Research: GPT-6 Astra can rewrite other models' weights directly
MIT researchers reported that OpenAI's GPT-6 Astra read and deliberately modified another model's weights, boosting its math accuracy without traditional training.
Read at The Decoder โ