AnalysisAI ModelsNovember 2, 2026

MIT CSAIL: GPT-6 Astra edits Qwen weights, lifts MATH-500 7.9% to 48.1%

Read original source →venturebeat.com

Four MIT CSAIL researchers report GPT-6 Astra natively read and wrote Qwen3.5-0.8B's weights, changing just 0.016% of them with no gradients or training loop. They also report instruction-following gains up to 3x, though the authors say the edit mostly removed a failure mode.

People · Rikiya Takehi, Omar Khattab

How this story unfolded

5 days · 5 reports · 10 community posts · 15 of 18 shown

  1. Nov 2
  2. Nov 3
  3. Nov 4
  4. Nov 7

More stories today

Open the live feed