MIT CSAIL researchers say GPT-6 Astra edited 0.016% of Qwen3.5-0.8B weights to fix reasoning loops, lifting MATH-500 from 7.9% to 48.1% without training.
Conclusion: The 0.016% weight edit fixed "premature termination due to thinking loops," and even unedited models can reach an ...
Space Programming".1. Weight-Space ProgrammingOn November 2, 2026, a research team at MIT CSAIL announced a new method called "Weight-Space Programming", in which an LLM directly reads and writes the ...
The MIT CSAIL team has proposed a "weight-space programming" method that enables GPT-6 Astra to directly read and modify ...
Simon Willison ran the 17-gigabyte open-weight model locally and fed it floating-point arrays from four reinforcement-learning policies. Without calculators or code execution, Qwen identified Breakout ...
GPT-6 Astra edited Qwen’s weights and raised MATH-500 from 7.9% to 48.1% in author-run tests. How weight-space programming works—and its limits.
A new MIT CSAIL blog post is making the rounds in AI circles this week, and it describes something that sounds almost ...
Complete logs of the experiments in the blog post 'LLMs can understand and write neural network weights' - RikiyaT/lm_understand-logs ...
Astra identifies game policies from raw weight matrices and fixes reasoning loops in Qwen3.5-0.8B by editing 0.016% of parameters, lifting MATH-500 scores from 7.9% to 48.1% without training loops.
Article by Shenchao TechFlowNo training required, no data needed, no gradient descent.Researchers at MIT fed the weights of ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results