One AI Just Rewrote Another AI

AI 📅 Nov 3, 2026 ✍️ Aamir Raza

MIT researchers had a frontier model edit another model's inner weights directly, with no training run at all. LexxWire

The Experiment Came From MIT

Researchers asked OpenAI's GPT-6 Astra to edit a smaller Qwen model to make it better at math. LexxWire

Astra changed just 0.016% of the weights. The small model's math score rose from 7.9% to 48.1%. LexxWire

The team calls the result very early, and some researchers dispute how much of the gain is truly new. LexxWire

The same day, Anthropic said AI may soon speed up its own research faster than humans can check it. LexxWire