MIT experiment: GPT-6 Astra directly modifies Qwen3.5 weights, mathematical ability improves by about 6 times
MIT's experiment allowed GPT-6 Astra to directly read and edit the weights of Qwen3.5-0.8B, changing only 0.016% of the parameters, without training and gradient descent, MATH-500 score improved from 7.9% to 48.1%, and instruction execution ability also increased 3 times.
MIT experiments show that cutting-edge models can directly read and write the weights of smaller models, providing a reproducible path for performance repair of gradient-free training.
Same story3 other outlets also reported this
- Hacker News · 300+··First
Astra can read and write neural network weights and improve Qwen3.5-0.8B
- QbitAI··+7 h 36 min·Reading now
MIT experiment: GPT-6 Astra directly modifies Qwen3.5 weights, mathematical ability improves by about 6 times
Adds:MATH-500 score improvement from 7.9% to 48.1%.
- ITHome · AI··+7 h 39 min
MIT CSAIL proposes weight space programming: GPT-6 Astra modifies Qwen3.5-0.8B's weights by about 0.016%, boosting MATH-500 score to 6 times
Adds:GSM8K score improvement from 13.8% to 67.1% and IFEval score improvement from 21.0% to 62.3%.
- Latent Space··+2 d
Latent Space reports on the latest trends in the AI field and in-depth interviews
In order of publication · later reports are added as they appear · Compared by AI from each outlet’s title and summary
Source: QbitAI · qbitai.com