MIT Researchers Used GPT-6 Astra To Edit Qwen's Open Weights Directly, Boosting Math Scores
2 44An anonymous reader quotes a report from VentureBeat: Researchers from the Massachusetts Institute of Technology (MIT) Computer Science & Artificial Intelligence Laboratory (CSAIL) reported today that they used OpenAI's GPT-6 Astra language model to raise a small, open source Qwen model's math score roughly sixfold by directly editing its weights -- the numerical parameters that govern its behavior. Rikiya Takehi, Jayden Teoh, Alex Zhang and Omar Khattab described their work in a research blog post published November 2. They call the approach "weight-space programming" and tested Astra's ability to interpret and alter networks without using conventional training procedures to generate the edits.
In the strongest language-model example, Astra's intervention raised Qwen3.5-0.8B's MATH-500 score from 7.9% to 48.1% while touching approximately 0.016% of its weights. The improvement came primarily from correcting repetitive reasoning—a distinction that shaped both the technical findings and the debate over their implications. Their experiments explored a broader possibility: AI systems that inspect, repair and construct other neural networks by working directly on their internal machinery. "This is cool as heck!" wrote Yuchen Jin, a technical staffer at Databricks and former co-founder and CTO of Hyperbolic Labs who was unaffiliated with the research, on X. "Imagine using a large LLM to directly edit a smaller model's weights and dramatically improve its capabilities. It's like gene editing for neural networks. Even cooler than distillation!"
Voyage AI staff researcher Chandan Singh, formerly of Microsoft AI, also reacted to the findings: "For anyone questioning whether this post is overhyped, I think they are actually *underselling* the approach."
2 comments
Oh great (Score: 5, Funny)
by TwistedGreen ( 80055 ) on Tuesday November 03, 2026 @11:06AM (#66407862)
They're breeding.
Misleading presentation (Score: 5, Informative)
by Samantha Wright ( 1324923 ) on Tuesday November 03, 2026 @11:30AM (#66407900)
When editing Qwen, Astra actually wrote and used a Python program to do the dirty work and required some trial and error. This is something humans are already doing; it just doesn't receive much publicity. Raw reading and writing was only done in other parts of the paper, for smaller tasks like detecting the purpose of a game-playing network and correcting a simple vehicle control feedback loop. Though the Qwen edit was successful (after some 50 tests), a similar edit on a smaller Gemma network to remove sycophancy wasn't.
See summary on honnibal.dev [honnibal.dev] for a breakdown.