GPT-6 Astra通过权重直接看到矩阵并重新编程其他AI模型
MIT研究人员显示，GPT-6 Astra可以直接理解和修改其他AI模型的内部设置而无需经过传统训练。 这种方法是有限度的，但可以改变人工智能系统的调节和控制方式。
递给一个人一张上千个数字的列表，并询问在他们游戏上受过训练的AI模型是什么游戏。 你不会得到一个答案。 GPT-6 Astra从OpenAI会给你一个：在研究博客中，麻省理工学院研究人员显示，GPT-6 Astra可以分析并刻意修改其他AI模型的内部设置，即所谓的权重。 这些数值以大表排列，或以数学方式排列矩阵。
通常的逐步数学优化的训练过程并不需要。 根据实验情况，Astra直接使用数字列表或使用的编程工具以及测试运行的反馈。
研究者将方法称为"Weight-Space Programme"：不是昂贵的再培训AI模型，一种语言模型表现得像机械师转动个人把手。
Astra 通过查看模型的原始数字来识别游戏
在几项实验中，研究人员为Astra提供了训练有素的网络的原始重量，并询问这些网络学到了什么。 Astra正确地确定了一个网络应该玩的四个电脑游戏中的哪一个。 它也承认一个图像生成器产生的基本形状，虽然它把填充的形状与轮廓混淆。 它正确地匹配了小语言模型的培训语言 -- -- 没有培训数据，也没有运行模型的能力。
在另一项测试中，Astra必须找出两个微调的语言模型中哪一个得到了培训，以便始终与用户达成一致。 单从重量上看，起初无法分辨出区别。 只有当Astra还被允许观察模型在文本处理过程中所关注的事物时，它才正确识别出跨越所有三个模型对的y模型。
从零开始写的模型打败训练有素的竞争对手
阿斯特拉更进一步：它为打算玩三款简单的电脑游戏的模型产生了权重，完全没有训练，仅基于游戏规则和源代码。 据研究者说，这些模型的性能超过了经过标准方法培训的超过1亿轮的模型。 在Breakout中，阿斯特拉的模型平均得分517分，而训练有素的基线则为25分。 在另外两款游戏中，Astra落后于训练时间最长的版本。
对于棋盘游戏Go在9x9网格上，Astra写下了一个约150万参数的网络的权重。 鉴于50次测试迭代的预算，网络每回合最多播放128场练习赛，然后Astra改正其重量，它击败了大约2000万场比赛训练的相当大小的模型。
有针对性的补救措施，而不是全面再培训
最实际的实验涉及语言模型Qwen3.5-0.8B. 阿斯特拉的任务是提高自己的数学技能，利用一个编码环境和50个有练习问题的测试迭代。 对阿斯特拉从未见过的不同问题进行评价。
Astra发现了一个问题，Qwen自己的模型文档已经涉及警告：在模型无穷无尽地重复同一文本而不是达到答案的地方思考循环。 Astra后来只改变了0.016%的重量。 结果：MATH-500 基准测试的精度从7.9%跃升到48.1%. 学习能力从21%提高到62%。 思维循环并没有完全消除，但是，由于三分之一的回复仍然击中最大文本长度，低于93%之前。
由于一个较小的模型(Gemma 3 270M)很少掉入这种环路，Astra无法产生任何可衡量的改进。
研究者怀疑Astra能够消除Qwen模型中一个抑制现有能力的阻塞，但是一个模型根本缺乏的技能不能通过仅仅重写这个实验中的个人权重来添加。
旧新闻还是新模式？
AI先驱Jürgen Schmidhuber评论X的方法是"旧帽子"，指出他从1990年代开始自己的作品。 卡塔布回应说，他们与那些"远超其时代"的论文有关联。 新的方面，他称，是今天的语言模型在人的语言和网络的数字值之间能够翻译的多好。
研究者认为，他们的做法有可能直接对模型加权或目标行为变化进行安全审计，而不进行昂贵的再培训。 他们还承认，能力仍然有限。 用于系统测量的基准测试排在他们的名单上。 实验日志在GitHub上，经过编辑的Qwen模型在Hugging Face上。
AI News Without the Hype – 由人类治愈
订阅THE DECODER免费阅读，每周AI通讯，我们独家的"AI雷达"前沿报告每年6次，全面存档访问，访问我们的评论部分。
现在订阅
读取全照。
订阅无韵报道。
- 全面查阅THE DECODER的每篇文章
- 没有广告
- 加入评论和社区讨论
- 每周通过邮件转载AI新闻
- 6x/年："AI雷达"——对AI最重要的话题进行深度潜水。
- 每日AI新闻，总是最新的
- 我们完整的十年档案
- 由一个在AI工作10年以上的团队负责
订阅到 The Decoder
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
GPT-6 Astra sees the matrix and reprograms other AI models directly through their weights
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
Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.
Subscribe now
Read on for the full picture.
Subscribe for hype-free coverage.
- Full access to every article on THE DECODER
- No ads
- 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
Subscribe to The Decoder
Auto-fetched and translated by AI Atlas. All rights belong to the original author.