Get Zlio on your home screen
One tap, full screen, like an app.
Suggests frontier models can directly reprogram other models, a notable capability for AI research.
MIT researchers found OpenAI's GPT-6 Astra can read and deliberately rewrite other models' weights without traditional training, boosting a Qwen model's math accuracy from 7.9% to 48.1% by changing 0.016% of its weights.