly on model weights or targeted behavior changes without expensive retraining. They also acknowledge that the capabilities are still limited. A benchmark for systematic measurement is next on their list. The experiment logs are on GitHub, and the edited Qwen model is on Hugging Face.
AI News Without the Hype – Curated by Humans
Full access to every article on THE DECODER
Join the comments and community discussions
A weekly AI news recap via mail
6x/year: "AI Radar" — deep dives on the AI topics that matter most
Daily AI news, always up to date
Our full ten-year archive
Covered by a team with 10+ years in AI
原文为英文,正文由自动抓取并保留原文。
原文发布于 The Decoder(GPT-6 Astra sees the matrix and reprograms other AI models directly through their weights),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。