MIT researchers report that GPT-6 Astra can alter the internal weights of another language model without a conventional training run. In a test on a Qwen model, math accuracy rose from about 7.9 percent to 48.1 percent.
MIT researchers report GPT-6 Astra rewrites another model's weights directly
MIT researchers report that GPT-6 Astra can alter the internal weights of another language model without a conventional training run. In a test on a Qwen model, math accuracy rose from about 7.9 percent to 48.1 percent.
Monitoring item. The full text is not distributed. Extract and source below.