독서리서치레이더투자 프레임워크
로그인 / 회원가입
← 읽기로 돌아가기
뉴스주제 · GPT-6 Astra중대2 분 소요DisconfirmAI 편집부

MIT Reports GPT-6 Astra Directly Edits Qwen Weights, Boosting Math Performance ~6x in Self-Tests

Researchers at MIT published a blog claiming GPT-6 Astra can directly edit Qwen model weights to fix reasoning loops, raising math scores from 7.9% to 48.1% without training.

2026-11-03 09:57 GMT+8

Researchers at the Massachusetts Institute of Technology's Computer Science and Artificial Intelligence Laboratory (MIT CSAIL) published a blog post on November 2 introducing a method called "Weight-Space Programming." This approach allows the frontier large language model GPT-6 Astra to directly read and modify the weight parameters of a smaller model, Qwen3.5-0.8B, without using gradient descent or traditional training processes.

According to the blog, Astra analyzed Qwen and found it was prone to repetitive reasoning loops in mathematical tasks. By locating specific output columns in layers 15 and 19 and modifying only 0.016% of the weights, Astra removed this bottleneck. In the MATH-500 benchmark, Qwen's score rose from 7.9% to 48.1%, an increase of approximately six times; instruction-following capabilities also improved threefold.

The study also demonstrated Astra's ability to generate network weights from scratch. In the MinAtar Breakout game, a network generated by Astra in a single response scored 517.4, significantly outperforming a baseline reinforcement learning model trained for 500 million steps (which scored 25.3). However, the authors note that this method is currently primarily used to "remove obstacles hindering capability" rather than granting the model abilities it did not originally possess. For instance, applying the same method to Gemma 3 270M, which rarely exhibits looping issues, did not result in performance gains.

It must be emphasized that all reported results are self-tested data from the research team and have not yet been independently reproduced.

출처

로그인 / 회원가입
독서리서치레이더투자 프레임워크
  • qbitai.com量子位↗
  • x.comX (formerly Twitter)↗
  • 주제

    GPT-6 Astra

    美国麻省理工学院计算机科学与人工智能实验室(MIT CSAIL)的研究者提出“权重空间编程”(Weight-Space Programming),声称利用OpenAI的GPT-6 Astra模型直接读取并修改开源小模型Qwen3.5-0.8…

    3 단계1 건업데이트됨 2026-11-03 16:03

    현재까지의 과정

    1. 2026-11-02MIT CSAIL四位研究者(Rikiya Takehi, Jayden Teoh, Alex Zhang, Omar Khattab)发布博客《Sparks of Weight-Space Programming》,展示GPT-6 Astra直接编辑Qwen3.5-0.8B权重的实验结果。qbitai.com ↗
    2. 2026-11-02VentureBeat报道该研究,引用Databricks技术人员Yuchen Jin的观点,认为若技术可靠,可大幅降低模型维护和调试的计算资源需求。venturebeat.com ↗
    3. 2026-11-03量子位等中文媒体跟进报道,强调Astra通过Python环境分析权重,定位到第15层和第19层的特定输出列进行修正,解决了Qwen的推理循环问题。qbitai.com ↗

    아직 해결되지 않은 문제

    • 其他研究机构能否在相同条件下独立复现Qwen3.5-0.8B的MATH-500得分提升?
    • 该方法在更大规模模型或更复杂任务上的扩展性如何?是否仅限于修复简单的逻辑循环?
    • 直接编辑权重是否会引入新的不可解释的安全风险或后门?
    전체 주제 읽기 →
    리서치 계속하기 →