MIT Computer Science and Artificial Intelligence Laboratory

Stories

MIT researchers use GPT-6 Astra to edit another model's weights directly

Published the Weight-Space Programming results

On November 2, researchers at MIT's Computer Science and Artificial Intelligence Laboratory reported that OpenAI's GPT-6 Astra edited the internal weights of Qwen3.5-0.8B without training or gradient descent. The edit changed 131,072 of the model's roughly 800 million parameters, or 0.016%, and raised its score on the MATH-500 benchmark from 7.9% to 48.1%.

Updated 1 hour ago