Research
MIT researchers show GPT-6 Astra can rewrite the weights of other AI models
In one experiment, Astra raised a Qwen model's math accuracy from 7.9 to 48.1 percent by changing 0.016 percent of its weights. The approach still has clear limits, according to the report.
Every outlet that ran it 1 article
The DecoderTech pressLead
GPT-6 Astra sees the matrix and reprograms other AI models directly through their weights